Large-scale irrigation and fertilization system optimization method based on remote sensing leaf area index

Abstract

The invention discloses a large-scale irrigation and fertilization system optimization method based on a remote sensing leaf area index, and the method comprises the steps: calibrating crop parameters of a field scale crop model based on field test data; then, by using an assimilation algorithm based on the remote sensing LAI, determining sensitive crop parameters of a regional scale, and verifying the rationality of determining the regional crop parameters by using the remote sensing LAI through regional crop growth simulation; and finally, constructing a regional scale crop model, and combining the regional scale crop model with a multi-target genetic algorithm to optimize the regional INS of different typical rainfall years. The regional INSs established by estimating regional crop parameters based on remote sensing data can effectively guide irrigation and nitrogen fertilizer application plans, the utilization efficiency of agricultural resources is improved, and an important reference is provided for regional agricultural management.

Description

Technical Field

[0001] The invention relates to the technical field of irrigation and fertilization, and in particular to a large-scale irrigation and fertilization system optimization method based on remote sensing leaf area index. Background Art

[0002] Water and nitrogen are key factors affecting crop growth and yield. Water and nitrogen must be properly allocated to improve the effectiveness of nitrogen fertilizer use and reduce environmental pollution. Therefore, appropriate and reasonable irrigation and nitrogen management measures must be formulated. The development of crop growth models such as the CERES-Maize model, the CERES-Wheat model, and the AquaCrop model enables researchers to simulate the crop growth process under different irrigation and nitrogen fertilizer application conditions. Regional agricultural water and fertilizer management and the use of crop models to formulate irrigation and nitrogen fertilizer application schemes (INSs) have become active research areas.

[0003] When using crop models to simulate crop growth and develop irrigation and nitrogen fertilizer application schemes (INS), first of all, it is crucial to obtain reliable crop parameters. Existing studies have used crop growth models calibrated based on field trial data to determine the best INS scheme, but when there are significant spatial differences in crop varieties and growth environments, directly applying the results of crop models at the field scale to the regional scale may lead to a series of errors. In addition, there is the problem of heterogeneity in regional soil, meteorological and field management conditions, and the crop variety coefficients in existing studies are usually derived from a single crop variety in a single field trial. When these variety coefficients are directly applied to the regional scale, it may lead to insufficient spatial representation.

[0004] The development of remote sensing technology has made the spatial expression of some simulation results of crop models (such as leaf area index, LAI) clearer. By combining remote sensing information with crop models, the spatial expression accuracy of variety coefficients can be improved. Data assimilation methods that combine crop growth models with remote sensing observations have developed rapidly to improve the spatial simulation accuracy of crop models. However, existing studies have only estimated crop model parameters through remote sensing LAI of a single pixel. Factors such as insufficient pixel remote sensing LAI data and remote sensing LAI errors may lead to instability of crop model parameters.

[0005] In summary, when crop parameters determined by field experiments are used at the regional scale, there will be a problem of insufficient spatial representation; and when regional-scale crop model parameters are derived pixel by pixel through pixel-level LAI, parameter instability may occur. Summary of the invention

[0006] The present application provides a large-scale irrigation and fertilization system optimization method based on remote sensing leaf area index to solve the problems in the prior art that INSs developed based on field experiments to determine crop parameters have low robustness at the regional scale and parameter instability exists when regional-scale crop model parameters are derived pixel by pixel through pixel-level LAI.

[0007] According to the first aspect, an embodiment provides a large-scale irrigation and fertilization system optimization method based on remote sensing leaf area index, the method comprising: The target study area is divided into multiple sub-areas according to the growth characteristics and growth environment of crops; The measured leaf area index (LAI) and microwave remote sensing data in the target study area were used to train the LAI prediction model based on support vector machine, and the trained LAI prediction model was used to predict the LAI estimation value of each sub-area based on microwave remote sensing data. Conduct sensitivity analysis on crop variety coefficients of regional scale crop growth models considering spatial variability of soil and meteorological parameters, and determine sensitive variety coefficients and non-sensitive variety coefficients; The coefficient values ​​of non-sensitive varieties are obtained by calibration with field-scale test data, and the coefficient values ​​of sensitive varieties are obtained by iterative updating using a global optimization algorithm with the goal of minimizing the gap between the LAI value simulated by the regional-scale crop growth model and the LAI estimated value predicted by microwave remote sensing data. Based on the regional-scale crop growth model, a multi-objective genetic algorithm was used to perform multi-objective optimization on the irrigation and nitrogen fertilizer application schemes in each sub-region in a typical rainfall year and determine the optimal solution, thus obtaining the optimal irrigation and nitrogen fertilizer application scheme for each sub-region.

[0008] Furthermore, the target research area is divided into multiple sub-areas, including: The target study area was divided into multiple sub-areas using a clustering algorithm based on the normalized difference vegetation index of the target crop and the clay and silt content of the soil.

[0009] Furthermore, the LAI prediction model based on support vector machine was trained using the measured leaf area index LAI and microwave remote sensing data in the target study area, including: The target research area is divided into multiple sampling areas, and multiple independent sampling points are set in each sampling area. The sampling indicators include the LAI of the target crop at different growth stages. The value of each sampling indicator in each sampling area is the average value of the measured values ​​of multiple sampling points. The vegetation index was calculated using the high spatial resolution Sentinel-2 data. The random forest classifier was used to classify the vegetation index and extract the planting area of ​​the target crop in the study area. Based on the planting area of ​​the target crop, the remote sensing data including backscatter intensity and incident angle of the corresponding planting area were extracted from the Sentinel-1 data. The measured LAI and the corresponding remote sensing data are divided into training sets and test sets, and the LAI prediction model based on support vector machine is trained and tested.

[0010] Furthermore, a sensitivity analysis of the crop variety coefficients of the regional scale crop growth model considering the spatial variability of soil and meteorological parameters was conducted to determine the sensitive variety coefficients and the non-sensitive variety coefficients, including: The EFAST method was used to conduct sensitivity analysis on crop variety coefficients of the regional-scale CERES-Maize model, and the coefficients were divided into sensitive variety coefficients and non-sensitive variety coefficients according to the calculated sensitivity index value of each variety coefficient.

[0011] Furthermore, the coefficient values ​​of non-sensitive varieties are calibrated through field scale test data, including: The field-scale CERES-Maize model was calibrated and verified using field trial data collected in different years, and the insensitive variety coefficient value was determined based on the variety coefficient value of the calibrated field-scale CERES-Maize model.

[0012] Furthermore, for the coefficient value of sensitive varieties, the goal is to minimize the gap between the LAI value simulated by the regional scale crop growth model and the LAI estimated value predicted by microwave remote sensing data, and the global optimization algorithm is used for iterative update, including: Divide the study area into different simulation units according to the preset grid; Construct the cost function: ; Where J is the cost function, X is the vector of calibrated sensitive parameters, is the LAI simulated by the regional-scale CERES-Maize model, is the LAI estimated from remote sensing data, Q is the number of observed LAI, K is the number of simulation units, Represents the mean LAI of all pixels in each simulation unit.

[0013] Furthermore, for the coefficient value of sensitive varieties, the goal is to minimize the gap between the LAI value simulated by the regional scale crop growth model and the LAI estimated value predicted by microwave remote sensing data, and the global optimization algorithm is used for iterative update, including: The SCE-UA optimization algorithm is used to minimize the cost function. If the optimized variety coefficient value is within the predefined normalized geometric range and the cost function does not decrease by more than the preset value after several iterations, the SCE-UA algorithm is considered to have converged.

[0014] Furthermore, based on the obtained regional-scale crop growth model, a multi-objective genetic algorithm was used to optimize the irrigation and nitrogen fertilizer application schemes in each sub-region in a typical rainfall year and determine the optimal solution, including: Based on the historical precipitation data of the target crops during the growing season, the typical rainfall years of each sub-region were identified, including wet years, normal years and dry years. In the three typical rainfall years, the irrigation and nitrogen fertilizer application schemes of each sub-region were optimized with multiple objectives. The multi-objective optimization of irrigation and nitrogen fertilizer application schemes is achieved by maximizing yield, water use efficiency WUE and nitrogen partial productivity PFPN, and the NSGA-III algorithm is used for multi-objective optimization. The regional-scale CERES-Maize model is used to calculate and achieve the optimization targets of crop yield, WUE, and PFPN:

[0015] Where Y is the average crop yield in kg / ha; K is the number of simulation units in each sub-region; j is the number of irrigation times; the units of WUE and PFPN in the formula are kg / m 3 and kg / kg; according to the experience of farmers in the study area, the number of irrigations j∈{1,2,3}; during the crop growing season, part of the nitrogen fertilizer is used as base fertilizer, and the remaining nitrogen fertilizer is dripped with irrigation water in equal amounts; t1–t j represents the irrigation date, w1–w j represents the irrigation amount, f1–f j represents the amount of nitrogen fertilizer applied, f b is basal nitrogen fertilizer; when j = 1, w2 = w3 = 0 and f2 = f3 = 0; when j = 2, w3 = 0 and f3 = 0.

[0016] Furthermore, based on the obtained regional-scale crop growth model, a multi-objective genetic algorithm was used to optimize the irrigation and nitrogen fertilizer application schemes in each sub-region in a typical rainfall year and determine the optimal solution, including: The TOPSIS method was used to determine the optimal solution for the Pareto solution set of maximizing yield, water use efficiency WUE and nitrogen partial productivity PFPN obtained by the multi-objective optimization algorithm, thereby obtaining the optimal irrigation and nitrogen fertilizer application plan.

[0017] According to the second aspect, an embodiment provides a large-scale irrigation and fertilization schedule optimization system based on remote sensing leaf area index, the system comprising: A regional division module is used to divide the target research area into multiple sub-areas according to the growth characteristics and growth environment of crops; The LAI estimation module is used to train the LAI prediction model based on the support vector machine using the measured leaf area index LAI and microwave remote sensing data in the target study area, and use the trained LAI prediction model to predict the LAI estimation value of each sub-area based on the microwave remote sensing data; Sensitivity analysis module, which is used to conduct sensitivity analysis on crop variety coefficients of regional scale crop growth models considering spatial variability of soil and meteorological parameters, and determine sensitive variety coefficients and non-sensitive variety coefficients; Variety coefficient acquisition module, used to obtain the coefficient value of non-sensitive varieties through field scale test data calibration, and for sensitive variety coefficient values, the gap between the LAI value simulated by the regional scale crop growth model and the LAI estimated value predicted by microwave remote sensing data is minimized, and the coefficient value is obtained by iterative update using a global optimization algorithm; The multi-objective optimization module is used to optimize the irrigation and nitrogen fertilizer application plans in each sub-region under typical rainfall years based on the obtained regional-scale crop growth model and determine the optimal solution using a multi-objective genetic algorithm to obtain the optimal irrigation and nitrogen fertilizer application plan for each sub-region.

[0018] This application provides a large-scale irrigation and fertilization system optimization method based on remote sensing leaf area index, and develops a framework for optimizing regional irrigation and nitrogen fertilizer application schemes (INS). The framework consists of the CERES-Maize model, microwave remote sensing data, and a multi-objective optimization algorithm, which can be used to optimize the INS of crops in large areas, and has the following beneficial effects: (1) First, microwave remote sensing data including backscatter intensity and incident angle are input into the support vector regression (SVR) model to obtain accurate LAI prediction. Compared with optical remote sensing data, microwave remote sensing data is less affected by clouds and precipitation. Therefore, microwave remote sensing data can be effectively used to estimate LAI in areas with heavy cloud cover and precipitation.

[0019] (2) Second, the estimated regional LAI was input into the assimilation algorithm to determine the key crop coefficients of the CERES-Maize model. The crop model parameters determined using remote sensing data can simulate LAI, plant height, soil water content (SWC), and yield more accurately at the regional scale than those determined using field experimental data.

[0020] (3) Third, the regional-scale CERES-Maize model was combined with a multi-objective genetic algorithm to optimize the INS of each sub-region under different typical rainfall years. The developed regional-scale CERES-Maize model fully considers the spatial variability of meteorological and soil parameters and the spatial representativeness of crop parameters. The optimized INS established by using regional-scale crop model parameters and remote sensing data can effectively guide irrigation and nitrogen fertilizer application plans, improve agricultural resource utilization efficiency, and provide an important reference for regional agricultural management. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A flowchart of a large-scale irrigation and fertilization system optimization method based on remote sensing leaf area index provided by an embodiment of the present invention; Figure 2 An overall roadmap for a large-scale irrigation and fertilization system optimization method based on remote sensing leaf area index provided by one embodiment of the present invention; Figure 3 A study area division and sampling point distribution map in a large-scale irrigation and fertilization system optimization method based on remote sensing leaf area index provided by an embodiment of the present invention; Figure 4 A dynamic spatiotemporal variation diagram of summer corn LAI in a large-scale irrigation and fertilization system optimization method based on remote sensing leaf area index provided by an embodiment of the present invention; Figure 5 The chromosome structure of NSGA-III in a large-scale irrigation and fertilization system optimization method based on remote sensing leaf area index provided by one embodiment of the present invention; Figure 6 A schematic diagram of irrigation dates in a large-scale irrigation and fertilization system optimization method based on remote sensing leaf area index provided by an embodiment of the present invention; Figure 7 An optimal solution for maximizing yield, water use efficiency (WUE) and partial productivity of nitrogen (PFPN) in different sub-regions in different typical rainfall years in a large-scale fertigation and irrigation system optimization method based on remote sensing leaf area index provided by an embodiment of the present invention; Figure 8 A schematic diagram of the logical structure of a large-scale irrigation and fertilization schedule optimization system based on remote sensing leaf area index is provided for one embodiment of the present invention. DETAILED DESCRIPTION

[0022] The present invention is further described in detail below by specific embodiments in conjunction with the accompanying drawings. Wherein similar elements in different embodiments adopt associated similar element numbers. In the following embodiments, many detailed descriptions are for making the present application better understood. However, those skilled in the art can easily recognize that some features can be omitted in different situations, or can be replaced by other elements, materials, methods. In some cases, some operations related to the present application are not shown or described in the specification, this is to avoid the core part of the present application being overwhelmed by too much description, and for those skilled in the art, it is not necessary to describe these related operations in detail, and they can fully understand the related operations according to the description in the specification and the general technical knowledge in the art.

[0023] In addition, the features, operations or characteristics described in the specification can be combined in any appropriate manner to form various implementations. At the same time, the steps or actions in the method description can also be interchanged or adjusted in a manner that is obvious to those skilled in the art. Therefore, the various sequences in the specification and the drawings are only for the purpose of clearly describing a certain embodiment and are not meant to be a required sequence, unless otherwise specified that a certain sequence must be followed.

[0024] The first embodiment of the present invention provides a large-scale irrigation and fertilization system optimization method based on remote sensing leaf area index. Figure 1 The overall technical route is as follows: Figure 2 shown.

[0025] like Figure 1 As shown, in step S100, the target research area is divided into multiple sub-areas according to the growth characteristics and growth environment of the crops.

[0026] In this example, the selected study area covers most of the Fenwei River Valley Plain (34.42°–35.72°N, 109.26°–111.48°E; see Figure 3 ), including 16 irrigation areas. The climate in the region is semi-arid, with an average annual precipitation of 550 mm, an average annual temperature of 14.4°C, and an altitude range of 350 to 800 meters. The land in the study area is mainly used for agricultural production, with winter wheat and summer maize cropping in rotation. The sowing period of summer maize is early June and the harvest period is early October.

[0027] It is very difficult to provide precise planting management plans for each farmer in a large area. Surveys show that farmers tend to choose similar corn varieties in a specific geographical area. Therefore, the study area can be divided into several sub-areas based on the growth characteristics and growth environment of corn, and it is assumed that farmers in the same sub-area use the same corn variety. Since the X-means algorithm can estimate the number of clusters, the algorithm is used to divide the study area into three sub-areas, MZ1, MZ2, and MZ3, based on the normalized difference vegetation index (NDVI) of summer corn and the clay and silt content of the soil ( Figure 3 ). Soil data are from the National Earth System Science Data Center (http: / / www.geodata.cn). Figure 3 The black dots in the figure represent regional sampling points, and the yellow dots represent field experiment stations.

[0028] like Figure 1 As shown, in step S200, the LAI prediction model based on the support vector machine is trained using the measured leaf area index LAI and microwave remote sensing data in the target study area, and the trained LAI prediction model is used to predict the LAI estimation value of each sub-area based on the microwave remote sensing data.

[0029] Measured LAI: To determine the crop parameters for the regional-scale crop growth model, a model was constructed using remote sensing data and measured leaf area index (LAI) to estimate the LAI of summer maize at different growth stages in the study area. In 2021, 56 sampling areas, each with an area of ​​10 m × 10 m, were established in the study area to monitor the growth of summer maize at the regional scale. The sampling areas were all located away from roads and trees to avoid interference from roads and trees on satellite signals. The center of the sampling area was determined by the global positioning system (GPS). Within each sampling area, three independent sampling points were randomly selected, each with an area of ​​1 m × 1 m. The values ​​of the sampling indicators (such as LAI, soil water content SWC, plant height, and yield) in each sampling area were the average of the values ​​measured at the three sampling points. LAI, plant height, and SWC at a depth of 0–20 cm were measured at the V5 (fifth leaf node), V7 (seventh leaf node), V10 (tenth leaf node), and R1 (tasseling) growth stages, respectively, and yield was measured at the maize maturity stage. In addition, data such as sowing date, sowing depth, plant spacing, row spacing, irrigation time, irrigation amount, irrigation method, fertilization time, fertilizer amount and fertilization method were collected for each sampling area.

[0030] Remote sensing data: To determine the distribution of summer maize, high spatial resolution Sentinel-2 data (with a resolution of 10 m for blue, green, red and near-infrared bands and 20 m for shortwave infrared and red-edge bands) were used to calculate vegetation indices such as Normalized Difference Vegetation Index NDVI, Normalized Difference Water Index NDWI and Normalized Difference Building Index NDBI). Sentinel-2 is able to provide data on land surface and coastal areas every 5 days, and its spectral range extends from the visible spectrum to the shortwave infrared region. The top atmosphere data of Sentinel-2 have been archived in Google Earth Engine and are radiometrically and geometrically corrected. Therefore, the maize planting areas in the study area during the summer maize growing season of 2021 were extracted using a random forest classifier and vegetation indices calculated from Sentinel-2 top atmosphere data.

[0031] Based on the extracted summer corn planting area, the leaf area index (LAI) of corn was estimated using the backscatter information of Sentinel-1 data. Since the microwave remote sensing system measurement of Sentinel-1 is less affected by solar radiation and has the ability to work around the clock, its revisit period is 12 days or 6 days. Therefore, in cloudy and rainy areas, Sentinel-1 images are very suitable for estimating the LAI of corn. This embodiment uses the Sentinel-1 ground-resolved detection (GRD) product provided by Google Earth Engine, which has been preprocessed, radiometrically calibrated, and terrain corrected. In order to reduce coherent speckle noise, the Lee filtering method is used to filter the GRD product. The backscatter intensity and incident angle information of vertical transmission-vertical reception (VV, cross-polarization) and vertical transmission-horizontal reception (VH, co-polarization) are extracted from the GRD product.

[0032] Regional LAI estimation based on support vector machine: In order to estimate the leaf area index (LAI) of summer corn in the study area, this example uses support vector regression to fit the relationship between backscatter intensity, incident angle, LAI measurement date and the measured LAI in 2021.

[0033] The LAI measurement date is converted into the day of year format. SVR has been widely used in the estimation of maize LAI. SVR is a statistical machine learning method that transforms a low-dimensional nonlinear problem into a high-dimensional linear regression problem by mapping variables into a high-dimensional space.

[0034] The code of the SVR algorithm was obtained from the LIBSVM online library (https: / / www.csie.ntu.edu.tw / ~cjlin / libsvm / index.html). The SVR model uses a kernel function to map nonlinear data from a low-dimensional space to a high-dimensional space. The Gaussian radial basis function (RBF) kernel is selected as the kernel function for constructing the SVR model in this embodiment because it has fewer hyperparameters, higher accuracy, and stronger approximation ability. In the training of the SVR model, the parameters and are two key hyperparameters. The grid search method and five-fold cross validation method are used to find the best hyperparameter combination.

[0035] After determining the optimal hyperparameters, 224 field-scale LAI observations from 56 sampling areas and their corresponding remote sensing data were randomly divided into training and test sets. The training set contained 140 sets of remote sensing data and their corresponding field-measured LAI values, which were used to train the SVR-based LAI model; the test set contained 84 sets of remote sensing data and their corresponding field-measured LAI values, which were used to evaluate the model performance. The coefficient of determination R 2 The performance of the LAI estimation model was evaluated using the root mean square error (RMSE).

[0036] After calculation, for the training set, the LAI predicted by the developed LAI prediction model is very close to the measured LAI, R 2 = 0.93, RMSE = 0.49. For the test set, R 2 and RMSE are 0.89 and 0.59, respectively. The measured and predicted LAI values ​​are close to the 1:1 line. The developed LAI estimation model can predict LAI with acceptable accuracy. Therefore, this example uses the model to estimate LAI on a regional scale ( Figure 4 ), Figure 4 The map in is obtained by calculating the LAI of each maize pixel and then performing spatial interpolation.

[0037] like Figure 4 As shown in the figure, as the growth of summer maize progressed, the LAI of each sub-region (MZ1, MZ2 and MZ3) gradually increased. At the V5 stage, the LAI values ​​of MZ1 and MZ2 were not significantly different, and the LAI of these sub-regions were mainly concentrated between 0.4 and 1. The LAI of MZ3 was higher than that of the other two sub-regions, mainly concentrated between 0.6 and 1.59. At the V7 and V10 stages, the LAI of maize showed obvious differences among the sub-regions. At the R1 stage, the LAI of MZ1, MZ2 and MZ3 reached the maximum value, and the LAI of MZ3 was slightly lower than that of the other two sub-regions.

[0038] like Figure 1As shown, in step S300, a sensitivity analysis is performed on the crop variety coefficients of the regional scale crop growth model considering the spatial variability of soil and meteorological parameters to determine the sensitive variety coefficients and the non-sensitive variety coefficients.

[0039] In this embodiment, the DSSAT-CERES-Maize model is selected to simulate the growth process of corn.

[0040] The CERES-Maize model is embedded in the decision support system DSSAT to simulate the dynamic changes of maize leaf area index (LAI) over time. In addition, the CERES-Maize model can describe the daily growth and development process of maize under temperature stress, water stress and nitrogen fertilizer stress. To calculate the daily soil water balance, the model uses a one-dimensional "tilted bucket" soil water balance method to predict soil water flow and root water uptake in up to 10 soil layers.

[0041] The operation of the CERES-Maize model requires input of meteorological data, soil data, and field management information.

[0042] The meteorological data of Wugong and Yangling experimental stations were obtained from China Meteorological Data Service Center (http: / / data.cma.cn / ) and Yangling Meteorological Station, respectively. The daily solar radiation was calculated from the daily sunshine hours using the Angstrom formula.

[0043] The basic soil data required by the CERES-Maize model include soil bulk density, organic carbon content, clay content, silt content, field capacity, wilting coefficient, saturated water content and soil pH, all of which are obtained through laboratory measurements.

[0044] Field management information was obtained from the experimental records of Wugong and Yangling experimental stations. Six variety coefficients were obtained by calibrating the field-scale CERES-Maize model (see Table 1). The model was calibrated using field trial data (including LAI, SWC, and yield) from 2016 and 2020, and field trial data from 2017 and 2021 were used to validate the model.

[0045] Table 1 Range of variety coefficients in the CERES-Maize model

[0046] The performance of the field-scale CERES-Maize model was evaluated by the coefficient of determination R 2 , root mean square error (RMSE) and average relative error (ARE), and the calculation formula is as follows:

[0047] in, is the number of observations; O i and S i are the observed value and simulated value of the i-th data pair respectively; and are the average of the observed values ​​and the average of the simulated values, respectively. 2 When the value is close to 1 and the RMSE and ARE values ​​are close to 0, it indicates that there is a high consistency between the observed and simulated values.

[0048] The regionalization of the crop model is achieved by running the crop model for each simulation unit in the study area. Therefore, the input parameters of the crop model also need to be regionalized. Considering the research purpose and minimizing the computational cost, this embodiment uses a 5 km × 5 km grid to divide the study area into different simulation units.

[0049] The interpolation of meteorological data uses ordinary kriging to estimate the meteorological station data for each grid.

[0050] For soil data, in order to ensure the consistency of the resolution of soil data and meteorological data, gridded soil data with a resolution of 5×5 km² was generated by the inverse distance weighted interpolation method.

[0051] Field management information was obtained through field measurements and interviews with local farmers. Different varieties of the same crop usually show certain similarities, which suggests that the crop parameters calibrated by the field-scale crop model can be used as a reference for the calibration of the regional-scale crop model.

[0052] However, this embodiment requires a sensitivity analysis to determine which variety parameters need to be spatialized and which variety parameters can directly adopt the parameters of the calibrated field-scale crop model.

[0053] Sensitivity analysis: Since the CERES-Maize model contains many crop parameters, and the irrigation and fertilization data of the sampling area are obtained through surveys, there is a certain degree of uncertainty in the model. In order to accurately obtain crop parameters, this embodiment performs a sensitivity analysis to distinguish sensitive parameters from non-sensitive parameters.

[0054] This embodiment uses extended Fourier amplitude sensitivity test to identify sensitive parameters. EFAST is a global sensitivity analysis method based on variance decomposition. In this test, the variance of the model simulation results is decomposed to calculate the sensitivity index of each parameter. The higher the sensitivity index of a parameter, the stronger its impact on the model output.

[0055] like Figure 1As shown, in step S400, the coefficient value of non-sensitive varieties is obtained by calibrating field-scale test data, and the coefficient value of sensitive varieties is obtained by iteratively updating using a global optimization algorithm with the goal of minimizing the gap between the LAI value simulated by the regional-scale crop growth model and the LAI estimated value predicted by microwave remote sensing data.

[0056] Specifically, non-sensitive parameters were identified using relatively fine-scale field-scale experimental results, while sensitive parameters were identified using remotely sensed estimated LAI data for the entire study area.

[0057] In this example, the sensitive parameters of the CERES-Maize model are iteratively adjusted through the assimilation algorithm to minimize the difference between the LAI estimated by microwave remote sensing data and the LAI simulated by the regional-scale CERES-Maize model. To quantify this difference, the following cost function is constructed:

[0058] where J is the cost function, X is the vector of calibrated parameters (sensitive parameters, such as P2, P5 and sowing date below), is the LAI simulated by the regional-scale CERES-Maize model, is the LAI estimated from remote sensing data, Q is the number of observed LAI, and K is the number of simulation units. Represents the mean LAI of all pixels in each simulation unit.

[0059] In order to identify the sensitive variety coefficient of corn, this embodiment uses the SCE-UA optimization algorithm to minimize the cost function in the above formula. This algorithm is a global optimization algorithm that combines the advantages of deterministic search, random search and competitive evolution, and is insensitive to the initial value of the optimization parameter. If the optimized parameter value is within the predefined normalized geometric range (0.001) and the cost function does not decrease by more than 0.01% after 10 iterations, the SCE-UA algorithm is considered to have converged.

[0060] In this example, according to the results of the sensitivity analysis, P2, P5, and sowing date, which have the highest sensitivity index values, are re-estimated during the data assimilation process, and other parameters, including P1, G2, G3, and PHINT, are obtained by the calibrated field-scale CERES-Maize model. To obtain the parameters of the field-scale CERES-Maize model (Table 2), the model was calibrated using field data from 2016 and 2020, and then validated using field data from 2017 and 2021.

[0061] Table 2 Estimated variety coefficients and sowing dates of the CERES-Maize model

[0062] After testing, the LAI and yield simulated using the calibrated field-scale CERES-Maize model during the calibration phase closely matched the measured LAI and yield, with the corresponding LAI R 2 and RMSE were 0.97 and 0.21 m² / m² respectively, and the R 2 and RMSE were 0.95 and 178.80 kg / ha, respectively. During the validation phase, the field-scale crop model also performed well in simulating LAI and yield, with the corresponding R 2 and RMSE were 0.94 and 0.29 m² / m² respectively, and the R 2 and RMSE were 0.94 and 316.03 kg / ha, respectively.

[0063] Subsequently, the sensitive parameters (sowing date, P2, and P5) affecting daily LAI were identified using the EFAST method. These three sensitive parameters were re-estimated for the MZ1, MZ2, and MZ3 sub-regions and the entire region through the data assimilation method (Table 2).

[0064] And by comparison, the regional-scale CERES-Maize model has a higher R in simulating summer maize growth indicators (LAI, plant height, soil water content SWC and yield). 2 , RMSE, and MRE, it performed better than the field-scale CERES-Maize model. In addition, when the study area was divided into three sub-regions, the simulation performance of the regional-scale CERES-Maize model was better than that when the sub-regions were not divided. 2 The values ​​increased by 0.02, 0.04, 0.10 and 0.15, respectively, while the RMSE values ​​decreased by 0.36 m² / m², 0.04 m, 0.01 cm³ / cm³ and 203.33 kg / hectare, respectively, and the MRE values ​​decreased by 33.12%, 2.67%, 5.23% and 1.67%, respectively. In summary, the regional-scale crop model that uses the data assimilation method to determine crop parameters and sowing dates can more accurately reflect the growth characteristics of summer corn at the regional scale when the study area is divided into three sub-regions than when the sub-regions are not divided. Therefore, this embodiment uses the regional-scale CERES-Maize model to determine the optimal irrigation and nitrogen fertilizer application scheme (INS) for each sub-region.

[0065] like Figure 1As shown, in step S500, based on the obtained regional-scale crop growth model, a multi-objective genetic algorithm is used to perform multi-objective optimization on the irrigation and nitrogen fertilizer application plans in each sub-region in a typical rainfall year and determine the optimal solution to obtain the optimal irrigation and nitrogen fertilizer application plan for each sub-region.

[0066] In this example, the calibrated regional-scale CERES-Maize model was used to optimize the summer maize irrigation and nitrogen fertilizer application scheme (INS) in three sub-regions, MZ1, MZ2 and MZ3, in different typical rainfall years using a multi-objective optimization algorithm.

[0067] Determine the typical rainfall year type: Multi-year historical precipitation data during the crop growing season is crucial to identifying typical rainfall years. In order to determine the typical rainfall year at the regional scale, this embodiment uses ordinary Kriging interpolation to generate daily precipitation data with a spatial resolution of 5 km × 5 km for the study area from 1970 to 2020 for a total of 51 years. Subsequently, the daily average precipitation data of the three sub-regions MZ1, MZ2 and MZ3 were calculated. Next, the daily precipitation data are accumulated to determine the precipitation of 51 summer corn growing seasons in each sub-region. These precipitation data are used to identify typical rainfall years. Through empirical frequency analysis, it is determined in this embodiment that 25%, 50% and 75% of the precipitation cumulative frequency curve of each sub-region represent wet years, normal years and drought years, respectively.

[0068] Irrigation and fertilization schedule optimization model: The multi-objective optimization of irrigation and nitrogen application scheme (INS) is achieved by maximizing yield (Yield), water use efficiency (WUE) and partial nitrogen productivity (PFPN), which reflects the goals of pursuing high crop yield, saving water, improving nitrogen use efficiency (NUE) and reducing environmental pollution.

[0069] In the adopted multi-objective optimization algorithm, irrigation frequency, irrigation time, irrigation amount, nitrogen fertilizer application time and nitrogen fertilizer application amount are set as decision variables. The optimization objectives of crop yield, WUE and PFPN can be calculated and achieved through the regional scale CERES-Maize model.

[0070]

[0071] Where Y is the average crop yield in kg / ha; K is the number of simulation units in each sub-region; j is the number of irrigation times; the units of WUE and PFPN in the formula are kg / m 3 and kg / kg; according to the experience of farmers in the study area, the number of irrigations j∈{1,2,3}; in the summer corn growing season, part of the nitrogen fertilizer is used as base fertilizer, and the remaining nitrogen fertilizer is dripped with irrigation water in equal amounts; t1–t jrepresents the irrigation date, w1–w j represents the irrigation amount, f1–f j represents the amount of nitrogen fertilizer applied, f b is basal nitrogen fertilizer; when j = 1, w2 = w3 = 0 and f2 = f3 = 0; when j = 2, w3 = 0 and f3 = 0.

[0072] In this embodiment, the above multi-objective optimization problem is solved by the NSGA-III algorithm. The algorithm combines the principles of evolutionary optimization and multi-objective optimization, and is very suitable for optimizing complex nonlinear goals. Therefore, it is a powerful multi-objective decision-making tool. NSGA-III uses a specific encoding method to convert the decision variable data in the solution space into genetic space data with a genotype string structure. The converted data is similar to chromosomes. To reduce coding redundancy, genes j, a1, a2, a3, w1, w2, w3, β and f T Considered as part of chromosomes ( Figure 5 ). The meaning of each gene is shown in Table 3.

[0073] Table 3 Meaning of each gene in NSGA-III

[0074] Note: a1, a2 and a3 reflect the duration of the irrigation interval. Irrigation rate: The numbers 1, 2, ..., 7 correspond to 60, 65, ..., 90 mm of irrigation, respectively. Nitrogen fertilizer application rate: The numbers 1, 2, ..., 9 correspond to 90, 120, ..., 330 kg / ha of nitrogen, respectively.

[0075] Before using the CERES-Maize model to calculate the objective function, the chromosome data needs to be decoded into decision variables (i.e., irrigation date, irrigation amount, and nitrogen fertilizer application amount). To avoid conflicts between the previous and next irrigation times, the following formula is used for decoding to determine the irrigation date. The schematic diagram of the irrigation date is shown in Figure 6 shown.

[0076]

[0077] where p d It is the summer corn planting season. GP is the total reproductive period length, h d is the harvest date. The harvest date of summer maize in each subplot is October 1. a1, a2, and a3 reflect the duration of the irrigation intervals, and t1, t2, and t3 represent the irrigation dates. In addition, considering the local irrigation level, the irrigation amount and nitrogen application rate are set to multiples of 5, and the number of irrigations is rounded to an integer.

[0078] PlatEMO is a MATLAB-based multi-objective optimization platform that can be used to execute the NSGA-III algorithm. In this embodiment, a MATLAB program was written to process the input and output of DSSAT. The program generates an experimental file (.MZX file) in the DSSAT standard format, which includes detailed information on irrigation date, irrigation amount, nitrogen fertilizer application date, and nitrogen fertilizer application amount. Through the executable file named DSCSM047.EXE, the MATLAB program can obtain the yield (Yield), water use efficiency (WUE), and nitrogen partial productivity (PFPN) of the experiment. Therefore, the objective function of NSGA-III can be concisely and elegantly integrated into DSSAT. When the hypervolume curve appears visually smooth, NSGA-III is considered to have converged. In addition, to improve computational efficiency, the INS optimization process is run on an AMD Ryzen 7 1700 eight-core processor that supports the MATLAB parallel computing toolbox.

[0079] Determine the optimal fertigation system: NSGA-III is a multi-objective optimization algorithm that obtains the Pareto solution set by determining the best trade-offs between multiple conflicting objectives. Since the optimal solution in the solution set cannot maximize all objectives at the same time, the TOPSIS method is used in this example to determine the optimal irrigation and nitrogen fertilizer application scheme (INS).

[0080] In the TOPSIS method, the weight of each objective has an important influence on the decision-making results. The weight can be determined by subjective or objective weighting methods. Since it is difficult to obtain the subjective preference of the importance of the three objectives (i.e., maximizing yield, WUE, and PFPN) in this example, an objective weighting method is used in this example to determine the weights of these objectives.

[0081] Researchers have developed a variety of objective weighting methods, including entropy, principal component analysis (PCA), coefficient of variation (CV), and standard correlation discriminant (CRITIC). In the entropy and CV methods, the discreteness of the solution is used to describe the importance of each target, but the correlation between targets is ignored. The PCA method handles the correlation between targets by converting multiple related variables into several unrelated comprehensive variables, but it cannot reflect the discreteness of the solution. There is both a certain discreteness and a certain correlation between the three targets in this embodiment. The CRITIC method measures the discreteness and correlation between targets by standard deviation and correlation coefficient, respectively, which can better take into account these two characteristics.

[0082]

[0083] Among them, w u is the weight of the u-th target, σu is the standard deviation of the u-th target, r uv is the correlation coefficient between the uth and vth targets, and V (= 3) is the number of targets.

[0084] After determining the weight of each objective, the optimal irrigation and nitrogen application scheme (INS) was determined using the TOPSIS method.

[0085] The basic idea of ​​TOPSIS is to sort the solutions by calculating the distances from the solutions to the positive ideal solution and the negative ideal solution. The optimal solution is the solution that is closest to the positive ideal solution and farthest from the negative ideal solution. In the TOPSIS method, the information of the original data can be fully utilized. In addition, TOPSIS can fully reflect the gaps between solutions and has the advantages of accuracy, reliability and strong versatility.

[0086] By using TOPSIS, the Pareto solution sets of yield, WUE and PFPN are normalized in this embodiment to eliminate the dimensional differences between them. Then, the normalized target values ​​are multiplied by their corresponding weights to construct the weighted target matrix Z. The matrix Z is an M × V matrix, where M is the number of Pareto solutions and V is the number of targets. The positive ideal solution Zbest and the negative ideal solution Zworst are determined according to. These solutions are row vectors reflecting the optimal and worst solutions for yield, WUE and PFPN. The relative closeness RCi is calculated based on the Euclidean distance between the Pareto solution and the positive and negative ideal solutions.

[0087]

[0088] in, and are the positive and negative ideal solutions of the vth objective respectively. is the Euclidean distance between the mth Pareto solution and the positive ideal solution, is the Euclidean distance between the mth Pareto solution and the negative ideal solution (m = 1, …, M); RC m represents the relative closeness of the mth Pareto solution. m The closer it is to 1, the closer the corresponding Pareto solution is to the positive ideal solution and away from the negative ideal solution. In addition, according to the RC m The values ​​are sorted to determine the optimal solution.

[0089] The multi-objective optimization of irrigation and nitrogen fertilizer application scheme (INS) in each sub-region was carried out in three typical precipitation years, and the Pareto solution set ( Figure 7). Water use efficiency (WUE) increases with increasing yield until it begins to decline after a certain point. The goal of maximizing PFPN conflicts with the goal of maximizing WUE and yield. Therefore, the weights of yield, water use efficiency, and nitrogen partial productivity were determined for different sub-regions in three typical rainfall years by using the CRITIC method, and it was found that the difference in weights in each case was small. In order to reflect the overall correlation and conflict of the three objectives in different typical rainfall years and different sub-regions, the Pareto solution sets in the nine cases were integrated to determine the weights, and the final weights of yield, water use efficiency, and nitrogen partial productivity were 0.41, 0.28, and 0.31, respectively. After the weights were determined, the optimal solutions for different sub-regions in drought years, normal years, and wet years were determined by using the TOPSIS method ( Figure 7 The corresponding optimal INS is given in Table 4. For comparison, Table 4 also includes the INS currently used by farmers.

[0090] Table 4 Optimal and current irrigation and nitrogen fertilizer application schemes (INS) for summer maize in different typical rainfall years in each sub-region

[0091] Note: Irrigation date is expressed as days after sowing. The numbers outside the brackets are the relevant parameters of the optimized irrigation and nitrogen fertilizer application scheme (INS) obtained in this example, and the numbers in the brackets are the relevant parameters of the INS currently used by farmers. The data of the INS currently used by farmers come from two reports (Administration, 2020; Shaanxi, 2020) and a survey of farmers in 56 sampling areas. The average nitrogen fertilizer application rates of MZ1, MZ2 and MZ3 were 283, 265 and 242 kg / ha, respectively, and the corresponding standard deviations were 43, 62 and 55 kg / ha, respectively.

[0092] Corresponding to the above-disclosed large-scale irrigation and fertilization system optimization method based on remote sensing leaf area index, the embodiment of the present invention also discloses a large-scale irrigation and fertilization system optimization system based on remote sensing leaf area index, such as Figure 8 As shown, it specifically includes: A regional division module is used to divide the target research area into multiple sub-areas according to the growth characteristics and growth environment of crops; The LAI estimation module is used to train the LAI prediction model based on the support vector machine using the measured leaf area index LAI and microwave remote sensing data in the target study area, and use the trained LAI prediction model to predict the LAI estimation value of each sub-area based on the microwave remote sensing data; Sensitivity analysis module, which is used to conduct sensitivity analysis on crop variety coefficients of regional scale crop growth models considering spatial variability of soil and meteorological parameters, and determine sensitive variety coefficients and non-sensitive variety coefficients; Variety coefficient acquisition module, used to obtain the coefficient value of non-sensitive varieties through field scale test data calibration, and for sensitive variety coefficient values, the gap between the LAI value simulated by the regional scale crop growth model and the LAI estimated value predicted by microwave remote sensing data is minimized, and the coefficient value is obtained by iterative update using a global optimization algorithm; The multi-objective optimization module is used to optimize the irrigation and nitrogen fertilizer application plans in each sub-region under typical rainfall years based on the obtained regional-scale crop growth model and determine the optimal solution using a multi-objective genetic algorithm to obtain the optimal irrigation and nitrogen fertilizer application plan for each sub-region.

[0093] It should be noted that for the detailed description of a large-scale irrigation and fertilization system optimization system based on remote sensing leaf area index provided in an embodiment of the present invention, reference can be made to the relevant description of a large-scale irrigation and fertilization system optimization method based on remote sensing leaf area index provided in an embodiment of the present application, which will not be repeated here.

[0094] The above specific examples are used to illustrate the present invention, which are only used to help understand the present invention and are not used to limit the present invention. For those skilled in the art to which the present invention belongs, some simple deductions, modifications or substitutions can be made according to the idea of ​​the present invention.

Claims

1. A large-scale irrigation and fertilization system optimization method based on remote sensing leaf area index, characterized in that: The method comprises: The target study area is divided into multiple sub-areas according to the growth characteristics and growth environment of crops; The measured leaf area index (LAI) and microwave remote sensing data in the target study area were used to train the LAI prediction model based on support vector machine, and the trained LAI prediction model was used to predict the LAI estimation value of each sub-area based on microwave remote sensing data. Conduct sensitivity analysis on crop variety coefficients of regional scale crop growth models considering spatial variability of soil and meteorological parameters, and determine sensitive variety coefficients and non-sensitive variety coefficients; The coefficient values ​​of non-sensitive varieties are obtained by calibration with field-scale test data, and the coefficient values ​​of sensitive varieties are obtained by iterative updating using a global optimization algorithm with the goal of minimizing the gap between the LAI value simulated by the regional-scale crop growth model and the LAI estimated value predicted by microwave remote sensing data. Based on the regional-scale crop growth model, a multi-objective genetic algorithm was used to perform multi-objective optimization on the irrigation and nitrogen fertilizer application schemes in each sub-region in a typical rainfall year and determine the optimal solution, thus obtaining the optimal irrigation and nitrogen fertilizer application scheme for each sub-region.

2. A large-scale irrigation and fertilization system optimization method based on remote sensing leaf area index as claimed in claim 1, characterized in that: The target study area is divided into several sub-areas, including: The target study area was divided into multiple sub-areas using a clustering algorithm based on the normalized difference vegetation index of the target crop and the clay and silt content of the soil.

3. The large-scale irrigation and fertilization system optimization method based on remote sensing leaf area index according to claim 1, characterized in that: Using the measured leaf area index LAI and microwave remote sensing data in the target study area, the LAI prediction model based on support vector machine is trained, including: The target research area is divided into multiple sampling areas, and multiple independent sampling points are set in each sampling area. The sampling indicators include the LAI of the target crop at different growth stages. The value of each sampling indicator in each sampling area is the average value of the measured values ​​of multiple sampling points. The vegetation index was calculated using the high spatial resolution Sentinel-2 data. The random forest classifier was used to classify the vegetation index and extract the planting area of ​​the target crop in the study area. Based on the planting area of ​​the target crop, the remote sensing data including backscatter intensity and incident angle of the corresponding planting area were extracted from the Sentinel-1 data. The measured LAI and the corresponding remote sensing data are divided into training sets and test sets, and the LAI prediction model based on support vector machine is trained and tested.

4. The large-scale irrigation and fertilization system optimization method based on remote sensing leaf area index according to claim 1, characterized in that: A sensitivity analysis of crop variety coefficients of regional scale crop growth models considering spatial variability of soil and meteorological parameters was conducted to determine sensitive variety coefficients and non-sensitive variety coefficients, including: The EFAST method was used to conduct sensitivity analysis on crop variety coefficients of the regional-scale CERES-Maize model, and the coefficients were divided into sensitive variety coefficients and non-sensitive variety coefficients according to the calculated sensitivity index value of each variety coefficient.

5. The large-scale irrigation and fertilization system optimization method based on remote sensing leaf area index according to claim 1, characterized in that: The coefficient values ​​for non-sensitive varieties are calibrated through field scale test data, including: The field-scale CERES-Maize model was calibrated and verified using field trial data collected in different years, and the insensitive variety coefficient value was determined based on the variety coefficient value of the calibrated field-scale CERES-Maize model.

6. The large-scale irrigation and fertilization system optimization method based on remote sensing leaf area index according to claim 1, characterized in that: For the coefficient value of sensitive varieties, the goal is to minimize the gap between the LAI value simulated by the regional scale crop growth model and the LAI estimated value predicted by microwave remote sensing data, and the global optimization algorithm is used for iterative update, including: Divide the study area into different simulation units according to the preset grid; Construct the cost function: ； Where J is the cost function, X is the vector of calibrated sensitive parameters, is the LAI simulated by the regional-scale CERES-Maize model, is the LAI estimated from remote sensing data, Q is the number of observed LAI, K is the number of simulation units, Represents the mean LAI of all pixels in each simulation unit.

7. A large-scale irrigation and fertilization system optimization method based on remote sensing leaf area index as claimed in claim 6, characterized in that: For the coefficient value of sensitive varieties, the goal is to minimize the gap between the LAI value simulated by the regional scale crop growth model and the LAI estimated value predicted by microwave remote sensing data, and the global optimization algorithm is used for iterative update, including: The SCE-UA optimization algorithm is used to minimize the cost function. If the optimized variety coefficient value is within the predefined normalized geometric range and the cost function does not decrease by more than the preset value after several iterations, the SCE-UA algorithm is considered to have converged.

8. The large-scale irrigation and fertilization system optimization method based on remote sensing leaf area index according to claim 1, characterized in that: Based on the regional scale crop growth model, a multi-objective genetic algorithm was used to optimize the irrigation and nitrogen fertilizer application schemes in each sub-region in a typical rainfall year and determine the optimal solution, including: Based on the historical precipitation data of the target crops during the growing season, the typical rainfall years of each sub-region were identified, including wet years, normal years and dry years. In the three typical rainfall years, the irrigation and nitrogen fertilizer application schemes of each sub-region were optimized with multiple objectives. The multi-objective optimization of irrigation and nitrogen fertilizer application schemes is achieved by maximizing yield, water use efficiency WUE and nitrogen partial productivity PFPN, and the NSGA-III algorithm is used for multi-objective optimization. The regional-scale CERES-Maize model is used to calculate and achieve the optimization targets of crop yield, WUE, and PFPN: ； Where Y is the average crop yield in kg / ha; K is the number of simulation units in each sub-region; j is the number of irrigation times; the units of WUE and PFPN in the formula are kg / m 3 and kg / kg; according to the experience of farmers in the study area, the number of irrigations j∈{1, 2,3}; during the crop growing season, part of the nitrogen fertilizer is used as base fertilizer, and the remaining nitrogen fertilizer is dripped with irrigation water in equal amounts; t1–t j represents the irrigation date, w1–w j represents the irrigation amount, f1–f j represents the amount of nitrogen fertilizer applied, f b is basal nitrogen fertilizer; when j=1, w2=w3=0 and f2=f3=0; when j=2, w3=0 and f3=0.

9. The large-scale irrigation and fertilization system optimization method based on remote sensing leaf area index according to claim 1, characterized in that: Based on the regional scale crop growth model, a multi-objective genetic algorithm was used to optimize the irrigation and nitrogen fertilizer application schemes in each sub-region in a typical rainfall year and determine the optimal solution, including: The TOPSIS method was used to determine the optimal solution for the Pareto solution set of maximizing yield, water use efficiency WUE and nitrogen partial productivity PFPN obtained by the multi-objective optimization algorithm, thereby obtaining the optimal irrigation and nitrogen fertilizer application plan.

10. A large-scale irrigation and fertilization system optimization system based on remote sensing leaf area index, characterized in that: The system comprises: A regional division module is used to divide the target research area into multiple sub-areas according to the growth characteristics and growth environment of crops; The LAI estimation module is used to train the LAI prediction model based on the support vector machine using the measured leaf area index LAI and microwave remote sensing data in the target study area, and use the trained LAI prediction model to predict the LAI estimation value of each sub-area based on the microwave remote sensing data; Sensitivity analysis module, which is used to conduct sensitivity analysis on crop variety coefficients of regional scale crop growth models considering spatial variability of soil and meteorological parameters, and determine sensitive variety coefficients and non-sensitive variety coefficients; Variety coefficient acquisition module, used to obtain the coefficient value of non-sensitive varieties through field scale test data calibration, and for sensitive variety coefficient values, the gap between the LAI value simulated by the regional scale crop growth model and the LAI estimated value predicted by microwave remote sensing data is minimized, and the coefficient value is obtained by iterative update using a global optimization algorithm; The multi-objective optimization module is used to optimize the irrigation and nitrogen fertilizer application plans in each sub-region under typical rainfall years based on the obtained regional-scale crop growth model and determine the optimal solution using a multi-objective genetic algorithm to obtain the optimal irrigation and nitrogen fertilizer application plan for each sub-region.

Images