Tea Farming System

Abstract

1. A computer-implemented system for managing a tea harvest, the system comprising: a data store storing data defining one or more growing areas of a tea plantation; for each of the growing areas of the tea plantation defined in the data store, a growth and yield prediction module configured to model a plurality of tea plants within the respective growing area, thereby predicting the growth of each tea plant within the respective growing area over a period of time, and therefore the yield of each growing area of ​​the tea plantation when harvested; the growth and yield prediction module further configured to predict a recommended harvest time to achieve a desired quality of tea and a desired yield of tea; and a user interface module configured to provide a user with the predicted yield of each growing area of ​​the tea plantation when harvested.

Description

[Technical Field]

[0001] The present invention is in the field of the technical field of cultivating and harvesting tea plants. [Background technology]

[0002] Tea is a beverage made from the leaves of the evergreen plant Camellia sinensis. In tea cultivation, new leaves from the shoot are preferred because their softness enhances the quality of the resulting tea. Tea plants are generally pruned to a low height to facilitate harvesting and allow access to this new growth at the top of the plant. Thus, tea plants can be harvested continuously throughout the year, if the season allows. Tea plants are harvested either through handpicking of new leaves, through harvesting with a reaper, or through mechanical harvesting using a cutter that removes the top layer of the shrub, including the new growth.

[0003] To support this continuous harvest, most tea cultivation is concentrated in certain geographic areas, which limits the amount of land available. Safe, resilient, and sustainable tea production requires informed and consistent management of the infrastructure of such production, including clonal selection, soil, nutrients, water, energy, labor, processing, and an efficient and well-connected logistics network across the supply chain.

[0004] Cranfield University Potential Productivity Analyser for Tea (CUPPA-Tea) enables modelling of tea yield based on characterising the growth of shoot populations. CUPPA-Tea is a process-based crop model for tea production (specifically for Camellia sinensis L.).

[0005] The model operates at daily time steps and includes routines describing shoot growth and development, dry matter production and partitioning, and the water balance of the crop. Summary of the Invention [Problem to be solved by the invention]

[0006] In situations where the amount of land available for tea cultivation is relatively limited, it is generally desirable to maximize the yield therefrom. [Means for solving the problem]

[0007] Thus, in a first aspect, an embodiment of the present invention provides a computer-implemented system for managing a tea harvest, the system comprising: a) a data store storing data defining one or more growing areas of a tea plantation; b) a growth and yield prediction module configured, for each of the growing areas of the tea plantation defined in the data store, to model a plurality of tea plants within the respective growing area, thereby predicting the growth of each tea plant within the respective growing area over a period of time, and therefore the yield of each growing area of ​​the tea plantation when harvested; c) a growth and yield prediction module further configured to predict a recommended harvest time to achieve a desired tea quality and a desired tea yield; d) a user interface module configured to provide the user with a predicted yield of each growing area of ​​the tea plantation when harvested.

[0008] Advantageously, the system may, for example, predict tea growth, yield, quality, and / or ideal harvest time, thereby enabling the user to make predictions about the parameters of the next tea harvest and / or choose when to harvest the tea in order to optimize these parameters as desired.

[0009] The user interface module may display the geographic data for each growth region, for example, as a map. Displaying the geographic data for each growth region facilitates recognition of a given region and allows the user to make selections and draw inferences based on the data. A map perspective, for example, a top-down view of the region and associated geographic data, may be a suitable way to display this.

[0010] The data may define a tea plantation comprising a plurality of growing areas, and the user interface module may be configured to allow a user to select a growing area from the plurality of growing areas, whereby the system displays the predictions made for the selected growing area.

[0011] The growing area may comprise or consist of a field of growing tea plants.

[0012] The growth and yield prediction module may be further configured to predict a recommended harvest interval, shoot number, and trellis height for each tea plant in each growing area. The recommended harvest interval may correspond to the time between harvests that maximizes the yield of the growing area. The trellis height may refer to the distance from the ground to the height of the pickable shoots.

[0013] Advantageously, this may allow for more accurate prediction of tea harvest parameters and optimisation of harvest timing.

[0014] The growth and yield prediction module may predict a first plucking event, a second plucking event, and a third plucking event, and if the third plucking event and the second plucking event occur within a first time period and the third plucking event and the first plucking event occur within a second time period, and the second time period is longer than the first time period, the second plucking event and the third plucking event are considered to be part of the same plucking event. The term "considered to be part of the same plucking event" may mean that characteristics of the second plucking event and the third plucking event are combined into a single plucking event. Advantageously, this modification to the growth and yield prediction module improves the accuracy of the harvest with respect to actual plantation conditions.

[0015] The growth and yield prediction unit may predict a quality index of the harvested tea plants, which may advantageously facilitate analysis of the quality of the harvest at one point in time as a comparison of the quality of the harvest at a different point in time or in response to variations in variables such as the amount of rainfall.

[0016] The quality index may be calculated as the percentage of shoots that are in a good condition for harvesting relative to the total number of shoots.

[0017] The growth and yield prediction module may predict the number of shoots to be harvested based on the number of shoots with more leaves than a predetermined threshold and / or the height of each shoot.

[0018] Preferably, the tea shoots are harvested when they have a desired number of leaves, preferably 0, 1, 2, 3, 4, 5 or more leaves. Preferably, the tea shoots are harvested when they are at, longer than, or shorter than a certain length, preferably 0 cm, 1 cm, 2 cm, 3 cm, 4 cm, 5 cm, 6 cm, 7 cm, 8 cm, 9 cm, or 10 cm.

[0019] Advantageously, the quality index or harvest forecast reflects the percentage of shoots in a favorable condition containing a desired number of leaves, allowing the user to make more accurate predictions about the expected quality of the tea.

[0020] The growth and yield prediction module may predict leaf emergence rate or phyllochron time. Advantageously, this prediction allows the module to more accurately predict the growth, yield, and / or quality of the shoots to be harvested.

[0021] The growth and yield prediction module may be configured to incorporate the effect of harvesting method on tea plant growth. Preferably, the harvesting method is mechanical harvesting, reaper harvesting, or hand picking. Advantageously, by taking into account the type of harvesting, the module may more accurately predict the growth, yield, and quality of the harvested tea, as the type of harvesting affects these parameters.

[0022] The system may include a data input module configured to receive data regarding each growth region, the data input module providing the data to the growth and forecasting module.

[0023] The data input module may comprise an environmental data module configured to receive environmental data for each growing area. This allows the model to take into account additional factors such as rainfall, temperature, or sunshine levels, which may be used to model shoot growth in the growth and prediction module. Advantageously, the prediction module may more accurately predict shoot growth, as the environment and these measured parameters affect tea growth.

[0024] The environmental data may comprise observed weather data or forecasted weather data. Advantageously, the forecasted weather data may be used to improve predictions of the impact that weather, such as increased or decreased rainfall, may have on tea growth. Advantageously, the observed weather data may be used to more accurately initialize the model to more accurately reflect the condition of the tea. It may be used to improve the accuracy of the predictions by replacing the forecasted weather data with the observed weather data as the weather actually occurs.

[0025] The observed or forecasted weather data may comprise one or more of precipitation data, temperature data, solar radiation data, wind data, water evaporation data, or mean vapor pressure data. The environmental data may be recorded by an automatic recording station. The automatic recording station may be an automatic weather station. The automatic recording station reduces the labor required to provide the environmental data to the model.

[0026] The data input module may include a remotely sensed data module configured to receive remotely sensed data comprising one or more observations of each growing area. The use of the remotely sensed data allows the system to reflect the ground truth of the plantation, which more accurately simulates the model.

[0027] A remote sensing module can detect the harvest status of each growing area, allowing the model to be accurately updated as to whether a given part of the plantation has been harvested.

[0028] The remote sensing module may detect a first area of ​​each growing area as harvested and a second area of ​​each growing area as unharvested. Unharvested may mean that the area has not been harvested during a given time frame; that is, the first area has been harvested, but the second area has not (at this time). This allows the growth and forecasting unit to model which shoots have been harvested and which have not been harvested in a given field within the plantation.

[0029] The remote sensing module can detect the type of harvesting used in each growing area, allowing the growth and prediction unit to model whether shoots were harvested by hand picking, manual harvesting, or some other technique. In hand picking, an individual manually selectively removes shoots suitable for harvesting. In mechanical harvesting, a machine removes the top layer of the tea plant in a non-selective process. Depending on the harvesting technique, the quality and yield of the harvested tea can be affected.

[0030] The remote sensing module can detect the level of tea plant growth in each growing region, which allows the growth and prediction unit to more accurately model the level of shoot growth in a given growing region, and the level of simulated shoot growth can be updated to reflect the ground truth.

[0031] The remote sensing module may receive data emanating from an orbiting satellite, which may be synthetic aperture radar (SAR) data, visible light data, or multispectral imaging data.

[0032] The synthetic aperture radar data may comprise X-band SAR data, C-band SAR data, or L-band SAR data. The synthetic aperture radar data may comprise one or more of HH-polarized backscatter coefficient values, VH-polarized backscatter coefficient values, HV-polarized backscatter coefficient values, or VV-polarized backscatter coefficient values. These backscatter types refer to either main polarization or cross-polarized backscatter coefficients in the vertical or horizontal direction (relative to the ground).

[0033] Multispectral imaging data can be used to calculate Normalized Difference Vegetation Index (NDVI) data.

[0034] The growth and yield prediction module may predict the impact of pruning on the growth and yield of the tea plant, which allows the growth and prediction unit to take pruning into account so that the predicted growth and yield more closely reflects the actual growth and yield.

[0035] The growth and yield prediction module may be configured to model each tea plant as having at least one shoot, the growth and yield prediction module comprising: a. Modeling shoot length and shortening shoot length; b. Modeling shoot developmental stages and setting developmental stages at early stages; and / or c. Model the base height between the bottom of the shoot and the soil surface, and reduce the base height of the shoot. The effects of pruning are modeled by one or more of the following steps:

[0036] These modifications to shoot parameters by growth and prediction units more accurately reflect the effects of pruning on tea plant growth and yield.

[0037] In a second aspect, embodiments of the present invention comprise the steps of obtaining, from a data store, data defining one or more growing areas of a tea plantation; modeling, by a growth and yield prediction module, for each of the growing areas of the tea plantation defined in the data store, a plurality of tea plants in the respective growing area, thereby predicting the growth of each tea plant in the respective growing area over a period of time, and therefore the yield of each growing area of ​​the tea plantation when harvested; providing a user via a user interface module with a predicted yield for each growing area of ​​the tea plantation when harvested; The present invention provides a computer-implemented method for managing tea harvesting, comprising:

[0038] The computer-implemented method of the second aspect may include any one, or any compatible combination, of the optional features as described in relation to the first aspect.

[0039] In a third aspect, embodiments of the present invention provide a computer-implemented method for predicting growth of a plurality of tea plants in a tea plantation, the method comprising: retrieving initial condition data indicative of initial nitrogen accumulation in a tea plantation comprising nitrogen retained as tea plant tissue, soil inorganic nitrogen, and soil organic nitrogen; predicting a nitrogen deficiency level, which indicates the level at which lack of nitrogen availability limits the growth of the tea plant; and predicting tea plant growth using the initial condition data and the nitrogen deficiency level.

[0040] Advantageously, the method predicts the effects of nitrogen deficiency on tea plant growth. Nitrogen deficiency is one of the main effects limiting growth and yield in tea plants. By determining when plants are limited by nitrogen deficiency, the model can more accurately predict tea plant growth and yield. This allows fertilizer to be applied at the optimal time and in the optimal amount to improve soil nitrogen, maximizing yield while controlling costs and minimizing harm to the environment.

[0041] The method further comprises: a. retrieving nitrogen addition data indicating the amount of nitrogen added to the soil of the tea plantation; b. predicting nitrogen transformation parameters that describe the transformation of nitrogen in the soil between different chemical forms by one or more of mineralization, nitrification, or denitrification; c. predicting tea plant nitrogen loss, indicating nitrogen lost by the tea plant due to one or more of tea plant harvesting, tea plant pruning, exudation, or abscission; or d. predicting soil nitrogen losses, which indicates nitrogen losses due to one or more of denitrification, ammonia volatilization, or leaching. may comprise one or more of:

[0042] Advantageously, nitrogen addition data can adjust the model to reflect the addition of fertilizer or other nitrogen sources, so that the model is more reflective of the amount of nitrogen in the soil and how that affects the level of nitrogen available to the tea plant, allowing for more accurate predictions of tea plant growth.

[0043] Advantageously, predicting nitrogen transformation parameters allows the model to reflect how nitrogen is chemically converted from one form to another and how this affects the level of nitrogen available to the tea plant, thereby allowing for more accurate prediction of tea plant growth.

[0044] Advantageously, predicting soil nitrogen loss allows the model to reflect how nitrogen is lost from the soil through various biological and chemical processes and how this affects the level of nitrogen available to the tea plant, thereby allowing for more accurate predictions of tea plant growth.

[0045] The initial nitrogen stock can be initialized with specified levels of inorganic soil nitrogen at a range of soil depths. Advantageously, this allows the model to better reflect what levels of nitrogen are available to different tea plants, allowing for more accurate predictions of tea plant growth.

[0046] Nitrogen addition data is a. amount of new carbon per soil layer; b. The proportion of carbon in new organic matter, and c. The amount of new organic nitrogen added to the soil per soil layer It may comprise:

[0047] Advantageously, this more detailed description of nitrogen sources in nitrogen addition allows the model to better reflect how those sources alter the nitrogen available to the tea plant, thereby allowing for more accurate predictions of tea plant growth.

[0048] The method may further comprise calculating the carbon to nitrogen ratio using the initial nitrogen accumulation and nitrogen addition data, which advantageously allows the model to more accurately model the extent to which the tea plant can uptake new nitrogen sources.

[0049] Predicting the growth of the tea plant may further comprise calculating using one or more of the potential growth rate of the tea plant, temperature, or soil moisture level.

[0050] Advantageously, by calculating the level of growth of the tea plant when unconstrained by environmental constraints (i.e., potential growth rate), it is possible to calculate how those constraints affect this potential growth. Temperature and available moisture can limit the growth of the tea plant. Advantageously, taking these parameters into account allows for more accurate predictions of tea plant growth.

[0051] The method may further comprise predicting the potential growth rate of the tea plant, the potential growth rate being calculated using one or more of the amount of solar radiation, the radiation use efficiency, the light extinction coefficient, or the leaf area index. Plants use leaves to absorb sunlight for photosynthesis, thereby producing substances for growth.

[0052] Advantageously, by using solar energy parameters and leaf size, the model can predict the potential growth rate of the plant. Taking these parameters into account allows for more accurate prediction of tea plant growth.

[0053] Predicting the nitrogen deficiency level may comprise calculating one or more of a critical nitrogen level for the canopy, a nitrogen percentage in the canopy, or a minimum nitrogen content level for the canopy.

[0054] Advantageously, the level of nitrogen deficiency can be predicted by considering the level of nitrogen in the tea plant canopy, or the level at which nitrogen begins to become a constraint on growth (critical nitrogen level). Taking these parameters into account allows for more accurate prediction of tea plant growth.

[0055] Predicting the nitrogen deficiency level may comprise predicting a nitrogen demand for each tea plant, wherein predicting the nitrogen demand comprises: a. predicting homeostatic nitrogen demand of a tea plant canopy, comprising calculating using one or more of canopy mass, critical nitrogen level, and nitrogen rate in the tea plant canopy; b. predicting the homeostatic nitrogen demand of tea plant roots, comprising calculating using one or more of root mass, critical nitrogen level, or nitrogen rate in the tea plant roots; c. predicting canopy growth nitrogen demand of tea plants, comprising calculating using the canopy potential growth rate and / or the canopy critical nitrogen level; or d. Predicting root growth nitrogen demand of tea plants, comprising calculating using root potential growth rate and / or root critical nitrogen level. It comprises one or more of:

[0056] Advantageously, by taking into account how different tissues need nitrogen, the model can more accurately predict tea plant growth.

[0057] Predicting tea plant growth may comprise predicting the growth of multiple leaves for each tea plant. Tea plants for harvest produce shoots with multiple leaves that are harvested at a preferred stage. Advantageously, predicting the growth state of multiple leaves allows the model to more accurately predict when the tea plants will be in optimal condition for harvest, as well as to more accurately predict tea plant growth.

[0058] The nitrogen conversion parameters may comprise parameters for describing nitrogen mineralization, and the method comprises calculating the nitrogen mineralization parameters using one or more of the amount of new organic nitrogen, the organic matter pool fraction, the total organic matter, a temperature coefficient, a moisture coefficient, a carbon-to-nitrogen ratio coefficient, a decay rate constant, the concentration of ammonium, the concentration of nitrate, or the amount of new humic nitrogen.

[0059] The nitrogen conversion parameters may comprise parameters for describing nitrification, and the method comprises calculating the nitrification parameters using one or more of the ammonium level, oxygen level, soil pH, temperature coefficient, or soil moisture coefficient of the substrate.

[0060] The nitrogen conversion parameters may comprise parameters for describing denitrification, and the method comprises calculating the denitrification parameters using one or more of ammonium concentration, pH, temperature, soil moisture content, or the amount of ammonium above a minimum value.

[0061] Advantageously, calculating parameters for nitrogen mineralization, nitrification, or denitrification allows the model to more accurately predict how nitrogen changes between chemical forms in the soil and how much nitrogen is available.

[0062] Predicting nitrogen loss may comprise predicting nitrogen volatilization, where predicting nitrogen volatilization comprises calculating using one or more of ammonia concentration in soil moisture, wind speed, soil temperature, or soil depth.

[0063] Advantageously, predicting nitrogen volatilization allows the model to more accurately predict how nitrogen will be lost from the soil, which in turn allows the model to more accurately predict tea plant growth or predict when nitrogen will be lost faster or slower from the soil, thereby optimizing nitrogen fertilizer application and preventing unnecessary or excessive use of fertilizer.

[0064] In a fourth aspect, embodiments of the present invention provide a computer program configured to perform the method according to the third aspect.

[0065] In a fifth aspect, embodiments of the present invention provide a storage medium comprising instructions for executing on a computer processor a method according to the third aspect.

[0066] It will be understood that the tea plants modeled using the system of the present invention may be wild-type tea plants, genetically engineered tea plants, tea plant clones, and combinations thereof. [Brief explanation of the drawings]

[0067] [Figure 1] FIG. 2 is a diagram illustrating the flow of information in an embodiment according to a first aspect of the present invention. [Figure 2] 1 is a flowchart showing the flow of information in the growth and yield prediction module. [Figure 3] 10A-10C are screenshots of the user interface provided by the user interface module. [Figure 4] Figure 1 shows the calculated average altitude of each field in the Kapgwen / Chemosit farm in Kenya. [Figure 5] FIG. 1 shows the calculated average altitude of each field on the Kapkorech farm in Kenya. [Figure 6] Figure 1 shows meteorological data from Kapkorech Research Stations from 2011 to 2013. [Figure 7] This figure shows meteorological data from the Chemosit station from 2011 to 2013. [Figure 8] 1 is a graph of cumulative growth from 12 CUPPA-Tea model runs testing soil susceptibility. [Figure 9] 1 is a graph of modeled yields for 12 test soils. [Figure 10] 1 is a graph of CUPPA-Tea performance with varying number of shoots and harvest intervals compared to recorded yield. [Figure 11] Figure 1 shows recorded monthly yields (kgMTha-1) for selected fields on the Kapkorech farm. [Figure 12] FIG. 1 shows recorded and modelled monthly yields from field 9 in Kapkorech tea plantation. [Figure 13] 1 is a graph of recorded and modeled monthly yields for field 28 of the Kapgwen Tea Estate. [Figure 14] Graph of modeled yield by harvest, recorded monthly yield, and soil moisture deficit for field 28 on the Kapgwen farm. [Figure 15] This is a graph of the average monthly yield of green tea leaves and total monthly rainfall from Kapkorech Estate from January 2014 to September 2016. [Figure 16]1 is a graph of the relationship between altitude, annual green tea leaf yield, and harvesting method for 2014 and 2015. [Figure 17] Graph of predicted yield using the "1997", "2004", "Recomp Pascal", and "C#" versions of the model for fully irrigated tea using the default input file (April 1989 to July 1990) for Ngwazi, Tanzania, with a harvest interval of 14 days. The difference in calculated soil moisture deficit in the "1997" and "C#" versions is also shown. [Figure 18] 10 is a graph comparing weekly yields predicted using the "C#" and "2004" versions of the CUPPA-Tea model against yields predicted for Ngwazi, Tanzania using the original input file. [Figure 19] 1 is a graph of yield for Ngwazi in Tanzania irrigated using the actual irrigation schedule versus the specified harvest date. [Figure 20] 10 is a graph of actual weekly yields from non-irrigated clone 6 / 8 and predicted yields from the 1997 and C# models for Ngwazi, Tanzania, for a specified harvest date. Soil moisture deficit is also shown. [Figure 21] 1 is a graph comparing water stress factors between the new and original versions of the Cuppa tea model. [Figure 22] FIG. 10 shows a screenshot of the output from the water model showing the effect of 6 mm of rainfall on the water stress factor. [Figure 23] Figure 1 shows actual yield reported for hand picked tea in Field 5, Kapkorech, Kenya. [Figure 24] 1 is a graph of actual yield reported for mechanically harvested tea in Field 37, Kapkorech, Kenya. [Figure 25] This is a graph of the dry matter constant (conversion rate of green tea leaf yield to finished tea yield) for the Kapkorech estate from March 2014 to September 2016. [Figure 26] 1 is a graph of grouped yields (expressed by week) of finished tea versus hand-picked tea from field 5. [Figure 27] 1 is a graph of grouped yields (expressed by week) of finished tea versus machine harvested tea from field 37. [Figure 28] 1 is a graph of weekly predicted and actual yields for hand-picked field 5. [Figure 29] 1 is a graph comparing predicted yield using different step heights for field 37 when the bed height is 75 cm. [Figure 30] 1 is a graph comparing predicted yield for Field 5 (handpicked) with different picking methods. [Figure 31] 1 is a graph comparing predicted yield for Field 5 (handpicked) with different shoot numbers. [Figure 32] 1 is a graph of the effect of soil type on predicted weekly yield of hand-harvested tea in Field 5. [Figure 33] Graph of the Mean Landsat-8 Normalized Difference Vegetation Index (NDVI) for field 57 in Kapgwen on September 9, September 25, and October 11, 2015. Error bars are the standard error of the mean. The discrepancy between the three dates is likely the result of differences in solar irradiation or atmospheric conditions. [Figure 34] FIG. 1 shows labeled photographs of tea stems. [Figure 35] Graph of the mean gamma nought (γvv 0) per field from two Sentienl-1 images. Error bars are the standard error of the mean. [Figure 36] Graph of the expected variation of γvv 0 in X-band. γvv 0 increases with the development of vertically oriented shoots. [Figure 37] A series of graphs showing the average minimum and maximum temperatures and rainfall from 2013 to 2015 compared to the 2016 values. [Figure 38]Figure 1 shows the timeline of the experiment reporting different satellite images (CosmoSkymed (CSK) VV and HH polarization, UK-DMC3), rainfall measurements, and harvest dates (dots) from monthly harvest reports, with hand-harvested (HP) in grey and machine-harvested (MH) in black. [Figure 39] This diagram shows how mechanically harvested furrows can be detected from their characteristic sinusoidal nature with a spatial period of 24 m or less. [Figure 40] FIG. 1 illustrates how automated detection of harvesting methods using SAR imagery can be achieved. [Figure 41] 1A and 1B are diagrams illustrating examples of scenarios in which the imaged scene remains coherent. [Figure 42] Figure 1 shows an example scenario where the imaged scene loses coherence. In a high-coherence scenario, the surface canopy changes slightly but remains correlated between image 1 and image 2. In a low-coherence scenario, the canopy is randomly permuted, which is what happens during harvesting. [Figure 43] 1 is a flowchart of the processing steps for detecting harvested areas between two SAR images using loss of coherence. Example for field 84 between August 15th and 22nd. [Figure 44] FIG. 10 illustrates a series of normalization steps to correct the backscatter coefficient for variations in moisture and local angle of incidence. [Figure 45] FIG. 10 illustrates the approach taken to compare the backscatter coefficient of an unharvested area with the backscatter coefficient of a harvested area. [Figure 46] FIG. 10 illustrates the approach taken to compare the backscatter coefficient of an unharvested area with the backscatter coefficient of a harvested area. [Figure 47] FIG. 10 illustrates the approach taken to compare the backscatter coefficient of an unharvested area with the backscatter coefficient of a harvested area. [Figure 48] This SAR image (August 23rd) shows the pattern of the rows after mechanical harvesting. [Figure 49]Pan-sharpened multispectral image (August 16th) showing the visible furrow pattern after mechanical harvesting. Figure 29 shows that the pattern is only visible after harvesting has taken place, suggesting that the pattern is disappearing as the shoots grow. [Figure 50] FIG. 1 shows how mechanical harvesting of 1.5 m width can result in wider swaths. [Figure 51] FIG. 10 shows a comparison of coherence loss and harvested area from a field sheet. [Figure 52] FIG. 10 shows the correlation between the harvested area detected between two SAR images using coherence and the yield for the corresponding period from the harvest report for mechanical harvesting. [Figure 53] Figure 1 shows the correlation between harvested area detected between two SAR images using coherence and yield for the corresponding period from harvest reports for hand-picked teas. Green, orange, and blue are data for Chinese tea, seedling tea, and clonal tea, respectively. The dashed black line is the overall linear regression using all tea types together. [Figure 54] FIG. 1 shows a comparison of harvested area backscatter (x-axis) and unharvested area backscatter (y-axis) for each hand-harvested field (blue dots) and each machine-harvested field (red dots) for VV polarized images. [Figure 55] This figure shows a comparison of harvested area backscatter (x-axis) and unharvested area backscatter (y-axis) for each hand-harvested field (blue dots) and each machine-harvested field (red dots) for HH polarization imagery. Gray dots highlight fields that were harvested on the day of the SAR imagery. These gray dots are expected to show the greatest difference between harvested and unharvested backscatter. In VV polarization, the gray dots are evenly spread (50%-50%) on both sides of the y = x diagonal; when all dots (red and blue) are considered, this split is 46%-54%. In HH polarization, for 86% of the gray dots, the harvested area backscatter is less than the unharvested area backscatter; when all dots are considered, this split is 70%-30%. [Figure 56] Figure 1 shows that using VV polarization, horizontally oriented leaves from new shoots are transparent and waves only interact with randomly oriented leaves of the older canopy. [Figure 57] Figure 1 shows that when HH polarization is used, the measured backscatter is a combination of direct scattering from new shoots and random scattering from the canopy, resulting in increased HH backscatter when young shoots are developing / developing. [Figure 58] FIG. 1 shows the variation in NDVI between the first UK-DMC3 image (August 16th) and the second UK-DMC3 image (September 10th) for field 29. [Figure 59] FIG. 1 shows the variation in NDVI between a first UK-DMC3 image (August 16th) and a second UK-DMC3 image (September 10th) for field 30. [Figure 60] FIG. 1 shows the variation in NDVI between the first UK-DMC3 image (August 16th) and the second UK-DMC3 image (September 10th) for field 32. [Figure 60A] FIG. 1 shows the variation in NDVI between the first UK-DMC3 image (August 16th) and the second UK-DMC3 image (September 10th) for field 33. [Figure 61] Figure 1 shows data from harvest reports of four fields and time for two UK-DMC3 images. [Figure 62] This figure shows pairwise scatter plots between (1) monthly average yield from January to September 2016 (harvest report), (2) average NDVI per field for the image on August 16, 2016, (3) pruning year, (4) tea type, and (5) harvesting method (MH = mechanical harvest, HP = hand-picked). [Figure 63] FIG. 1 shows a series of images, all cloudy but alternating to reveal different fields. [Figure 64] Predicted monthly and annual yields are shown. [Figure 65]This is a screenshot of the Cuppa Tea Tool showing annual yields in kg mt / ha. The slider bar at the bottom left allows for year selection (background image is DigitalGlobe, 23 January 2015). [Figure 66] This is a screenshot of the Cuppa Tea Tool showing data at the field level (background image is DigitalGlobe, January 23, 2015). [Figure 67] FIG. 10 is a graph showing shoot type development and quality index for Field 1 in Kericho from January 1 to December 31, 2018, assuming a manual harvesting routine using CUPPA-Cloud. [Figure 68] Figure 14 shows the calculated number of shoots for Field 1 from January 1 to December 31, 2018, assuming a default minimum harvested shoot length of 5 cm. The majority of harvestable shoots have more than two leaves and more than three leaves, and the quality scores of harvested leaves are similar to harvestable leaves. [Figure 69] 1 is a graph showing that assuming a minimum harvested shoot length of 4 cm (compared to 5 cm) increases the number of small shoots and reduces the number of some larger shoots. Quality and yield are improved. [Figure 70] 1 is a graph showing that assuming a minimum harvested shoot length of 3 cm (compared to 4 cm) increases the number of small shoots, and therefore improves quality. Yield also improves. [Figure 71] 1 is a graph showing that assuming a minimum harvested shoot length of 2 cm (compared to 3 cm) increases the number of 1+ shoots and decreases the number of 3+ shoots. Quality scores improve. The effect on yield is variable. [Figure 72] 10 is a graph showing that assuming a minimum harvested shoot length of 1 cm (compared to 2 cm) makes little difference to the number of 2+ shoots. However, the number of growing1+ and banjhi1+ shoots increases slightly. [Figure 73]Figure 1 shows the calculated number of shoots (one or more shoots) for field 1 from January 1 to December 31, 2018, assuming a minimum harvested shoot length of 0 cm. The quality score, expressed as (1banj + 1grow + 2grow) / (1 to 5banj + 1 to 5grow), drops to approximately 50 at harvest. However, this yield comes from only 2 and 3 shoots and buds, so the actual harvest quality is much lower. [Figure 74] 1 is a graph showing the hypothesized effect of temperature on the development rate for fenoclone and phylloclone by day in CUPPA-Tea. [Figure 75] 11 is a graph showing the effect of days since harvest (on December 11, 2018) on predicted quality index and shoot number in Field 1 after harvest, assuming a minimum shoot length of 0 cm. [Figure 76] 11 is a graph showing the effect of days since harvest (on December 11, 2018) on predicted quality index and shoot number in Field 1 after harvest, assuming a minimum shoot length of 1 cm. [Figure 77] 11 is a graph showing the effect of days since harvest (on December 11, 2018) on predicted quality index and shoot number in Field 1 after harvest, assuming a minimum shoot length of 2 cm. [Figure 78] 11 is a graph showing the effect of days since harvest (on December 11, 2018) on predicted quality index and shoot number in Field 1 after harvest, assuming a minimum shoot length of 3 cm. [Figure 79] 11 is a graph showing the effect of days since harvest (on December 11, 2018) on predicted quality index and shoot number in Field 1 after harvest, assuming a minimum shoot length of 4 cm. [Figure 80] 11 is a graph showing the effect of days since harvest (on December 11, 2018) on predicted quality index and shoot number in Field 1 after harvest, assuming a minimum shoot length of 5 cm. [Figure 81]10 is a graph showing the effect of days since harvest (on November 24, 2018) on predicted quality index and shoot number in field 14 after harvest, assuming a minimum shoot length of 0 cm. [Figure 82] 1 is a graph showing the effect of days since harvest (on November 24, 2018) on predicted quality index and shoot number in field 14 after harvest, assuming a minimum shoot length of 1 cm. [Figure 83] 1 is a graph showing the effect of days since harvest (on November 24, 2018) on predicted quality index and shoot number in field 14 after harvest, assuming a minimum shoot length of 2 cm. [Figure 84] 1 is a graph showing the effect of days since harvest (on November 24, 2018) on predicted quality index and shoot number in field 14 after harvest, assuming a minimum shoot length of 3 cm. [Figure 85] 1 is a graph showing the effect of days since harvest (on November 24, 2018) on predicted quality index and shoot number in field 14 after harvest, assuming a minimum shoot length of 4 cm. [Figure 86] 1 is a graph showing the effect of days since harvest (on November 24, 2018) on predicted quality index and shoot number in field 14 after harvest, assuming a minimum shoot length of 5 cm. [Figure 87] Schematic diagram showing one phyllochron (from Burgess & Carr, 1998). [Figure 88] Graph showing the effect of potential soil water deficit (SWD) at the beginning of the period on the ratio of actual to estimated leaf emergence rate (f) for 3- to 4-year-old tea plants (from (Burgess & Carr, 1998)). [Figure 89] 1 is a graph showing predicted leaf emergence in Kericho based on temperature measurements from January 1, 2016 to December 31, 2018. [Figure 90]This graph shows the predicted leaf emergence rate in Kericho based on temperature measurements from January 1 to December 31, 2018. The average temperature was 17.95°C. [Figure 91] FIG. 10 is a graph showing soil moisture deficit calculated from CUPPA-Tea from January 1 to December 31, 2018, indicating drought stress early in the year. [Figure 92] FIG. 12 is a graph showing calculated adjusted leaf emergence rates in CUPPA-Tea from January 1 to December 31, 2018, indicating drought stress early in the year. [Figure 93] 1 is a graph showing phyllochron accumulation (calculated manually and using CUPPA-Tea) for Field 1 in 2018. [Figure 94] FIG. 1 shows the cumulative number of days between harvests reported for Field 1 in 2018. [Figure 95] 1 is a graph showing that the increase in phyllochron ratio can be shown as a step change according to the year of pruning. [Figure 96] 1 is a graph showing that the Philochron multiplier can be expressed as a logarithmic relationship. [Figure 97] 1 is a graph showing that the Philochron multiplier can be expressed as a linear relationship. [Figure 98] 1 is a graph showing the initial relationship between actual harvest intervals for hand-harvested leaves and predicted harvest intervals based on phyllochrons according to the year of pruning, based on the reported harvesting methods in the first report. [Figure 99] 1 is a graph showing actual and predicted harvest intervals for hand-harvested field 35 (pruned in April 2019). [Figure 100] Figure 10 shows the actual and predicted harvest intervals (based on phyllochron) for hand-harvested field 24 (pruned in August 2017). Note that the maximum QI_3cm is constrained to 32 days. [Figure 101]1 is a graph showing the actual and predicted harvest intervals (based on phyllochron) for hand-harvested field 12 (pruned in February 2017). Note that the value of QI_3cm is fixed to have a maximum value of 32 days. [Figure 102] 1 is a graph showing actual and predicted harvest intervals (based on phyllochron) for hand-harvested field 19A (pruned on August 15, 2018). [Figure 103] 1 is a graph showing actual and predicted harvest intervals (based on phyllochron) for reaper-harvested field 48A (harvested in March 2019). [Figure 104] Graph showing the relationship between phyllochron predicted spacing and actual harvest spacing for reaper-harvested leaves. Originally, we identified that the predicted spacing for first-year pruning appeared to be about 27% too low, but it now appears more likely that an incorrect harvesting method was assumed. [Figure 105] 1 is a graph showing actual and predicted harvest intervals (based on phyllochrons) for Field 1, assuming harvesting by reaper. However, the field may have been mechanically harvested in 2018. [Figure 106] This graph shows the actual and predicted harvest intervals (based on phyllochrons) for field 13, assuming it was harvested mechanically. However, it may have been harvested by hand, and the wrong multiplier was used. [Figure 107] 1 is a graph showing actual and predicted harvest intervals (based on phyllochrons) for mechanically harvested field 52 prior to pruning in June 2019. [Figure 108] 1 is a graph showing actual and predicted harvest intervals (based on phyllochrons) for machine-harvested field 62 in the year of pruning. [Figure 109] Figure 1 shows the relationship between actual harvest intervals for machine harvesting and predicted harvest intervals with predicted phyllochron for unpruned and pruned teas (before pruning correction). [Figure 110]1 is a graph showing the relationship between average predicted harvest spacing using phyllochron and QI when the actual average spacing is 3 cm for machine harvested teas (before pruning correction). [Figure 111] 1 is a graph of the effect of changing the "minimum harvestable shoot height" or "step height" from 5 cm to 3 cm on increasing yield overall for Field 5. [Figure 112] FIG. 1 is a schematic diagram of the meaning of "base height" and how it may increase over time. [Figure 113] 10 is a graph of the effect of including new base height from pruning in field 5. This correction compensates for the higher yield and synchronization of growth immediately after pruning (assuming a constant 650 shoots / m2). [Figure 114] Graph of the impact of including new base height from pruning for field 13 to correct for higher yield on February 22, 2019 after pruning in November 2018 (assuming a constant 650 shoots / m2). [Figure 115] 10 is a graph showing the effect of including new base height from pruning for field 25 on reducing the high yield immediately after pruning on December 23, 2017 (assuming a constant 650 shoots / m2). [Figure 116] Yield graph showing that assuming a 75cm platform height reduced to 60cm is the same as assuming a 70cm platform height reduced to 55cm. [Figure 117] 10 is a graph of updated yield after creating a new base height. [Figure 118] 1 is a graph of platform and chute height over time. [Figure 119] Graph of base height in field 5 at the end of 2019. [Figure 120] 10 is a graph of the effect of enabling "break-back" on yield in field 5. [Figure 121a]1 is a graph showing the effect of no "breakback" on shoot base height after pruning. [Figure 121b] 1 is a graph showing the effect of "breakback" effectiveness on shoot base height after pruning. [Figure 122] FIG. 10 shows a screenshot of an Excel table of shoot growth stages in field 5. [Figure 123] Graph of the effect of coefficient of variation of shoot elongation and development on predicted yield (Field 5; minimum harvestable shoot length of 3 cm; pruning from 75 cm to 60 cm). [Figure 124] 10 is a graph of the effect of setting last leaf to 7 when base height is less than stand height in field 5. [Figure 125] 10 is a graph showing how the updated algorithm for when base height < platform height showed that all shoots had moved to platform height in about 9 months. [Figure 126] 10 is a graph of the effect of setting last leaf to 7 or 10 in field 5 for pruning events only. [Figure 127] This graph shows how the updated algorithm for pruning events alone meant that all shoots moved to platform height in about 9 months. With lastleaf=7, all shoots moved from a base height of 60 cm within 12 months of pruning. With lastleaf=10, all shoots moved from a base height of 60 cm within 10 months of pruning. [Figure 128] 10 is a graph showing the effect of setting last leaf to 7 for field 13 when base height is less than stand height. [Figure 129] 10 is a graph showing the effect of pruning with last leaf set to 7 on assumed base height for field 13. [Figure 130]10 is a graph showing the effect of pruning with last leaf set to 10 on assumed base height for field 13. [Figure 131] 10 is a graph showing the effect of setting the last leaf to 7 or 10 on pruning for field 25. [Figure 132] 10 is a graph showing the effect of pruning with last leaf set to 7 on assumed base height for field 25. [Figure 133] 10 is a graph showing the effect of pruning with last leaf set to 10 on assumed base height for field 25. [Figure 134] 10 is a graph of the effect of setting last leaf to 7 or 10 on pruning for field 8. [Figure 135] 10 is a graph of the effect of setting last leaf to 7 or 10 on pruning for field 29. [Figure 136] 10 is a graph of the effect of setting last leaf to 7 or 10 on pruning for field 55. [Figure 137] 10 is a graph of the effect of setting last leaf to 7 or 10 on pruning for field 58. [Figure 138] 1 is a graph showing cumulative leaf emergence over time for field 58 before and after a pruning reset is applied. [Figure 139] Figure 1 is a flowchart showing the major storage and flow of nitrogen between soil and tea plants. [Figure 140] 1 shows two graphs showing the response of a) the C:N ratio coefficient (cnrf) to the soil C:N ratio and b) the soil temperature coefficient (Tf) to the soil temperature (Tsoil) as assumed in the CUPPA-Tea mineralization routine. [Figure 141] 1 is two graphs showing the response of a) pH coefficient (pHN) to pH and b) soil temperature coefficient (Tf) to temperature in the CUPPA-Tea nitrification routine. [Figure 142] Two graphs showing a) the soil moisture index (PWP: permanent wilting point, FC: field capacity, SAT: saturated water content, adapted from (Godwin & Jones, 1991)) used in CUPPA-Tea to correct soil nitrogen transformation rates, and b) the soil temperature coefficient response used in the CUPPA-Tea denitrification routine. [Figure 143] 1 is a map showing that tea yield data for calibration was derived from validation data from Tanzania (Ngwazi Tea Estate) and Kenya (Kericho Tea Estate). [Figure 144] Two graphs showing linear regression analysis of predicted yields using the CUPPA-Tea model along with observed yields from experiments in Tanzania (Ngwazi; 1989–1995) and Kenya (Kericho; 2015–2020) (dashed line: 1:1, RMSE: root mean square error, EF: modeling efficiency, shape indicates year and color indicates nitrogen treatment). [Figure 145] Three graphs showing the predicted effect of soil organic carbon (SOC) content (0%, 1.6%, 4%, and 8%) on tea yield response to nitrogen application in unpruned years (average yield in unpruned years) for a) rain-fed tea and b) fully irrigated tea (I4) in Tanzania, and c) rain-fed tea in Kenya. The default SOC values ​​for Ngwazi and Kericho were 1.6% and 4%, respectively. [Figure 146] Graph showing a comparison of the goodness of fit of CUPPA-Tea for Ngwazi and Kericho when the nitrogen routine is on (dashed line: 1:1, RMSE: root mean square error, EF: modeling efficiency, shape indicates year and color indicates nitrogen treatment). [Figure 147] 10 is a graph showing a comparison of CUPPA-Tea goodness of fit for Ngwazi and Kericho when the nitrogen routine is turned off. [Figure 148]Four graphs showing the projected revenues and costs of applying various levels of nitrogen fertilizer in Kericho based on measured and modelled yields of unpruned and pruned tea in Kenya, assuming a soil organic carbon (SOC) content of 4% (assuming 108KSh = 1USD). [Figure 149] 1 shows three graphs showing simulated daily soil moisture content in Kericho (rain-fed conditions in the top 15 cm) and Ngwazi (rain-fed conditions in the top 25 cm and I4 treatment). The solid line indicates the permanent wilting point, the dashed line indicates field capacity, and the dotted line is saturation moisture content. [Figure 150] A series of graphs showing monthly rainfall (blue bars), average temperature (red line), and average solar radiation (orange dashed line) for Ngwazi Tea Estate (Tanzania). [Figure 151] A series of graphs showing monthly rainfall (blue bars), average temperature (red line), and average solar radiation (orange dashed line) for Kericho Tea Estate, Kenya. [Figure 152] FIG. 1 shows the relationship between green tea leaf yield and time since pruning. [Figure 153] This graph shows the effect of adjusting the shoot count multiplier based on time since pruning (resetting to 467 to achieve 650 at 730 days). Pruning occurred on February 15, 2018. This procedure results in a falsely large yield in the first harvest after pruning. [Fig. 154] Graph showing that the number of modeled shoots increases from 0 to 1 per day of pruning, but the predicted number of shoots is multiplied by the output. [Figure 155] 10 is a graph showing the response of predicted and actual yield and predicted number of harvestable shoots to pruning on February 15th, assuming that the basal height remains at 75 cm and the modeled shoots increase by 1 per day. [Figure 156] FIG. 1 shows the modeled increase in shoot number from 0 to 2 per day of pruning. [Figure 157] Graph showing the response of predicted and actual yield and predicted number of harvestable shoots to pruning on February 15th (4 years since last pruning), assuming basal height remains at 75 cm and modeled shoots increase by 2 per day. DETAILED DESCRIPTION OF THE INVENTION

[0068] Aspects and embodiments of the present invention will now be discussed with reference to the accompanying drawings. Further aspects and embodiments will be apparent to those skilled in the art.

[0069] 1 is a diagram illustrating the flow of information in a tea harvest management system 110. Data about the tea plantation is communicated to a model 120 from a combination of a data store 130, user input 140, and sensor data 150. The model 120 executes on a processor 111 and establishes a simulated shoot 160 in a memory 112 using parameters based on an integration of information from the data store 130, from the user input 140, and from the sensor data 150.

[0070] The model 120 runs iterations, updating the parameters of the simulated chute 160. At each iteration, the model produces output data based on the parameters of the simulated chute and the collective information about them. This output data is sent to the output module 170, which processes the output data and prepares it for display to the user. This information is communicated to the user in the form of graphs, charts, and tables on the display 180, or in printed form (not shown).

[0071] definition "Tea" can refer to the raw material harvested from the plant Camellia sinensis, suitable for making the beverage of the same name. Camellia sinensia is also sometimes called "tea plant."

[0072] A "plantation" may refer to multiple plants under cultivation. It will be understood that a "plantation" may refer to a single growing area in which multiple plants are under cultivation, or to several combined growing areas in which multiple plants are under cultivation.

[0073] "Geographic data" may refer to data and information that has an implicit or explicit correlation to a location relative to the Earth (a geographic position or location). It may also be referred to as "geospatial data," "georeferenced data," "geodata," or "geographic information."

[0074] "Yield" may refer to the amount of tea (usually by weight) harvested per area of ​​land.

[0075] A "growing area" may refer to a geographically defined area where plants are grown, such as a field, greenhouse, or cultivation facility.

[0076] A "picking event" may refer to the harvesting of a tea plant, particularly when a given tea plant is harvested.

[0077] "Environmental data" may refer to data and information about weather and other ambient conditions that may affect the growth, yield, quality, or condition of tea or the tea plant.

[0078] "Remote sensing" may refer to obtaining information about an object or phenomenon without physical contact with the object. Specifically, it may include using satellite- or aircraft-based sensor technology to detect conditions in or changes to growing areas. It may include active remote sensing, such as radar technology, and passive remote sensing, such as photographic sensors that detect reflected sunlight. Aircraft may include unmanned aerial vehicles or drones.

[0079] "Ground truth data" may refer, in contrast, to data or information obtained through close, direct observation and recording of an object. Specifically, it may include manual recording of data about tea plants by one or more people through visual observation, through the use of hand tools or other short-range instruments.

[0080] "Nitrogen dynamics" can refer to the movement of nitrogen through atmospheric, terrestrial, and marine ecosystems through various biochemical forms and the biochemical processes that convert between them.

[0081] "Soil inorganic nitrogen" can refer to nitrogen compounds in the soil that are inorganic, i.e., the compounds do not contain carbon. Such forms include nitrate compounds (NO3 - ), ammonia (NH3), and ammonium compounds (NH4 + ) is included.

[0082] "Soil organic nitrogen" can refer to nitrogen compounds in soil that are in organic form, i.e., the compounds contain carbon. Such forms include fresh organic matter (FOM), such as carbohydrates, cellulose, and lignin. It also includes soil humus, a substance produced from the decaying remains of plants and animals.

[0083] "Critical nitrogen level" may refer to the level of available nitrogen below which tea plant growth slows.

[0084] "Abscission" can refer to the natural shedding of a part of a plant, typically a dead leaf.

[0085] "Banjhi" refers to the state of growth dormancy experienced by tea shoots.

[0086] "Chipping" refers to the first few removals of tea plants when the shoots (first) grow on the plant after pruning or training. The plants are chipped or topped off at a predetermined height parallel to the ground.

[0087] "Plantation" can refer to a large farm for growing crops, especially tea.

[0088] CUPPA-Tea model This section describes the use of a modified version of the CUPPA-Tea model with spatial data. The developed tool (Figure 2) has four key components: the CUPPA-Tea model, a user interface, data files, and a graphing / output module. This section begins with a list of data files and a description of the user interface. The data files are then described.

[0089] The original executable version 1.0 of the model, dated April 30, 1997, was used in the development leading to the 2015 version of the tool.

[0090] A problem with version 1.0 was that it could not handle any years after 1999. Therefore, for weather data and harvest dates after 1999, the tool was modified so that all years in the input had 40 years removed before running the model. The 40 years were added back afterwards so that the output would be correct.

[0091] The user interface is shown in Figure 3. In this version, the Kapkorech tea plantation in Kenya was used as an example.

[0092] Daily temperatures from the weather station were corrected for each field based on field altitude by subtracting 0.6°C for every 100 m increase in altitude. Temperature = Weather station temperature - 0.6 * (Field altitude (m) - Weather station altitude (m) / 100)

[0093] All outputs from the model were read by the tool and converted into CSV format.

[0094] To perform a simulation of the model, the following steps may be followed. -User selects an Estate from a drop-down list Weather stations and year start dates from the Estate are copied to the Simulation Run Start and end years based on the weather data from the farm's weather station ·Map zooms to the Estate -User selects a field or fields via a map · Default values ​​from Field are copied to Simulation Run User selects Run type (Estate, Selection of Fields or Field) For fields, users can choose Normal or Scenarios User selects the start and end years of the simulation based on available weather data Run Title defaults to run type and field selection Users can override the Irrigation, Plucking management, and Shoot initialisation options except when scenarios are selected. Scenarios can be selected by the user via a predefined tab-delimited text file called scenarios.txt -User presses the "Run Simulation" button to run the model The simulation will run - the output will be saved to a series of csv files in a new folder in the "Output" folder, and a new panel will appear allowing you to manipulate the various graphs. The graphs to be displayed are selected via a predefined list in csv format called graphlist.csv · User can press the Export to Excel button to export all data and graphs to an Excel spreadsheet

[0095] There are 12 data files with the following names: ·xxxxxx.geodatabase - Overview of Geographic Information Systems (GIS) for plantations and tea fields Field.csv - data related to individual fields Estate.csv - Data related to the estate, such as weather stations Tea.csv - detailed data related to tea genotypes SoilChar.csv - General soil properties SoilProfile.csv - soil profile properties xxxxYYYY.WTH - weather data related to weather station xxxx for year YYYY GumTea.INP - Data passed directly to the model from the data sources above along with default model parameters Scenarios.txt - a list of predefined scenarios to be executed Colours.csv - list of colours to use for the scenario in the graph Graphlist.csv - list of graphs to display Codelist.csv - column headings for the Excel spreadsheet

[0096] The input data used are a set of three files describing i) GIS layers, ii) plantations, iii) fields, iv) tea clones, v) soil, vi) weather, vii) initial tea conditions, viiii) crop management, ix) scenarios, and x) the form of the output, which are then explained.

[0097] GIS layer (xxxxxx.geodatabase) Field boundaries were digitized using ArcGIS and satellite imagery embedded in ArcGIS. Farm plans in pdf format were used to help determine individual fields and field numbers.

[0098] Three estates, Kapkorech, Kapgwen, and Chemosit, were digitized for use within the tool, along with other Kenyan tea estates, weather stations, and soil sampling points. ArcGIS map documents were then created for each estate so that runtime packages for each tea estate could be created for use within the CUPPA-Tea tool. However, digital maps of all Kenyan and Rwandan estates are now included in the tool.

[0099] Estate (Estate.csv) The data related to each farm is Farm name, weather station code (xxxx), weather station description, start year (day of year).

[0100] The starting year defaulted to January 1st because the recorded yield data for all comparisons was monthly / annual.

[0101] The weather station for Kapkorech was set up at Research and Development because there was no single weather station with sufficient weather data near the Kapkorech tea plantation. For Kapgwen and Chemosit, the weather station was set up at Chemosit because this station had more complete weather data and was closer to both plantations.

[0102] Field (Field.csv) Data related to each field on each farm is Field id, Farm name, Soil identifier, Soil description (e.g. medium sandy clay, medium sandy clay loam), Area (ha), Altitude (m), Dominant clone (genotype name) (e.g. BB35), Irrigation (Yes / No), Year of pruning (Yes / No), Harvesting date (date), Harvesting interval (days), Harvesting date (comma separated string).

[0103] Elevations were taken from the Aster V2 DEM and retrieved from the online Data Pool (NASA LP DAAC, 2015). The average elevation across each field was taken from the DEM using ArcGis statistics and then compared to the survey points (Figures 4 and 5). Most elevations were close to the survey points, with the exception of field 25 in Kapgwen and field 5 in Kapkorech.

[0104] The area in hectares of each field was taken from the digitized field boundaries, but these values ​​are only approximate and depend on the accuracy of the digitization.

[0105] Soil depth was taken from "Soil property maps of Africa at 250 m" (Hengl et al., 2015), giving an average depth of 175 cm across Kapgwen and Kapkorech. Therefore, the soil profile for each field was set to Medium (150 cm).

[0106] Soil texture was calculated from soil sampling points across the Kapkorech and Kapgwen farms. The soils were mainly sandy clay loam or sandy clay, and these sampling points were used to assign texture to the rest of the field.

[0107] For the three Kapgwen fields, the harvest date in 20115 was used to calculate similar harvest dates in 2011, 2012, and 2013, and the remaining days in October, November, and December were set to regular 28-day intervals.

[0108] Once a field is harvested, harvest information can be entered into the model. Data such as when the field was harvested, what method was used, yield, and field pruning status can be recorded. This can be recorded by manual means such as a paper form. It can also be recorded through a web form such as a Microsoft PowerApps card interface.

[0109] The remaining parameters were left to their default approximate values. The dominant clone was set to BB35. Irrigation was set to No, as these farms do not use irrigation. Year of pruning was not used in this version of the tool.

[0110] Tea (tea.csv) The model parameters defined by the original model for clone 6 / 8 (Table 1), except for the clonal coefficient, were calculated to account for other clones and new clones in the plantation. Tea cultivars grown in the tea plantation include BB35, MRTM1, 31 / 8, and seedlings. The first fitting model parameters for BB35 were set the same as those for 6 / 8, but 20% was added to "base temperature for development," "base temperature for elongation," "stage 2 development rate," and "stage 2 elongation rate." The shoot number was set by comparing the model run with recorded annual yield data. [Table 1]

[0111] Soil (SoilChar.csv and SoilProfile.csv) The five soils described in the original model were silty clay, silty loam, sandy loam, sand, and "Ngwazi Tea Research Unit" (NTRU). These were defined for three depths. Tables 2 and 3 list the required parameters and which of them are currently used by the model. Table 4 was also provided in the original data so that parameters for other soil textures could be derived. Sampling points across the tea plantation yielded two soil textures: sandy clay and sandy clay loam. As a result, new soil descriptions IB00000014 to IB00000019 were derived using median soil property values ​​and weighted averages of the soil profile based on the existing soil parameters and Table 4.

[0112] The derived soil data was modelled and may not necessarily represent the soils in tea plantations, so a best approximation was used. [Table 2] [Table 3] [Table 4]

[0113] Weather(xxxxYYYY.wth) Tables 5 and 6 show the data used for each weather station. Stable daily weather data from a single weather station was not available for any of the three estates. Therefore, weather data was taken from several sources, and the one closest to each tea estate was selected when possible. The model uses at least two years of weather data to initialize the shoots. [Table 5] [Table 6]

[0114] For all three tea estates, the only available solar radiation data were monthly values ​​from the Tea Research Foundation of Kenya. These monthly values ​​were converted to daily values ​​using a 30-day moving average algorithm.

[0115] For the Kapkorech farm, minimum and maximum temperatures were taken from recorded monthly data for 2011 and from recorded daily temperatures for 2012 and 2013. For monthly data, a 30-day moving average algorithm was used to generate daily values. Monthly rainfall data for 2011–2013 were taken from the Kapkorech weather station.

[0116] Figure 6 shows the weather data used for the Kapkorech farm.

[0117] For the Kapgwen and Chemosit farms, minimum and maximum temperatures were taken from monthly data from Kapgwen for 2011 and 2012, and from daily temperatures from Chemosit for 2013. For monthly data, a 30-day moving average algorithm was used to generate daily values. Daily rainfall was taken from the Chemosit weather station for 2011, 2012, and 2013.

[0118] Figure 7 shows the weather data used for the Kapgwen and Chemosit farms.

[0119] The altitude of each weather station is important as it is used to determine temperature differences in the field. Therefore, Kapkorech weather data was set to that of Unilever Research Station (2067 m) for all years, and Chemosit data was set to the altitude of Kapgwen (1902 m) for 2011 and 2012 and to the altitude of Chemosit (1830 m) for 2013.

[0120] Initial conditions (Gumtea.inp) Table 7 shows the data for the initial conditions, which are derived from the original model data. These parameter values ​​were not changed except for the number of shoots, which was derived from the genotype as listed in tea.csv. [Table 7]

[0121] crop management Table 8 shows the crop management data, which is primarily derived from user input into the user interface or from the scenarios text file. Some of these parameters can also be derived from selected field data (field.csv). Population density and row spacing remained unchanged. [Table 8]

[0122] Scenarios (Scenarios.txt) The scenario text file is a tab-delimited file containing the following data: Genotype name, e.g. BB35 Title, the title that should be used on the graph Irrigation, 0 for no irrigation or 2 for automatic irrigation If Irrigation=2, Irrigation threshold (mm) Harvest interval (days) Manual, True for hand-picked or FALSE for machine-picked Pluck Min Leaves, 1 to 5 Pluck Max Leaves, 1 to 5 Shear height (mm) Break back, TRUE or FALSE Break leaves, 1 to 4 if Break=TRUE Minimum shoot size (cm) Maximum shoot size (cm) Shoot initial, 0 for Zero all shoots, or 1 for automatic initial setting Harvest dates, a list of dates in dd / mm / yyyy or yyyy / mm / dd format Harvest mode - usually 0, or 1 for selected dates

[0123] Tool Setup Data Three additional csv files contain detailed information required by the tool on how to configure and adapt the tool's output. These files (Table 9 and Table 10) Colours.csv - list of colours to use for scenarios in the graph Graphlist.csv - list of graphs to display Codelist.csv - column headings for the Excel spreadsheet is. [Table 9] [Table 10A] [Table 10B]

[0124] Output data This section describes the output data. The tool creates several output files, which are then used to create a series of graphs. A single Summary file listing all model runs An overview file for each model run listing model run parameters, crop details, and irrigation details A growth file for each model run listing data on crop growth and yield Water files for each model run for soil water balance and crop water uptake and use

[0125] Summary The Summary file (Table 11) describes the output, such as annual yield. [Table 11]

[0126] Overview The Overview output file describes important characteristics of the run (Table 12), harvest (Table 13), irrigation (Table 14), water (Table 15), and tea growth (Table 16). [Table 12] [Table 13] [Table 14] [Table 15] [Table 16]

[0127] Soil Sensitivity Analysis To test the sensitivity of the model to soil profile parameters, the first soil defined in the original model, with a silty clay texture, was copied to new soil IDs (IB00000014 to IB00000022). Table 17 shows the general soil properties used for all test soils. These soils were then modified to test three of the soil parameters: root growth, apparent density, and organic carbon. Table 18 shows the modified soil profile properties for the test soils. In addition, three different depths were tested for the silty clay. [Table 17]

[0128] The CUPPA-Tea model was then run for each of the 12 soils in a single field at the Kapkorech tea plantation, and cumulative growth and yield were plotted (Figure 8). Root growth, apparent density, and organic carbon parameters appear to have little effect on the results. The key parameter affecting the output is soil depth, with deeper soils showing greater cumulative growth than shallower soils. Figure 9 shows the modeled yields for the 12 test soils. [Table 18]

[0129] Sensitivity to tea shoot density Several test runs were conducted on a single field at the Kapkorech tea estate to determine the most appropriate number of shoots for the tea varieties grown on the estate. The modeled annual yield of field 9 was then compared with the recorded annual yield of the same field. For the BB35 variety, the values ​​were copied from the default and adjusted by 20%. The number of shoots was 1200 m -2 750m from -2and at a 21-day harvest interval, predicted annual yields were similar to recorded values ​​(Table 19).

[0130] Figure 10 shows runs from CUPPA Tea, with variable shoot numbers and harvest intervals compared to recorded yields. Figure 11 shows recorded monthly yields (kg MT ha-1) for selected fields on the Kapkorech tea estate, and Figure 12 compares modelled yield per harvest with recorded yield per month. There may be differences in amounts as there could be more than one harvest per month. [Table 19]

[0131] Modelled and recorded monthly yields in Kapkorech Figure 11 shows recorded monthly yields for selected fields on the Kapkorech tea estate from January 2011 to December 2013. By using proportional yield coefficients for each month (Table 20), the annual yield per field appears to be distributed over 12 months, so the field data follow a consistent seasonal pattern. This distribution appears to be assumed for each year and each field regardless of pruning. Therefore, the CUPPA-Tea output (Figure 12) shows significantly more variation than the field data. [Table 20]

[0132] Modeled and recorded yields at Kapgwen Harvest dates for 2015 were recorded for some fields on the Kapgwen tea estate. Field 28 was selected because it was recorded as cultivating the BB35 variety. The dates given for field 28 were: 24th, 23rd, 25th, 24th, 24th, 25th, 23rd, 24th, 19th, 23rd, 26th, and 24th.

[0133] These dates were used to create the same specific dates for 2011, 2012, and 2013, and the model was then run across several scenarios. Figure 13 shows the modeled monthly yields in different scenarios against the recorded yields for 2013. Some variation was expected, as it was unlikely that the same harvest dates were used in 2013 as in 2015.

[0134] The yield reduction measured in the August 2013 field is interesting in that it does not appear to have been caused by temperature or drought effects. Possible reasons for the yield reduction could be i) hail damage or ii) some pruning activity in parts of the field.

[0135] In Figure 14, the modeled output for field 28 in 2013 is presented along with the field's recorded yield (green) and calculated soil moisture deficit. The modeled yield (red) behaves similarly to what would be expected. In an environment such as Kericho, where temperatures are relatively stable, yield depends primarily on changes in soil moisture deficit. Note that yield per month and yield per harvest are not the same, as there can be more than one harvest per month.

[0136] The modified model is presented in a new user interface that can handle spatial data.

[0137] The need for individual harvest data Initial results suggest that the CUPPA-Tea model can predict reasonable seasonal variations in individual harvests of clonal tea in Kenya. Integration of the model with the use of remote sensing to support yield forecasting has been achieved. The extent to which remote sensing can be used to support tea yield forecasting is discussed below.

[0138] Analyzing the model Analysis of Kapkorech harvest data Field-level data for the Kapkorech Farm in Kenya was analyzed. Field data were provided for the period January 2014 to September 2016.

[0139] During this period, monthly rainfall varied from a minimum of 1 mm in January 2015 to a maximum of 370 mm in November 2016. Between January and March, rainfall was 303 mm in 2014, only 67 mm in 2015, and 327 mm in 2016 (Figure 15).Thus, although temperatures in Kenya can be relatively stable, there can be considerable variability in rainfall from year to year.

[0140] The total fresh weight tea yield in 2015 (11,390 kg / ha) was only 77% of that in 2014 (14,700 kg / ha). The main reason for the low yield in 2015 was the low yield in February, March, and April 2015, which was related to the low rainfall levels mentioned above. For example, only 870 kg / ha of green tea leaves were harvested between February and April 2015, compared to 2,790 kg / ha in February and April 2014 and 3,240 kg / ha in February and April 2016. Figure 15 clearly shows that although there was significant rainfall in April 2015, yields did not recover until May 2015. Thus, there is a delay between the onset of rainfall and the new shoots reaching harvestable size.

[0141] Between May and September, the average monthly yield of fresh tea in each of the three years was relatively stable, with monthly green tea leaf totals between 926 kg / ha and 1,417 kg / ha. There was a spike in tea production in October 2014, which may suggest something synchronized tea growth in the preceding months.

[0142] An analysis of the relationship between green tea leaf yield, temperature, and harvesting method (Figure 16) reveals that all tea grown above 2150 m altitude is mechanically harvested. Below 2150 m, about half of the tea is mechanically harvested and half is hand-harvested. In 2015, the average yield from mechanically harvested fields (13,240 kg / ha) was about 29% higher than the average yield from hand-harvested fields (10,270 kg / ha).

[0143] Previous studies have suggested that tea yields decrease with increasing altitude due to lower temperatures and slower shoot growth rates (Carr & Stephens, 1992), an effect that is not immediately apparent from this initial analysis of the data.

[0144] Further development of the CuppaTea model Validating a new version of the model As discussed above, the 1997 CUPPA-Tea model was modified to provide a new user-friendly front end and was also adapted to incorporate geographic data. Below, a further modified version of the 2004 CUPPA-Tea model is discussed, which also had source code written in the Pascal programming language. The source code was also converted to C# to allow further development of the model in common modern programming environments.

[0145] Thus, there were four versions of the model compared. 1. The original version from 1997 (1997) 2. Version submitted with source code from 2004 (2004) 3. The version resulting from recompiling the source code with the Pascal compiler (Recomp Pascal) 4. Model recoded in C# based on source code (C#)

[0146] This new C# version of the model is 1. Run the 2004 executable recompiled pascal code to ensure that the conversion to C# did not introduce errors. 2. To investigate what changes were introduced into the model between 1997 and 2004, it was tested against the original 1997 version.

[0147] Comparing the output of the C# model and the 2004 model Different versions of the model were compared using original data from the Ngwazi Tea Research Unit in Mufindi, Tanzania. Parameter options were set for a regular 14-day harvest interval from April 13, 1989, to July 19, 1990, without irrigation. Results (Figure 17) showed good agreement between the 2004 and new C# versions of the model. In fact, the correlation between these two versions is strong (Figure 18).

[0148] In contrast, the results from the 1997 version differ from the 2004 and C# versions (Figure 18). The 1997 version shows an earlier peak in yield, in December 1989, after the end of drought stress, compared to the later 2004 version. This is in part a result of the 1997 version of the model calculating soil water deficit differently than subsequent versions (Figure 17). The new model also limits the number of leaves that can open during the elongation phase. Once the number of leaves reaches the "last leaf" parameter for an individual shoot, the new model resets the growth phase to 0, thus preventing the emergence of new leaves.

[0149] The difference in output between the C# and 2004 Pascal versions, highlighted in Figure 17, was relatively small. A detailed analysis of the output on a particular day (day 114) showed that for 500 shots, the 2004 model: Five more shoots with more than two leaves, so five more shoots to remove; Fewer shoots with only two leaves, Fewer shoots with two leaves and growth stage set to 1 It was shown that the prediction

[0150] One possible reason for this small difference is that these two models use different methods to calculate random numbers, so the small difference found between the C# and Pascal versions of the model is likely due to some random number generation of the shoot development parameters.

[0151] Comparison of models for irrigated tea The outputs from the 2004 model and the C# model were compared with the actual yields of fully irrigated clone 6 / 8 measured at the Ngwazi Tea Research Unit. In general, the timing of the first yield peak coincides with the actual yield peak (Figure 19). Therefore, the new version of the model appears capable of predicting the yield distribution of fully irrigated tea in Tanzania. Because the original 1997 model did not perform correctly with irrigation dates / amounts, it was excluded from this analysis.

[0152] Comparison of models for non-irrigated tea The outputs of the "1997" and "C#" models were compared with the actual yield from non-irrigated tea in Tanzania from January 24, 1989, to October 19, 1990 (Figure 20). In this case, both model versions showed an earlier crop peak than the actual results, in December 1989, after the removal of drought conditions. This earlier predicted peak for these models resulted in two more crop peaks, compared to only one in the actual results.

[0153] CUPPA-Tea uses a water balance model to calculate water stress factors, which are then used to determine shoot development. Figure 21 shows the seasonal evolution of the modeled water stress factors for the period from January 1989 to September 1990. First, the 1997 version of the model incorrectly indicates water stress early in 1989. Then, during the main dry season, the water stress factors in the 1997 version show a slower response than the new version, which includes a delay of approximately 20 days in the shoot growth response after the arrival of rain. The adjusted water stress is incorporated into this delay factor.

[0154] The role of water stress factors The water stress factor is an important parameter derived from the water balance model within CUPPA-Tea. Although the water stress factor recovers slower than in the 1997 version (when rainfall starts to occur again), the recovery of the water stress factor, and therefore of shoot growth, appears to be faster than that indicated by the actual yield data.

[0155] As referenced in Appendix A, a case study from Tanzania noted that 6 mm of rainfall in one day, followed by no rain for 11 days before and 5 days after, could increase the water stress factor from 0 to 0.255. Since there was little rain before or after, the water stress factor is expected to remain at 0. The calculation of soil moisture and the calculation of actual water uptake based on the initial setting of the lower limit appear to be the key factors that cause changes in the water stress factor (Figure 22). The water balance model is described below in Table 21. [Table 21]

[0156] Comparison of modelled yields with Kenyan yields Daily yield data for each field on the Kapkorech estate from January 2014 to March 2016 was analyzed to determine the best way to compare yields predicted by the model. Two fields were selected for this purpose: Field 5, which contains an area of ​​hand-picked seedling tea, and Field 37, which contains an area of ​​mechanically harvested clone 35. Field 5 is at an altitude of 2096 m, and Field 37 is at an altitude of 2095 m.

[0157] The average green tea leaf yield for each harvest from the field is shown in Figures 23 and 24. The CUPPA-Tea model assumes an individual tea plant and a single harvest date spaced over a specified period. However, actual harvest data shows that each field is harvested over several days, making it difficult to determine a single harvest date.

[0158] To justify the harvest date (which is necessary for comparison with the model), the actual data were processed as follows. 1. Actual daily yield data were grouped to create yield data with a single harvest date for the entire field per harvest event. 2. Green tea leaf yield was converted to dry leaf yield.

[0159] Harvest date grouping Initially, actual yields were grouped, with harvests occurring on consecutive days or every other day, and the harvest date set to the date on which the highest yield occurred. However, the number of harvest days was still too frequent, so yields were regrouped so that harvests occurring over a 7-day period were considered a single harvest event. The harvest date was set as the average date. However, the model could also be configured to reset to the harvest start date.

[0160] When the model was run for other fields, it was noted that the initial algorithm meant that some fields were harvested daily for weeks at a time, which resulted in a single very large yield on the average harvest date. Therefore, the following new grouping algorithm was developed: Harvested within 7 days of each other and less than 14 days from the first harvest date were considered part of the same harvest event. The harvest date was set as the average date.

[0161] Conversion to dry leaf yield The CUPPA-Tea model predicts yield in kilograms of finished tea per hectare, while Kapkorech's field data are green tea leaf yields. To convert green tea leaf yields to dry leaf (or finished tea), a dry matter conversion constant was used. This was calculated monthly as "total finished tea" divided by "factory weight." Average monthly values ​​generally varied between 20% and 25% (Figure 25). However, in March 2015, a particularly dry month, the constant exceeded this range by 28%. (Stephen & Carr, 1993) noted seasonal variations in dry matter content for Tanzanian tea plants, with figures ranging from 18% to 32%.

[0162] By grouping harvests and using dry matter conversions, it was possible to derive yield profiles for each field, which could be compared with the output of the CUPPA-Tea model (Figures 26 and 27).

[0163] Comparison of modeled and hand-harvested yields The CUPPA-Tea model output for hand-harvested field 5 shows some agreement with actual yield (Figure 28). As reported by Matthews and Stephens (1989), the first few months of output from an uncalibrated model tend to result in a series of peaks that are artifacts of new shoot development. The model predicted low yields that occurred between February and April 2015, which were associated with drought stress. Subsequent yields followed a relatively consistent pattern through January 2016. The model appears to predict a stronger response to drought between February and April 2016, which is not evident in the actual data for field 5. Absolute yield levels are generally good.

[0164] Comparison of modelled and mechanically harvested yields Unlike harvested tea, the initial parameterization of the CUPPA-Tea model was not able to model the actual yield from the machine-harvested field (Figure 29). This occurred despite changing the assumed height of the "step" from 5 cm to 50 cm. CUPPA-Tea appears to predict higher yields from machine-harvested tea when there is a larger harvest interval that allows the shoots (remaining on the tea plant at the last harvest) to reach a harvestable size.

[0165] Sensitivity analysis of CUPPA-Tea parameters Sensitivity analyses were conducted for selected parameters in the CUPPA-Tea model. The parameters tested were variation in step height (mechanical harvesting), number of leaves picked (hand harvesting), shoot number, soil type, and soil depth.

[0166] Step height In mechanical harvesting, varying the step height is one way to manage the proportion of shoots removed in each harvest. As demonstrated in Figure 29, choosing a larger step height tends to lead to lower yields because a higher proportion of harvestable shoots remain on the tea plant. Varying the step height does not appear to change the timing of the modeled peaks and troughs in yield.

[0167] Number of leaves on harvested shoots Ground truth data from remote sensing surveys (shown in Appendix B) indicated that picking regimes in hand-picked fields were variable but tended to be two to three leaves. When hand-picked tea was modeled and the criteria for acceptable hand-harvested shoots were extended from shoots with at most three leaves to, say, shoots with at most five leaves, there was a tendency for yield peaks to become more pronounced (Figure 30). Yields during yield troughs remained relatively stable.

[0168] Number of shots Since varying the number of shoots in CUPPA-Tea effectively changes the number of shoots per square meter, increasing the number of shoots increases yield without changing the timing of peaks and troughs (Figure 31).

[0169] soil type Varying the assumed type had a relatively small effect on modeled yields (Figure 32). However, significant differences emerged during the dry season in February 2016, when the "medium sand," "medium sandy clay," and "medium silty clay" soils produced greater drought stress responses than the "medium sandy clay loam," "medium silty loam," and "medium sandy loam." This is likely due to the higher available soil moisture content in the latter soils.

[0170] Investigating options for weather station and meteorological data integration The incorporation of current and forecast weather data into the CUPPA-Tea model has advantages. The key climate parameters in the CUPPA-Tea model are: 1) Daily solar radiation (MJ / m 2 ) 2) Daily rainfall (mm) 3) Daily minimum temperature (℃) 4) Daily maximum temperature (℃) is.

[0171] Additionally, the following may be locally useful for tea plantation managers: Saturation deficit, as high values ​​(>2.3 kPa) can reduce yield. Hail, which on average across Kenya's tea-growing regions can reduce annual yields by 10% and result in the loss of crops for up to seven weeks. -Frost occurrence.

[0172] The frequency of updates to current and forecast data in CUPPA-Tea depends on how the model is used. For example, if the fiscal year is January to December, running the model with the latest weather information from June will show the impact of dry seasons on annual yields. Alternatively, CUPPA-Tea can be updated ad-hoc by reading comma-separated files in a designated PC folder in a familiar format expected by the software.

[0173] Accuracy of meteorological data to CUPPA-Tea is important. Preferably, data should be fully validated and, if necessary, corrections made before use. Therefore, human intervention in collecting meteorological parameters can be advantageous, as it can help enable regular quality control and ensure each sensor is working. The problem with manual processing of data is the time required to record details, ensure the data is recorded in a consistent format, and upload the data to the correct folder with the correct name.

[0174] Automatic Weather Station An automated weather station (AWS) is a collection of sensors and data loggers for recording information from the sensors. The sensors record raw values ​​that must be calibrated to allow conversion to specified weather parameters. Typically, two sets of values ​​are recorded by the logger: the raw sensor measurements and conversion factors, which can then be summarized into an archive. Alternatively, some logging systems use third-party software on a PC directly connected to the data logger to directly process the real-time data from the sensors and summarize it into an archive.

[0175] Raw values ​​and weather data can either be "pushed" by a logger to a specific location for retrieval, or "pulled" to the AWS by dialing. The appropriate approach depends on the type of AWS and logger, as well as the associated infrastructure surrounding the AWS. For example, an AWS could be combined with a cellular modem and solar panel power to enable communication with the AWS anywhere cellular reception is available.

[0176] One advantage of communicating directly with the data logger is that only the needed data at the selected frequency is accessed. The data does not need to be pre-processed by any third-party software, which can have delays and drawbacks. One possible drawback is that although the logger is accessed directly, it may not be able to send real-time data to any PC-based software interface often provided with AWS. Additionally, custom software would need to replicate the functionality of the PC software, including data management and archiving.

[0177] The data format is system dependent, but could be a comma separated value (CSV), which is easily readable by scripts, however there is no established standard for the content so this will vary between logger builds.

[0178] There are many different brands of weather stations such as ADCON Telemetry or Davis, or bespoke solutions such as Cwi Technical Ltd, which will be designed to suit the project requirements and intended AWS infrastructure / location. The cost and type of AWS will depend on many factors such as what data is required, the accuracy required, how often it is required and whether the AWS will need to be reprogrammed or queried from the UK.

[0179] If automated weather stations are implemented, full validation of the data should be investigated. A good starting point for this is the paper published on the World Meteorological Organization's community page on quality control procedures for AWS (Zahumensky, 2004). Alternatively, there are many third-party software packages available to process and present AWS data, which should incorporate data validation, data management, and archiving.

[0180] Forecast Data To predict potential annual yields, it would be useful to provide CUPPA-Tea with forecast weather data from the end of the current data through the end of the current year. There are many companies advertising forecast data for Kenya. However, the accuracy of such forecasts should be questioned.

[0181] An alternative to purchasing weather data is to use weather data from existing weather stations averaged over the past five years.

[0182] Figure 37 shows the average maximum and minimum temperatures and rainfall from 2011 to 2015 and a comparison with the current data for 2016. The average temperatures show a good approximation, but the average rainfall does not capture extreme / unusual rainfall events.

[0183] Remote Sensing Using remote sensing to assist tea yield forecasting This section evaluates the use of remote sensing (RS) to support tea plantation yield predictions using the CUPPA-Tea model. We focus on two tea plantations located in Kenya: Kapgwen Estate and Kapkorech Estate.

[0184] Kenyan plantations are located close to the equator, which means that seasonal variations in temperature and day length are minimal, allowing tea to be harvested all year round.

[0185] Previous research Previous studies have considered using remote sensing to monitor tea plantations in India and Sri Lanka. While tea is harvested year-round in Sri Lanka and southern India, the harvest season in northern India lasts only from late March to early November.

[0186] (Rajapakse et al., 2010) described the spectral reflectance of tea canopies in Sri Lanka using in situ measurements with a spectrophotometer. They reported that the reflectance peak at near-infrared wavelengths (NIR) (800 nm) increased with time since the last pruning. They also reported that the spectral reflectance in the red band (500-600 nm) decreased with time since the last leaf pruning. They also derived an empirical relationship between the Normalized Difference Vegetation Index (NDVI) and the Leaf Area Index (LAI).

[0187] (Dutta, 2011) used MODIS imagery to delineate flooded tea areas in Northeast India and established an empirical relationship between MODIS NDVI and leaf area index of tea plantations. The relationship between NDVI and leaf area index was analyzed over a period of one year. Dutta concluded that NDVI measurements alone cannot explain the seasonal variations in tea yield.

[0188] (Banerjee, 2012) also evaluated the use of L-band Synthetic Aperture Radar (SAR) to estimate tea plant biomass, also in Northeast India. He used a water cloud model to correlate fully polarimetric ALOS PALSAR data with surface biomass. The measured root mean square error between the estimated biomass and available ground measurements was 25m 2 The authors attributed the limited performance of the results to the inability of the model to cope well with variations in shrub structure.

[0189] Remote Sensing Option Tea harvest, or the young, perishable shoots, usually occurs at intervals of 2 to 4 weeks. To predict the seasonal distribution of tea yield, the CUPPA-Tea model requires 25 crop parameters (Table 1) and 13 harvest parameters (Table 8). The number of shoots per unit area and the harvest interval are important determinants of the modeled yield.

[0190] Results reported by (Rajapakse et al., 2010) and (Dutta, 2011) suggest that NDVI measurements do not appear to be sensitive to canopy variations between harvests. NDVI may only be sensitive to significant changes in overall tea plant growth and leaf area index. Therefore, optical near-infrared remote sensing may only be able to detect year-to-year variations in crop parameters.

[0191] Another limitation of optical near-infrared remote sensing is that it requires minimal cloud cover. The frequent cloud cover over Kenyan tea plantations is evident from an examination of Landsat-8 archive imagery. Cloud cover over Kapgwen Plantation appears to be less than over Kapkorech Plantation, and it was possible to derive average NDVI values ​​for each field on the plantation where field boundaries were available (Figure 33).

[0192] NDVI was calculated directly from the three digital images. Because leaf harvesting occurs on different days in each field, the NDVI variation for these three dates would be expected to be different for each field. Instead, there is an overall deviation between the three NDVI dates. The NDVI on September 25 was lower than the NDVI on September 9. This could be the result of pruning activities, but could also be due to differences in solar irradiation and / or atmospheric conditions. A longer time series of measurements is needed to confirm the sensitivity of NDVI to tea plant growth or variation between harvests. Cloud cover prevented NDVI calculations for the Kapkorech farm.

[0193] Synthetic Aperture Radar Remote Sensing (SAR RS) may be a more suitable technique due to its all-weather capability. Radar wavelength is an important parameter in SAR RS; the longer the wavelength, the greater the penetration depth (Ulaby et al., 1981). In L-band (i.e., in the range of 1 GHz to 2 GHz), the incident waves can penetrate the canopy and some of the waves also interact with the soil. This is why (Banerjee, 2012) used L-band imagery to estimate aboveground biomass of tea.

[0194] L-band may appear to be an ideal wavelength for capturing information about tea crops, since the response signal is a composite of energy scattered by the upper canopy, leaves, and soil surface. However, the different scattering mechanisms throughout the wave round trip and their complex interactions also mean that they are difficult to model. Different scattering contributions are difficult to separate when there is no dominant process. However, L-band measurements can identify changes in tea plant size, since greater volume scattering can produce stronger depolarization of the incident wave. Increased depolarization can be attributed to the cross-polarized backscatter coefficient

[0195]

number

number

[0196] Tea bushes have fairly dense canopies, so when using the X-band (7-11.2 GHz range), incident electromagnetic waves are expected to be scattered by the upper part of the canopy with little penetration. This may make SAR measurements easier to interpret. Harvested shoots can be 2-8 cm in length. Therefore, measurements in the X-band, at a wavelength of approximately 3 cm, may be expected to be very sensitive to shoot leaves and may be able to detect when harvesting has occurred.

[0197] Combining different radar polarizations at X-band can provide valuable information. Vertically polarized incident waves interact preferentially with vertically oriented scatterers, while horizontally polarized incident waves interact specifically with horizontally oriented scatterers. Young leaves have a more vertical orientation than mature leaves, so the vertical co-polarized backscatter coefficient

number

number

[0198] The orientation of the leaves on the tea stem can be seen in Figure 34. The young leaves on the tea shoot are more vertically oriented than the mature leaves below.

[0199] Measurements in the intermediate C-band (4-8 GHz range) may also be able to detect changes in the upper canopy, but their sensitivity may be lower than X-band measurements. The main advantage of C-band is the open image access from Sentinel-1, which allows for preliminary testing. Figure 35 shows the average gamma noughts for each field on Kapgwen Farm on September 15 and October 9, 2015, i.e., a 24-day interval.

number

[0200] Between these two dates, it is expected that leaf harvesting must have occurred in some parts of the field during the 24-day period.

number

number

number

[0201] In summary, NDVI and L-band measurements can be used to extract information about the overall condition of the tea plant, however, initial analysis suggests that only X-band measurements, which are sensitive to upper canopy structure, will be able to detect the effects of individual crops.

[0202] Examples of measurement activities Number of shots from L-band SAR To field test the use of L-band remote sensing, ground-based measurements of canopy cover and shoot density need to be made at the time of L-band SAR acquisition. Standard information about the weather (rain, temperature) is also important for the interpretation of the measurements. Ideally, monthly SAR acquisitions over a year (possibly including pruning events) would improve the probability of capturing changes in tea plant condition.

[0203] The L-band radar PALSAR-2 onboard ALOS-2 can provide fully polarimetric images (Suzuki et al., 2009). The 6m resolution attenuates the effects of speckle noise and allows sufficient averaging per field to produce an accurate radar vegetation index. A plot of approximately 50km is sufficient to cover two farms in a single image. To limit the variation in measurements over time due to differences in imaging geometry (different incidence angles), images should be acquired from repeated ground tracks. The repeat period of ALOS-2 is 14 days, allowing monthly acquisition.

[0204] Figure 36 shows the X-band

number

number

[0205] Harvesting time from X-band SAR To maximize the probability of detecting the effect of leaf harvest on X-band measurements,

number

[0206] As an alternative, to test for harvest detection, the first image may be acquired before harvest 1 and the second image may be acquired immediately after harvest 2 (Figure 36). In either case, the radar acquisition must be synchronized with the harvest cycle of the largest number of fields. This should not be a problem as COSMO-SkyMed provides daily imaging opportunities. SAR measurements are very sensitive to water, so image acquisition should not be performed on days when rainfall is expected, as liquid water carried by the canopy can significantly affect the measurements. γ with shoot development 0 To better understand the evolution of γ over time, images are ideally taken every two days over several months. Such a time series shows the evolution of γ over time. 0 allows for a complete characterization of

[0207] Regarding imaging modes, there are trade-offs between polarization, spatial resolution, and coverage. Dual polarization (vv, hh) is desirable because vv and hh measurements are expected to vary in opposite directions. The canopy scattering index vv / (vv+hh) then reveals variations in the structure of the upper canopy (Haldar et al., 2012). Dual polarization is available in stripmap Ping Pong mode, which has a coverage of 30 km and a resolution of 15 m. Although the distance between the two farms is slightly less than 30 km, the SAR image may not be centered to cover both farms simultaneously. Therefore, individual images can be used to perform analysis on both farms. Other imaging modes offer higher resolution (3 m for 40 km coverage or 1 m for 10 km coverage) but single polarization. Higher resolution allows for further averaging per field to attenuate speckle noise, but this may not be significant, as with Sentinel-1's 14m resolution

number

[0208] Finally, ground data should include the exact date of harvest, the density of harvested shoots, and photographs of the canopy before and after harvest for each field. Additional information on canopy size and height is also useful in understanding any field-to-field variability in SAR measurements.

[0209] Synthetic aperture radar (SAR) remote sensing, rather than optical / near infrared (NIR) remote sensing, appears to be the best method to aid in the prediction of tea yields in Kenya. · Unlike optical / NIR radiometers, SAR can image through clouds. · L-band SAR can provide information comparable to NDVI regarding the overall growth of tea plants. ALOS can provide L-band imagery, one image per month over the course of a year, with all images acquired from the same repeating ground track. X-band SAR appears to be the best option for detecting changes in the upper canopy structure due to leaf harvest. Such measurements can be used to detect the time when harvesting took place for each field. X-band imagery can be provided by the COSMO-SkyMed constellation. Dual-polarized images acquired before and after several harvests, or a two-day time series, will demonstrate the X-band's ability to detect harvests. · All SAR images (L-band or X-band) should not be acquired on days when rain is expected to occur, to prevent liquid water trapped in the canopy from affecting the measurements. Ground data should include for each field the exact date of harvest, the number of shoots harvested, photographs of the canopy before and after harvest, information on the size of the tea plants (height, width), and standard weather information (temperature, rainfall).

[0210] Remote detection of tea Here, the results of an experiment evaluating the ability of satellite remote sensing to provide information about tea plantations are discussed. The overarching goal is to use remote sensing to improve plantation yield estimation and management based on reliable, timely, and spatially distributed information.

[0211] Previous studies on satellite remote sensing of tea have been mainly limited to classifying land cover to detect tea areas. Here, the goal is to explore the link between remote sensing measurements and biophysical parameters of tea plants. As discussed herein, remote sensing can monitor shoot growth and · Study the response of tea growing to changing climatic conditions; Detect areas where tea may be affected by disease or hail damage; - Estimate the optimal time for harvesting and allocate the right number of pickers to the right locations; -Estimate yields from the stage when the field is harvested, Current plant condition and growth rate, along with additional weather data, can be used to predict future yields.

[0212] This last point can be achieved by combining remotely sensed measurements with the CUPPATea model. While satellite remote sensing cannot always provide absolute measurements of biophysical parameters, it can provide indicators that reflect plant condition. Such indicators can be used to calibrate the output of models such as CUPPATea.

[0213] Two types of remotely sensed data are tested: (i) multispectral data from UK-DMC3 and (ii) X-band SAR data from CosmoSkeymed. The ability of multispectral remote sensing to monitor crop stage using the Normalized Difference Vegetation Index (NDVI) has been highlighted in numerous scientific papers (Wiegand et al., 1991). Here, we hypothesize that NDVI may be able to detect small changes in canopy volume between harvested and unharvested plants. In comparison, X-band SAR is expected to capture changes in canopy surface structure because at this wavelength, there is very little penetration through the canopy and the measured signal is theoretically more sensitive to surface scattering.

[0214] Although monitoring shoot growth is the most valuable remote sensing output, results are also presented for the detection of management activities, such as harvesting method (mechanical harvesting (MH) or hand-harvesting (HP)), which relates to yield, and surface area harvested, which is important since fields are often only partially harvested on a given day.

[0215] Satellite imagery Table 22 lists all satellite images (10 SAR images and 2 multispectral images) acquired throughout the experiment. The SAR images were systematically acquired around 18:00 local time (descent path), so each image may capture harvesting that took place on the same day. Because UK-DMC-3 is on a different type of orbit, the local time for the multispectral images was around noon. Therefore, they may only capture a portion of the harvesting activity that took place on that acquisition day. [Table 22]

[0216] SAR image - CosmoSkymed CosmoSkymed is a constellation of four identical satellites equipped with X-band SAR (wavelength 3.1 cm). The imaging mode selected for the experiment provides images in a single polarization (VV or HH) with a spatial resolution of 2.2 m and a coverage of approximately 30 × 40 km. All images were acquired with the same observation geometry (incidence angle 57 degrees). The SAR measurements (backscatter coefficient, σ) between images with different incidence angles were compared. 0 ) fluctuations should be avoided.

[0217] Multispectral Imagery - UK-DMC3 UK-DMC3 is a constellation of three identical satellites carrying multispectral radiometers. The images have four multispectral bands (blue, green, red, and near-infrared) with a resolution of 3.2 m, and a panchromatic band with a resolution of 0.8 m, for a coverage of 10 x 10 km. Each UK-DMC3 image corresponds to the most cloud-free image available within a one-month observation horizon. As a constellation, UK-DMC3 may occasionally acquire daily images across the region of interest, but in practice, frequent cloud cover severely limits the opportunities for clean image acquisition.

[0218] Field survey Ground information was collected on the day of each SAR image to understand and validate the information derived from the satellite imagery. Note that ground information was not collected for the UK-DMC3 imagery, as the exact date of acquisition could not be known in advance (multiple attempts were made to collect cloud-free imagery within the one-month observation window).

[0219] On the day of each SAR image, a local agronomist took photographs of all fields harvested that day and recorded them on a field sheet. For each harvested field, six photographs were taken according to Table 23. The field sheet included a map of the tea plantation to report (i) the location where the photograph was taken, (ii) what fraction of the field was harvested, (iii) the harvesting method and typical number of leaves harvested, (iv) the field boundary error used for analysis, and (v) meteorological observations. [Table 23]

[0220] Ancillary Data Other data of interest included monthly harvest reports (number of leaves harvested daily for each field, harvesting method), and meteorological data (rainfall) from an automated weather station.

[0221] Figure 38 shows the timeline with the most important data. The no-harvest period between July 12 and July 24 corresponds to a plucker strike. To limit the number of images during this inactive period, one SAR acquisition was intentionally postponed during the strike. This explains the relatively coarse time sampling at the beginning of the experiment. Overall, SAR images were acquired every 4–8 days.

[0222] Satellite image preprocessing CosmoSkymed images: For better control of image processing, CSK images were acquired in raw format (Single Look Complex - SLC), so the images were (i) calibrated (digital numbers to calculate the backscatter coefficient σ 0(ii) coregistration had to be performed (realigning all images to a common grid), and (iii) terrain correction and ground projection were performed (correcting for geometric artifacts due to the terrain). These steps were performed with the Sentinel Application Platform (SNAP v4.0), a free software capable of processing a wide range of satellite imagery.

[0223] UK-DMC3 images: UK-DMC3 images are already orthorectified. The only pre-processing step is to convert the image from digital numbers to reflectance σ and then calculate NDVI (NDVI = (ρ NIR -ρ Red ) / (ρ NIR +ρ Red The rest of the image processing and analysis was done in the programming language Python.

[0224] Harvesting method detection On satellite imagery, mechanical harvesting can be identified from the furrow patterns left by machines dragged over the tea canopy. A method implemented for automatic detection of harvesting methods using SAR imagery is described below. In multispectral imagery, the harvesting method is only identified visually, but a similar method can be implemented for automatic detection.

[0225] Figure 40 summarizes the main steps of automated detection in SAR images. The technique consists of detecting mechanically harvested furrows as a sinusoidal pattern with a spatial period of approximately 24 m (Figure 39). This detection was performed in the frequency domain by taking the 2D Fourier transform of a small image clip (101 × 101 pixels) moving across the SAR image. After the Fourier transform, the sinusoidal component appeared as a peak at a spatial frequency equivalent to the 24 m period. The moving image chip samples a given field multiple times, and the detection algorithm can return conflicting outputs when the furrow pattern is not clear. This is why the final decision on whether a field is mechanically harvested is obtained by applying a majority filter to each field. This filter simply returns the most frequent output of the automated detection.

[0226] Detection of harvested areas from SAR coherence Between two SAR image dates, the harvested area can be detected from the interferometric coherence γ. Coherence is the magnitude of the complex correlation coefficient between two image chips (9 × 9 pixels) from two SAR images acquired using the same observation geometry (Preiss & Stacy, 2006).

[0227] Coherence is a dimensionless quantity between 0 and 1 that quantifies the amount of random relocation of scatterers between two image chips (Figures 41 and 42). Coherence tends to decrease over time due to temporal decorrelation. Overall, when there is a period of several days between two images, bare loam and urban areas are characterized by high coherence values ​​(0.7-0.8), while vegetated areas have low coherence values ​​(0.15-0.3) due to wind-induced vegetation disturbances.

[0228] Still, it is possible to distinguish unharvested tea areas, with a coherence close to 0.3, from harvested tea areas, with a coherence close to 0.15. By walking across the entire tea field, pickers disturb the canopy surface more than natural perturbations (e.g., wind). Finally, for each tea field, harvested area detection is based on an averaging filter (to remove noisy fluctuations in coherence) followed by a simple threshold filter to determine whether a given pixel is harvested or not (Figure 43). The threshold is 0.24, which was manually adjusted until the detection output matched the harvested area reported in the field sheet. Pixels with a coherence less than 0.24 are detected as harvested.

[0229] Normalization of SAR measurements Following the detection of harvested and unharvested areas using coherence, the backscatter coefficient σ of the harvested area was used to evaluate the sensitivity of SAR to the canopy surface condition. 0 but σ in the unharvested area 0 To ensure that the observed variations are related only to the harvest conditions, σ 0 should first be normalized for variations in local incidence angle and water content, two parameters that affect SAR measurements.

[0230] This normalization technique takes advantage of the spatial and temporal characteristics of local incidence angle and moisture (Figure 44). First, moisture can be reasonably assumed to be uniform across a given field, so it only varies over time and from one image to another. Therefore, for each image, the normalization to moisture is calculated by the σ 0 The value of σ across the field 0The local angle of incidence at a given pixel is then the same for all images since they are all acquired using the same observation geometry. Therefore, for each pixel, normalization for the local angle of incidence can be done by dividing the pixel value (already normalized for moisture) by the average value for this pixel over the time series of images. For simplicity, the final normalized quantity is called the backscatter coefficient.

[0231] After normalization, the backscatter coefficient of the unharvested area is plotted against the backscatter coefficient of the harvested area (Figure 45). The harvested area is obtained from a coherence analysis of the SAR image pair. Our hypothesis is that the backscatter value of the harvested area should be lower than the backscatter value of the unharvested area. This difference should be greatest when the harvesting occurred just before the second image in the image pair used to calculate the coherence, as new shoots have not had time to grow.

[0232] Therefore, when harvesting occurs shortly after the date of the first image (green dots) (Figure 46), there will be enough time for new shoots to grow until the date of the second image, when backscattering is measured, so the difference in backscattering between harvested and unharvested should be small (green dots on the y=x diagonal) (Figure 47). In comparison, when harvesting occurs just before the date of the second image, the difference in backscattering between harvested and unharvested should be greatest (black dots away from the y=x diagonal) (Figure 47).

[0233] Harvesting method detection The pattern left by mechanical harvesting is relatively clear in both the SAR image (Figure 48) and the multispectral image (Figure 49). During harvesting, the machine is dragged across the canopy surface, trimming the leaves in a dominant direction. Adjacent bands result from the machine trimming the canopy in opposite directions. Images, whether they are terrestrial photographs, multispectral images, or SAR images, capture reflected electromagnetic radiation, which depends on the overall orientation of the leaves. The harvesting machine is about 1.5 m wide, but the band on the satellite image is about 12 m wide. This suggests that one band corresponds to eight adjacent passes of the machine in the same direction. In reality, the machine is likely operated as shown in Figure 50. With time and shoot regrowth, this trimming effect diminishes, and the furrow pattern becomes less visible on the satellite image (Figure 49).

[0234] Automated furrow detection was tested on the first five SAR images. Accuracy assessment (Table 24) indicates that 67% and 58% of machine-harvested and hand-harvested fields were correctly detected. Because furrow patterns decay with shoot growth, better detection can be obtained by including more images. Nevertheless, visual inspection of all SAR images suggests that some MH fields never exhibit furrow patterns, even when harvesting occurred on the day of SAR acquisition. Bidirectional scattering / reflection effects can be complex, and these hidden furrows may have become more visible if the images had been acquired using a different observation geometry. The effect of furrows may also be masked by other parameters that vary across the field, such as the local angle of incidence if the field is in highly undulating terrain. [Table 24]

[0235] The data used as ground truth may be inaccurate. Harvesting methods were extracted from harvest reports for January, February, and March 2016, which have identical harvesting methods for each field. These data were updated along with ground surveys (field sheets and field photographs). However, because not all fields were visited, the harvesting methods in the ground truth dataset may be outdated for some fields. Overall, 41 of 91 fields were surveyed. Of the surveyed fields, 8 fields (20%) had different harvesting methods from the January / February / March reports. Applying the same percentage to the 50 unsurveyed fields suggests that the harvesting methods for approximately 10 fields may still be incorrect. This is acceptable for an initial accuracy assessment, but the ground truth should be updated before attempting further refinement of the detection method. Harvesting methods can also be detected from multispectral imagery, as long as cloud-free imagery is available.

[0236] Detection of harvested areas from SAR coherence Figure 51 shows how loss of coherence for a given field (field 32) can relate to harvested area using a field sheet. The SAR coherence image (left) shows the level of coherence measured between SAR detections made on July 30 and August 3 for field 32. Darker areas represent areas of relatively low coherence (approximately 0.15), while lighter areas represent areas of relatively high coherence (approximately 0.30). A snapshot of the field sheet for field 32 on August 3 (right) shows shaded areas representing harvested areas and unshaded areas representing unharvested areas. There is a very high degree of correlation between areas of high / low coherence and harvested / unharvested areas.

[0237] The main advantage of coherence analysis is that it captures all harvests that took place between two dates, as opposed to analyzing pixel values ​​from a single image, which may only indicate which fields were harvested near the image date. However, the drawback of coherence is that anything that alters the alignment of the canopy surface (wind, rain, animals) leads to a loss of coherence. Thus, there is a possibility of false positives, especially when the temporal separation between the two images is long. Temporal decorrelation gradually reduces coherence across any vegetated area.

[0238] Accuracy Assessment Ideally, accuracy assessment would compare the detected harvested area with the true harvested area. Unfortunately, the harvest report only provides the weight of harvested leaves. As an initial assessment, the harvested area was ignored, and it was simply verified that the field report confirms that harvesting occurred when a field is detected as harvested between two dates. Table 25 shows that the highest detection accuracy reached 85% (pair number 7), where the two images were only four days apart and there was no significant rainfall. In comparison, pair number 4 had an accuracy of only 27%, as heavy rain led to a loss of coherence across the field. Overall, detection accuracy was found to be relatively dependent on weather conditions. [Table 25]

[0239] Yield estimation In this section, the feasibility of estimating yield from harvested area detection was investigated. The analysis was limited to pairs 5, 6, 7, and 8, which had the highest detection accuracy. False positives (harvest detected but not reported as harvested) were also removed, as were false negatives (harvest not detected but reported as harvested). Thus, the analysis focused on the correlation between yield and detected harvested area, ignoring possible errors from harvested area detection. Figure 52 shows that there was a reasonable correlation between yield and harvested area, and that this relationship differed between machine-harvested tea (Figure 52) and hand-picked tea (Figure 53). Regarding tea varieties, seedling and clonal teas had similar yields per harvested surface area, but only machine-harvested Chinese tea had a higher yield.

[0240] The data are widely scattered around the regression line, so a linear model cannot be used to estimate short-term yields (Table 26). However, for the entire period at hand (August 15–September 8, 24 days), the error in total yield is only 8% (Table 27). When yields are estimated over a long enough period, overestimation and underestimation compensate each other.

[0241] The large yield difference between mechanized and hand-harvested crops highlights the need for accurate knowledge of the harvesting method for each field. In summary, these results suggest that monthly yields for the entire farm can be reasonably estimated from estimates of the area harvested. [Table 26] [Table 27]

[0242] Monitoring shoot growth The sensitivity of remotely sensed measurements to canopy condition was verified to enable shoot growth to be monitored.

[0243] CosmoSkymed Images. In this section, the output of the coherence analysis (harvested and unharvested areas for each field) was used to compare the backscatter coefficient of the harvested portion of a given field with the backscatter coefficient of the unharvested portion, following the methodology shown in Figure 45. Image pairs 1 and 4 were excluded from the analysis due to inaccurate coherence analysis.

[0244] Figure 54 shows that using VV polarization imagery, there is no significant difference between harvested and unharvested backscatter. This suggests that VV polarization is insensitive to canopy surface conditions. In comparison, Figure 55 shows that using HH polarization imagery, backscatter in harvested areas tends to be smaller than backscatter in unharvested areas for at least 70% of the points. Note that because the analysis is based on detecting harvested areas using SAR coherence, which is prone to false positives, perfect segmentation cannot be expected.

[0245] Nevertheless, even for the few cases where the detected harvested area is accurate (validated against the field sheet, see e.g., Figure 55), the difference between harvested and unharvested backscatter is less than 0.5 dB, which is of the order of magnitude of the noise level in SAR images. In summary, HH polarized backscatter is sensitive to the canopy surface conditions, but this sensitivity is too weak to monitor shoot growth.

[0246] The observed difference between HH and VV polarization can be explained by considering leaf orientation. In HH polarization (Figure 57), waves interact preferentially with horizontally oriented leaves / stems, while in VV polarization (Figure 56), waves interact primarily with vertically oriented leaves / stems. Because leaves from new shoots open in the horizontal plane (to capture more sunlight), only HH-polarized waves are scattered by new shoots (backscattering increases with shoot growth). In comparison, young shoots pass VV-polarized waves, which are scattered randomly in a dense, mature canopy with randomly oriented leaves.

[0247] UK-DMC3 images. Frequent cloud cover severely limited the acquisition of cloud-free images. Nevertheless, two UK-DMC3 images (August 16 to September 10) show within-field variations in NDVI that are undoubtedly related to harvest. Analysis was limited to visual interpretation of the few cloud-free fields in the two available images.

[0248] Figure 58 shows the NDVI variation between two image dates for four different fields, along with a timeline of the amount of harvested leaves for these fields (extracted from harvest reports). In the first image in Figure 58, the top half of field 29 was recently harvested, while the bottom half will not be harvested until the next two days. The NDVI levels in the harvested half are clearly lower than those in the unharvested half. In the second image in Figure 58, both leaves have had time to regrow, leading to uniformly high NDVI levels. Analysis of fields 30 (Figure 59) and 33 (Figure 60) is less clear, but the NDVI clearly indicates that these fields were harvested in sections at different times. Some sections have increased NDVI between the two dates, while other sections have decreased NDVI. Field 32 (Figure 60) is the only field on the farm that was not harvested between the two satellite acquisitions (it was last harvested on August 12). Regardless of the initial level, NDVI increases between the two dates. In summary, shoot yield and growth are captured by NDVI through fluctuations between about 0.7 and about 0.8. From these observations, time-series NDVI images can be used to monitor shoot growth throughout the year.

[0249] The range of NDVI values ​​will differ for each field. This naturally stems from differences in the overall density of the tea canopy. For example, a recently pruned tea field would be expected to have a less dense canopy (lower overall NDVI) than older tea fields. For the same reason, differences in NDVI may also be related to tea variety.

[0250] Figure 62 attempts to reveal the connection between the average field NDVI and data from harvest reports (yield, pruning year, tea type, and harvest method). For a given tea classification (point color), NDVI tends to increase with pruning year. Chinese teas tend to have higher average NDVI and yield than seedling and clonal teas. There is no clear connection between average NDVI and yield. Chinese teas had higher yields than clonal teas, which had higher yields than seedling teas. This is related to harvest method (Chinese teas are all MH, clonal teas are mainly MH, and seedling teas are mainly HP).

[0251] X-band SAR remote sensing Analysis of CosmoSkymed images showed that: 1. X-band SAR can detect harvesting methods from the patterns of furrows left by mechanical harvesting. 2. X-band SAR can provide an estimate of the harvested area between two image dates that are several days apart, using the loss of coherence caused by harvesting. A linear model between the detected harvested area and the corresponding yield could provide a reliable monthly yield for the entire farm. 3. HH polarized backscattering decreased after harvest, but VV polarized backscattering did not change significantly. The difference in behavior between HH and VV may be related to the horizontal orientation of harvested young leaves. However, the sensitivity of backscattering to harvest is too weak to monitor shoot growth.

[0252] Multispectral Remote Sensing Multispectral remote sensing has capabilities similar to SAR, but is operationally severely limited by cloud cover. 1. Harvesting methods can be detected from multispectral and panchromatic bands. 2. NDVI shows good sensitivity to canopy condition and can be used to monitor shoot growth. 3. NDVI can also be used to detect harvested areas, at least for areas that were harvested immediately prior to image acquisition.

[0253] GIS data development and automatic weather updates Tea farm digital interface (interactive map) A digital representation was created for the Kapkorech farm, showing the field boundaries and incorporating existing data about the field (yield, soils, etc.) into the Cuppa Tea model output.

[0254] Digitization of field boundaries Using paper maps, satellite images in ArcGIS, and aerial photographs above Kapkorech, field boundaries were digitized and the following tea areas were digitized: [Table 28]

[0255] Many of these paper maps had no field identifiers or contained outdated field identifiers, so they were corrected by adding previously digitized line drawings containing Kapkorech. This correction process was repeated to refine the field boundaries of the new tea fields in each district.

[0256] Extending the model data to all Kericho tea areas To provide a repeatable framework for creating the GIS data required by the Cuppa tea model and tools, a set of GIS models was created in ArcGIS Model Builder to combine and add tea plantation data. The first step was to combine the tea areas into a single shapefile.

[0257] The next step was to incorporate the field data held in an Excel spreadsheet. To uniquely identify each field within every tea zone, a key field was created based on the farm name and field ID, added to an Excel spreadsheet, and then exported as a dbf file. The dbf file was then merged with the newly created Kenyan tea shapefile to create a shapefile containing all UTK data at the field level. This step allowed the data to be visualized in a GIS tool.

[0258] The final step was to generate the data required by the CUPPA Tea model. The model data was calculated as follows: Weather Stations - Data from 14 weather stations were collated. In Cheboswa, Chemogo, and Kimoi, temperature data was incomplete and was copied from the nearest weather station. Each tea field was assigned a weather station based on the nearest station. Elevation – taken from the Aster V2 DEM and retrieved from the Online Data Pool (NASA LP DAAC, 2015). The median elevation across each field was taken from the DEM using ArcGIS Statistics. Soil - soil depth was taken from the "Soil property maps of Africa at 250m" (Hengl et al., 2015), giving an average depth of 175cm across Kapgwen and Kapkorech. Therefore, the soil profile for each field was set to medium (150cm). The default soil type was also taken from the Soil map of Africa and set to "sandy clay loam". Data from a previous project for Kapkorech and soil sample data were used to correct for specific fields. ·Harvesting interval - Based on previous projects, the harvesting interval was set at 17 / 18. Manual harvest - In this example, all fields were set to manual harvest. Updates to the model now allow fields to be set to the mechanical harvest algorithm as well. · Irrigation - Based on current knowledge, all fields were set to non-irrigation. Harvest Date - Since not all fields are harvested on the same day, the harvest date is used in conjunction with the harvest interval to vary the day harvesting starts. This is set to 1 by default. Tea Type - Tea type was set as either Seedling Tea, China Tea, or 6 / 8 based on data in a 2014 Excel spreadsheet.

[0259] Extending Cuppa Tea's tools to represent field-level data on GIS tools The updated CUPPA-Tea tool discussed above has been updated to include a GIS interface that allows for the integration of field-level data. The mapped data is based on previous data provided on an Excel spreadsheet and combined with digitized field boundaries.

[0260] The tool uses ArcGis Runtime to display GIS data. The data to be visualized was loaded as a layer in an ArcMap document based on an augmented field boundary shapefile and then exported as a runtime geodatabase for use in the CUPPA-Tea tool. Through the CUPPA-Tea tool, the following field data can be viewed: 1. Altitude 2. Soil type 3. Annual yield (1985-2013) (kg mt / ha) 4. Average yield in 2016 (kg mt / ha) 5. Number of plants per hectare 6. Year of pruning as of 2014 7. Year of vegetation 8. Types of Tea 9. Field Boundaries

[0261] As the tea zones were organised by factory, the location of each factory was added to the map along with the location of the weather station. You can now see the current month and next month along with other field data.

[0262] Figures 64 and 65 show the predicted monthly and yearly yields. The images are initially displayed as a splash screen, as loading the yield values ​​can delay the display of the tool.

[0263] The map can be zoomed in and individual fields can be selected. Once a field is selected, the main model data is shown at the top of the window and the full field level data is displayed via a pop-up window (Figure 66).

[0264] The Simulation tab allows you to select / change the parameters of the CUPPA-Tea model and run the model.

[0265] Updates weather to generate monthly yield forecasts In order for the CUPPATea model to perform yield prediction more effectively, · Integration of the CUPPATea model with current weather and weather forecasts has been carried out. Using summarized weather or new weather forecast data from 2014-2016, a comparison of monthly forecasts for 2016 was undertaken and forecasts for 2017 were generated.

[0266] 1. Download weather data from an automated weather station, 2. Update CUPPA-Tea weather data based on current weather data and downloaded automatic weather data, 3. A further tool was developed that runs the CUPPA-Tea model on all fields and produces monthly yield prediction csv files.

[0267] Download automatic weather data A new automated weather station was prepared for testing, which automatically sends the latest weather updates every five minutes.

[0268] Once created, weather station data was available every 15-30 minutes.

[0269] A sample python script was created to test the automatic download of daily summary data and solar radiation. This was converted to c# and included in the new Update tool. The new Update tool lists all weather stations in a csv file. The last update date is obtained from the last line of the station csv file named xxxxyyyy.csv, where xxxx is the four-character station id and yyyy is the year. If the date is prior to today, the station is highlighted. When "update" is pressed, the tool will attempt to download data for each station and append it to the station csv file.

[0270] Cuppa Tea Weather Data Update The CUPPA-Tea model looks for weather files named xxxxyyyy.wth, where xxxx is the four-character weather station id and yyyy is the year.

[0271] The intent of the tool was to demonstrate how these CUPPA-Tea weather files can be updated by automatic weather station data, downloaded by the tool, or manually collected data via csv files placed in specific folders.

[0272] When the tool starts, it reads the date in the last line of each automatic weather station's csv file, and the earliest date among all automatic weather stations, whether manual or not, is the date for which the weather data for that station will be copied into all CUPPA-Tea weather files.

[0273] For both automated and manual weather data to be copied into CUPPA-Tea, the weather files must be in a csv named xxxxyyyy.csv, where xxxx is the four-character weather station id and yyyy is the year, reflecting the name of the cuppa weather file. For automated weather stations, these files are created by the tool when data is downloaded from the web. For manual weather stations, these files are created according to the following format: 1. Date 2. Solar radiation (MJ / m2) 3. Maximum temperature (℃) 4. Minimum temperature (℃) 5. Rainfall (mm)

[0274] Creating a monthly forecast Once the weather data was updated, the CUPPA-Tea model was run for each field in the Kenya Shapefile using the model data discussed above in the Shapefile for each field.

[0275] The weather year for each weather station was calculated so that a common set of years was used for each field. This means that when a new automated weather is acquired, a set of historical data based on surrounding weather stations should be created to allow the model to run for at least two years.

[0276] In addition, the following model parameter values ​​were used: ·Plucking = "2-3 leaves" Start of year = 1 st Jan Break Back = true Break back leaves = 1 Shoot Size = 0-100cm Zero all shoots = true

[0277] The model first looks for harvest date information by field in a simple csv file called "Harvestyyyy.csv", where yyyy is the year. The model fills in blanks if the entire year is missing from a date, but otherwise considers any blanks to be intentional.

[0278] As the model is run, predicted yields for each field in kg mt / ha are accumulated into monthly values ​​for each year the model is run and saved in a csv file called "Yieldyyyy.csv", where yyyy is the year.

[0279] Corrected harvest interval calculation Rationale Improving the decision support tool using CUPPA-Tea required understanding why the tool predicted an unexpectedly large number of fields ready for harvest.

[0280] Several model runs were generated for fields 1 and 14 from January 2016 to December 31, 2018, and then projected forward to February 1, 2019 (Figure 67; Table 28).

[0281] The default quality index (QI) for harvested leaves is based on the number of shoots. QI=(1grow+1banj+2grow) / all shoots It was defined as follows: [Table 29]

[0282] Where do mature shoots come from? CUPPA-Tea models 500 shoots. It produced outputs for days 823, 824, 825, and 826. Individual shoot data were collated for April 3 (day 823) (Table 29), April 4 (day 824) (Table 30), and April 5 (day 825) (Table 31). Harvest occurred on April 4. [Table 30]

[0283] The algorithm for calculating the harvested shoots is as follows: If hand plucked: If (leaf count > 0) and (leaf count > pluck to leaf) and (observed length > shear height) and (base height + base position of previous leaf >= table height): If 'leaf count' > 0, AND ·If 'leaf count' > 'pluck to leaf' AND ·If 'observed length' > 'shear height' AND ·'base height' + 'base position of previous leaf' >= 'table height' where 'pluck to leaf'=1, 'shear height'=5cm, and 'table height'=75cm.

[0284] On day 823 (Table 29), shoots 6 and 7 are harvestable by the above criteria. However, they are not harvested until the harvest date, which in this case was day 824 (April 4) (Table 30). On day 824, shoot 154 has eight leaves (Table 30), but is not harvested because its "observed length" is less than the "reaper height" of 5 cm. In contrast, on day 825 (Table 31), shoot 154 reaches an "observed length" of over 5 cm and is ready for harvest. However, it is not harvested until the next harvest date. [Table 31] [Table 32]

[0285] Effect of minimum harvestable shoot length on CUPPA-Tea introduction It was not possible to explain why the "5 cm shoot" with nine open leaves was not marked for harvest by the model. In the context of hand harvesting, the term "reaper length" does not refer to the length of the reaper, but to the minimum harvestable shoot length.

[0286] method The model was re-run for fields 1 and 14 with the following criteria: Field 1 is clone 225 and is mechanically harvested. The default system harvests shoots assuming a minimum harvestable shoot length (reaper height) of 5 cm. If hand plucked: If (leaf count > 0) and (leaf count > pluck to leaf) and (observed length > shear height) and (base height + base position of previous leaf >= table height): If 'leaf count' > 0, AND ·If 'leaf count' > 'pluck to leaf' AND ·If 'observed length' > 'shear height' AND ·'base height' + 'base position of previous leaf' >= 'table height' The model was run with the following five options: If 'pluck to leaf'=1, 'shear height'=5cm, and 'table height'=75cm If 'pluck to leaf'=1, 'shear height'=4cm, and 'table height'=75cm If 'pluck to leaf'=1, 'shear height'=3cm, and 'table height'=75cm If 'pluck to leaf'=1, 'shear height'=2cm, and 'table height'=75cm If 'pluck to leaf'=1, 'shear height'=1cm, and 'table height'=75cm If 'pluck to leaf'=1, 'shear height'=0cm, and 'table height'=75cm

[0287] result The output of the analysis is shown in Figures 68 to 73. 1) The effect of lowering the "minimum harvestable shoot length" (reaper height) parameter is to increase the number of shoots considered harvestable, specifically the number of shoots with one leaf (1+banjhi bud and 1+growing bud). 2) Lowering the "minimum harvested shoot length" substantially increased the quality index, and quality measures became more predictable. 3) Note that the quality score of the harvested leaves on the day of harvest is lower than the day before harvest because shoots with less than two leaves are not harvested in the model. Theory of harvest intervals There are various methods for calculating the appropriate harvest interval for tea.

[0288] Determining harvest intervals from phyllochrons (Burgess & Carr, 1998) described using the rate of leaf opening (or phyllochron), which varies with temperature and drought stress levels, as a way to determine harvest intervals. If tea is harvested so that all shoots with two or more leaves are removed, leaving only shoots with one open leaf, waiting three phyllochrons means that all shoots have fewer than four leaves. Maximum HI to have shoots < 4 leaves = Harvest 1 leaf + 3 Phyllochrons

[0289] The average temperature in Kericho is about 18°C. (Burgess & Curr, 1998) reported that the leaf emergence rate (LAR) of clone 6 / 8 was 0.171-1.6*0.817. Tmean Thus, a temperature of 24°C would result in 0.159 phyllochrons per day, which means a phyllochron takes 6.2 days. A temperature of 18°C ​​would suggest a leaf emergence rate of 7.7 days.

[0290] CUPPA-Tea leaf opening rate The CUPPA-Tea model also predicts the rate of leaf opening. The model assumes that leaf opening occurs from the first true leaf; that is, leaf opening marks the end of the bud stage. The CUPPA-Tea model predicts the rate of leaf opening as a function of temperature.

[0291] (Matthews & Stephens, 1998) reported a basal temperature for growth of clone 6 / 8 of 8°C, an optimum temperature of 24°C, and a critical temperature of 40°C (Figure 74).

[0292] (Matthews & Stephens, 1998) explains that phyllochron is equivalent to 4.52 fenoclones. Therefore, at a temperature of 24°C, CUPPA-Tea assumes that a shoot can unfold a leaf every 4.5 days. At a temperature of 18°C, phyllochron = 0.221 * 18 / 24 = 0.166, which suggests that it takes 6 days for a leaf to unfold. Note that this leaf emergence rate is approximately 78% of the value reported by (Burgess & Carr, 1998).

[0293] Theory of harvest interval prediction using CUPPA-Tea When the minimum shoot length is set to 0 cm, the default setting in the CUPPA-Tea model assumes that all shoots with more than one open leaf are harvested. Therefore, the maximum developmental stage of an individual shoot at harvest is one day before the shoot opens its second leaf.

[0294] Using the above information, we can predict when harvestable shoots will reach a quality index of 0.5. Consider a theoretical example in which all shoots with one or more open leaves are harvested on day 0. Assume that the shoots are perfectly evenly developed and each shoot progresses 0.166 phyllochrons each day. Thus, the first shoot (Shoot 1) will open its first leaf on day 1, its second leaf on day 7, and its third leaf on day 13.

[0295] The default quality index (QI) is based on the number of shoots and is defined as: QI=(1grow+1banj+2grow) / all shoots

[0296] Assuming that all shoots with two leaves are growing shoots, day 13 is the first day that the quality index is less than 1.00.

[0297] If n<=2LAR, QI=100% If n>2LAR, then QI=2LAR / n It can be demonstrated that:

[0298] When the quality threshold QI = 0.5, 0.5 = 2 LAR / n, so 0.5n = 2 LAR, and therefore n = 4 LAR. That is, when the harvest interval is 4 phyllochrons, the quality threshold is 50%.

[0299] In practice, CUPPA-Tea is set up so that shoots with only one leaf are not harvestable. In this case, the QI value will be 0.5 at an earlier date. Interval to reach QI=0.5=4LAR-LAR=3LAR. Assuming LAR=6 days, the harvest interval will be 18 days.

[0300] Impact of predicted harvest interval / assumed minimum harvested shoot length in field 1 Predicted quality indices and shoot composition were examined for field 1 after harvest on December 11, 2018, for various assumptions regarding minimum shoot length.

[0301] The average temperature was about 19°C, so the phyllochron was 19 / 24*=0.174. Therefore, the modeled leaf emergence rate in CUPPA-Tea was about 5.75 days.

[0302] Figures 75 to 80 show some of the important features of the reaction.

[0303] Assuming a minimum shoot length of 0 cm (Figure 75), the quality index does not begin to decline until the first phyllochron is completed around day 7, similar to the theoretical example above.

[0304] The first two leaves and buds appeared on day 1, the first three leaves and buds on day 7, and the first four leaves and buds on day 13. This is similar to the theoretical example above.

[0305] Assuming there is no minimum size for harvestable shoots and that all shoots with two leaves are harvested (Figure 75), on the day of harvest, 100% of the remaining shoots have one open leaf. On day 18, when the QI theoretically falls below 0.5, the percentage of shoots with one leaf (39.5%) and the percentage of developing shoots with two leaves (20.5%) is higher than the percentage of banjhi shoots (8%), shoots with three leaves (22%), and shoots with four leaves (10%). Therefore, the QI remains at 60%.

[0306] We wanted to understand why the number of shoots with one leaf (39.5%) was higher than the number of shoots with two leaves (28.5%) and the number of shoots with three leaves (22%). The model assumes that shoots do not become banjhi and immediately develop more leaves. Although some banjhi shoots eventually emerge from dormancy, the majority of banjhi shoots remain dormant. Thus, quality declines more slowly than expected. The impact of banjhi shoots is to extend the predicted period with a QI < 0.5 from 18 days to approximately 24 days (+33%) (Table 32). [Table 33]

[0307] Should banjhi shoots with two open leaves be acceptable? Managers rarely know whether a sprout is a banjhi or growing when the second leaf opens. Therefore, some argue that a more consistent quality index is ((1banj + 2banj + 1grow + 2grow) / all). Assuming a minimum harvestable shoot length of 0 cm, including banjhi shoots with two leaves increases the predicted harvest interval. However, if the minimum harvestable shoot length is set to the default of 5 cm, the results are very similar to the current values ​​of the quality index (Table 33).

[0308] The current setup of CUPPA-Tea assumes a minimum harvestable shoot length of 5 cm and that shoots should have more than one open leaf. Assuming a minimum harvestable shoot length of 5 cm means that most shoots with only one leaf are unharvestable (Figure 80). In this example, the absence of shoots with more than one open leaf means that the quality index drops sharply, reaching 0.5 after about 12 days, or two phyllochrons (Table 33). This raises the question of why a quality measure of a shoot with one open leaf is used when the model assumes that shoots with one open leaf are not harvested. An alternative measure could be (2grow + 2banj) / (2grow + 2banj + 3grow + 3banj). This analysis predicts that a longer period of 16 days, rather than 12 days, is required for shoots with three leaves to outnumber shoots with two leaves (Table 33).

[0309] Effect of time since pruning: Current projected harvest intervals in Kericho range from about 8-10 days in the first year, to 14 days in the second year, and 18 days in the fourth year. [Table 34]

[0310] Impact of predicted harvest interval and assumed minimum harvested shoot length in field 14 The effect of various assumed minimum harvest lengths on field 14 is shown in Figures 81 to 86. The minimum shoot length is 0 cm in Figure 81, 1 cm in Figure 82, 2 cm in Figure 83, 3 cm in Figure 84, 4 cm in Figure 85, and 5 cm in Figure 86.

[0311] Predicted quality index and shoot composition for field 14 after harvest on November 24, 2018. Since the average temperature was approximately 19°C, the phyllochron was 19 / 24*=0.174, and therefore the leaf emergence rate was approximately 5.75 days. Therefore, the harvest interval equivalent to 2 phyllochrons should be approximately 12 days.

[0312] In contrast to the results in field 1, the distribution of shoots that become harvestable in field 14 is more uneven. In fact, very few shoots are assumed to open one leaf between 1 and 37 days after harvest. The main reason for this is that there was no harvest between September 29 and November 24. New shoots are assumed to emerge only approximately 40 days after harvest, following the effect of harvest on releasing lateral buds from apical dominance.

[0313] The predicted harvest interval assuming a minimum harvestable shoot length of 0 cm is 21 days, which is about 16% greater than the theoretical value of 18 days based on 3 phyllochrons. Assuming a default minimum harvestable shoot length of 5 cm results in a predicted harvest interval of 12 days, which is equivalent to 2 phyllochrons (Table 34). [Table 35]

[0314] Initial conclusions and next steps Minimum harvestable shoot length is an important parameter in the model This analysis reveals the importance of the "minimum harvestable shoot length" (named "reaper height") in the CUPPA-Tea model. The current default setting of a minimum shoot length of 5 cm (compared to shorter lengths) results in more stable estimates of harvest intervals. The quality index falls below 0.5 more quickly assuming a minimum harvestable shoot length of 5 cm compared to shorter lengths. Should two banjhi shoots be acceptable on the quality scale? Managers rarely know whether a bud after the second leaf opens is a banjhi or developing. Therefore, it may be more consistent to claim that a banjhi shoot with two open leaves is an acceptable harvestable shoot. If the minimum harvestable shoot length is less than 5 cm, including 2 banjhi shoots extends the expected harvest interval. However, if the minimum harvestable shoot length is 5 cm, the impact of including 2 banjhi in the quality measure appears minimal because the majority of 2 banjhi shoots are less than 5 cm (Table 33 and Table 34).

[0315] Appropriateness of quality measures Currently, the CUPPA-Tea model is set up to harvest only shoots with more than one leaf, i.e., shoots with less than two leaves are excluded. This raises the question of whether a quality index (QI) that includes shoots with one leaf is appropriate.

number

[0316] If we include 2banj shoots as acceptable and assume that the number of shoots with one open leaf is the same as the number of shoots with more than three open leaves, the scale is one that is more consistent in terms of what is harvested. If Σ1banj+1grow=Σ4banj+4grow+≧5grow+≧5banj,

number

[0317] This quality measure appears to predict longer harvest intervals and appears to be less robust than current measures.

[0318] Calculating harvest intervals using the phyllochron calculation The current use of a quality index of 0.5 as an indicator of optimal harvest intervals is relatively unstable. Given that phyllochron abundance varies according to harvesting method, the question was raised as to whether it would be more reasonable to base predicted harvest intervals on a calculated number of phyllochrons (determined directly from temperature and drought stress measurements). Therefore, it was proposed to use phyllochron calculations to determine harvest intervals and CUPPA-Tea to calculate shoot composition and yield.

[0319] Philochron's Calculations Temperature effects (Burgess & Carr, 1998) reported that the average phyllochron for six clones was related to the mean temperature and potential soil moisture deficit (Figure 87, Table 35). The reciprocal of phyllochron is the leaf emergence rate (LAR). Equation 2 below describes the average response of the six clones.

number

[0320] Effects of drought Experimental measurements in Tanzania have shown that leaf emergence rates can slow down when soil moisture deficits exceed approximately 50 mm. The formula established from four clones of 3-4 year old plants is shown in the formula below and in Figure 88. Formula 3: f=1.218-0.161×1.00584SWD Implementing the code for Kericho Equations 2 and 3 were included in CUPPA-Tea to describe the accumulation of phyllochrons. / / Leaf appearance rate double dailyLAR = 0; if (weatherData.Tmean < 12) {dailyLAR = 0;} else {dailyLAR = 0.186 - 4.6 * Math.Pow(0.765, weatherData.Tmean);} double f = 0; if (waterData.Swdef < 52.15) {f = 1;} else {f = 1.218 - 0.161 * Math.Pow(1.00584, waterData.Swdef);} _LAR += (dailyLAR * f);

[0321] It should be noted that the leaf emergence rate described by (Burgess & Carr, 1998) differs from the current observed developmental stages within CUPPA-Tea. / / Observed dev stage double S2DRwf = Utility.Limit(0.364 + 0.636 * _wSFadj, 0, 1); _DVS += _s2DevRateI * (DDIdev * DLfac * S2DRwf);

[0322] The algorithm shows that leaf emergence rates are relatively stable for most years (Figure 89). However, the output reveals a step change (likely due to a change in temperature methodology from May 1, 2017) and some low readings due to erroneous temperature readings in November 2017. To remove these effects, the remaining analysis focused on 2018 (Figure 90).

[0323] Drought Impact. The CUPPA-Tea model calculates soil moisture deficit (Figure 91). The soil moisture deficit reduced the predicted leaf emergence rate for January 2018, but the overall impact was relatively small (Figure 92).

[0324] Using the values ​​in Figure 91, it is possible to accumulate the number of phylloclones between harvests. As shown in Figure 93, in Field 1 in 2018, the number of phylloclones reached 2 to 4 before harvest. It can be argued that this value should be consistent.

[0325] Field 1 has clone 225 which is mechanically harvested and the tea was pruned in 2017. Therefore, in 2018 it was the second year since pruning.

[0326] (Burgess & Carr, 1998) reported that the optimal harvest interval for hand-harvested tea is approximately 2 phyllochrons, which is consistent with the current practice in Kericho of harvesting at 14-day intervals (assuming a 7-day phyllochron). Mechanically harvested tea can be harvested at longer intervals because smaller shoots are harvested (Burgess et al., 2006). (Burgess et al., 2006) recommended that tea harvested with a reaper be harvested at intervals of 2.5 phyllochrons. Full mechanical harvesting removes even smaller leaves, so intervals of 3 phyllochrons may be possible.

[0327] The standard harvest intervals used in Kericho were as shown in Table 36. [Table 37]

[0328] Should harvest intervals increase with the number of years since pruning? Table 36 suggests that the standard practice in Kericho is to extend the harvest interval as the number of years since pruning increases. The reason for this may have more to do with labor productivity than with the physiological response of the tea plant. Under experimental conditions in Tanzania, it was successful in maintaining a common harvest interval for all tea plants in the second, third, fourth, and fifth years after pruning.

[0329] Conclusion: The use of phyllochrons provides a physiological and environmental basis for scheduling harvest intervals. Because temperatures are stable in Kericho, seasonal variation in phyllochrons is small. An interval of 2 phyllochrons is appropriate for hand-harvested tea, and reaper-harvested tea can be harvested at intervals of 2.5 phyllochrons. Machine-harvested tea can potentially be harvested at intervals of 3.0 phyllochrons. The reason for extending the harvest interval in the third and fourth years after pruning appears to be primarily the need to maintain labor productivity at later stages of the pruning cycle, rather than the physiology of tea growth.

[0330] The next step was to determine the average harvest interval in Kericho for different field types.

[0331] Follow-up experiments Under experimental conditions, the clonal field trials at Kericho were hand-harvested at consistent intervals of 10–11 days regardless of the year of pruning.

[0332] However, under commercial conditions, managers increased the harvest interval as the time since the last pruning increased. This appears to be primarily due to smaller shoots, rather than shoots that leaf out later. Therefore, there is a tendency to increase harvest intervals to maintain income for harvesters working hand-picked per-piece rates. Under mechanical harvesting conditions, increasing harvest intervals is also practiced, as smaller shoots increase the likelihood that the machine will harvest a higher proportion of coarse mother leaves.

[0333] When using hand-held harvesting with 25 mm steps, Kericho used the same spacing as described for hand harvesting in Table 36.

[0334] There are several ways to present recommended harvest intervals, either stepwise from year to year (Figure 95) or continuously (Figures 96 and 97).

[0335] The phyllochron multiplier can be expressed as a logarithmic or linear relationship. The goodness of fit for a linear relationship for machine-harvested tea is higher than for a logarithmic relationship, so a linear relationship is recommended. For simplicity, it has been argued that the relationship should also be linear for hand-harvested tea. Philochron multiplier (mechanical harvest) = 0.0006694 x number of days since pruning + 2.73 Philochron Multiplier (reaper harvest) = 0.00118 x number of days since pruning + 1.26 Philochron Multiplier (hand harvested) = 0.00118 x days since pruning + 1.26

[0336] (Burgess & Carr, 1998) reported the spread of leaves from the second to the third leaf. (Jayasinghe et al., 2014) has some indication that the phyllochron from the second to the third leaf and the phyllochron from the phyllochron to the first leaf are both 7 days, and that the time from the first to the second may be shorter at 5 days. The relationship with drought stress in (Burgess & Carr, 1998) is based on potential soil moisture deficit, but the model assumes it is based on actual soil moisture deficit.

[0337] Validating the use of phyllochron to predict harvest intervals in 2018 method We developed an algorithm to predict the planned harvest interval for each field in Kericho in 2018, determined using phyllochron and the following multipliers: Philochron multiplier (mechanical harvest) = 0.0006694 x number of days since pruning + 2.73 Philochron Multiplier (reaper harvest) = 0.00118 x number of days since pruning + 1.26 Philochron Multiplier (hand harvested) = 0.00118 x days since pruning + 1.26

[0338] These results were compared with the actual intervals. Predicted intervals were also calculated based on the current Quality Index Scale (QI_5cm), which is based on a minimum harvestable shoot height of 5 cm as calculated by CUPPA-Tea. Data for the Quality Index Scale (QI_3cm), which is based on a minimum harvestable shoot length of 3 cm, were also included. Due to calculation limitations, QI-based intervals that were greater than 32 days were reported as 32 days.

[0339] Effect of pruning on harvest interval: After tabulating the following results, we investigated the possibility that the next harvest interval would include a routine that would occur at an interval of 8 phyllochrons if pruning had been carried out. This can be demonstrated by taking the values ​​of the thermal hours of the phyllochron and the values ​​of the shoot renewal period (SRC) from the release of apical dominance to the opening of the third leaf, at the same base temperature. Clone BBK35 has a thermal time to phyllochron of 60°C d (degree days) above the base temperature of 9°C. BBK35 is 1 / SRC=0.0017T mean -0.015 (Burgess, 1992) The shoot regeneration cycle (SRC) in Kenya was 427°C / day, above a base temperature of 9°C.

[0340] Therefore, the number of phyllochrones per shoot renewal cycle is approximated as 427 / 60 = 7.11 phyllochrones. At a temperature of 18°C, a phyllochron is 7 days, so the above value should result in an interval of approximately 49 days to reach three leaves and a bud at the existing height. However, this analysis ignores the increase in height at harvest. If we assume that tea shoots also need to have additional leaves, this period should be further extended from 7 to 8 phyllochrones. The value of 8 is also attractive because it is recommended that four generations of shoots should be maintained in tea plants.

[0341] A note about harvest method: The harvest method for each field was initially determined from the first set of records in 2018. However, an updated field metadata set received in March 2020 indicates a different harvest method in 2018. While the predictions for hand-harvested leaves and for reaper-harvested leaves should be similar, assuming machine-harvested leaves are hand-harvested, and vice versa, introduces errors. Fields in the subsequent tables with differences are shown in bold.

[0342] Hand-harvested tea (several years before pruning) It was assumed that the phyllochron multiplier (hand harvested) = 0.00118 × number of days since pruning + 1.26. The hand harvested tea fields pruned in 2018 can be grouped according to the date of pruning (Figure 98).

[0343] For teas hand-harvested in the year before pruning, the predicted harvest interval of 19 days is broadly similar to that tea's actual harvest interval of 17 days (seeding field 35, Figure 99). It is interesting to note that the change in actual harvest interval between recently pruned and not recently pruned teas is minimal.

[0344] The results suggest that the actual harvest intervals fluctuated around the phyllochron-based predictions (Figure 99). The QI_5cm results produced some very low values. The QI_3cm results also showed some variation around the actual results. [Table 38]

[0345] Tea harvested by hand the year after pruning Originally, there were 12 fields shown as hand-picked in 2018; these were pruned in 2017. However, secondary records indicate that all but two fields (24 and 25) may have been pruned with a harvester in 2018 (Table 38). In fields 24 and 25 (which are seedling fields), the actual and predicted harvest intervals are very similar. Also, there was no apparent increase in the harvest interval in February 2018 (Figure 101). The results show a large variation in the quality index harvest intervals. [Table 39]

[0346] In other fields, which are often clonal, tea may have been reaper harvested and there is usually an extended harvest interval in February (Figure 101). However, after this peak, the actual and phyllochron harvest intervals become similar.

[0347] We investigated whether there was a relationship between location and longer intervals in February 2018. There is a spatial pattern to this variation, but it appears to be more related to clonal type than location per se.

[0348] Hand-harvested tea from the year of pruning The teas pruned in 2018 can be divided into three groups (Table 39). Teas pruned in February 2018 served primarily as year 1 fields. Teas pruned in May 2018 appeared to take a particularly long time to return to leaf after pruning, with an average maximum interval of 119 days. This indicates that including an 8-phyllochron interval in pruning events may work (see field 19A) (Figure 102). [Table 40]

[0349] Reaper Harvested Tea The predicted spacing of leaves in the reaper harvest was shorter than actual in the year following pruning and similar in the year before pruning (Table 40). Philochron Multiplier (reaper harvest) = 0.00118 x number of days since pruning + 1.26

[0350] Our original conclusion was that this relationship was acceptable for the last years after pruning (e.g., field 48A) but was an underestimate for reaper-harvested tea in the first year after pruning (e.g., field 1) (Table 40). However, updated harvesting method records suggest that this difference is the result of assuming an incorrect harvesting method. [Table 41]

[0351] Unpruned, machine-harvested tea Our original analysis identified eight fields where the actual interval for many fields was 18 days or less, but discussions with Kenyan tea farmers indicated that the minimum planned interval for machine-harvested tea was 18 days (Table 41). However, our subsequent analysis suggests that these fields were misassigned. Those fields shown as having short actual harvest intervals (except for field 91) are actually identified as hand-harvested fields. These fields include, for example, field 13. In contrast, the prediction for field 52 is consistent with the actual situation. [Table 42]

[0352] Pruned and machine-harvested tea The average spacing predicted using phyllochrons is broadly similar to that predicted for machine-harvested leaves in the year of pruning (Table 42). Including a spacing correction of 8 phyllochrons for the year of pruning can work very well (Figure 108). [Table 43]

[0353] Predicted harvest intervals based on phyllochron are much more stable than those achieved in practice (Figure 109). The correlation between predicted and actual harvest intervals for machine-harvested tea is close to a 1:1 line when forced through the origin. Before pruning correction, the initial phyllochron calculations did not address the long intervals immediately after pruning. The average predicted harvest interval per field using the phyllochron method is broadly similar to that predicted using a quality index based on a minimum harvestable shoot length of 3 cm (Figure 110).

[0354] conclusion The QI_3cm method in CUPPA-Tea may provide intervals similar to the actual and phyllochron measurements, suggesting that harvest quality is relatively stable.

[0355] The phyllochron method for scheduling harvest intervals results in a robust interval scheduling program based on tea's response to temperature and drought stress.

[0356] The presented phyllochron method assumes different algorithms for hand-harvested, reaper-harvested, and machine-harvested tea, but hand-harvested tea and reaper-harvested tea are assumed to be the same.

[0357] Using a multiplier with the phyllochron method takes into account the time since pruning. Comparing harvest intervals predicted using phyllochron with actual harvest intervals identified some outliers. However, subsequent analysis indicates that this may be explained by a misidentification of the harvest method. Encouragingly, using updated data on harvest methods from 2018 addresses the main outliers.

[0358] A routine was used that set the first harvest after pruning at 8 phyllochrons, which appears to work well as an initial approach for harvest interval scheduling.

[0359] The observation remains that the phyllochron method is often not sensitive enough to drought. There remains the possibility of recalibrating the experimental response to potential SWD to one based on actual SWD. However, the harvest intervals of some teas (especially seedling teas) lasted for a similar length to the drought period, possibly making some clones more sensitive. This may be one aspect of clonal variation that should be investigated in the future.

[0360] The changes to the CUPPA-Tea code are presented in Appendix D.

[0361] Modifying the model to simulate the effects of pruning Modifications to address yield after pruning The height of the picking surface increases during tea production. Under experimental conditions in Tanzania, the annual height increase is 5–10 cm for hand-harvested tea and 1–14 cm for reaper- and machine-harvested tea (Burgess, 1992). Under commercial conditions, the height increase is generally greater. (TRFK, 1986) reported an annual height increase of 20 cm for hand-harvested tea, reaching a height of 120–150 cm after four years. These height increases mean that it is advisable to prune the tea canopy back to a height between 45 and 75 cm, or an average of 60 cm, typically every four years.

[0362] (TRFK, 1986) recommends that chipping be done at a height of 100 mm above the pruning level. Thus, if pruning is done at 600 mm, picking should be done at 700 mm. Currently, CUPPA-Tea assumes a default constant picking height of 750 mm. If 700 mm is assumed as the initial value and increases by 150 mm per year, it will reach 1300 mm after 4 years.

[0363] Modifications to CUPPA-Tea to improve how the model deals with pruning are discussed herein.

[0364] Early yield and basal height after pruning The original CUPPA-Tea model was developed to describe shoot growth in unpruned tea plants, and one effect of this is that there is a long interval before the first harvest after pruning, resulting in a false peak in yield (Figure 111).

[0365] Update of change in minimum harvestable shoot length In an analysis of harvest interval data (Burgess et al., 2020), it was found that a "minimum harvestable shoot length" of 5 cm prevented shoots with more than four leaves from being harvested, which increased variability in quality calculations. Varying the minimum harvestable shoot length of 3 cm reduced the proportion of shoots with more than four leaves and improved quality index calculations. (Burgess et al., 2020) also showed that harvest intervals calculated using the "Quality Index QI_3cm method" in CUPPA-Tea predicted intervals similar to actual measurements and phyllochron measurements.

[0366] The effect of changing the minimum harvestable shoot length from 5 cm to 3 cm is shown for field 5 in Figure 111. The main effect is an increase in yield because more shoots are harvested. However, the peak yield in May 2018 after pruning in February 2018 is still evident.

[0367] Base height and platform height As observed in Figure 111, the default CUPPA-Tea model predicts a falsely large yield at the first chipping event after pruning. In the default model, the base height and basal height before and after pruning are about 75 cm. The "basal height" of the tea shoot is the height of the base of the tea shoot from the soil surface. The "base height" defines the distance from the ground to the harvest height.

[0368] Initially, the base height of the new shoot is assumed to be equal to the height of the platform (Figure 112). At some stage, most of the shoots will be ready for harvest.

[0369] The algorithm for calculating the number of harvested shoots from hand-harvested shoots is as follows: If 'leaf count' > 0, AND ·If 'leaf count' > 'pluck to leaf' AND ·If 'observed length' > 'shear height' AND ·'base height' + 'base position of top leaf' >= 'table height'

[0370] The original parameters in CUPPA-Tea were 'pluck to leaf' = 1, 'shear height' = 5 cm, and 'table height' = 75 cm. As explained above, 'shear height' is the minimum harvestable shoot length, which was reset to 3 cm in the modified CUPPA-Tea.

[0371] If the shoot is picked back to the first leaf, the base height of the shoot may increase after the first harvest. The exact point of picking is determined (l: cm from the base of the shoot, which is picked from just above the "fish leaf" or first leaf on the left side of the tea plant). New shoots will come from the axil of this leaf. Fresh weight is calculated from the shoot stage (leaf number), and dry weight is calculated from the fresh weight using DM%. If the shoot has more than three leaves, three leaves are picked and the rest are "broken back" to the fish leaf. [Table 44]

[0372] The effect of altering shoot base height after pruning After initial attention to altering shoot number, a proposed approach to addressing high yields after pruning was to reset the base height. Initial analysis showed that a 15cm change in base height, from 75cm to 60cm, looked promising.

[0373] The changes shown in Figures 113, 114, and 115 are: 1. On the day of pruning, change the average base height of all shoots to 600mm and change the average base height at pruning in the config file to 750mm. 2. The platform height should remain at 750mm for the first three harvests (chippings). The surface height at the chippings should remain constant at this time.

[0374] Resetting the base height eliminates the yield peak that would otherwise occur in the first harvest after pruning (Figures 113, 114, and 115).

[0375] Figure 113 shows the effect of including the new base height from pruning in field 5. This correction is responsible for the high yield and synchrony of growth immediately after pruning (a constant 650 shoots / m 2 (assuming that

[0376] Field 13 is a 13.7 ha field (planted in 1929) containing hand-picked seedling tea. The tea was pruned on 21 November 2018. In Figure 114, it is shown that for field 13, the inclusion of the new base height from pruning corrects for the higher yield on 22 February 2019 after pruning in November 2018 (a constant 650 shoots / m 2 (Assume that

[0377] Field 25 is a 7.6 ha seedling tea planted in 1952. It was pruned on August 28, 2017, hand-picked in 2018, and harvested with a reaper in 2019. In Figure 115, it is shown that for field 25, the inclusion of new base height from pruning reduces the high yield (constant 650 shoots / m) immediately after pruning on December 23, 2017. 2 (Assume that

[0378] What is the effect of different platform heights? A simple analysis was completed for field 5 to determine if the yield when the 70 cm stand height is reduced to 55 cm is similar to the yield when the 75 cm stand height is reduced to 60 cm. Analysis of the data, shown in Figure 116, suggests that the results are exactly the same. Figure 116 shows that the yield when the 75 cm stand height is assumed to be reduced to 60 cm is the same as the yield when the 70 cm stand height is assumed to be reduced to 55 cm.

[0379] Analysis of continuously declining yield after pruning Creating a new base height after pruning appears to eliminate the falsely high yields in the first yield after harvest (Figure 113, Figure 114, Figure 115). However, it was noted that CUPPA-Tea predicted lower yields in the second year after pruning. For example, in field 5, the predicted yield (excluding the first yield) was 23% lower than before the correction (Figure 117).

[0380] Further analysis demonstrated that the base height of approximately 12% of the shoots (60 shoots) remained at 60 cm. There were two shoots with higher base heights, but the majority of the shoots remained at approximately 73-85 cm (Figures 118 and 119). Figure 112 shows that after pruning to a base height of 60 cm in Field 5, the base height of some shoots remained at 65 cm. Figure 120 shows the distribution of base heights in Field 5 at the end of 2019.

[0381] Based on the above analysis, it was doubly confirmed that the retention of 60 shoots at 60 cm was not due to a "break-back" routine. Break-back occurs, for example, when a harvester, faced with a shoot with five open leaves, picks the shoot with three open leaves and then breaks back the rest of the shoot, removing the two older leaves.

[0382] The default CUPPA-Tea model is parameterized so that shoots with more than three leaves are plucked as shoots with three leaves, and the rest of the leaves break back. This should be true for hand-harvested leaves, but is less likely to occur for machine-harvested leaves. The lack of break back is noted as one of the main differences between hand-harvesting and reaper-harvesting.

[0383] In the default setup of CUPPA-Tea, breakback is disabled. Adding "breakback" (e.g., breakback = true) meant that the recorded yield was lower than before (Figure 117), since the proportion of leaves with more than three leaves was not included. The effect of adding the breakback routine was that the basal height remained close to 60 cm (Figure 118). Therefore, at this stage, it is appropriate that the breakback routine was disabled.

[0384] With reference to Figure 120 and Figures 121(a) and (b), it can be seen that the effect of enabling "breakback" along with pruning in field 5 was to result in significantly lower yields. Also, the effect of including "breakback" was to stop the increase in shoot base height after pruning.

[0385] To determine why approximately 12% of the shoots did not elongate beyond a base height of 50 cm, the growth of one such shoot (number 5 in field 5) is shown in Figure 122. This shoot had a very slow development rate. The following equation means that within the shoot growth routine, the growth stage is reset to 0 as soon as the shoot comes out of dormancy (i.e., as a bud): If (c_nlv = last leaf) (gs = 0; c_nlv = 0)

[0386] Figure 122 shows that if the assumed development and elongation rates are very low, the growth stage of some shoots will automatically reset to banjhi as soon as they leave banjhi.

[0387] One possible approach was to modify the coefficient of variation (default = 0.3) for shoot elongation and development rate. Choosing a lower coefficient of variation (e.g., 0.2) appears to result in higher yields after pruning, while choosing a higher coefficient of variation (e.g., 0.4) appears to result in lower yields. Figure 123 shows an analysis of the effect of the coefficient of variation for shoot elongation and development on predicted yield (Field 5; minimum harvestable shoot length of 3 cm; pruning from 75 cm to 60 cm).

[0388] A further approach was to set the "last leaf" for pruning to i) 7 and ii) 10. As described in the CUPPA Tea Technical Manual (Matthews & Stephens, 1996), there is a function called "meanlastleaf" that defines the average number of open leaves before a shoot goes dormant. Immediately after a tea plant is pruned, because there are fewer shoots, individual shoots tend to elongate further for a period than when the tea plant is unpruned. When the shoot base height is reset after pruning, a higher "meanlastleaf" needs to be allowed so that the shoot can reach the new picking height. Therefore, for the first 10 months after pruning, the "last leaf" was raised to 7 or 10.

[0389] If the base height was less than the platform height, the model was first run with the last leaf set to 7 or 10. The results were stored in a spreadsheet.

[0390] [Table 45]

[0391] This method makes it possible to eliminate the first yield peak after pruning (Figure 124). There will then be a period when yields are higher than those predicted before pruning, but yields will follow a similar pattern to that observed previously. The shoot base height will grow back to the level of the base within about 9 months (Figure 125).

[0392] Figure 126 shows an analysis of the effect of setting last leaf to 7 when base height is less than platform height in field 5. Figure 127 shows that the updated algorithm for when base height is less than platform height means that all shoots have moved to platform height within about 9 months.

[0393] The second approach was to set the last leaf to 7 or 10 only if it was a pruning event. The data were stored in a spreadsheet.

[0394] Figure 126 shows an analysis of the effect of setting last leaf to 7 or 10 in field 5 for pruning events only. Figure 127 shows that the updated algorithm for pruning events only means that all shoots moved to platform height within approximately 9 months. With a setting of lastleaf=7, all shoots moved from a base height of 60 cm within 12 months of pruning. With a setting of lastleaf=10, all shoots moved from a base height of 60 cm within 10 months of pruning.

[0395] It was determined that resetting the last leaf only at the time of pruning was more physiologically consistent.

[0396] Testing the algorithm in other fields The proposed pruning algorithm (resetting the last leaf to 7 or 10 only at the time of pruning) was investigated on six other fields: first fields 13 and 25, then fields 8, 13, 25, 29, 55, and 58.

[0397] Field 13 was a hand-harvested field. Resetting the base height at pruning eliminated the yield peak for the first harvest after pruning, and the corrected last leaf value increased the yield after pruning (Figure 128). For the next year, shoots only emerged just below the base height.

[0398] Figure 128 shows an analysis of the effect of setting last leaf to 7 when the base height is less than the stand height for field 13. Figures 129 and 130 show an analysis of the effect of setting last leaf to 7 or 10 (respectively) when pruning on an assumed base height for field 13.

[0399] Field 25 is a reaper-harvested field. Resetting the basal height at pruning eliminates the yield peak for the first crop after harvest, and the corrected last leaf value maintains yield the second year after pruning (Figure 131). In this case, choosing a last leaf value of 7 appears to give better results for the first yield after pruning than assuming a last leaf value of 10.

[0400] Figure 131 shows an analysis of setting the last leaf at pruning to 7 or 10 for field 25.

[0401] In the analysis of field 25, assuming 7 last leaves instead of 10, the increase in basal values ​​appears to be more gradual (Figures 132 and 133).

[0402] 132 and 133 show graphs illustrating the effect of setting the last leaf to 7 or 10 (respectively) at pruning on assumed base height for field 25.

[0403] Figure 8 is clone 225, which was mechanically harvested and pruned on May 5, 2017 (Figure 134). It seems reasonable to assume a last leaf of 7.

[0404] Field 29 contains a mix of clones including 3 1 / 8, harvested with a reaper and pruned on May 24, 2017 (Figure 135). A last leaf of 7 seems reasonable to assume.

[0405] Field 55 contains clone MRT, which is reported as being mechanically harvested and pruned on July 5, 2019 (Figure 136). However, harvest is reported on July 26, August 9, and August 21. Analysis of the data suggests that pruning occurred on April 12, 2019. It seems reasonable to assume a last leaf of 7.

[0406] Field 58 contains seedling tea that is reported as being hand harvested and pruned on July 15, 2019 (Figure 137). The yield assuming a last leaf of 7 appears better than the yield assuming a last leaf of 10.

[0407] Time from pruning to first harvest The seven surveyed fields can be used to determine the average time from pruning to first harvest (Table 45). The time between the last harvest before pruning and the first harvest after pruning is approximately 106 days. This is similar to the value of 90 days reported by our Kenyan tea farms for the timing from pruning to first chipping. [Table 46]

[0408] In an earlier analysis, the time from pruning to first harvest was assumed to be 8 phyllochrons (Burgess et al., 2020). At a temperature of 18°C, a phyllochron takes 7 days. Therefore, 8 phyllochrons is approximately 56 days. However, the original analysis did not consider the time it takes for the shoot to elongate from a base height of 60 cm to 75 cm. Assuming a maximum leaf-to-node length of 3 cm, an additional 5 phyllochrons are required before the shoot is ready for harvest. Therefore, 13 phyllochrons are required between pruning and first chipping. Therefore, at 18°C ​​(and assuming no drought stress), 13 × 7 = 91 days are required.

[0409] Therefore, there is a logical argument to increase the phyllochrons from pruning to chipping to 13 phyllochrons (8 phyllochrons for bud to shoot development and 5 phyllochrons for height development). conclusion 1. The default CUPPA-Tea model produces high yields for the first harvest after pruning. predicted because the model assumes constant platform height and base height for the shoot. 2. CUPPA-Tea has been modified so that shoot length and developmental stage are reset at pruning, and a reduction in base height to 15 cm below platform height appears to provide adequate yield for the first harvest after pruning. 3. This approach to addressing high initial yields appears to be more physiologically consistent than the original approach that focused on modifying shoot number (Appendix E). Because harvest intervals change with time since pruning, it would be prudent to determine the effect of varying harvest intervals only on yields before adjusting shoot number. 4. The important parameter is the change in basal height. The predicted yield when basal height is reduced from 75 cm to 60 cm is the same as the yield when basal height is reduced from 70 cm to 55 cm. 5. Early studies showed that long-term yield reductions were caused by resetting the basal height as approximately 12% of shoots that originated at 50 cm and did not grow beyond a pedestal height of 75 cm. This was not the result of assumptions about breakback. It was due to a formula within the developmental routine that caused the slowest-growing shoots to become banjhi and reset such shoots to banjhi even when they left dormancy. A way to prevent this is to assume that the last leaf for each shoot initiated at pruning is fixed at a large number. The results of some early analyses suggest that entering a last leaf number of 7 resulted in lower yields than 10. There is an argument that the phyllochrons between pruning and first chipping should be increased from 8 to 13. Code changes Underlined text -modified code Black text = existing code Summary of changes that will be made: 1. Add base height to model parameters 2. If the simulation date coincides with the pruning date, reset all shoots and shoot base heights to the base height from the model parameters. 3. Add a True / False flag to reset shoots to indicate whether the prune was true or false. 4. If prune is true in reset shoot, reset lastleaf to 7 instead of calculating Data to be passed to CUPPA Add Prune Base Height to the model parameters under Initial Conditions and set it to 60cm 1. “PruneBaseHeight”: 60, Code changes TeaModel.cs 1. In the GetInputData routine, add the base height: no3 = config[“InitialConditions”][“NO3”]; teaGrowthobj._PruneBaseHeight = config["InitialConditions"]["PruneBaseHeight"]; Teagrowth.cs 1. Add prune base height as a class variable / / AH 26 / 02 / 2020 add prune date public DateTime _prunedate { get; set;} public DateTime _startDate { get; set;} public double _PruneBaseHeight { get; set;} 2. In Growth, after checking Prunedate, reset the baseheight to the prune base height parameter from the model input parameters, and also change the phyllochron multiplier from 8 to 13. / / {if date of pruning then prune bushData} if (PruneDate(simDate)) { _bushobj.Prune(); _phenoCode = “Pr”; _phenoString = “Pruning”; _Phyllomult = 13 ; _bushobj.Prune(_PruneBaseHeight); } Bush.cs 1. In FrostDamage, pass false as a new parameter to reset shoots: _shoot[i].ResetShoot(_shoot[i].BaseHeight, a, b, c, d, _meanLastLeaf, _s1DevRateI, _s1DevRateMin, false ); / / {reset new shoot parameters} twenty three 2. In Prune, reset shoots and base height: public void Prune( double resetpruneheight ) { double a = 0; double b = 0; double c = 0; double d = 0; / / reset every shoot back to base height foreach (Shoot sht in _shoot) { GetNewParameters(ref a, ref b, ref c, ref d); sht.ResetShoot(resetpruneheight, a, b, c, d, _meanLastLeaf, _s1DevRateI, _s1DevRateMin, true); / / {reset new shoot parameters} } } 3. In Pluck, pass false as a new parameter to reset shoots: sht.ResetShoot(sht.BaseHeight + new_bp, a, b, c, d, _meanLastLeaf, _s1DevRateI, _s1DevRateMin, false ); / / {reset new shoot parameters} Shoot.cs 1. In Shoot, pass false as a new parameter to reset shoots: sht.ResetShoot(sht.BaseHeight + new_bp, a, b, c, d, _meanLastLeaf, _s1DevRateI, _s1DevRateMin, false ); / / {reset new shoot parameters} 2. In ResetShoot, add a new parameter flag to indicate whether prune is true or false: public void ResetShoot(double ht, double s1DR, double s2DR, double s1ER, double s2ER, double meanLastLeaf, double s1DevRateI, double s1DevRateMin, Boolean prune ) 3. In ResetShoot, if prune is true, reset lastleaf to 7: if (meanLastLeaf < 0) { _lastLeaf = 7;} else { if (prune==false) { _lastLeaf = (int)Math.Round(Utility.Limit(meanLastLeaf * (s1DR - s1DevRateMin) / (s1DevRateI - s1DevRateMin), 1, 10)); } else { _lastLeaf = 7;}

[0410] Resetting the Leaf Emergence Rate (LAR) when pruning Leaf Emergence Rate (LAR) is an accumulation of daily leaf emergence rates. It is reset to 0 at picking, but not at pruning. The modification tested was to also reset the LAR at pruning so that the number of days to harvest would be from pruning rather than from harvest. Field 58 was selected, with a harvest date of 7 / 11 / 19 and a pruning date 10 days later, 7 / 21 / 19.

[0411] Figure 138 and Table 46 and Table 47 show the accumulated LAR from two model runs, one before the change and one after adding a reset of the LAR on pruning. Before the change, the accumulated LAR reached 13 phylochrons on October 7, 2019. After changing the model to also reset on pruning, the accumulated LAR reached 13 phylochrons 10 days later, on October 17, 2019.

[0412] This change successfully moved the date when 13 phylochrons was reached. The code change required for this fix is ​​presented below and consists of just one line of code to set LAR=0 when the pruning date is reached. Changes to the code Underlined Text of - modified code Plain text = existing code 1. In TeaGrowth.cs a. In Growth, add a reset of LAR after checking if the pruning date is the current date: / / {if date of pruning then prune bushData} if (PruneDate(simDate)) { _phenoCode = “Pr”; _phenoString = “Pruning”; _Phyllomult = 13; / / 26 / 03 / 21 AH reset accumulated Leaf appearance rate on prune LAR = 0; [Table 47-1] [Table 47-2] [Table 48-1] [Table 48-2]

[0413] Modelling the effects of soil organic matter content and pH on tea yield response to nitrogen fertiliser Sustained high tea yields depend on a supply of nitrogen from soil reserves, typically maintained by nitrogen (N) fertilizer. The modifications discussed herein improved the CUPPA-Tea model, which simulates tea shoot development and growth, by including the effects of N deficiency. The updated model accurately explained 78% of the annual yield variation in the nitrogen and irrigation experiment in Tanzania and 79% of the annual yield variation in the fertilizer experiment in Kenya. The slopes of the relationships were 0.84 and 0.73, respectively, with a root mean square error of 681 kg ha -1 and 507 kg ha -1 and the modeling efficiencies were 0.80 and 0.75, respectively.

[0414] The model, which predicted high soil organic carbon content (4.0%) at the Kenyan site, resulted in a smaller yield response to nitrogen application than the model in Tanzania, where soil organic carbon content was lower (1.6%). The model also predicted small losses of nitrogen due to denitrification and leaching due to acidic soil conditions (pH < 4.5) in Kenya. In Tanzania, irrigation was predicted to result in approximately 10% higher nitrogen uptake than non-irrigated conditions. CUPPA-Tea proved to be a useful model for supporting decision-making and providing accurate tea yield estimates, as well as for predicting the fate of nitrogen within the soil-plant-atmosphere continuum.

[0415] Currently, there are no prior art process-based tea models that simulate the effect of nitrogen on tea yield. Therefore, this disclosure describes the inclusion of a nitrogen algorithm in the CUPPA-Tea model, as well as its calibration and validation using measured tea yields in Tanzania and Kenya.

[0416] Model Development The original CUPPA-Tea model included draft code for nitrogen, but the algorithm was not calibrated or validated and did not work (in that it did not follow the laws of physics and could not predict the effect of nitrogen on tea growth).

[0417] Initial nitrogen accumulation The updated model disclosed herein assumes initial nitrogen stocks that include nitrogen retained in plant tissues, soil inorganic nitrogen, and soil organic nitrogen. - and NH4 + Initial levels of soil inorganic nitrogen as (mg per kg of soil) are specified for depths of 15, 50, 200, 320, and 500 cm (Appendix G - Table 58). Initial levels of organic carbon are also specified for similar depth increments. For surface crop residues, the model is based on an initial C:N ratio of 40:1 and an initial C:N ratio of 1000 gm -2 The model assumes an initial canopy aboveground biomass of woody and green stems (2325 gm each). -2 and 200gm -2 ), which is multiplied by the initial assumed canopy nitrogen fraction of 0.0325 to give 114 gNm -2 The initial canopy nitrogen value is 1245gm -2 The initial weight is 190g -2 The fine root weight of 200gm -2 Assuming a root residue weight (weight of dead roots) of 0.01, these values ​​are 16 g N m -2 Therefore, the initial plant nitrogen content is approximately 130 g Nm -2, i.e. 1300 kg Nha -1 is.

[0418] Adding nitrogen to the system Additional nitrogen to a cropping system can be added as inorganic or organic fertilizer, which can be added to the surface of the top soil layer. In the model, nitrogen, held in the form of nitrate and urea, is assumed to be able to move through the soil with the movement of water from the top to the bottom layers of the soil. Ammonium is assumed not to be transported (Ritchie, 1985) (Godwin & Jones, 1991). The model includes the ability to include nitrogen from atmospheric deposition, but in this study, that input was assumed to be zero because specific measurements of nitrogen deposition were not available. Previous studies have estimated nitrogen deposition to be between 2.2 and 10.3 kg N ha -1 year -1 It has also been shown that the ratio tends to be smallest at low temperatures (Laouali et al., 2016) (Bakayoko et al., 2021). The model calculates the dynamic C:N ratio (cnr) (Equation 4) of carbon as litter and dead roots to nitrogen in their residues that is available as inorganic nitrogen. Carbon contributes to the new organic matter pool (OMnew) by dividing the carbon fraction (Cfrac new ;0.4) to obtain the new organic nitrogen pool (OrgN new ;kgNha -1 ) (Equation 4) and the total ammonium and nitrate in the soil (Ntot; kgNha -1 The carbon pool is calculated by accumulating the amount of each pool (carbohydrate, cellulose, and lignin) per soil layer (i). new It is updated daily on.

number

[0419] For fresh organic matter, the assumed nitrogen fraction is 0.02. The value of 0.02 is the product of the fraction of carbon in fresh organic matter (40%), the biological efficiency of microbial carbon turnover (40%), and the microbial N:C ratio (0.125:1) (Godwin & Jones, 1991). For humus, the carbon and nitrogen fractions are 0.4 and 0.024, respectively (Moran et al., 2005).

[0420] Nitrogen uptake, use, and loss by crops The model is based on the assumption that the crop is exposed to inorganic ammonium (NH4 + ) ions and nitrate (NO3 - ) ions. The amount of nitrogen taken up is assumed to be constrained by either the nitrogen demand of the crop or the nitrogen supply. The demand for nitrogen has two components: the demand by the existing canopy and roots, and the demand for canopy or root growth. The demand from the canopy is calculated as the canopy weight (W canopy ) and the difference between the critical canopy nitrogen level (CCNP; default value: 0.04) (i.e., the level below which growth slows) and the actual fraction of nitrogen in the canopy (CANC; %) (Equation 5). A similar relationship exists for roots based on the critical root nitrogen level (RCNP) and the actual fraction of nitrogen in the roots (RANC) (Equation 6). Nitrogen demand for new canopy growth (NewCanopyDem) depends on the canopy potential growth rate and the value of CCNP (Equation 7), and a similar relationship was assumed for roots (Equation 8). Equation 5: CanopyDem = W canopy ×(CCNP-CANC) where: CanopyDem is the demand for nitrogen by the tea canopy, W canopy is the canopy weight, CCNP is the critical nitrogen level for the canopy; CANC is the percentage of nitrogen in the canopy. Formula 6: RootDem=W root ×(RCNP-RANC) where: RootDem is the demand for nitrogen by tea plant roots; W root is the root weight, RCNP is the critical nitrogen level for roots; RANC is the proportion of nitrogen in the roots. Equation 7: NewCanopyDem = PotCanopyGrowth × CCNP where: NewCanopyDem is the demand for nitrogen for new growth in the tea canopy, PotCanopyGrowth is the potential growth rate of the canopy. Equation 8: NewRootDem = PotRootGrowth × RCNP where: NewRootDem is the demand for nitrogen for new root growth of the tea plant; PotRootGrowth is the potential growth rate of roots. Formula 9: TotCropNDem=CanopyDem+RootDem+NewCanopyDem+NewRootDem

[0421] Adding these demands together gives the total crop nitrogen demand (Equation 9). The potential nitrogen uptake across the soil profile (TotPotNUpt) is calculated by adding NH4 + and NO3 - The potential uptake of each soil layer is calculated by adding up the potential uptake of each soil layer (Equation 10 and Equation 11). [i] ), this is the NH4 per unit length of the root + and NO3 - The maximum daily intake of (which is 0.006 mg N (cm root) -1 ), root density (RLv[i]; cm root(cm 3 soil)-1 ), NH4 + and NO3 - The availability coefficients fNH4 and fNO3 range from 0 to 1 (Eqs. 12 and 13), and the soil NH4 + and NO3 - Concentration (mg kg -1 soil). Equation 10: PotNH4Upt = 0.006 * 100 * RLv [i] *Layer [i] *fNH4*smdfr 2 Formula 11: PotNO3Upt = 0.006 * 100 * RLv [i] *Layer [i] *fNO3*smdfr 2 Equation 12: fNH4=1.0-exp(-0.03*[NH4] [i] ) Equation 13: fNO3=1.0-exp(-0.03*[NO3] [i] )

[0422] A soil-water coefficient (smdfr) between 0 and 1 reduces potential nitrogen uptake as calculated from the relative availability of soil water (Equations 14 and 15).

number

number

[0423] In situations where the potential nitrogen supply (PotNSup, defined as the sum of PotNH4Upt and PotNO3Upt) is greater than TotCropNDem, the NUF coefficient is used to reduce the nitrogen uptake from each layer to the level of demand (Equation 16).

[0424] The effect of nitrogen on plant growth is related to the nitrogen deficiency index (Nfac), which is usually in the range of 0 and 1 and depends on the actual fraction of N in the canopy (CANC), the CCNP below which growth slows, and the minimum nitrogen content below which nitrogen content never falls (CMNC default value: 0.01) (Equation 17).

number

[0425] It was hypothesized that nitrogen stress could reduce both the rate of dry matter production and leaf area expansion, and leaf senescence. The rate of dry matter production by the crop (DMP; kg DMha) -1 d -1 , Equation 19) is the potential crop growth rate (PCGR; kg DMha -1 d -1 18), the temperature coefficient (Tfac), and the water stress multiplier (swdf1; calculated from the water stress coefficient) or the value of Ndef1, whichever is smaller (Equation 19), and the value of Ndef1 is defined by Equation 20. In Equation 18, S is the solar radiation (MJha -1 ) and RUE refers to the radiation utilization efficiency (gMJ -1 ), kext represents the attenuation coefficient, and LAI represents the leaf area index (m 2 m -2 ) Equation 18: PCGR=S*RUE(1-exp(-kext×LAI)) where: PCGR is the potential crop growth rate of the tea plant; S is the amount of solar radiation, RUE is the radiation utilization efficiency, kext is the damping coefficient, LAI is the leaf area index. Equation 19: DMP=PCGR×Tfac×Min(swdf1,Ndef1) where: DMP is a dry matter product of tea plants, Tfac is the temperature coefficient, swdf1 is the soil water deficit coefficient; Ndef1 is the nitrogen deficiency value calculated using the following formula: Formula 20: Ndef1 = 0.9952 × Nfac 2 -0.0354×Nfac+0.252 where Nfac is the nitrogen deficiency index.

[0426] The relationship between Ndef1 and the nitrogen deficiency index was based on the response described by (Godwen & Jones, 1991). However, a y-axis intercept of 0.252 was added to ensure that dry matter production could be positive after a pruning event even if no nitrogen was applied.

[0427] The nitrogen deficiency index was also used to correct leaf growth potential for its effect on leaf area expansion and leaf senescence. Leaf expansion rate (LeafLength; cmd -1 ) depends on the shoot elongation rate (SER), a coefficient for temperature in degree days (DdIlen; values ​​between 0 and 1), day length (Dlfac; values ​​between 0 and 1), and the minimum of a coefficient for saturation deficit (Sdfac), a water stress reduction coefficient (SERwf), and a nitrogen deficiency index (Nfac) (Equation 21). A similar relationship is assumed for leaf area growth rate (Equation 22), where PotLeafArea is the potential leaf area. Equation 21: LeafLength=S2ER×DDIlen×DLfac×Min(SDfac,Min(SERwf,Nfac)) where: LeafLength is the rate of leaf expansion, S2ER is the shoot elongation rate; DDIlen is the coefficient of temperature in degree days, Dlfac is photoperiod, Sdfac is the coefficient of undersaturation, SERwf is the water stress reduction factor; Nfac is the nitrogen deficiency index. Equation 22: LeafAgr=PotLeafArea×DDIlen×DLfac×Min(SDfac,Min(SERwf,Nfac)) where: LeafAgr is the leaf area growth rate, PotLeafArea is the potential leaf area, DDIlen is the coefficient of temperature in degree days, Dlfac is photoperiod, Sdfac is the coefficient of undersaturation, SERwf is the water stress reduction factor; Nfac is the nitrogen deficiency index.

[0428] Finally, it was hypothesized that nitrogen can be lost from the tea plant in four major ways. First, nitrogen can be lost from the system as leaves are harvested. Crop nitrogen can also be lost from the abscission of mature leaves and through pruning. The canopy nitrogen fraction was calculated daily as the canopy nitrogen content divided by the canopy weight. On the day the tea is pruned, the pruned canopy and associated nitrogen were assumed to be added to the topsoil, adding carbon and nitrogen to the fresh organic matter and organic nitrogen pools, respectively. Finally, crop nitrogen can be lost through root exudation, while organic nitrogen is added back to the fresh organic matter pool, where it can be mineralized and made available again to the plant.

[0429] Nitrogen flow in soil Nitrogen flows within the soil are considered in terms of mineralization, nitrification, denitrification, volatilization, and leaching (Figure 139).

[0430] Mineralization The conversion of soil organic nitrogen to inorganic nitrogen in CUPPA-Tea occurs through the mineralization of four organic matter pools (OMPools) (Eq. 23). These pools comprise fresh organic matter (FOM) (which is further subdivided into carbohydrates, cellulose, and lignin) and the more stable humus (HUM). For each component of the pool, there is a designated decay coefficient (rdecr; Eq. 23), which are 0.2, 0.05, and 0.0095 for carbohydrates, cellulose, and lignin, respectively (Seligman & van Keulen, 1981). Mineralization is also modified by the soil moisture factor (mf; Figure 141a), the C:N ratio factor (cnrf; Figure 140a), and the soil temperature factor (Tf; Figure 140b). For humus, which is more resistant to decay, the potential decay rate constant is small (8.3*10 -5 d -1 ). It is also adjusted by Tf and mf, but cnrf is not used. Another index (HumMinFactor) can be used to adjust the humus mineralization rate for atypical soils (e.g., some volcanic ash soils where mineralization is low, and some newly cultivated, unused soils where mineralization is high).

[0431] Mineralization rate (N mineralise[t+1] ) is the mineralization rate of the previous day and the new organic nitrogen (OrgN new ;kgha -1 ) and the fraction of the organic matter pool relative to total organic matter, plus the reduction factor (I indicates the type of pool; Equation 23). new The level of fixed NH4 + and NO3 - From the sum of N mineralise[t] and new humus nitrogen (Eq. 24), and assume that 20% of the nitrogen released from new organic matter is added to humus.

number

number

[0432] The amount of nitrogen fixed each day (N imm ;kgha -1 ) is first, 0.5mg NH4 + (kg soil) -1 The amount of NH4 that exceeds the minimum level + (N imm NH4, Equation 28), and then the minimum level (minimum NO3 - Concentration: 0.25mgkg -1 soil) - Pool (N imm NO3, taken from Eq. 29).

number

number

[0433] nitrification Nitrification occurs under aerobic conditions and produces NH4 + NO2 - Furthermore, NO3 -It is a process of biological oxidation of the substrate NH4 + , oxygen, soil pH, and temperature. The environmental limit of nitrification (Elnc; Eq. 31) was determined from the minimum of the temperature coefficient (Tf), the soil moisture coefficient for nitrification (wdf, Figure 142a), and the coefficient Sanc (Eq. 30). The soil moisture coefficient for nitrification increases from 0 at permanent wilting to 0.9 at field capacity, then decreases to 0 at saturation (Figure 143a). The coefficient Sanc, which ranges from 0 to 1, is related to the ammonium pool (sNH4, kg ha -1 ) (Equation 30). Formula 30: Sanc=1-exp(-0.01363×sNH4)

number

[0434] The value of a (Equation 35) is determined by the minimum of a series of environmental indices ranging from 0 to 1 related to pH, temperature (Figure 140b), and soil moisture content (Figure 142a) and rp2. The assumption in CUPPA-Tea is that there is no nitrification at pH below 4.5 and above 9.0, and at soil temperatures below 5°C (Figure 141) (Yao et al., 2011). The value of rp2 is included to account for delays in the event that previous (last 2 days) conditions were not favorable for nitrification. As shown in Equation 34, this is assumed to depend on Elnc and the relative microbial nitrification potential (cni) of the layer on the previous day, where cni was assumed to have an initial value of 0.1.

number

[0435] Calculations are performed for the current day, the previous day, and the next day. Sarnc is a coefficient between 0 and 1 for ammonium availability (Equation 36). The least restrictive of the current day's and previous day's moisture coefficients (wfd and wfy, respectively) and temperature coefficients (Tf and Tfy, respectively) are used in calculating the new value of cni (Equations 37 through 39). Equation 36: Sarnc=1-exp(-0.1363×sNH4) Formula 37: xw = Max(wfd, wfy) Equation 38:xt=Max(Tf,Tfy) Equation 39:cni=rp2×Min(xw,xt,Sarnc)

[0436] The above effect is to suppress the influence of a single day of low soil temperature or drought stress on the value of microbial nitrification capacity (cni). The relative magnitudes of exp(2.302 × Elnc) and Min(xw, xt, Sarnc) determine whether the relative nitrification capacity increases or decreases over the short term (Godwin & Jones, 1991).

[0437] In nitrification, most of the oxidized ammonium is converted to nitrite (NO2 - ) via nitrate (NO3 - ), and a small proportion of nitrogen can be released as NO. The fraction of NO emissions associated with nitrification is modeled as 0.5% in the ExpertN model and 2% in the CENTURY model (Frolking et al., 1998). Within CUPPA-Tea, we assumed an overall nitrification loss of 1%, as used in the VISIT model (Inatomi et al., 2019).

[0438] denitrification Denitrification refers to the reduction of nitrate to gaseous products such as NO, N2O, and N2. (Parton et al., 2001) reported that denitrification typically occurs when water-filled pore space exceeds 55%, reaching a maximum when the soil is saturated and soil microorganisms use nitrate as an oxygen source. Therefore, CUPPA-Tea predicts that denitrification will occur when the soil moisture content exceeds field capacity (Figure 142a), soil temperature is higher than 5°C (Figure 142b), and the nitrate concentration is 1 mg NO3 kg -1 Denitrification is assumed when the water temperature is higher than the soil temperature. denitrify ;kgNha -1 ) is 0.000006kgNha -1 In addition, the moisture coefficient (Figure 142a), temperature coefficient (Figure 142b), and nitrate concentration NO3 - (mgkg -1 soil), and total water extractable carbon (cw; mgCkg -1 The value of cw is assumed to be 24.5 mg (kg soil) -10.31% of soil carbon (SoilC, kgha -1 ) (assumed to be the 58% fraction of stable humus) and the C content (Cfrac New is assumed to be 0.4; Equation 41) and is calculated as (Rolston et al., 1980). Formula 40:N denitrify =6×10 -6 ×fw×Tf×NO3(L)×DLAYR(L)×cw where DLAYR(L) is the depth of the soil layer Equation 41: cw=24.5+(SoilC×0.0031+Cfrac New ×OMpools[1])

[0439] Volatilization The volatilization routine is based on the method described by (Liu et al., 2020) and is used in the DNDC model. Volatilization of ammonia (NH3) to the atmosphere is calculated by the ammonium concentration [NH4 + ], pH, and temperature, and is derived from the ammonia concentration in soil water, NH3(l) (Eq. 42).

number

[0440] where fwind is a function of wind speed (Equation 44, wind is 2ms -1 is assumed to be constant at soil ; °C) (Equation 45). There is also a soil depth function (fdepth), where q is the number of soil layers and j is the layer of application, which is assumed to be 1 when fertilizer is applied to the soil surface (Equation 46).

number

[0441] Nitrate flow Within the soil, only the movement of nitrate and urea is considered. Ammonium is assumed not to be transported, as it is retained by the clay component of the soil (Table 48). Nitrate and urea movement, which is assumed to depend on the movement of water from one layer to the next, was calculated using the same cascade approach, where nitrogen lost from one layer is added to the layer below. A portion of the mass of nitrate present in each layer (sNO3) is assumed to move with each drainage event. Within the water routine of the model (Matthews & Stephens, 1998), the volume of water (FLUX) that moves from one layer (L) to the layer below is calculated. [L] ) is calculated. Before drainage, the volume of water present in each layer (swc [i] ×Layer [i] ) and the water (FLUX [L] ) is used to calculate the nitrate lost from each layer (OutN; Equation 47). NO3 - The concentration of 1 mg kg -1 Once the nitrate falls to the bottom, no further leaching from that layer is assumed. Nitrates that migrate from a layer (UpN) are transported upward (FLOW) in the same manner as leaching. [L] ) is calculated as a function of (Equation 48).

number

[0442] Updated CUPPA-Tea Calibration and Validation The model was first calibrated using annual tea yields collected between 1989 and 1995 from a replicated irrigation and fertilization experiment at the Ngwazi Tea Research Unit in Tanzania (8°33′S, 35°10′E, 1840 m altitude) (Figure 143). The experiment involved clones 6 / 8 planted in the late 1960s on sandy clay overlying clay, where the percentage of pore space filled with water at field capacity ranged from 46% to 50%. Average annual rainfall ranged from 839 to 1121 mm, with most falling between late November and May (Table 48). Field measurements indicated that tea plants were rooted to a depth of 500 cm. Measured soil organic carbon content ranged from 1.61% at 15 cm to 0.70% at 70 cm (Burgess & Sanga, 1993), and the organic carbon content was assumed to be 0% at 500 cm. pH was assumed to range from 4.6 at 15 cm to 4.2 at 70 cm (Burgess & Sanga, 1993), and to 6.0 at 500 cm. [Table 49]

[0443] The six nitrogen treatments were 0, 150, 225, 300, 375, or 450 kg N ha of N, P2O5, and K2O applied in a 2:1:1 ratio. -1 The treatments included annual applications of 100 mg / kg of fertilizer. There were also five irrigation treatments, ranging from no irrigation (treatment I0) to tea plants that received 50 mm of irrigation whenever the soil moisture deficit reached 50 mm (treatment I4). In the irrigated treatments, fertilizer applications were evenly divided between applications on January 1 and July 1, while in the non-irrigated treatment I0, all annual fertilizer applications were made on January 1 during the rainy season. The main forms of nitrogen fertilizer applied were urea and ammonium nitrate. The tea plants were pruned on November 25, 1990.

[0444] Validation of the nitrogen algorithm was completed using annual tea yield data collected from an area of ​​clone MRTM1 on the Kericho Tea Estate in Kenya (latitude 0°36'S, longitude 35°28'E, altitude 2002 m) between 2015 and 2020 (Figure 141). Average annual rainfall ranged from 1,341 mm to 3,576 mm (Table 48; Figure 151). The soil was defined as a medium sandy clay loam, with an assumed soil depth of 300 cm and 59% water-filled pore space at field capacity. Soil organic carbon content was assumed to be 4.0% at 10 cm, decreasing to 1.3% at 300 cm. pH was assumed to be approximately 4 throughout the soil profile. Fertilizer was applied at 0, 90, 180, or 270 kg Nha -1 The tea plants were pruned on June 1, 2015, and February 20, 2019. The site-specific analytical model assumptions for crop management, initial conditions, clonal parameters, and soil parameters are presented in Appendix G.

[0445] Statistical analyses were performed using R for both the calibration and validation datasets (R Core Team, 2022). Linear regression was performed using R. 2 , root mean square error (RMSE), and modeling efficiency (EF). EF allows comparing the efficiency of the chosen model with its efficiency in describing the data as a measure of observation (Smith et al., 1997). EF values ​​can be positive or negative, with a maximum value of 1. Positive values ​​indicate that the simulated values ​​describe the trends of the measured data better than the average of the observations. Smaller and negative values ​​indicate that the simulated values ​​describe the data less well compared to the average observations.

[0446] Model calibration and validation The inclusion of the nitrogen algorithm enabled the CUPPA-Tea model to predict annual yields over 6 nitrogen treatments x 5 irrigation treatments x 7 years (210 data points), which explained 79% of the observed yield variation in Ngwazi, Tanzania (Figure 144). Measured tea yields ranged from 542 to 6,291 kg ha -1 The slope of the best-fit linear regression was 0.84, and the RMSE error was 660 kg ha -1 and the modeling efficiency (EF) was 0.77.

[0447] Same R of 79% 2 Values ​​were obtained when predicted yields for the four nitrogen treatments x 6 years (24 data points) for the Kericho experiment were compared with measured values ​​(Figure 144). Measured yields ranged from 1,440 to 5,412 kg ha -1 The slope of the best-fit linear regression was 0.72, and the RMSE error was 507 kg ha -1 and the modeling efficiency (EF) was 0.75. At each site, the lowest yield occurred in the year after pruning (1991 in Tanzania and 2016 in Kenya).

[0448] Nitrogen balance in the system The CUPPA-Tea model was used to derive the initial nitrogen stocks and then the annual flow of nitrogen between organic and inorganic forms in the soil, and the nitrogen retained by the tea plants for both sites (Tables 49 and 50). The 2015 Kenyan experiment showed that the majority of nitrogen in the tea agroecosystem was in the form of soil organic nitrogen (46,672 kg N ha -1 ), mainly humus (46,157 kg Nha -1 ) (Table 49).

[0449] The next largest accumulation was nitrogen stored in the crop (1,565 kg N ha -1 ), and the model assumes rapid uptake of inorganic nitrogen, so the initial amount of inorganic nitrogen is minimal (1 kg N ha -1) was assumed to be 270 kg / ha -1 In treatments receiving fertilizer, the annual addition of nitrogen was 270 kg ha -1 The amount of nitrogen removed as harvested leaves was assumed to be 90 kg N ha in 2015. -1 (When the tea was pruned on June 1st) to 212kgNha in 2020 -1 (after pruning in February 2019). Volatile losses were nearly zero, while N2O losses through denitrification and nitrification were zero due to the assumed low pH of less than 4.1 (Table 48). Leaching losses were also essentially zero due to the assumption that any nitrogen released from urea hydrolysis was added to the soil ammonium pool, which was assumed to be fixed in the soil profile (Table 62).

[0450] Although losses from the system were minimal, the model predicted changes between nitrogen stored in the tea plants and soil organic and inorganic nitrogen. The level of nitrogen stored in the tea plants was 1,567 kg N ha in early January 2016. -1 From 2,488 kg / Nha in January 2019 before pruning in February 2019 -1 Inorganic nitrogen levels were low (typically 1 kg N ha) except in January 2020 after the previous year's pruning. -1 ) was predicted. In Kenya, all of the nitrogen applied in the experiment was added in July. Levels of fresh organic nitrogen were higher in January 2016 than in January 2015, before the June 2015 pruning, and higher in January 2020 than in January 2019, before the February 2019 pruning. Levels of soil organic nitrogen stored as humus remained relatively stable. [Table 50]

[0451] 300 kg / ha under rain-fed conditions in Tanzania -1 By completing a similar analysis on tea received from Kenya (46,000 kg ha -1) than in the area where soil organic nitrogen levels are lower (approximately 13,500 kg ha -1 ), which is directly related to the assumed lower levels of soil organic carbon (0-1.6% compared to 1.3-4.0%). -1 ) in Kenya (1565-2488 kg Nha -1 However, the inorganic nitrogen levels on January 1 were significantly higher than in Kenya (182-195 kg N ha -1 ), because in the no-irrigation treatment, it was assumed that all of the nitrogen fertilizer was applied on January 1. In Tanzania, tea was pruned in November 1990 and 92.4 kg Nha -1 Large leaching losses of 1000kJ / kg were predicted in 1991. The pH in Tanzania was assumed to be 4.2-6.0, compared with about 4.0 throughout the soil profile in Kenya, so that nitrification of ammonium to nitrate was possible (Table 50). [Table 51]

[0452] Nitrogen response curves and sensitivity to soil organic carbon Within the modified CUPPA-Tea model, it was possible to predict the effect of soil organic carbon (SOC) content on the response of tea yield to nitrogen application. To investigate this, a sensitivity analysis was conducted for different initial SOC values ​​per tea plantation. Figure 145 presents how the default nitrogen response changed when the initial SOC was i) 0, ii) 1.6%, iii) 4%, and iv) 8%. Although the soil depths were 500 cm and 300 cm, the default SOC content in the top 15 cm was 1.6% and 4.0% in Ngwazi and Kericho, respectively (Table 60).

[0453] In general, modeled yields at 1.6% SOC (default value in Tanzania) matched measured yields in Tanzania (Figure 145a and b), and modeled yields at 4% SOC (default value in Kenya) matched measured yields in Kenya (Figure 145c). At both sites, CUPPA-Tea predicted that higher SOC content resulted in higher average tea yields, with the effect being greatest when no fertilizer was applied (Figure 145). Under rainfed conditions and without fertilizer application, the average modeled yield at Kericho at 4% SOC (3,033 kg ha -1 ) was slightly higher than Ngwazi (2,318 kg ha ) possibly as a result of higher rainfall levels in Kenya. -1 ).

[0454] The predicted effect of increasing SOC from 0% (red line) to 8% (orange line) under rainfed conditions and without fertilizer is 1,783 kg ha in Tanzania. -1 and 2,180 kg ha in Kenya. -1 Under irrigation and without fertilizer, raising SOC from 0% to 8% in Tanzania would require 4,000 kg ha -1 Higher levels of nitrogen application reduced the effect of SOC, resulting in a predicted yield benefit of 450 kg N ha -1 The yield response to changes in SOC for rainfed tea in Tanzania at application rates of was minimal (Figure 145).

[0455] Results are discussed with respect to the importance of including nitrogen dynamics in crop models, the importance of developing nitrogen response tools, understanding environmental losses of nitrogen, organic tea production and limitations, and future research.

[0456] Including nitrogen in crop models Nitrogen (N), especially inorganic N, is typically the most limiting nutrient for tea growth and development (LeBauer & Treseder, 2008). Given that crop modeling began approximately 50 years ago (Hoogenboon et al., 2020), crop models that consider N dynamics exist but are still relatively rare (e.g., Jones et al., 2003; Wolf, 2012; Holzworth et al., 2014). Such models that adequately describe the daily interactions of the soil-plant-atmosphere interface can help managers understand the functioning of farm systems and assist them in decision-making. Herein, we demonstrate that it is possible to successfully develop a working model of N responses in tea, calibrated for experiments in Tanzania and validated for experiments in Kenya. The calibrated and validated model was able to explain 78–79% of the variation in tea yield across different years, different fertilizer and drought treatments, and different stages of pruning. Across the two sites, the slope of the best-fit regression line ranged from 0.72 to 0.84, suggesting a tendency to overestimate yields in the years of pruning and underestimate yields in the highest yields. This effect of moderating the model so that it tends not to predict the full range of field observations appears to be a common feature of a wide range of models (Giannitsopoulos et al., 2021) (Beka et al., 2022).

[0457] The integration of nitrogen responses into CUPPA-Tea allows for the investigation of interactions between fertilizer application rates, climate, drought, and soil type. By including nitrogen responses, the model improved estimates compared to simulations where, for example, the nitrogen routine was turned off (Table 51). [Table 52]

[0458] The interaction between water availability and nitrogen is shown in Table 52. The model predicted that for a given SOC, the yield response to nitrogen application would be greater under water-unconstrained conditions. This is because water is essential to enable soil nitrogen uptake by the crop, and well-watered conditions can also enhance soil mineralization and high levels of microbial activity (Schimel et al., 2007; Plaza-Bonilla et al., 2022).

[0459] For example, in Tanzania, as expected, the high irrigation treatments produced higher yields than rainfed tea across all N applications. It is noteworthy that when SOC is 0% (Figure 145a and b; red lines) and N application is high, fertilizer utilization is better under fully irrigated (I4) conditions than under rainfed (I0) conditions, with predicted yields of 5,442 kg ha−1 and 5,442 kg ha−1, respectively. -1 and 3,964 kg ha -1 On average, the amount of nitrogen taken up by tea plants under fully irrigated conditions was predicted to be about 10% higher than under rainfed conditions (Table 52). [Table 53]

[0460] The transport of mobile ions in soil depends on the soil water content, both allowing simple diffusion and mass flow driven by transpiration (Plett et al., 2020). This suggests that the potential soil water deficit was higher in Tanzania (average 818 mm) than in Kenya (average 425 mm) for a 464 kg Nha2 field study. -1 Compared to 345kg / ha less -1This was also evident in both sites under rainfed conditions, where it was associated with total nitrogen uptake by the crop (Table 52). It is noteworthy that when soil moisture content remains permanently high, oxygen deficiency can affect the function of aerobic microorganisms, thereby reducing the conversion of organic to inorganic nitrogen (Moyano et al., 2012).

[0461] Development of financial nitrogen reaction tools There are various methods to estimate the optimal level of fertilizer application. Generally, agronomists recommend implementing the "4R" practice to optimize nitrogen use, which is applying the right amount of fertilizer with the right nutrient source at the right time in the right place in the soil (Sposari & Flis, 2017). At a minimum, a decision can be made to simply replace the nitrogen removed by leaf harvesting. In Kenya, 270 kg Nha -1 At this nitrogen application rate, in years when the plant is not pruned, 145 to 212 kg Nha -1 are removed as harvested shoots, which is equivalent to 53% and 79% of the applied nitrogen. Ensuring that such levels of nitrogen application are carried out ensures that the total amount of nitrogen in the system is not reduced.

[0462] An alternative approach is to determine the yield response curve to nitrogen fertilizer, consider any effects of nitrogen on tea quality, and include assumptions about the price of nitrogen fertilizer and the value of the tea harvested. Using the response curves modeled for pruned and unpruned tea in Kenya, it is possible to establish marginal revenue and cost curves for unpruned and pruned tea. The assumptions included a fertilizer cost value of US$1.00 per kg of urea, a dry matter content of US$0.41 per kg of green tea leaves harvested, a dry matter content of 0.22, and transportation costs to the factory of US$0.005 per kg of green tea leaves harvested (assuming a distance to the factory of 10 km; Table 61). The results show that, based on these assumptions, tea profitability is higher at 300 kg Nha for unpruned tea. -1By spraying 70-150 kg / ha of tea leaves, and for pruned tea, -1 (Figure 148) The above assumed costs for fertilizer at the beginning of 2023 are significantly higher than the value for 2019 (US$ 0.20 per kg of urea; (Index Mundi, 2022)).

[0463] Nitrogen fertilizer is commonly applied to tea worldwide, and in Zhejiang province, China, the average annual application rate is 520 kg Nha -1 In practice, nitrogen applications have reached 1,200 kg Nha, which is 2 to 2.5 times higher than the recommended rate (Ma et al., 2013). -1 year -1 There are examples such as Japanese tea, where the temperature can reach 100°C (Nakasone, 2009).

[0464] Environmental consequences of nitrogen loss In addition to financial impacts, nitrogen losses can have negative environmental consequences (Rebello et al., 2022). Therefore, optimal nitrogen application may also depend on the social cost of nitrogen losses to the environment. Ideally, nitrogen fertilizer should be applied in a way that allows it to be utilized by the tea plant, rather than being lost to the aquatic or atmospheric environment, where it can be a major source of pollution (Farzadfar et al., 2021).

[0465] When and where nitrogen is lost can depend on rainfall intensity, soil cover, and the physical state of the soil (Vagstad et al., 1997) (Patil et al., 2010). Nitrogen losses in soil systems can include nitrate and ammonium leaching, ammonia volatilization, and denitrification (emission of NO) (Musyoka et al., 2019) (Zou et al., 2021). In the model, surface runoff losses of nitrogen were not considered because the experimental site was on relatively flat ground. In reality, tea is often grown on steep slopes, so it may not be correct to assume there are no runoff losses.

[0466] Leaching: Within the CUPPA-Tea model, predicted rates of nitrate leaching are typically low due to i) low levels of inorganic nitrogen in the soil, ii) acidic soils that prevent nitrification, and iii) relatively deep tea roots (3-5 m). In Kericho, under a default pH of 4.0, average nitrogen losses across all nitrogen applications and years were zero (Table 53). The algorithm used in the CUPPA-Tea model assumed that soil nitrification was ineffective at pHs below 4.5 and above 9.0.

[0467] At a slightly higher pH of 4.51, the model showed that in years when tea was not pruned, 1-18 kg / ha -1 year -1 The annual leaching rates of 1000 to 10 ... - concentrations, resulting in less nitrogen transport through leaching (Table 53). In Lake Taihu, China (Zhou et al., 2022) found concentrations of 209 and 174 kg N ha -1 When urea was applied in March and October, respectively, the yields were 73, 200, and 240 kg / ha in dry, normal, and wet years, respectively. -1 reported a nitrogen leaching loss of 270 kg N ha in the year of pruning (e.g., 2019) (when crop nitrogen uptake is reduced and nitrogen release from pruning and root turnover can provide additional nitrogen to the soil). In that study, the average annual temperature and precipitation over a 60-year period were 15.9°C and 1,157 mm, respectively, the tea was over 15 years old, the soil texture was silty loam, and the average pH reached 6.15. High pH is a major factor in high leaching losses. If a higher pH of 4.51 is assumed for soils in Kenya, the model predicts that 270 kg N ha in the year of pruning (e.g., 2019) (when crop nitrogen uptake is reduced and nitrogen release from pruning and root turnover can provide additional nitrogen to the soil) -1 If sprayed, 174 kg / ha -1 year -1We estimate that there may be leaching losses of 1000 mg / kg (Table 53).

[0468] Volatilization: In this study, predicted volatilization losses in Kenya were lowest at a pH of about 4 (Table 53) and remained lowest at a pH of 5 (0.16 kg Nha -1 year -1 Using the model with an assumed pH of 7.0, the predicted average annual volatilization rate over several years is 16 kg N ha -1 (data not shown). There are relatively few field studies on volatile losses of tea. In a greenhouse experiment, (Liyanage et al., 2015) found that the average percentage of ammonia volatilized in Sri Lanka was 1.48-1.53 ​​g cm −1 at pH values ​​of 4.46-4.84 and 1.48-1.53 ​​g cm −1. -3 In UK temperature conditions, (Bishop & Manning, 2010) showed that approximately 19% of applied nitrogen (average application rate: 130 kg N ha ) from greenhouse and arable soils, respectively, can be applied as urea. -1 ) and 17% of the applied nitrogen (average application rate: 100 kg N ha -1 reported the average nitrogen loss to the environment through NH3 volatilization of 1000 m3. N2O losses: Denitrification and nitrification rates were also low in this study, partly due to low soil pH values ​​and generally well-aerated soil conditions (Figure 149). Annual denitrification rates in Zhou et al. (2022) for Chinese tea fields ranged from 1.76 to 2.30 kg N ha -1 year -1 Our average calculated N2O emission rate from nitrification was 270 kg N ha-1 at pH 4.0 to 5.0. -1 Assuming a spray of 0 to 1.21 kg / ha -1 year -1 The range of 510 kg / ha for tea in Japan with soil pH of 3.4 was 510 kg / ha (Table 53). -1 year -1 10.6 kg Nha from spraying -1 year -1Higher N2O emissions have been reported at higher application rates, including (Hirono & Nonaka, 2012). In Hunan, China, at a soil pH of 4.5, the no-fertilizer treatment and conventional (410 kg N ha -1 year -1 7.1 and 17.2 kg / ha for the spraying treatments, respectively. -1 year -1 have been observed (Fu et al., 2012). The reported global average direct emission factor of N2O induced by fertilizer nitrogen from tea plantations worldwide is 2.31%, which is twice the IPCC default value of 1% (Hergoualc'h et al., 2019); (Wang et al., 2020). [Table 54]

[0469] organic tea The updated CUPPA-Tea model can be useful in supporting management decisions for organic tea, i.e., tea that does not receive inorganic nitrogen applications. The nitrogen model algorithm can be used to predict nitrogen derived from decomposition into fresh and recalcitrant organic matter (humus), and using the model with experimental calibration and validation can provide useful insights into the impact of adding organic fertilizer. One potential impact of adding organic fertilizer is to increase soil pH (Yan et al., 2018) and reduce the level of soil acidification in tea plantations (Xie et al., 2021). This increase in soil pH can increase leaching and volatilization losses (Benbi & Nieder, 2003). Application of organic matter (e.g., residuals) with a high C:N ratio can also temporarily sequester nitrogen within soil microorganisms (immobilization). However, this nitrogen source can become available again to the crop when the microorganisms die. In a 41-year experiment in Germany, soil carbon stocks in various crops, such as rye, wheat, maize, and barley, appeared to be higher when organic inputs were combined with inorganic fertilizer compared to organic inputs alone or inorganic fertilizer alone (Hijbeek et al., 2017). Additionally, (Bagchi et al., 2020) reported that organically grown tea may also have a higher polyphenol content than tea grown with inorganic fertilizer.

[0470] The results in Figure 145 demonstrate that yield losses due to the elimination of inorganic nitrogen applications can be partially offset by ensuring high soil organic carbon content. In contrast, in a global meta-analysis focusing on experimental results for maize, (Oldfield et al., 2019) reported that a steeper nitrogen response curve can be achieved with higher levels of SOC. One possible reason for this contrasting result is that increasing soil organic carbon content can provide additional benefits beyond just nitrogen supply, which was the focus of this study. For example, high levels of organic matter can improve soil structure, increase water retention capacity, and soil biodiversity, and reduce the risk of soil erosion (Johnston 1986; Lal 2009).

[0471] Continuous production of high tea yields depends on a steady supply of nitrogen from the soil to meet the tea plant's needs. Field use of the CUPPA-Tea model revealed that yield responses to applied nitrogen were particularly sensitive to changes in soil organic carbon content, soil water availability, and pH. Including a nitrogen algorithm in the model resulted in accurate predictions of annual tea yields in two experiments, and CUPPA-Tea estimated low nitrogen losses to the environment as long as pH levels were approximately 4.0. The model revealed that, in the absence of inorganic nitrogen application, higher tea yields were obtained in sites with high soil organic carbon content than in sites with low soil organic carbon content. The yield benefit of high soil organic carbon content was less clear at higher inorganic nitrogen application rates. Thus, tea grown in soils with low soil organic carbon content tended to respond more strongly than tea grown in soils with high soil organic matter content. CUPPA-Tea also demonstrated the importance of water-limited conditions in reducing crop nitrogen uptake and, therefore, tea yield. Using such a model as CUPPA-Tea can help estimate financially optimal nitrogen requirements and can also generate marginal revenue and cost curves after considering similar assumptions. Modeling different scenarios and testing various hypotheses can help users make the right decisions regarding sustainable tea production, targeting profitable yields while simultaneously protecting the natural environment.

[0472] The systems and methods of the above embodiments, in addition to the structural components and user interactions described, may be implemented in a computer system (specifically in computer hardware or computer software).

[0473] The term "computer system" includes hardware, software, and data storage devices for embodying a system or performing a method according to the embodiments described above. For example, a computer system may include a central processing unit (CPU), input means, output means, and data storage. A computer system may have a monitor for providing a visual output display. Data storage may include RAM, a disk drive, or other computer-readable medium. A computer system may include multiple computing devices connected by a network and capable of communicating with each other via the network.

[0474] The methods of the above embodiments may be provided as a computer program, or as a computer program product, or as a computer-readable medium carrying a computer program, which is adapted to perform the methods described above when executed on a computer.

[0475] The term "computer-readable medium" includes, but is not limited to, any non-transitory medium that can be read and accessed directly by a computer or computer system. The medium may include, but is not limited to, magnetic storage media such as floppy disks, hard disk storage media and magnetic tape, optical storage media such as optical disks or CD-ROMs, electrical storage media such as RAM, ROM, and memory including flash memory, and hybrids and combinations of the above, such as magnetic / optical storage media.

[0476] While the present disclosure has been described in conjunction with the exemplary embodiments set forth above, many equivalent modifications and variations will be apparent to those skilled in the art given this disclosure. Accordingly, the exemplary embodiments of the present disclosure set forth above are to be considered illustrative rather than limiting. Various changes to the described embodiments may be made without departing from the spirit and scope of the present disclosure.

[0477] In particular, although the methods of the above embodiments are described as being performed on the systems of the described embodiments, the methods and systems of the present disclosure need not be implemented in conjunction with each other, and each may be implemented on alternative systems or using alternative methods.

[0478] The features disclosed in the description, or in the following claims, or in the accompanying drawings, and expressed in terms of specific forms or means for carrying out a disclosed function, or methods or processes for obtaining a disclosed result, may be used, as appropriate, separately or in any combination of such features to realize the disclosure in diverse forms thereof.

[0479] While the present disclosure has been described in conjunction with the exemplary embodiments set forth above, many equivalent modifications and variations will be apparent to those skilled in the art given this disclosure. Accordingly, the exemplary embodiments of the present disclosure set forth above should be considered illustrative rather than limiting. Various changes to the described embodiments can be made without departing from the spirit and scope of the present disclosure.

[0480] For the avoidance of any doubt, any theoretical explanations provided herein are provided for the purpose of improving the understanding of the reader, and the inventors do not wish to be bound by any of these theoretical explanations.

[0481] Any section headings used herein are for organizational purposes only and should not be construed as limiting the subject matter described.

[0482] Throughout this specification, including the claims that follow, unless the context requires otherwise, the terms "comprise" and "include," and variations such as "comprises," "comprising," and "including," will be understood to imply the inclusion of stated integers or steps or groups of integers or steps and not the exclusion of other integers or steps or groups of integers or steps.

[0483] It should be noted that, as used in this specification and the appended claims, the singular forms "a," "an," and "the" include plural references unless the context clearly dictates otherwise. Ranges may be expressed herein as from "about" one particular value and / or to "about" another particular value. When such a range is expressed, another embodiment includes from the particular value and / or to the other particular value. Similarly, when values ​​are expressed as approximations by use of the antecedent "about," it will be understood that the particular value forms another embodiment. The term "about" with respect to numerical values ​​is optional and may mean, for example, + / - 10%.

[0484] Disclosure Terms 1. A computer-implemented method for predicting the growth of a plurality of tea plants in a tea plantation, comprising: retrieving initial condition data indicative of initial nitrogen accumulation in the tea plantation, comprising nitrogen retained as tea plant tissue, soil inorganic nitrogen, and soil organic nitrogen; predicting a nitrogen deficiency level, which indicates the level at which lack of nitrogen availability will limit the growth of the tea plant; and predicting tea plant growth using the initial condition data and nitrogen deficiency levels. 2. The method further includes: a. retrieving nitrogen addition data indicating the amount of nitrogen added to the soil of the tea plantation; b. predicting nitrogen transformation parameters that describe the transformation of nitrogen in the soil between chemical forms by one or more of mineralization, nitrification, or denitrification; c. predicting tea plant nitrogen loss, indicating nitrogen lost by the tea plant due to one or more of tea plant harvesting, tea plant pruning, exudation, or abscission; or d. The method according to clause 1, comprising one or more steps of predicting soil nitrogen loss, indicating nitrogen loss due to one or more of denitrification, ammonia volatilization, or leaching. 3. A method according to either clause 1 or 2, in which the initial nitrogen stock is initialized using defined levels of soil mineral nitrogen at a range of soil depths. 4. Nitrogen addition data a. The amount of new carbon in each soil layer; b. Percentage of carbon in new organic matter; c. A method according to clauses 2 or 3, providing for the amount of new organic nitrogen added to the soil per soil layer. 5. The method according to any one of clauses 2 to 4, further comprising the step of calculating a carbon to nitrogen ratio using the initial nitrogen accumulation or nitrogen addition data. 6. The method according to any of clauses 1 to 5, wherein the step of predicting the growth of the tea plant further comprises calculating using one or more of the potential growth rate of the tea plant, temperature, or soil moisture level. 7. The method according to clause 6, further comprising the step of predicting the potential growth rate of the tea plant, wherein the potential growth rate is calculated using one or more of the amount of solar radiation, the radiation use efficiency, the light extinction coefficient, or the leaf area index. 8. The method according to any of clauses 1 to 7, wherein the step of predicting the nitrogen deficiency level comprises calculating using one or more of a critical canopy nitrogen level, a percentage of nitrogen in the canopy, or a minimum canopy nitrogen content level. 9. The step of predicting a nitrogen deficiency level comprises predicting a nitrogen demand for each tea plant, the step of predicting the nitrogen demand comprising: a. predicting homeostatic nitrogen demand of the tea plant canopy, comprising calculating using one or more of canopy mass, critical nitrogen level, and nitrogen rate in the tea plant canopy; b. predicting the homeostatic nitrogen demand of the tea plant roots, comprising calculating using one or more of root mass, critical nitrogen level, or nitrogen rate in the tea plant roots; c. predicting canopy growth nitrogen demand of the tea plant, comprising calculating using the canopy potential growth rate and / or the canopy critical nitrogen level; or d. The method according to any of clauses 1 to 8, comprising one or more steps of predicting the root growth nitrogen demand of the tea plant, comprising calculating using the root growth potential rate and / or the root critical nitrogen level. 10. The method according to any of clauses 1 to 9, wherein predicting the growth of the tea plants comprises predicting the growth of a plurality of leaves for each tea plant. 11. The method according to any one of clauses 2 to 10, wherein the nitrogen conversion parameters comprise parameters for expressing nitrogen mineralization, and the method comprises calculating the nitrogen mineralization parameters using one or more of the amount of new organic nitrogen, the organic matter pool fraction, the total organic matter, the temperature coefficient, the moisture coefficient, the carbon:nitrogen ratio coefficient, the decay rate constant, the concentration of ammonium, the concentration of nitrate, or the amount of new humic nitrogen. 12. The method according to any one of clauses 2 to 11, wherein the nitrogen conversion parameters comprise parameters for describing nitrification, and the method comprises calculating the nitrification parameters using one or more of substrate ammonium level, oxygen level, soil pH, temperature coefficient, or soil moisture coefficient. 13. The method according to any one of clauses 2 to 12, wherein the nitrogen conversion parameters comprise parameters for describing denitrification, and the method comprises calculating the denitrification parameters using one or more of ammonium concentration, pH, temperature, soil moisture content, or the amount of ammonium above a minimum value. 14. The method according to clause 12 or 13, further comprising calculating a parameter to represent loss from N2O. 15. The method according to any one of clauses 2 to 14, wherein predicting nitrogen loss comprises predicting nitrogen volatilization, and predicting nitrogen volatilization comprises calculating using one or more of ammonia concentration in soil moisture, wind speed, soil temperature, or soil depth. 16. A computer program configured to carry out the method of any one of clauses 1 to 15. 17. A storage medium containing instructions for carrying out a method according to any one of clauses 1 to 15 on a computer processor.

[0485] (References)

[0486] appendix Appendix A - Calculation of Water Stress Factors Please refer to Figure 22. The calculated water stress factor is recorded below as it appeared to go from drought too quickly with only a small amount of rainfall. 1. 6 mm of rain on November 14, 1989 meant that the water stress factor rose from 0 to 0.255. There had been no rain in the 11 days prior to this or the 5 days after. Water stress Factor (Wsf) = actual water uptake / potential transpiration on 07 / 11 / 1989 Wsf = 0 as in drought: Wsf = 0 / 4.7375 on 14 / 11 / 1989 wsf = 0.255 due to 6mm of rain: wsf = 0.7348 / 2.878 The difference between the two dates is that the actual fluid intake was no longer zero on the later day. 2. Calculate your actual fluid intake: MaxInflow = 0.03; swcon1 = 0.00264; swcon2 = 62; swcon3 = 6.68; pwu = 0 SoilData for each soil zone (20 zones) if soilwater > LowerLimit inflow = SWcon1 * Exp(lower of (Soilwater-LowerLimit or 10)) / SWcon3 - Log(r-otdensity) else inflow = 0 inflow must be no greater than MaxInflow rwu = inflow * rootdensity * Zone depth * 10 * (0.18+0.00272 * sqr(rootdensity -18) rwu = lower of rwu or (soilwater-LowerLimit)*zone depth * 10 pwu = pwu + rwu if pwu > 0 and potential transpiration > 0 ep = lower of pwu and potential transpiration wuf = Ep / pwu = 1 for each soil zone (20 zones) rwu = rwu * wuf soilwater = soilwater - rwu / (-epth * 10) When SoliWater<=LowerLimit, the actual water intake is set to 0. On July 11, 1989, soliwater was <=lowerlimit, but on November 14, 1989, soliwater was >lowerlimit. 3. Soil moisture and lower bounds are read from the input file and then interpolated between the 20 zones. [Table 55]

[0487] 4. In three locations, only soilwater is adjusted by the model: Soil moisture distribution b. Soil water evaporation c. Actual fluid intake (highlighted above in 2) 5. Calculation of soil moisture distribution: Procedure for calculating soil moisture distribution in a soil profile after water has been added by irrigation or rain or snow. Drainage (if any) is also calculated. All water downward flow is in mm. SWCON is the soil moisture drainage constant (fraction drained per day) from the Soil data file. for each soil zone (20 zones) fac = Zone * 10 Qsat = SatWC * fac Qupper = Upper limit * fac Qactual = soilwater * fac Downflux = (Qactual - Quppe-) * SWCON (if Qactual > Qupper) Downflux = Downflux + (qactual + Waterin) - Qsat - (if Qactual + WaterIn > Qsat) if downflux > 0 draining = true Qactual = qactual + waterin - downflux waterin = downflux soilwater = Qactual / fac Waterin = 6mm of rain 6. Soil evaporation: Procedure for calculating soil evaporation (Es) in mm. The inputs are LAI (Leaf Area Index), Es1max (upper limit of stage 1 evaporation), WaterIn (rain + irrigation + snow), sumEs1 (cumulative stage 1 soil evaporation), sumEs2 (cumulative stage 2 soil evaporation), Eo (potential evaporation), t (time counter for stage 2 soil evaporation), swc (soil water content array). The model was based on Ritchie's (1972, Water Resources Res. 8:1204), but modified to further reduce Es when the SWC in the upper layers reaches a fixed low threshold (swef). This was needed when simulating stratified soils to prevent the soil surface from drying out too much when roots are also removing water from near the surface. Swef is set to 0.9 - 0.00038 * (depth of top soil layer - 30) * (depth of top soil layer - 30) = 0.89050000000000007. The problem appears to be either with the initial setting of the lower limit or with the recalculation of soil moisture. The tea growth model uses an adjusted water stress factor that includes a maximum delay of 5 days after re-watering.

[0488] Appendix B - Field data on number of shoots harvested

[0489] [Table 56A] [Table 56B]

[0490] Appendix C - Machine Harvest Calculations When Mode=0, mechanical harvesting is simulated. The only other important parameter in this case is the variable CutHt, which is the height (in cm) above the ground at which harvesting takes place. This value is obtained by adding the reaper height (in cm) to the platform height (in cm). All other parameters are not important and are set to 0 or default values ​​for mechanical harvesting. The "cut-point", the position at which the shoot is cut relative to its base height, is determined by subtracting the base height of the shoot from CutHt, giving the length of the remaining shoot. The number of leaves per shoot that are harvested is determined according to whether their base position is higher than this cut-point. It is currently assumed that the leaves themselves are not cut off by the harvesting process. - Description of machine harvesting from manual This gives the following: [Table 57]

[0491] 1. Regarding field 17, June 15, 2015: a. The mechanical harvesting routine predicted a yield of 12.46 kg / ha per week b. A manual harvesting routine predicted a yield of 159 kg / ha per week c. Actual yield was 355 kg / ha per week 2. Comparison of hand-picked and mechanically harvested grapes on June 15, 2015 a. The machine routine picked 359 of the 500 shoots. b. The hand-picking routine resulted in 187 of 500 shoots being picked. 3. However, the machine requires much fewer leaves to harvest: 100 leaves compared to 1,698 leaves picked by hand. 4. Mechanical harvest - 38 of 500 shoots at growth stage 2 5. Hand-picked - 258 of 500 shoots at growth stage 2 Whenever a shoot is picked, it resets its leaves to 0 and its growth stage to 0. Growth stages and leaves increase based on: 1. ob_S1DR * (DDIdev * DLfac * s1DRwf) 2. ob_S2DR * (DDIdev * DLfac * s2DRwf) Here, on June 15, 2015: ddidev = 0.637260232, daily increase in development days for phenology dlfac = 0.992648663(daily) ... (1-ShootDLsensitivity*(ShootDLcrit-Photoperiod)) s1drwf = 1, water stress reduction factor s2drwf = 1, water stress reduction factor s1dr = between 0.025 and 0.06, stage 1 developmental rate s2dr = between 0.1 and 0.35, stage 2 developmental rate

[0492] Appendix D. Proposed changes in the CUPPA-Tea Code The changes include modified methods for determining phyllochron-based harvest intervals, minimum harvestable shoot height, and additional quality indicator outputs. Red text - modified code Black text - existing code Overview of the changes that will be made 1. Add the last day of pruning to the model parameters 2. Add harvesting method to the model parameters 3. Change ShearHeight from 5 to 3 and include additional quality indices 4. Calculate the days since the initial pruning in teagrowth initial settings 5. Calculate daily leaf emergence rate based on temperature and soil moisture deficit 6. Calculate the phyllochron multiplier based on days since pruning and harvesting method 7. Accumulate leaf emergence rate and set it to 0 at harvest time. 8. Compare the accumulated leaf emergence rate with the phyllochron multiplier Data to be passed to CUPPA-Tea Adding the last pruning date and harvesting method to the model parameters 1. Last pruning date in DateTime format, for example: "12 / 04 / 2019" 2. Harvesting method set to 1, 2, or 3, where: a. 1 = mechanical harvest b. 2=Reaper c. 3=hand 3. config file "CropManagementParameters":{ "ShearHeight": 5, from "CropManagementParameters":{ "ShearHeight": 3, Change to Changes to DSS 1. Harvesting interval is when "SumLAR" is >="Phyllo" 2. To predict the following quality parameters output forward: QI2toall = (number of 2 banj + number of 2 grow) / (number of 2 banj + number of 2 grow + number of 3 banj + number of 3 grow + number of 4 or greater banj + number of 4 or greater grow) Code changes TeaModel.cs 1. In the TeaModel routine, add the start date to the tea growth _teaGrowthobj._endDate = endDate; _teaGrowthobj._startDate = startDate; 2. In the GetInputData routine, add the pruning date and harvesting method. _waterobj.IrrigLevel = config["CropManagementParameters"]["IrrigationThreshold"]; _teaGrowthobj._prunedate = config["CropManagementParameters"]["PruneDate"]; _teaGrowthobj._harvestMethod = config["CropManagementParameters"]["HarvestMethod"]; OutputData.CS 1. In Update Overview, add new accumulated leaf emergence rate, phyllochron multiplier, and daily leaf emergence rate: record.Add("5+b", Math.Round(teaData.PluckDist[5], 0)); record.Add("SumLAR", Math.Round(teaData._LAR, 2)); record.Add("Phyllo", Math.Round(teaData._Phyllomult, 2)); record.Add("DailyLAR", Math.Round(teaData._dailyLAR, 2)); Teagrowth.cs 1. Add leaf emergence rate, phyllochron multiplier, harvest method, daily LAR, pruning date, and start date as class variables. public DateTime _endDate { get; set;} public double _LAR { get; set;} public double _Phyllomult { get; set;} public Int16 _harvestMethod { get; set;} public double _dailyLAR { get; set;} / / AH 26 / 02 / 2020 add prune date public DateTime _prunedate { get; set;} public DateTime _startDate { get; set;} 2. Add a private variable to the class for the pruning cycle private double _aSenescedWt = 0; / / 26 / 02 / 2020 Calculate days from prune private int _daysfromprune = 0; const int _prunecycle = 4; / / number of years in a prune cycle 3. Number of days since initial pruning in Init Calculate double TRW; double THtSD; int NS; / / Calculate days from prune, start from day before sim start date as day added in growth routine / / and allow for prune dates before simulation start date if (_prunedate > _startDate) { _daysfromprune = (_prunecycle * 365) + (_startDate - _prunedate).Days - 1;} else { _daysfromprune = (_startDate - _prunedate).Days - 1;} 4. At the beginning of the Growth routine: public void Growth(DateTime simDate, int das, int cYear, Weather weatherData, Water waterData) { / / AH 26 / 02 / 20 Days from prune _daysfromprune += 1; if (simDate == _prunedate) { _daysfromprune = 0;} 5. In Growth, after ShootDevelopment and before Prunedate check, calculate the leaf emergence rate and phyllochron multiplier, and set the multiplier to 8 if the simulation day is a pruning day: _bushobj.ShootDevelopment(DDIdev, DDIlen, DLfac, DeltaDL, _sDfac, _wSFadj, _ndef2, das); _bushobj.AllocateDM(_dryMatterOn); / / {dry matter balance} / / {with bushData^ do / / begin / / writeln(das:6,pwt:6:1,ShootWeight:6:1,ShootGR:6:1,PluckWt:6:1,BreakWt:6:1, / / MaintenanceLeafWt-mlw:6:1, / / ' ',ShootWeight-pwt-ShootGR+PluckWt+BreakWt+(MaintenanceLeafWt-mlw):12:8); / / pwt=ShootWeight; mlw=MaintenanceLeafWt; / / end;} / / Leaf appearance rate double dailyLAR = 0; if (weatherData.Tmean < 12) { dailyLAR = 0; else { dailyLAR = 0.186 - 4.6 * Math.Pow(0.765, weatherData.Tmean);} double f = 0; if (waterData.Swdef < 52.15) { f = 1;} else { f = 1.218 - 0.161 * Math.Pow(1.00584, waterData.Swdef);} _LAR += (dailyLAR * f); _dailyLAR = (dailyLAR * f); / / default multiplier to 8 if in prune cycle and before first harvest if (_phenoCode == "Pr") { _Phyllomult = 8; } else { if (_harvestMethod == 1) { _Phyllomult = 2.73 + _daysfromprune * 0.0006694;} else if (_harvestMethod == 2) { _Phyllomult = 1.26 + _daysfromprune * 0.00118;} else { _Phyllomult = 1.26 + _daysfromprune * 0.00118;} } / / {if date of pruning then prune bushData} if (PruneDate(simDate)) { _bushobj.Prune(); _phenoCode = "Pr"; _phenoString = "Pruning"; _Phyllomult = 8; } 6. In Growth, reset leaf appearance rate to 0 on harvest day. / / {if date of plucking then pluck bushData} if (PluckDate(simDate) && (das != 0)) { / / with bushData do _bushobj.Pluck(_maxPluckLeaves, _pluckDist, _pluckBanji, das); _phenoCode = "Pl"; _phenoString = "Plucking"; _LAR = 0; } 7. If the simulation date is the pruning date in the PruneDate routine check: private bool PruneDate(DateTime SimDate) { / / return false; return _prunedate == SimDate; }

[0493] Appendix E - Explanation of initial focus on shot counts The first approach attempted to address the impact of yield change with pruning was to vary the shoot number multiplier in CUPPA-Tea. However, this approach was not developed because the basal height approach seemed more promising and it is good practice to vary only one parameter at a time. In theory, adding a harvest interval that differs by the time since pruning (Burgess et al., 2020) would address some pruning effects. The analyses performed are briefly reported here for the record. Test B1 - Recalculate shoot number based on time since pruning The CUPPTA-Tea model was developed for tea in the second or third year after pruning. Analysis of actual yield data from the Kericho estate in 2015, 2016, and 2017 showed that green tea leaf yield increased with the number of years since pruning (Figure 153). Equation 49: Green tea leaf yield = 8464 kg / ha + 1657 * years since pruning This relationship indicates that on the day of harvest (year=0), the yield is 0.7186 (i.e., 8464 / (8464+3314)) of the yield two years after pruning. method The first test was to simply correct shoot number based on time since pruning. Field 5 was used to test the changes, assuming a pruning date of February 15, 2018, 225 clones, mechanical harvesting, and a default shoot number of 650 shoots per tea plant. At the beginning of the model run: 1. Calculate the number of days since pruning: If the pruning date is before the model start date, then the number of days since pruning = (5 years * 365) + the difference in dates between the start date and the pruning date - 1 b. If the pruning date is later than the model start date, then the number of days since pruning = the difference in dates between the start date and the pruning date - 1 2. Remember the original number of shots 3. Calculate the number of shoots based on the original number of shoots and the number of days since pruning using the formula: Number of shoots = original number of shoots / 1.392 + 0.000386 * original number of shoots * number of days since pruning. This is 0.718SN original +(1-0.718) / (2*365)*SN original *Equivalent to the number of days since pruning. 4. The above formula was derived from a yield formula derived from actual harvest data from Kericho for 2015-2017, based on the assumption that the number of shoots currently being used is the shoots 24 months after pruning: y=1657.6x+8464.3, y=yield and x=year of pruning result The above procedure should result in a steady increase in yield over time since pruning. Immediately after pruning, the shoot number multiplier is 467 (right-hand scale) compared to the default shoot number of 650 (Figure 147). The result of this analysis is that from the pruning on February 15, 2018, all modeled shoots are synchronized, with a very large spurious yield peak at the first harvest. Test B2 - Modeled Chute Adjustment The second trial focused on reducing the modeled number of shoots to 0 after pruning and adding one new modeled shoot each day until the shoot number reached 500. In trial 3, the results were also multiplied by the predicted number of shoots. method In this test, the number of modeled shoots is reduced to 0 on the day of pruning. 1) The base height remains at 75cm. 2) One additional shoot is modeled each day after pruning (starting from 0) until 500 shoots are reached (Figure 148). 3) Predicted yield is determined by multiplying the modeled shoots by the number of shoots. result Adding an extra shoot each day removes the synchronization of the shoots, and the modeled predicted yield appears similar to the actual yield (Figure 149). Test B3 - Test 2, but assuming a 4-year cycle and adding two shoots each day Trial 3 was to assume a 4-year pruning cycle instead of 5 years and to investigate the effect of adding two modeled shoots each day. method In this test, the modeled shoot number is reduced to 0 on the day of pruning. The basal height remains at 75 cm. However, the procedure 1) Two additional shoots are modeled each day after pruning (starting from 0) 2) Pruning is assumed to follow a 4-year pruning cycle instead of a 5-year cycle. Assuming a 4-year pruning cycle reduces the assumed number of shoots immediately prior to pruning from 924 to 833 shoots / plant (Figure 150). 3) The actual number of shoots is the modeled number of shoots until it reaches 500 shoots / tea plant. This differs from Test 2 in that respect. result Moving to the assumption of pruning at 4 years instead of 5 years reduces the predicted yield before pruning to the limit (Figure 151). Test B3 has a much smaller shoot multiplication coefficient, resulting in a lower initial predicted yield after pruning.

[0494] Appendix F - Crop Management Parameter Description [Table 58]

[0495] Appendix G - Parameter Table [Table 59] [Table 60] [Table 61] [Table 62] [Table 63] [Table 64] [Table 65] [Explanation of symbols]

[0496] 110 Tea Harvest Management System 111 processors 112 memory 120 model 130 Datastores 140 User Input 150 sensor data 160 shots 170 Output Module 180 display

Claims

1. 1. A computer-implemented system for managing a tea harvest, comprising: a. a data store storing data defining one or more growing areas of a tea plantation; b. a growth and yield prediction module configured for each of the growing areas of the tea plantation defined in the data store to model a plurality of tea plants within the respective growing area, thereby predicting the growth of each tea plant within the respective growing area over a period of time, and therefore the yield of each growing area of ​​the tea plantation when harvested; c. a growth and yield prediction module further configured to predict a recommended harvest time to achieve a desired tea quality and a desired tea yield; d. a user interface module configured to provide a user with the predicted yield of each growing area of ​​the tea plantation when harvested; A system comprising:

2. The system of claim 1 , wherein the user interface module displays the geographic data of each growth area, preferably as a map.

3. 3. The system of claim 1, wherein the data defines a tea plantation comprising a plurality of growing areas, and the user interface module is configured to enable a user to select a growing area from the plurality of growing areas, and to cause the system to display predictions made for the selected growing area.

4. 4. The system of claim 1, wherein the growth and yield prediction module is further configured to predict a recommended harvest interval, shoot number, and harvestable shoot height for each tea plant within each growing area.

5. 5. The system of claim 1, wherein the growth and prediction module predicts a quality index of the harvested tea plants, preferably the quality index being calculated as the percentage of shoots in a favorable condition for harvesting to the total number of shoots.

6. 6. The system of claim 1, wherein the growth and yield prediction module predicts the number of shoots to be harvested based on the number of shoots having more leaves than a predetermined threshold and / or the height of each shoot.

7. 7. The system of claim 1, wherein the growth and yield prediction module predicts leaf emergence rate or phyllochron time.

8. 8. The system of claim 1, wherein the growth and yield forecast is configured to incorporate the effect of harvesting method on the growth of the tea plant, preferably the harvesting method being mechanical harvesting, reaper harvesting, or hand picking.

9. 9. The system of claim 1, wherein the system comprises a data input module configured to receive data relating to each growth region, the data input module providing the data to the growth and forecasting module.

10. 10. The system of claim 9, wherein the data input module comprises an environmental data module configured to receive environmental data for the respective growing area, preferably wherein the environmental data comprises observed or forecasted weather data.

11. 12. A system according to claim 10 or 11 dependent on claim 8, wherein the leaf emergence rate or phyllochron time is calculated based on the environmental data, preferably based on air temperature or soil water deficit level.

12. the data input module comprising a remotely sensed data module configured to receive remotely sensed data comprising one or more observations of the respective growth regions; Preferably, the remote sensing module detects a harvesting status of the respective growing area; 13. The system of any one of claims 9 to 12, wherein the remote sensing module detects a first area of ​​the respective growing area as being to be harvested and a second area of ​​the respective growing area as not to be harvested.

13. 14. The system of any one of claims 12 to 13, wherein the remote sensing module detects the type of crop used in the respective growing area or detects the level of growth in the respective growing area.

14. the growth and yield prediction module predicts the effect of pruning on the growth and yield of the tea plant; Preferably, the growth and yield prediction module is configured to model each tea plant as having at least one shoot, and the growth and yield prediction module is configured to: a. modeling the length of the chute and shortening the length of the chute; b. modeling the developmental stage of the shoot and setting the developmental stage to an early stage; and / or c. modeling a base height between the bottom of the chute and the soil surface and lowering the base height of the chute.

14. The system of claim 1, wherein the system models the effects of pruning by one or more of:

15. obtaining data defining one or more growing areas of a tea plantation from a data store; - by a growth and yield prediction module, for each of the growing areas of the tea plantation defined in the data store, modelling a plurality of tea plants within the respective growing area, thereby predicting the growth of each tea plant within the respective growing area over a period of time, and therefore the yield of each growing area of ​​the tea plantation when harvested; predicting, by a growth and yield prediction module, a recommended harvest time to achieve a desired tea quality and a desired tea yield; and providing a user via a user interface module the predicted yield of each growing area of ​​the tea plantation when harvested.

16. 1. A computer-implemented method for predicting growth of a plurality of tea plants in a tea plantation, comprising: retrieving initial condition data indicative of initial nitrogen accumulation in said tea plantation, comprising nitrogen retained as tea plant tissue, soil inorganic nitrogen, and soil organic nitrogen; predicting a nitrogen deficiency level, the level indicating a level at which lack of nitrogen availability will limit the growth of the tea plant; and predicting the growth of the tea plant using the initial condition data and the nitrogen deficiency level.

17. a. retrieving nitrogen addition data indicative of the amount of nitrogen added to the soil of the tea plantation; b. predicting nitrogen transformation parameters that describe the transformation of nitrogen in the soil between different chemical forms by one or more of mineralization, nitrification, or denitrification; c. predicting tea plant nitrogen loss indicating nitrogen lost by the tea plant due to one or more of harvesting the tea plant, pruning the tea plant, exudation, or abscission; or d. predicting soil nitrogen loss, which indicates nitrogen loss due to one or more of denitrification, ammonia volatilization, or leaching.

17. The method of claim 16, comprising one or more of:

18. 18. The method of claim 16 or 17, wherein the initial nitrogen accumulation is initialized at designated levels of soil inorganic nitrogen at a range of soil depths.

19. The nitrogen addition data is Amount of new carbon per soil layer, b. the proportion of carbon in new organic matter, and c. The amount of new organic nitrogen added to the soil per soil layer 18. The method of claim 16 or 17, comprising:

20. 20. The method of any one of claims 16 to 19, further comprising calculating a carbon to nitrogen ratio using the initial nitrogen accumulation or the nitrogen addition data.

21. The step of predicting the nitrogen deficiency level comprises predicting the nitrogen demand of each tea plant, wherein predicting the nitrogen demand comprises: a. predicting the homeostatic nitrogen demand of the canopy of the tea plant, comprising calculating using one or more of a canopy weight, a critical nitrogen level, and a nitrogen rate in the canopy of the tea plant; b. predicting the homeostatic nitrogen demand of the roots of the tea plant, comprising calculating using one or more of the root weight, the critical nitrogen level, or the nitrogen rate in the roots of the tea plant; c. predicting the canopy growth nitrogen demand of the tea plant, comprising calculating using the canopy potential growth rate of the canopy and / or the critical nitrogen level; or d. predicting the root growth nitrogen demand of the tea plant, comprising calculating using the root potential growth rate of the roots and / or the critical nitrogen level.

21. The method of any of claims 16 to 20, comprising one or more of:

22. The nitrogen conversion parameter is a. providing parameters for describing nitrogen mineralization, the method comprising calculating the nitrogen mineralization parameters using one or more of the amount of new organic nitrogen, the organic matter pool fraction, the total organic matter, a temperature coefficient, a moisture coefficient, a carbon-to-nitrogen ratio coefficient, a decay rate constant, the concentration of ammonium, the concentration of nitrate, or the amount of new humic nitrogen; b. parameters for describing nitrification, the method comprising calculating the nitrification parameters using one or more of the ammonium level, oxygen level, soil pH, temperature coefficient, or soil moisture coefficient of the substrate; c. parameters for describing denitrification, the method comprising calculating the denitrification parameters using one or more of ammonium concentration, pH, temperature, soil moisture content, or the amount of ammonium above a minimum value; and / or dN 2 22. The method of claim 16, comprising a parameter for expressing losses from O.

Images