Water stress estimation method, water stress estimation device, machine model learning method, machine model, water stress estimation program, and recording medium

Abstract

To estimate a water stress value in consideration of water absorption from a ground part of a tree when a root area is dried in outdoor cultivation of citrus.SOLUTION: A water stress estimation method (S1) includes a water stress value acquiring step (S11) of acquiring an actually measured water stress value of a fruit tree on a certain date, a meteorological datum acquiring step (S12) of acquiring an actually measured and / or predicted meteorological datum including a leaf wet time which is a time during which surfaces of leaves of the fruit tree are wet with water, and an estimating step (S13) of estimating the water stress value of the fruit tree after the last date of the time series datum by inputting the water stress value and the actually measured and / or predicted meteorological datum into a machine model.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to a water stress estimation method, a water stress estimation device, a machine model learning method, a machine model, a water stress estimation program, and a recording medium. [Background technology]

[0002] Citrus fruits such as unshu mandarins are in high demand in the market. It is known that these fruits can be produced with high sugar content by applying water stress during the summer. However, excessive water stress can lead to high oxidation of the fruit and a decline in tree vigor due to a lack of acid reduction, so it is necessary to maintain a moderate level of water stress.

[0003] Technological developments for methods of estimating water stress from meteorological conditions, etc., are progressing in greenhouse cultivation, and methods have been developed to estimate the degree of water stress based on meteorological conditions within the facility, such as temperature, humidity, vapor pressure deficit, and light intensity, as well as work details such as irrigation volume, and biological data such as stem diameter (e.g., Patent Documents 1 and 2).In addition, a method has been proposed for estimating and calculating necessary agricultural parameters in open-field environments using meteorological conditions, including precipitation (e.g., Patent Document 3). [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Re-tabled publication 2019 / 156070 [Patent Document 2] Japanese Patent Application Publication No. 2019-193583 [Patent Document 3] Patent Publication No. 2021-190108 Summary of the Invention [Problem to be solved by the invention]

[0005] One cultivation method for inducing water stress in citrus is to prevent rainwater from entering the root zone of the crop and dry out the root zone, thereby inducing water stress. However, although plants primarily absorb water from underground, if the above-ground parts are wet due to rainfall or other factors, they can supplementarily absorb water through leaves, etc. Conventional water stress estimation models for greenhouse cultivation do not take rainfall into account, making it difficult to directly introduce conventional water stress estimation models into the above-mentioned citrus cultivation method, which involves drying out the root zone despite being grown outdoors.

[0006] According to the inventors' findings, there are currently no models that can estimate water stress from meteorological conditions for outdoor-grown crops. While it may be possible to apply existing models that estimate other parameters to water stress, only citrus plants are subjected to treatments that impart severe water stress, and most crops allow rainwater to flow into their root zones. For this reason, existing models for outdoor cultivation do not need to take into account supplemental water absorption from leaves, making them difficult to apply to the above-mentioned citrus cultivation system.

[0007] One aspect of the present invention aims to provide a technology for estimating water stress values ​​in outdoor cultivation of citrus trees when the root zone is dried, taking into account water absorption from the above-ground parts of the tree. [Means for solving the problem]

[0008] In order to solve the above problem, a water stress estimation method according to one embodiment of the present invention includes a water stress value acquisition step of acquiring the measured water stress value of a fruit tree on a certain date; a weather data acquisition step of acquiring measured and / or predicted weather data, which is time series data of weather from the certain date onwards, including leaf wetness time, which is the time when the surface of the leaves of the fruit tree is wet with water; and an estimation step of inputting the water stress value and the measured and / or predicted weather data into a machine model and estimating the water stress value of the fruit tree after the last date of the time series data has passed.

[0009] A water stress estimation device according to one embodiment of the present invention comprises an acquisition unit that acquires the measured water stress value of a fruit tree on a certain date and measured and / or predicted weather data from the certain date onwards, the measured and / or predicted weather data including leaf wetness time, which is the time during which the surface of the leaves of the fruit tree is wet with water; a mechanical model that estimates the water stress value of the fruit tree after the last date of the time series data has passed from the water stress value and the measured and / or predicted weather data; and an output unit that outputs the estimated value estimated by the mechanical model.

[0010] A method for training a machine model according to one embodiment of the present invention includes the steps of acquiring multiple sets of training data, each set consisting of the measured water stress value of a fruit tree on a certain date, measured time series data on the weather from the certain date onwards, the measured weather data including leaf wetness time, which is the time when the surface of the leaves of the fruit tree is wet with water, and the measured water stress value of the fruit tree after the last date of the time series data has passed; and a learning step of updating the parameters of the machine model so that the value output when the measured water stress value of the fruit tree on the certain date and the measured weather data from the certain date onwards are input into the machine model approaches the measured water stress value of the fruit tree after the last date of the time series data has passed.

[0011] A second water stress estimation method according to one embodiment of the present invention includes a weather data acquisition step of acquiring measured and / or predicted weather data, which is time series data on weather from a certain date onwards, including leaf wetness time, which is the time during which the surface of the leaves of a fruit tree is wet with water, and an estimation step of inputting the measured and / or predicted weather data into a machine model and estimating the amount of change in the water stress value of the fruit tree after the last date of the time series data has passed.

[0012] A second water stress estimation device according to one embodiment of the present invention comprises an acquisition unit that acquires measured and / or predicted weather data, which is time series data on weather from a certain date onwards, including leaf wetness time, which is the time during which the surface of the leaves of a fruit tree is wet with water; a mechanical model that estimates the amount of change in the water stress value of the fruit tree from the measured and / or predicted weather data after the last date of the time series data has passed; and an output unit that outputs the estimated value estimated by the mechanical model.

[0013] A second machine model learning method according to one embodiment of the present invention includes a step of acquiring multiple sets of learning data, each set consisting of the measured water stress value of a fruit tree on a certain date, measured time series data on the weather from the certain date onwards, the measured weather data including leaf wetness time, which is the time when the surface of the leaves of the fruit tree is wet with water, and the measured water stress value of the fruit tree after the last date of the time series data has passed; and a learning step of updating the parameters of the machine model so that the value output when the measured weather data from the certain date onwards is input into the machine model approaches the change in the measured water stress value of the fruit tree after the last date of the time series data has passed from the measured water stress value of the fruit tree on the certain date.

[0014] The water stress estimation device according to each aspect of the present invention may be realized by a computer. In this case, the control program that realizes the water stress estimation device on a computer by causing the computer to operate as each part (software element) of the water stress estimation device, and the computer-readable recording medium on which the control program is recorded, also fall within the scope of the present invention. [Effects of the Invention]

[0015] According to one aspect of the present invention, when the root zone of citrus trees is dried in outdoor cultivation, the water stress value can be estimated taking into account water absorption from the above-ground parts of the tree. [Brief explanation of the drawings]

[0016] [Figure 1] 1 is a block diagram showing the configuration of a water stress estimation device 1 according to a first embodiment. [Figure 2] 1 is a flowchart showing the flow of a water stress estimation method S1 according to the first embodiment. [Figure 3] FIG. 10 is a schematic diagram illustrating three estimation patterns of estimation method S1. [Figure 4] 10 is a flowchart showing the flow of a machine model learning method S2 according to the first embodiment. [Figure 5] FIG. 10 is a block diagram showing the configuration of a water stress estimation device 1A according to a second embodiment. [Figure 6] 10 is a flowchart showing the flow of a water stress estimation method S3 according to the second embodiment. [Figure 7] 10 is a flowchart showing the flow of a machine model learning method S4 according to the second embodiment. [Figure 8] This is a graph showing the results of estimating water stress values ​​using a machine model trained using weather data excluding leaf wetness time. [Figure 9] 10 is a graph showing the results of estimating water stress values ​​using a machine model trained using meteorological data excluding precipitation. [Figure 10] 10 is a graph showing the results of estimating water stress values ​​using a machine model trained to take into account water stress values ​​at the start of a date. [Figure 11] 10 is a graph showing the results of estimating water stress values ​​using a machine model trained without considering water stress values ​​at the start of the date. DETAILED DESCRIPTION OF THE INVENTION

[0017] [Embodiment 1] An embodiment of the present invention will be described in detail below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of a water stress estimation device 1 according to this embodiment. The water stress estimation device 1 according to this embodiment is a device that estimates the water stress value of citrus trees grown in an open-field cultivation method that dries the root zone by referring to predicted and / or measured weather data. The phrase "A and / or B" means both A and B, or either one of them. An open-field cultivation method that dries the root zone is a cultivation method that, although open-field, covers the ground surface around the tree trunk with a sheet or the like to prevent rainwater and other elements from seeping into the ground. Examples of such cultivation methods include sheet mulch cultivation, shielding mulch cultivation (S. mulch), and root-zone restricted cultivation. These methods involve covering the ground surface with a covering material or planting trees in a water-tight frame with a mound of soil. All of these methods prevent rainwater from entering the root zone. However, as mentioned above, even if waterproofing treatment is applied to the root zone, it is necessary to consider supplementary water absorption from the above-ground parts of the tree, such as the trunk and leaves. The water stress estimation device 1 is a device that can estimate water stress values ​​by taking into account water absorption from the trunk, leaves, etc.

[0018] (Water stress value) First, the water stress value will be described. The water stress value indicates the degree of water deficiency within a plant. The water stress value is measured, for example, by pressurizing a leaf with gas using a pressure chamber method and observing the air pressure at which water escapes from the petiole. Because the water stress value varies depending on the time of day, it is typically measured early in the morning (3:00 to 5:00 AM), when the plant is most supplied with water. However, since measuring every day during this time period is a laborious task, it is expected that the water stress value will be estimated mechanically. Conventionally, a large amount of intermittent water stress value data measured over a period of time has been accumulated as actual measurements. Therefore, there is a wealth of data available for training a mechanical model using the intermittent measured water stress values. In other words, there is an advantage in that training data for the water stress estimation device 1 according to this embodiment can be easily collected.

[0019] (Water stress estimation device 1) As shown in FIG. 1 , the water stress estimation device 1 includes an acquisition unit 10, an estimation unit 20, an output unit 30, a communication unit 40, and a control unit 60. The water stress estimation device 1 may also include a learning unit 50. An external display device 70 may be connected to the water stress estimation device 1. Alternatively, the water stress estimation device 1 may also include a display unit (not shown). The water stress estimation device 1 may also be connectable to the Internet 80 or the like. The acquisition unit 10 may also be capable of acquiring water stress value data 91 and / or weather data 92 and / or irrigation work data 93 via the Internet 80 or the like and storing the data in the memory 62. Furthermore, each of the above-mentioned units may be located in a single device, or may be distributed across multiple devices. Furthermore, at least some of the units, except for the communication unit 40, may be located on the cloud.

[0020] The acquisition unit 10 includes a water stress value acquisition unit 11 and a meteorological data acquisition unit 12. The water stress value acquisition unit 11 acquires the water stress value of the fruit tree on a certain date based on a user operation. The certain date is the start date and time of a certain continuous period, and may be, for example, a date. On the other hand, the last date is the end date and time of a certain continuous period, and may be, for example, a date. The meteorological data acquisition unit 12 acquires time-series data of meteorology from a certain date onward, including actually measured and / or predicted meteorological data including leaf wetness time, which is the time during which the surface of the fruit tree's leaves is wet with water, based on a user operation. The time-series data of meteorology is meteorological data including chronological information (year, month, day, etc.) for a continuous period. The meteorological data may be an accumulation of daily data for a continuous period. The time-series data is, for example, data in which daily data is arranged chronologically. The certain date may be, for example, a date preceding the predicted harvest date by a predetermined period. The predetermined period may be, for example, a period during which water stress affects fruit growth.

[0021] The acquisition unit 10 may further include an irrigation operation data acquisition unit 13. The irrigation operation data may include, for example, the amount or duration of irrigation water from aboveground piping, or the amount or duration of rainfall or sprinkler irrigation when mulch is intentionally folded. Unlike sprinklers, aboveground piping is intended to irrigate the ground surface and not to deliver water to aboveground parts (tree parts) such as leaves. In this case, the irrigation operation data includes the irrigation time and / or the amount of rainfall and irrigation water. In this case, the leaves may or may not be wet. In other words, while a growth method that dries the root zone is generally used, the irrigation operation data includes data obtained when an exceptional process of wetting the ground is performed to prevent excessive water stress on the tree body. If the irrigation operation data acquisition unit 13 is included, the irrigation operation data acquired by the irrigation operation data acquisition unit 13 may be created as time-series data for a certain date, similar to the water stress value and weather data described above, and used for machine model optimization and water stress estimation.

[0022] In this embodiment, the weather data includes leaf wetness time, which is the time during which the surface of the leaves of the fruit tree is wet with water. As mentioned above, in open-field cultivation methods that dry the root zone, the intrusion of water from the ground surface is limited, but the amount of water absorbed by leaves and other parts cannot be limited. Therefore, leaf wetness time is used as an index representing the amount of water absorbed by leaves and other parts. Note that water may also be absorbed from parts other than leaves, such as the trunk, branches, and fruit, and leaf wetness time is used as an index of the amount of water absorbed from parts other than the roots.

[0023] The leaf wetness time is the time that the surface of fruit tree leaves is wet with water due to rainfall or sprinkler watering (hereinafter simply referred to as "watering"). However, it is difficult to actually measure the duration for which a large number of leaves remain wet. Therefore, a predetermined calculation method for calculating the leaf wetness time is set and used. The leaf wetness time is the time from when the leaves begin to get wet due to rainfall or watering to when the rainfall or watering ends and the leaves dry. Therefore, for example, the leaf wetness time may be expressed as the sum of the rainfall time and the rainwater remaining time after rainfall. Furthermore, when watering is performed, the leaf wetness time expressed as the sum of the watering time and the watering remaining time after watering may be added to the leaf wetness time due to rainfall. The watering remaining time may be calculated using the same concept as the rainwater remaining time. The leaf wetness time may be calculated by the meteorological data acquisition unit 12 using the above-mentioned calculation formula. The leaf wetness time may also be measured using a known leaf wetness sensor (see, for example, https: / / www.metergroup.co.jp / product / e_LWS_Leafwetness.html).

[0024] Here, we explain the difference between "irrigation" and "sprinkler watering." Even when using open-field cultivation methods that dry the root zone, as described above, the underground root zone may be wetted (watered) to prevent excessive water stress. When the ground surface is covered with a sheet, as in sheet mulch cultivation or shielding mulch cultivation, a normal sprinkler cannot supply water to the underground root zone. On the other hand, when the ground surface is not covered with a sheet and water is supplied from aboveground pipes, or when water is supplied by burying pipes or tubes underground in the root zone, water can be supplied to the underground. To clarify this distinction, we will refer to the case where only the aboveground parts of the tree are wetted, as with a sprinkler, as "sprinkler watering," and the case where water can be supplied to the underground parts of the tree as "irrigation."

[0025] The rainfall duration is the actual duration of rainfall, which can be obtained as meteorological data. The rainwater remaining time may be a predetermined fixed time or a time obtained by multiplying the rainfall time by a predetermined coefficient. The same applies to the watering remaining time. That is, the leaf wetting time can be the time obtained by adding a predetermined bonus time to the rainfall time or watering time. The bonus time can be set to, for example, one hour or two hours. The leaf wetting time can also be the time obtained by multiplying the rainfall time or watering time by a predetermined bonus coefficient. The bonus coefficient can be set to, for example, 1.1, 1.2, etc. The bonus coefficient can also be a power of the rainfall time or watering time (e.g., 1.5 to the power of 3 / 4). The bonus time or bonus coefficient can be set taking into account conditions such as the intensity of the rainfall, the time of day or night, and wind speed. If rainfall and watering occur and are performed on the same day, the sum of both values ​​can be used. The extra time or extra coefficient may be learned as a parameter when the machine model is trained.

[0026] The estimation unit 20 includes a machine model 21. The machine model 21 is a trained machine model that receives a water stress value and measured and / or predicted weather data as input and outputs an estimated value of the water stress value of the fruit tree after the last date of the time-series data has passed. The estimation unit 20 inputs the water stress value and weather data acquired by the acquisition unit 10 to the machine model 21 based on a user operation, and acquires the estimated value of the water stress value output by the machine model 21. If the acquisition unit 10 includes an irrigation work data acquisition unit 13, the estimation unit 20 may input the irrigation work data acquired by the irrigation work data acquisition unit 13 to the machine model 21. By including the irrigation work data in the input data, the machine model 21 can estimate water stress even in citrus cultivation using irrigation equipment. A method for training the machine model 21 will be described later.

[0027] The output unit 30 outputs the estimated value of the water stress value estimated by the mechanical model 21 of the estimation unit 20 based on a user operation to, for example, the display device 70. The output unit 30 may output other data such as input data together with the estimated value.

[0028] The communication unit 40 is an interface for the water stress estimation device 1 to communicate information with the outside world. The communication unit 40 may be, for example, a short-range wireless communication interface such as Wi-Fi (registered trademark) or Bluetooth (registered trademark), or may be a wired connection interface such as a USB connector.

[0029] The control unit 60 controls the entire water stress estimation device 1. The control unit 60 includes at least one processor 61 and at least one memory 62. The processor 61 can be configured using a general-purpose processor such as at least one MPU (Micro Processing Unit) or CPU (Central Processing Unit). The memory 62 may include multiple types of memory such as ROM (Read Only Memory) and RAM (Random Access Memory). As an example, the processor 61 realizes the functions of each unit by loading various control programs recorded in the ROM of the memory 62 into RAM and executing them. The processor 61 may also include a dedicated processor configured using an ASIC (Application Specific Integrated Circuit), FPGA (Field Programmable Gate Array), PLD (Programmable Logic Device), or the like.

[0030] (weather data) Next, details of the weather data will be described. The weather data includes measured weather data observed (measured) in the past and / or forecast weather data predicted for the future. The measured weather data may be, for example, past observation data published by a public institution such as the Japan Meteorological Agency or a private company such as the Japan Weather Association. Alternatively, it may be weather data observed by the user himself. The forecast weather data may be, for example, data created based on weather forecast data published by a public institution such as the Japan Meteorological Agency or a private company such as the Japan Weather Association. The weather data may be data published by the AMeDAS (Automated Meteorological Data Acquisition System) that can acquire local forecast data, and may be used with permission. For example, the weather data may further include at least one of temperature, humidity (relative humidity), and precipitation in addition to the leaf wetness time described above. The weather time series data may be created as daily values. This is because, as mentioned above, water stress values ​​are usually obtained at one data point per day. For example, temperature and humidity may be daily average values. Precipitation may be the accumulated precipitation amount per day. As mentioned above, data on the duration of precipitation is required to calculate the duration of leaf wetting, and this data can be obtained as meteorological data. The acquisition unit 10 of the water stress estimation device 1 may acquire the water stress value data 91 to be input into the trained machine model 21 and the measured and / or predicted meteorological data 92 via the Internet 80.

[0031] (Water stress estimation method S1) Next, an estimation method S1 for estimating a water stress value using the water stress estimation device 1 will be described with reference to Fig. 2. Fig. 2 is a flowchart showing the flow of the water stress estimation method S1 according to this embodiment. As shown in the figure, the water stress estimation method S1 includes steps S11 to S13, which are executed by at least one processor. Each step may be executed by a processor (each unit) of the water stress estimation device 1.

[0032] Step S11 is a water stress value acquisition step for acquiring the water stress value of the fruit tree on a certain date. The water stress value acquisition step is executed by the acquisition unit 10 of the water stress estimation device 1.

[0033] Step S12 is a weather data acquisition step for acquiring time-series weather data from a certain date onward, including actually measured and / or predicted weather data including leaf wetness time, which is the time when the leaf surfaces of fruit trees are wet with water. The weather data acquisition step is executed by the acquisition unit 10 of the water stress estimation device 1.

[0034] It should be noted that further irrigation work data may be acquired in step S12 or after step S12. The acquired irrigation work data may be input to the machine model 21 in step S13.

[0035] Step S13 is an estimation step in which the water stress value and measured and / or predicted weather data are input into the machine model 21 to estimate the water stress value of the fruit tree after the last date of the time-series data has passed. The estimation step is executed by the estimation unit 20 of the water stress estimation device 1. As described above, the data input to the machine model 21 may further include measured irrigation work data. When data including irrigation work data is input, the machine model 21, which has been trained using data including irrigation work data, can estimate the water stress value taking the irrigation work data into consideration. Although not shown, the method may include an output step, following the estimation step, in which the output unit 30 outputs the estimated water stress value to the outside.

[0036] Here, the estimation patterns of estimation method S1 will be described. FIG. 3 is a schematic diagram illustrating three estimation patterns of estimation method S1. Pattern 1 is a pattern in which a future water stress value (estimated water stress value) is estimated using a current water stress value (actual measurement value) and predicted weather data from the present to the future (estimation time) as inputs. Pattern 2 is a pattern in which a current water stress value (estimated water stress value) is estimated using a past water stress value (actual measurement value) and measured weather data from the past to the present (estimation time) as inputs. Pattern 3 is a pattern in which a future water stress value (estimated water stress value) is estimated using a past water stress value (actual measurement value) and measured weather data and predicted weather data from the past to the future (estimation time) as inputs. In this way, estimation method S1, which estimates a water stress value using a machine model 21, may input either a past or current water stress value (actual measurement value), and may input either measured data or predicted data, or both. The estimated water stress value may be either a current value or a future value.

[0037] (Machine Model 21) Next, the machine model 21 will be described. The machine model 21 is a machine model trained using the water stress value of the fruit tree on a certain date, time-series data of the weather measured from that date onward, which is actual weather data including the leaf wetness time, which is the time when the surface of the fruit tree's leaves is wet with water, and measured (actually measured) data of the water stress value after the last date of the time-series data has passed. The time-series data of the weather used during training is not a predicted value but an actual value. The measured data of the water stress value after the last date of the actually measured time-series data has passed is also an actual value. As described above, actually measured irrigation work data may be added to the input data of the machine model 21. In this case, the machine model 21 estimates the water stress value taking into account the input irrigation work data.

[0038] The machine model 21 is a model that can estimate water stress values ​​taking into account the temporal context of measured and / or predicted time-series data. Examples of such machine models include LSTM (Long Short Term Memory), GRU (Gated Recurrent Unit), which is a simplified version of LSTM, and a Transformer model. These machine models can output estimation results that take into account changes in the data over time by inputting data in chronological order, for example, on a daily basis.

[0039] When using the machine model 21 to estimate a water stress value for a long period in the future, the period may be divided and estimated. For example, a certain period may be divided into three periods, and the water stress value after the last day of the first period has passed (e.g., the next day) may be estimated. Next, the water stress value after the last day of the first period has passed (e.g., the next day) may be used as the water stress value at the beginning of the second period to estimate the water stress value after the last day of the second period has passed (e.g., the next day). Finally, the water stress value after the last day of the second period has passed (e.g., the next day) may be used as the water stress value at the beginning of the third period to estimate the water stress value after the last day of the third period has passed (e.g., the next day). In this way, a long period may be divided and water stress values ​​may be estimated.

[0040] (Methods for learning machine models) Next, a training method for the machine model 21 will be described with reference to FIG. 4. FIG. 4 is a flowchart showing the flow of the machine model training method S2 according to this embodiment. The training method S2 includes steps S21 and S22, which are executed by at least one processor. Step S21 is an acquisition step in which multiple sets of training data are acquired, each set including the measured water stress value of the fruit tree on a certain date, measured time-series data on weather from the certain date onward, including measured weather data including leaf wetness time, which is the time during which the leaf surfaces of the fruit tree are wet with water, and the measured water stress value of the fruit tree after the last date in the time-series data. Next, step S22 is a training step in which the parameters of the machine model 21 are updated using the multiple acquired training data sets so that the value output when the measured water stress value of the fruit tree on a certain date and the measured weather data from the certain date onward are input to the machine model 21 approaches the measured water stress value of the fruit tree after the last date in the time-series data. Here, the leaf wetness time may be expressed as the sum of the rainfall time and the time the rainwater remains after the rainfall, as described above, and in this case, the extra time or extra coefficient may be trained as a parameter (or hyperparameter). In this way, the machine model 21 is trained.

[0041] As described above, machine model 21 may be trained using actually measured irrigation work data as well. That is, training data including actually measured irrigation work data may be acquired in step S21 or after step S21. Then, in step S22, the parameters of machine model 21 are updated so that the value output when training data including irrigation work data is input to machine model 21 approaches the actually measured water stress value of the fruit tree after the last date in the time-series data has passed. By training in this manner, it is possible to generate machine model 21 that estimates a water stress value taking irrigation work data into consideration.

[0042] The water stress estimation device 1 may include a learning unit 50. The learning unit 50 trains the untrained machine model 21. In this case, the acquisition unit 10 acquires externally stored measured water stress value data 91 and measured weather data 92 via the Internet 80, and executes an acquisition step of generating multiple training data sets. The learning unit 50 then executes a learning step of training the machine model 21 using the multiple training data sets. The learning step is then repeated until the difference between the output value and the measured water stress value falls within a predetermined error range, or until a predetermined number of learning attempts is reached.

[0043] The date may be a day, as described above. In this case, the training data includes the water stress value of the fruit tree on a certain day, time-series data of weather measured from that day onwards, which is daily measured weather data including leaf wetness time, which is the time when the surface of the fruit tree leaves is wet with water, and the measured data of the water stress value after the last day of the time-series data has passed (for example, the next day). Learning is performed by updating the parameters of the machine model 21 when the water stress value of the fruit tree on a certain date and daily time-series data of weather from that date onwards are input so that the output approaches the water stress value of the day after the last day of the time-series data has passed.

[0044] The trained machine model 21 trained using the above-mentioned method is a machine model that inputs the above-mentioned water stress value and time series data, which is measured and / or predicted weather data including leaf wetness time, and outputs the estimated water stress value of the fruit tree after the last date of the time series data has passed.

[0045] According to the water stress estimation device 1, water stress estimation method S1, and trained machine model 21 having the above-mentioned configuration, when the root zone is dried out in outdoor cultivation of citrus fruits, the water stress value can be estimated taking into account the water absorption from the above-ground parts of the tree.

[0046] [Embodiment 2] (Water stress estimation device 1A) Next, another embodiment of the present invention will be described in detail with reference to the drawings. The same components as those described in the first embodiment are denoted by the same reference numerals, and their description will be omitted. FIG. 5 is a block diagram showing the configuration of a water stress estimation device 1A according to a second embodiment. As shown in the figure, the water stress estimation device 1A includes an acquisition unit 10A, an estimation unit 20A, an output unit 30, a communication unit 40, and a control unit 60. The acquisition unit 10A includes a meteorological data acquisition unit 12A. The acquisition unit 10A may further include an irrigation work data acquisition unit 13. The water stress estimation device 1 may also include a learning unit 50.

[0047] The water stress estimation device 1A may be connected to an external display device 70. Alternatively, the water stress estimation device 1A may be equipped with a display unit (not shown). The water stress estimation device 1A may be connectable to the Internet 80 or the like. The acquisition unit 10A may be capable of acquiring water stress value data 91 and / or weather data 92 and / or irrigation work data 93 via the Internet 80 or the like and storing them in the memory 62.

[0048] The weather data acquisition unit 12A of the acquisition unit 10A acquires measured and / or predicted weather data, which is time-series data on weather from a certain date onwards, including leaf wetness time, which is the time when the surface of the leaves of a fruit tree is wet with water. The time-series data, leaf wetness time, and weather data are as described in the first embodiment. The acquisition unit 10A does not need to acquire the water stress value of the fruit tree on a certain date. If the water stress estimation device 1 includes a learning unit 50, the acquisition unit 10A may acquire the actually measured water stress value of the fruit tree.

[0049] The estimation unit 20A includes a machine model 21A. The machine model 21A estimates the amount of change in the water stress value of the fruit tree after the last date in the time-series data has passed from weather data (which may further include irrigation work data). Based on user operation, the estimation unit 20A inputs the weather data acquired by the acquisition unit 10A to the machine model 21A, and acquires an estimate of the amount of change in the water stress value of the fruit tree after the last date in the time-series data has passed, which is output by the machine model 21A. The amount of change is the difference between the water stress value of the fruit tree on a certain date and the water stress value of the fruit tree after the last date in the time-series data has passed.

[0050] The configurations of the output unit 30, communication unit 40, control unit 60, and learning unit 50 of the water stress estimation device 1A are the same as those described in the water stress estimation device 1 of embodiment 1, so description thereof will be omitted here.

[0051] The water stress estimation device 1A of this embodiment differs from the water stress estimation device 1 described in the first embodiment in that the input values ​​do not include the "water stress value of a fruit tree on a certain date." Therefore, the output value of the machine model 21A is the "amount of change in water stress value."

[0052] (Water stress estimation method S3) Next, an estimation method S3 for estimating a water stress value using the water stress estimation device 1A will be described with reference to Fig. 6. Fig. 6 is a flowchart showing the flow of the water stress estimation method S3 according to embodiment 2. As shown in the figure, the water stress estimation method S3 includes steps S31 and S32, which are executed by at least one processor.

[0053] Step S31 is a weather data acquisition step for acquiring time-series weather data from a certain date onward, including actually measured and / or predicted weather data including leaf wetness time, which is the time when the surface of fruit tree leaves is wet with water. The weather data acquisition step is executed by the acquisition unit 10A of the water stress estimation device 1A.

[0054] Note that further measured irrigation work data may be acquired in step S31 or after step S31. In this case, the measured irrigation work data may be further input to the machine model in step S32. The machine model can estimate the amount of change in the water stress value taking the irrigation work data into account.

[0055] Step S32 is an estimation step in which measured and / or predicted weather data is input into the machine model 21A to estimate the amount of change in the water stress value of the fruit tree after the last date of the time-series data has passed. The estimation step is performed by the estimation unit 20A of the water stress estimation device 1A. Although not shown, the process may include an output step, following the estimation step, in which the output unit 30 outputs the estimated water stress value to an external display device 70.

[0056] (Methods for learning machine models) Next, a learning method for the machine model 21A will be described with reference to FIG. 7. The learning of the machine model 21A is performed by the learning unit 50. In this case, the acquisition unit 10A may also include a water stress value data acquisition unit (not shown). FIG. 7 is a flowchart showing the flow of the machine model learning method S4 according to the second embodiment. The learning method S4 includes steps S41 and S42 executed by at least one processor. Step S41 is an acquisition step of acquiring multiple sets of learning data, each set including a measured water stress value of a fruit tree on a certain date, measured time-series data of weather from the certain date onwards, including leaf wetness time, which is the time when the surface of the fruit tree's leaves is wet with water, and a measured water stress value of the fruit tree after the last date in the time-series data has passed. The acquisition step is executed by the acquisition unit 10A.

[0057] Note that further measured irrigation work data may be acquired in step S41 or after step S41. In that case, the further measured irrigation work data may be input to the machine model in step S42 for learning.

[0058] Next, step S42 is a learning step in which the parameters of the machine model are updated using the acquired multiple learning data sets so that the value output when measured weather data from a certain date onwards is input approaches the amount of change in the measured water stress value of the fruit tree from the measured water stress value of the fruit tree on a certain date after the last date in the time-series data has passed. The learning step is then repeated until the difference between the amount of change in the measured water stress value and the output value falls within a predetermined error range or until a predetermined number of learning cycles has been reached. The learning step is executed by the learning unit 50.

[0059] The water stress estimation device 1A, water stress estimation method S3, and trained machine model 21A configured as described above can estimate a water stress value by taking into account water absorption from the aboveground parts of the tree when the root zone is dried out in outdoor cultivation of citrus trees. It is also possible to estimate a water stress value by taking into account irrigation work data. [Example]

[0060] Next, the results of estimating water stress values ​​using the method described in this embodiment will be described with reference to the drawings. FIG. 8 is a graph showing the results of estimating water stress values ​​using a machine model trained using weather data excluding leaf wetness duration. FIG. 9 is a graph showing the results of estimating water stress values ​​using a machine model trained using weather data excluding precipitation. In both graphs, the vertical axis represents the machine model's estimate of water stress values ​​after the last date has passed, and the horizontal axis represents the actual measured water stress values ​​after the last date has passed. The weather data common to both FIGS. 8 and 9 are the average daily temperature, average daily relative humidity, and the measured water stress value on a certain date; FIG. 8 further includes precipitation, and FIG. 9 further includes leaf wetness duration.

[0061] The coefficient of determination (R 2 ) was approximately 0.81, and the coefficient of determination for the graph in Figure 9 was approximately 0.91. In other words, it can be seen that including leaf wetness time in the training data can generate a machine model with better estimation accuracy than when not including it.

[0062] Figure 10 is a graph showing the estimation results of a machine model that uses the water stress value on the first day (date) of a certain period and weather data for that period to estimate the water stress value after that period. Figure 11 is a graph showing the estimation results of a machine model that uses only weather data for that period to estimate the change in water stress value before and after that period. The coefficient of determination for the graph in Figure 10 was approximately 0.81. On the other hand, the coefficient of determination for the graph in Figure 11 was approximately 0.48. Although the estimation accuracy of the machine model trained using only weather data was inferior to that of the machine model trained using the water stress value and weather data for the first day of a certain period, it was found that a certain degree of estimation was possible. It is believed that estimation accuracy can be further improved by considering the type of weather data used for training.

[0063] [Software implementation example] The functions of the water stress estimation device 1 and the water stress estimation device 1A (hereinafter referred to as the "device") can be realized by a water stress estimation program, which is a program for causing a computer to function as the device, and which causes a computer to function as each part executed by the control unit of the device.

[0064] In this case, the device includes a computer having at least one control device (e.g., processor 61) and at least one storage device (e.g., memory 62) as hardware for executing the water stress estimation program. The functions described in each of the above embodiments are realized by executing the water stress estimation program using this control device and storage device.

[0065] The water stress estimation program may be stored non-transitory on one or more computer-readable storage media. The storage media may or may not be included in the device. In the latter case, the program may be supplied to the device via any wired or wireless transmission medium.

[0066] Furthermore, some or all of the functions of the control blocks can be realized by logic circuits. For example, an integrated circuit in which a logic circuit that functions as each of the control blocks is formed is also included in the scope of the present invention. In addition, the functions of the control blocks can also be realized by, for example, a quantum computer.

[0067] (Additional notes) A water stress estimation method according to a first aspect of the present invention includes the steps of: a water stress value acquisition step of acquiring a water stress value of a fruit tree on a certain date; a meteorological data acquisition step of acquiring time-series data of meteorological conditions from the certain date onward, which is actual and / or predicted meteorological data including leaf wetness time, which is the time when the surface of the leaves of the fruit tree is wet with water; an estimation step of inputting the water stress value and the measured and / or predicted weather data into a machine model to estimate the water stress value of the fruit tree after the latest date of the time series data has passed; Includes:

[0068] (Aspect 2) In the water stress estimation method according to a second aspect of the present invention, in the first aspect, the leaf wetting time is expressed as the sum of the rainfall time and the rainwater remaining time after rainfall.

[0069] (Aspect 3) A water stress estimation method according to aspect 3 of the present invention is, in aspect 2, the rainwater remaining time being at least one of a predetermined fixed time, a time obtained by multiplying the rainfall time by a predetermined coefficient, and a time actually measured by a sensor.

[0070] (Aspect 4) A fourth aspect of the present invention provides a water stress estimation method in any one of the first to third aspects, wherein the meteorological data further includes at least one of temperature, humidity, and precipitation.

[0071] (Aspect 5) A fifth aspect of the present invention provides a water stress estimation method in any one of the first to fourth aspects, wherein the time-series data is acquired as a value per day.

[0072] (Aspect 6) A water stress estimation method according to aspect 6 of the present invention is any one of aspects 1 to 5, in which the water on the surface of the leaves further includes sprinkler watering, and the meteorological data further includes at least one of watering data and the leaf wetting time taking watering into account.

[0073] (Aspect 7) A water stress estimation method according to aspect 7 of the present invention is any one of aspects 1 to 6, in which the machine model estimates the water stress value taking into account the temporal context of the measured and / or predicted time series data.

[0074] (Aspect 8) A water stress estimation method according to aspect 8 of the present invention is any one of aspects 1 to 7, in which irrigation work data is further acquired in the meteorological data acquisition step, and the irrigation work data is further input into the mechanical model in the estimation step.

[0075] (Aspect 9) A water stress estimation device according to aspect 9 of the present invention comprises an acquisition unit that acquires the water stress value of a fruit tree on a certain date and measured and / or predicted weather data from the certain date onwards, the measured and / or predicted weather data including leaf wetness time, which is the time during which the surface of the leaves of the fruit tree is wet with water; a mechanical model that estimates the water stress value of the fruit tree after the last date of the time series data has passed from the water stress value and the measured and / or predicted weather data; and an output unit that outputs the estimated value estimated by the mechanical model.

[0076] (Aspect 10) A learning method for a machine model according to aspect 10 of the present invention includes the steps of acquiring multiple sets of learning data, each set consisting of the measured water stress value of a fruit tree on a certain date, measured time series data on the weather from the certain date onwards, the measured weather data including leaf wetness time, which is the time when the surface of the leaves of the fruit tree is wet with water, and the measured water stress value of the fruit tree after the last date of the time series data has passed; and a learning step of updating the parameters of the machine model so that the value output when the measured water stress value of the fruit tree on the certain date and the measured weather data from the certain date onwards are input into the machine model approaches the measured water stress value of the fruit tree after the last date of the time series data has passed.

[0077] (Aspect 11) A machine model according to an eleventh aspect of the present invention is a machine model trained using the machine model training method according to the tenth aspect.

[0078] (Aspect 12) A program according to aspect 12 of the present invention is a water stress estimation program for causing a computer to function as the water stress estimation device described in aspect 9, and is a water stress estimation program for causing a computer to function as the acquisition unit, the mechanical model, and the output unit.

[0079] (Aspect 13) A recording medium according to a thirteenth aspect of the present invention is a computer-readable non-transitory recording medium having the water stress estimation program according to the twelfth aspect recorded thereon.

[0080] (Aspect 14) A second water stress estimation method according to aspect 14 of the present invention includes a weather data acquisition step of acquiring measured and / or predicted weather data, which is time series data on weather from a certain date onwards, including leaf wetness time, which is the time during which the surface of the leaves of a fruit tree is wet with water, and an estimation step of inputting the measured and / or predicted weather data into a machine model and estimating the amount of change in the water stress value of the fruit tree after the last date of the time series data has passed.

[0081] (Aspect 15) In the water stress estimation method according to aspect 15 of the present invention, irrigation work data is further acquired in the meteorological data acquisition step, and the irrigation work data is further input to the machine model in the estimation step.

[0082] (Aspect 16) A second water stress estimation device according to aspect 16 of the present invention comprises an acquisition unit that acquires measured and / or predicted weather data, which is time series data on weather from a certain date onwards, including leaf wetness time, which is the time during which the surface of the leaves of a fruit tree is wet with water; a mechanical model that estimates the amount of change in the water stress value of the fruit tree from the measured and / or predicted weather data after the last date of the time series data has passed; and an output unit that outputs the estimated value estimated by the mechanical model.

[0083] (Aspect 17) A second machine model learning method according to aspect 17 of the present invention includes the steps of acquiring multiple sets of learning data, each set consisting of the measured water stress value of a fruit tree on a certain date, measured time series data on the weather from the certain date onwards, the measured weather data including leaf wetness time, which is the time when the surface of the leaves of the fruit tree is wet with water, and the measured water stress value of the fruit tree after the last date of the time series data has passed; and a learning step of updating the parameters of the machine model so that the value output when the measured weather data from the certain date onwards is input into the machine model approaches the change in the measured water stress value of the fruit tree after the last date of the time series data has passed from the measured water stress value of the fruit tree on the certain date.

[0084] (Aspect 18) In the second machine model learning method according to aspect 18 of the present invention, irrigation work data is further acquired in the acquiring step, and the irrigation work data is further input to the machine model in the learning step.

[0085] The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the present invention. [Explanation of symbols]

[0086] 1, 1A...Water stress estimation device 10, 10A…Acquisition part 11...Water stress value acquisition unit 12, 12A...Weather data acquisition section 20, 20A... Estimation section 21, 21A... Mechanical model 30...Output section 40…Communications Department 50...Study Department 60...Control unit 61...Processor 62...Memory 70…Display device 80…Internet 91...Water stress value data 92...Weather data

Claims

1. a water stress value acquisition step of acquiring an actually measured water stress value of the fruit tree on a certain date; a meteorological data acquisition step of acquiring time-series data of meteorological conditions from the certain date onward, the meteorological data being measured and / or predicted, including leaf wetness time, which is the time during which the surface of the leaves of the fruit tree is wet with water; an estimation step of inputting the water stress value and the measured and / or predicted weather data into a machine model to estimate the water stress value of the fruit tree after the latest date of the time series data has passed; A method for estimating water stress, including:

2. The water stress estimation method according to claim 1 , wherein the leaf wetting time is expressed as a total value of a rainfall time and a rainwater remaining time after the rainfall.

3. 3. The water stress estimation method according to claim 2, wherein the rainwater remaining time is at least one of a predetermined fixed time, a time obtained by multiplying the rainfall time by a predetermined coefficient, and a time actually measured by a sensor.

4. The water stress estimation method according to claim 1 , wherein the meteorological data further includes at least one of temperature, humidity, and precipitation.

5. The water stress estimation method according to claim 1 , wherein the time-series data is acquired as a daily value.

6. The water stress estimation method according to any one of claims 1 to 3, wherein the water on the surface of the leaves further includes sprinkler watering, and the meteorological data further includes at least one of watering data and the leaf wetting time taking watering into account.

7. The water stress estimation method according to claim 1 , wherein the machine model estimates the water stress value taking into account the temporal context of the measured and / or predicted time-series data.

8. The water stress estimation method according to claim 1 , wherein the meteorological data acquisition step further acquires irrigation work data, and the estimation step further inputs the irrigation work data into the machine model.

9. an acquisition unit that acquires an actually measured water stress value of a fruit tree on a certain date and time-series data of weather from the certain date onward, which is actually measured and / or predicted weather data including leaf wetness time, which is the time when the surface of the leaves of the fruit tree is wet with water; a machine model that estimates a water stress value of the fruit tree after the last date of the time series data has passed from the water stress value and the measured and / or predicted weather data; an output unit that outputs an estimated value estimated by the machine model; A water stress estimation device comprising:

10. a step of acquiring multiple sets of learning data, each set being made up of a measured water stress value of a fruit tree on a certain date, measured time-series data of weather after the certain date, the measured weather data including leaf wetness time, which is the time when the leaf surfaces of the fruit tree are wet with water, and a measured water stress value of the fruit tree after the last date of the time-series data has passed; a learning step of updating the parameters of the machine model so that the value output when the measured water stress value of the fruit tree on the certain date and the measured meteorological data after the certain date are input to the machine model approaches the measured water stress value of the fruit tree after the last date of the time series data has passed; Methods for learning machine models, including:

11. A machine model trained using the machine model training method according to claim 10.

12. A water stress estimation program for causing a computer to function as the water stress estimation device according to claim 9, the water stress estimation program causing a computer to function as the acquisition unit, the mechanical model, and the output unit.

13. A computer-readable non-transitory recording medium having the water stress estimation program according to claim 12 recorded thereon.

14. a meteorological data acquisition step of acquiring time-series data of meteorological conditions from a certain date onward, which is actual and / or predicted meteorological data including leaf wetness time, which is the time when the surface of the leaves of the fruit tree is wet with water; an estimation step of inputting the measured and / or predicted weather data into a machine model to estimate a change in the water stress value of the fruit tree after the last date of the time series data has passed; A method for estimating water stress, including:

15. The water stress estimation method according to claim 14, characterized in that in the meteorological data acquisition step, irrigation work data is further acquired, and in the estimation step, the irrigation work data is further input into the mechanical model.

16. an acquisition unit that acquires time-series data of weather from a certain date onward, which is actual and / or predicted weather data including leaf wetness time, which is the time when the surface of the leaves of fruit trees is wet with water; a machine model that estimates, from the measured and / or predicted weather data, the amount of change in the water stress value of the fruit tree after the last date of the time series data has passed; an output unit that outputs an estimated value estimated by the machine model; A water stress estimation device comprising:

17. a step of acquiring multiple sets of learning data, each set being made up of a measured water stress value of a fruit tree on a certain date, measured time-series data of weather after the certain date, the measured weather data including leaf wetness time, which is the time when the leaf surfaces of the fruit tree are wet with water, and a measured water stress value of the fruit tree after the last date of the time-series data has passed; a learning step of updating the parameters of the machine model so that a value output when the measured weather data after the certain date is input to the machine model approaches the amount of change in the measured water stress value of the fruit tree from the measured water stress value of the fruit tree on the certain date after the last date in the time series data has passed; Methods for learning machine models, including:

18. 18. The method for learning a machine model according to claim 17, wherein in the acquiring step, irrigation work data is further acquired, and in the learning step, the irrigation work data is further input to the machine model.

Images