Automatic irrigation decision-making system based on WOFOST crop growth model

Abstract

The invention discloses an automatic irrigation decision-making system based on a WOFOST crop growth model, and relates to agricultural intelligent irrigation. Comprising a data acquisition module, a WOFOST model processing module, a remote sensing correction module, a water demand calculation module, a machine learning prediction module and an irrigation control module. The method comprises the following steps: acquiring and preprocessing multi-source data, simulating a crop growth process by using a WOFOST model, assimilating remote sensing inversion LAI by combining a data assimilation algorithm so as to improve simulation precision, calculating actual water demand based on corrected LAI and a Penman formula, predicting water demand in the future 3-7 days by using a machine learning algorithm, and automatically controlling operation of irrigation equipment. Organic fusion of a mechanism model and a data driving model is achieved, the problems that an existing system is poor in adaptability, low in prediction precision, lack of perspectiveness in decision making and the like are solved, the water saving rate reaches 25%-35%, the crop yield is increased by 12%-18%, and the system has the multi-crop adaptation capacity and is suitable for modern agricultural precise irrigation.

Description

Technical Field

[0001] This invention relates to the field of intelligent agricultural irrigation technology, and in particular to an automatic irrigation decision system based on the WOFOST crop growth model. Background Technology

[0002] Traditional irrigation systems often rely on fixed times or soil moisture thresholds for control, lacking dynamic responses to crop growth status, which can easily lead to over- or under-irrigation. While the WOFOST (World Food Studies Model), a classic mechanistic model, can simulate crop growth processes such as photosynthesis, respiration, and dry matter accumulation, its fixed parameters result in poor adaptability to complex climatic environments, different crop varieties, and varying growing seasons, limiting its simulation accuracy.

[0003] In existing technologies, remote sensing can retrieve vegetation growth parameters such as leaf area index (LAI). However, remote sensing monitoring and crop growth models are independent of each other and fail to be effectively integrated to dynamically correct model simulation results. This leads to discrepancies between model outputs and actual crop growth, making continuous real-time tracking impossible. Meanwhile, traditional irrigation decisions rely on current or historical soil moisture data, lacking forward-looking predictions of future water demand, resulting in low precision in irrigation timing and water volume control. While machine learning methods have advantages in prediction, they are often directly used for water demand prediction and are not closely integrated with crop physiological mechanism models, leading to prediction results lacking physiological significance and decreasing reliability under drastic environmental changes. Furthermore, the utilization of multi-source heterogeneous data from meteorology, soil, crops, and remote sensing is insufficient. Existing systems lack dynamic parameter calibration and feedback optimization mechanisms, making long-term performance prone to degradation and difficult to adapt to the needs of multiple crop types.

[0004] In summary, existing technologies suffer from problems such as poor crop model adaptability, insufficient integration of remote sensing and models, lack of foresight in irrigation decisions, disconnect between mechanism and data-driven models, insufficient system adaptability, and inadequate utilization of multi-source data. There is an urgent need for an integrated system that can achieve accurate prediction and automatic irrigation control. Summary of the Invention

[0005] The purpose of this invention is to address the lack of dynamic response to crop growth status in existing traditional irrigation systems by providing an automatic irrigation decision-making system based on the WOFOST crop growth model. This system integrates remote sensing data with machine learning algorithms to achieve accurate prediction of crop water requirements and automatic irrigation control.

[0006] To achieve the above-mentioned objectives, the present invention provides the following technical solution:

[0007] An automatic irrigation decision-making system based on the WOFOST crop growth model includes:

[0008] Data acquisition module: used to collect meteorological data, soil moisture data, crop growth data and remote sensing image data;

[0009] WOFOST model processing module: used to receive output data from the data acquisition module, simulate the crop growth process, invert and output the crop LAI value;

[0010] Remote sensing correction module: Used to receive the simulated LAI value from the WOFOST model processing module and the measured LAI value retrieved from remote sensing image inversion, and output the assimilated and corrected LAI value through the data assimilation algorithm;

[0011] Water demand calculation module: It is used to receive the corrected LAI value from the remote sensing correction module, and calculate the actual water demand of crops using the corrected Penman formula, which is used as the irrigation target value.

[0012] Machine learning prediction module: used to train models based on historical soil moisture data and meteorological data, use machine learning algorithms, input future meteorological data, and output predicted values ​​of crop water requirements for future periods;

[0013] Irrigation control module: Based on the received water demand forecast and the effective water storage capacity of the soil, it determines the timing and amount of irrigation, automatically controls the start and stop of irrigation equipment, and achieves precision irrigation.

[0014] Furthermore, the meteorological data of the data acquisition module includes solar radiation, air temperature, relative humidity, wind speed, precipitation, etc.; soil moisture data is the volumetric water content at a depth of 0-60cm in the crop root zone; and the remote sensing image data uses Sentinel-2 multispectral imagery.

[0015] Furthermore, the WOFOST model processing module is deployed based on the PCSE framework, and automatically runs daily to output simulated LAI values, biomass, and developmental stage data.

[0016] Furthermore, the data assimilation algorithm used in the remote sensing correction module is the ensemble Kalman filter (EnKF) algorithm.

[0017] Furthermore, the calculation formula of the water demand calculation module is as follows: , where f (Kc (LAI)) is the crop coefficient function driven by LAI, K_s is the water stress coefficient, and ET0 is the reference crop evapotranspiration.

[0018] Furthermore, the machine learning prediction module employs either the XGBoost algorithm or the LightGBM algorithm, and the evaluation metrics for model training include mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²) to ensure prediction accuracy.

[0019] Furthermore, the machine learning prediction module can output crop water requirement predictions 3 to 7 days in advance, and issue early warnings 24 to 48 hours in advance in case of abnormal situations.

[0020] Furthermore, the irrigation control module sends control commands to the irrigation controller via a wireless communication network (such as LoRa or 4G) to control the opening and closing of the solenoid valve and the operating time of the water pump.

[0021] Furthermore, the system supports outputting crop water requirements on a daily or hourly basis to adapt to different irrigation strategies.

[0022] Furthermore, the system also includes a crop parameter library for storing growth parameters such as effective accumulated temperature, specific leaf area dynamic table, and assimilate allocation table for different crops, supporting adaptation to multiple crop types.

[0023] The technical problems solved by this invention include:

[0024] 1. Problem of poor adaptability of traditional crop models: Existing crop growth models such as WOFOST have fixed parameters under complex environmental conditions, different crop varieties and variable growing seasons, resulting in low simulation accuracy and insufficient adaptability, and cannot accurately reflect the real growth status and water demand of crops.

[0025] 2. Problem of insufficient integration of remote sensing data and crop model: In the existing technology, remote sensing monitoring and crop growth model are independent of each other. The vegetation parameters (such as LAI) retrieved by remote sensing cannot be effectively used to dynamically correct the crop model, resulting in deviation between the model simulation results and the actual situation, and it is impossible to achieve continuous and real-time tracking of crop growth status.

[0026] 3. Lack of foresight in irrigation decision-making: Traditional irrigation systems rely heavily on current or historical soil moisture data for decision-making, lacking accurate predictions of future water demand. This results in imprecise timing and volume control of irrigation, making it impossible to achieve true on-demand irrigation.

[0027] 4. The problem of disconnect between machine learning applications and mechanistic models: Existing machine learning methods are mostly used directly for water demand prediction, and are not closely integrated with crop physiological mechanism models. The prediction results lack clear crop physiological significance, and the reliability of predictions is significantly reduced when environmental conditions change drastically.

[0028] 5. Problem of insufficient system adaptability: Existing irrigation systems generally lack effective feedback optimization mechanisms, and cannot dynamically adjust model parameters and control strategies according to crop growth process and environmental changes, resulting in a long-term decline in system performance.

[0029] 6. Insufficient utilization of multi-source data: Multi-source heterogeneous data such as meteorological, soil, crop, and remote sensing data have not been effectively integrated, resulting in insufficient data value mining and limiting the accuracy and reliability of irrigation decisions.

[0030] Compared with the prior art, the present invention has the following outstanding advantages and technical effects:

[0031] 1. Technological Integration and Innovation Advantages: This invention achieves four-dimensional data fusion of meteorological data, soil data, crop parameters, and remote sensing imagery by constructing a complete data acquisition and preprocessing module. Experiments show that this fusion scheme increases data utilization to 92.3%, approximately 45% higher than traditional single-data source methods. This invention innovatively combines the WOFOST mechanistic model with machine learning algorithms, preserving the physical meaning of crop physiological processes while fully utilizing the advantages of machine learning in handling nonlinear relationships. Test results show that this hybrid modeling method achieves an R² of 0.875 in LAI simulations during the cotton growing season, approximately 32% higher than the single WOFOST model.

[0032] 2. Significantly Improved Prediction Accuracy: This invention utilizes an assimilation algorithm based on ensemble Kalman filtering to effectively fuse remotely sensed LAI (Local Area Image) with WOFOST-simulated LAI. Experimental data shows that the mean absolute error (MAE) of the assimilated LAI decreased from 68.5 to 45.74, a reduction of 33.2%; the root mean square error (RMSE) decreased from 89.2 to 50.55, a reduction of 43.3%; and the coefficient of determination (R²) increased from 0.712 to 0.875. By using the XGBoost / LightGBM machine learning model to predict future water demand, this invention achieves the following technical effects: Daily prediction: MAE=45.74, RMSE=50.55, R²=0.875; Hourly prediction: MAE=8.68, RMSE=16.17, R²=0.986; Prediction timeliness: Crop water demand can be predicted 3-7 days in advance, providing sufficient preparation time for irrigation decisions.

[0033] 3. Improved Irrigation Precision and Efficiency: Enhanced Scientific Decision-Making: Based on accurate water demand prediction and combined with a soil moisture balance model, the system achieves: Irrigation Timing Accuracy: 95.2% (compared to approximately 70% for traditional methods); Irrigation Quantity Control Accuracy: ±8.7% (compared to over ±25% for traditional methods); Reduction in Ineffective Irrigation Frequency: 67.3%. Significantly Improved Water Resource Utilization Efficiency: In practical applications, the system demonstrates excellent water-saving effects: Overall Water Saving Rate: 25%–35%; Water Use Efficiency (WUE): Increased by 28.6%; Irrigation Water Productivity increased from 1.32 kg / m³ to 1.78 kg / m³.

[0034] 4. System Adaptability and Intelligence: Through a continuous feedback optimization mechanism, the system has the following characteristics: automatic model parameter calibration cycle: 7-15 days; after calibration, the model prediction accuracy remains consistently high: R²>0.85; environmental adaptability: it performs stably in different climate zones and soil types.

[0035] 5. Economic benefits and practicality: Labor management costs are reduced by 60%–70%; fertilizer utilization efficiency is increased by 18.5%, and overall production costs are reduced by 20%–25%. Cotton yield is increased by 12%–18%; fiber quality indicators are improved: length uniformity is increased by 5.2%; crop growth uniformity is improved: the coefficient of variation is reduced to below 0.15.

[0036] 6. Direct effects of technological innovation: Breaking through the limitation of traditional Kc coefficient lookup methods, which cannot be used for real-time continuous estimation of crop evapotranspiration throughout the entire growing season, the system achieves: Temporal resolution: supports hourly continuous estimation; Spatial resolution: achieves field-scale estimation by combining remote sensing data; Estimation continuity: uninterrupted monitoring throughout the entire growing season. Through the machine learning prediction module, the system possesses: Decision lead time: 3–7 days; Decision reliability: confidence level >90%; Anomaly warning: issues alerts 24–48 hours in advance. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the structural composition of an embodiment of the present invention.

[0038] Figure 2 This is a flowchart of the WOFOST model and remote sensing LAI assimilation process.

[0039] Figure 3 This is a flowchart of machine learning prediction and irrigation control. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the following embodiments will be used in conjunction with the accompanying drawings to further illustrate the invention. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0041] Example 1: Precision Irrigation System for Cotton Based on WOFOST and Remote Sensing Assimilation

[0042] This embodiment uses the cotton-growing region of Xinjiang as an example to describe in detail the implementation of this system. The system consists of six core parts connected in sequence, and its data flow and control flow are shown in the system structure block diagram (…). Figure 1 As shown in the diagram. Specifically:

[0043] (1) Data acquisition and preprocessing section

[0044] Meteorological data: Automatic weather stations were deployed in cotton-growing areas of Xinjiang to continuously record daily total solar radiation (IRRAD), daily maximum and minimum temperatures (TMAX, TMIN), relative humidity, wind speed, and precipitation.

[0045] Soil data: Install soil moisture sensors in the crop root zone (e.g., 0-20cm, 20-40cm, 40-60cm) to monitor volumetric water content in real time.

[0046] Crop parameters: Establish a local crop parameter library. For cotton, key parameters include: the effective accumulated temperature TSUM1 required from emergence to flowering (set at 800℃·d), the effective accumulated temperature TSUM2 required from flowering to maturity (set at 1550℃·d), as well as the specific leaf area dynamic table SLATB and the assimilate allocation table FRTB.

[0047] Remote sensing data: Sentinel-2 multispectral images are acquired through programming interfaces (such as Google Earth Engine) and preprocessed, including radiometric calibration, atmospheric correction, and cloud masking.

[0048] Data acquisition and preprocessing ensure the diversity, real-time nature, and accuracy of input data, providing a high-quality data foundation for subsequent model simulation and machine learning.

[0049] (2) WOFOST model simulation part

[0050] The WOFOST model is a core simulator for crop growth mechanisms. The WOFOST model is deployed using the PCSE framework. It runs automatically once daily, taking as input the aforementioned meteorological data, soil type, crop parameters, and initial conditions, and outputting simulated leaf area index (LAI), biomass, and developmental stage. The WOFOST model provides a dynamic, mechanism-driven simulation of crop growth processes; however, its simulated LAI values ​​may be biased under complex environments.

[0051] (3) Remote sensing data assimilation part

[0052] The remote sensing data assimilation process is crucial for improving the accuracy of model simulations, using data assimilation algorithms to correct the model. This includes:

[0053] Remote sensing LAI inversion: Using the PROSAIL physical model or the NDVI / LAI empirical conversion model, field-scale LAI observations are inverted from preprocessed remote sensing images.

[0054] Ensemble Kalman Filter Assimilation: The EnKF algorithm is employed. The LAI state set predicted by the WOFOST model is fused with the LAI observations retrieved from remote sensing. By calculating the Kalman gain, the model predictions and observational information are optimally combined to output corrected LAI analysis values ​​that more closely approximate the actual state. This process significantly reduces the uncertainty of the model simulation.

[0055] Experiments show that remote sensing data assimilation improves the simulation accuracy (R²) of LAI from about 0.71 before assimilation to over 0.88, achieving complementary advantages between the model and observations.

[0056] (4) Crop water requirement calculation section

[0057] The crop water requirement calculation section converts the growth status into a key indicator for irrigation decisions—water requirement. Based on the assimilated high-precision LAI, a modified Penman-Montes formula is used to calculate the actual crop evapotranspiration ETc (i.e., actual water requirement). A dynamic functional relationship is established between the crop coefficient Kc and LAI, and a water stress coefficient K_s based on real-time soil moisture content is introduced. The calculation formula is simplified as follows: The crop water requirement calculation yields a high-resolution and reliable estimate of the crop's current actual water requirement, which will serve as the target variable for training the machine learning model.

[0058] (5) Machine learning prediction part

[0059] Figure 3 A flowchart of machine learning prediction and irrigation control is presented. The machine learning prediction component endows the system with forward-looking decision-making capabilities. First, a feature dataset is constructed, including historical weather data, soil moisture data, crop phenological stages, and historical water requirements. Then, the XGBoost algorithm is employed. Using the feature dataset as input and the ETC output from the water requirements calculation component as the target variable, the prediction model is trained. The trained model can be input with future weather forecasts and other data, outputting predicted crop water requirements for the next 1 to 7 days. Machine learning prediction achieves high-precision prediction of future water requirements (RMSE of 50.55 and R² of 0.875 for predictions 7 days in advance), providing valuable lead time for irrigation decisions.

[0060] (6) Intelligent irrigation control section

[0061] The intelligent irrigation control unit is the system's execution terminal, used to achieve precision irrigation. It compares the predicted future water demand with the current effective soil water storage, and based on the principle of soil moisture balance, determines the timing of irrigation and calculates the required irrigation amount. A lower limit threshold for soil moisture is set; irrigation is triggered when the predicted soil moisture falls below this threshold. Control commands generated by the decision engine are sent to the irrigation controller in the field via a wireless communication network (such as LoRa or 4G), automatically controlling the opening and closing of solenoid valves and the duration of water pumps. Intelligent irrigation control enables on-demand irrigation. In its application in cotton-growing areas of Xinjiang, it has saved 28.5% of water compared to traditional irrigation methods, while increasing cotton yield by 15.3%.

[0062] This embodiment fully demonstrates the closed-loop intelligent decision-making process from data acquisition to irrigation execution. By assimilating the WOFOST model with remote sensing data, the problem of insufficient adaptability of mechanistic models in complex environments is solved; machine learning prediction enables forward-looking irrigation decisions; and finally, integrated control achieves significant water-saving and yield-increasing effects.

[0063] Example 2: Crop growth monitoring system based on WOFOST and remote sensing LAI assimilation

[0064] Figure 2 The WOFOST model and remote sensing LAI assimilation flowchart shown below illustrate the specific implementation of this invention in cotton crop growth monitoring.

[0065] 1. System Initialization Phase

[0066] System Startup and Parameter Loading: After system startup, the WOFOST model runtime environment is initialized first. Cotton-specific crop parameters are loaded: accumulated temperature parameters: TSUM1 = 800℃·d (emergence to flowering), TSUM2 = 1550℃·d (flowering to maturity); specific leaf area dynamic table SLATB: establishing the functional relationship between LAI and specific leaf area; assimilate allocation table FRTB: defining the dry matter allocation coefficients for each organ; meteorological data: including daily total solar radiation (IRRAD), daily maximum temperature (TMAX), daily minimum temperature (TMIN), etc.; soil parameters are configured: soil type is loam, and the initial soil moisture content is set to 70% of field capacity.

[0067] 2. Model simulation cycle: Daily growth simulation. The model enters the daily cycle simulation process:

[0068] 2.1) Time step progression: The simulation time is advanced in days;

[0069] 2.2) Photosynthesis simulation: Calculate potential photosynthetic yield based on daily solar radiation and temperature;

[0070] 2.3) Calculation of respiration: Consider the consumption of growth respiration and maintenance respiration;

[0071] 2.4) Dry matter accumulation: conversion of net photosynthetic products into biomass;

[0072] 2.5) Organ growth allocation: Assimilates are allocated to organs such as roots, stems, leaves, and seeds according to FRTB;

[0073] 2.6) Dynamic simulation of LAI: Calculate the daily LAI value based on leaf dry weight and specific leaf area.

[0074] 3. Remote sensing data processing: Acquisition of remote sensing LAI and parallel processing of remote sensing data:

[0075] 3.1) Acquire multispectral remote sensing imagery: Acquire Sentinel-2 satellite imagery every 5 days;

[0076] 3.2) Radiometric calibration and atmospheric correction: Converting digital quantization values ​​into surface reflectance;

[0077] 3.3) Cloud masking and quality control: Remove cloud-contaminated pixels to ensure data quality;

[0078] 3.4) Vegetation index calculation: Calculate vegetation indices such as NDVI and EVI;

[0079] 3.5) Remote sensing LAI inversion: LAI observations are obtained by inversion based on the PROSAIL radiative transfer model.

[0080] 4. Core processes of data assimilation:

[0081] Ensemble Kalman filter assimilation is performed when remote sensing data is available:

[0082] 4.1) Generate a set of state variables: Create a set of 50 members, considering the uncertainty of model parameters;

[0083] 4.2) Application of observation operators: Mapping the LAI state simulated by the model to the observation space;

[0084] 4.3) Calculate the Kalman gain: Calculate the optimal weights based on the forecast error covariance and the observation error covariance;

[0085] 4.4) State variable update: Perform assimilation update equation: Where Xa is the analyzed value, Xf is the predicted value, Y is the observed value, and K is the Kalman gain;

[0086] 4.5) Uncertainty analysis: Evaluate the uncertainty of the assimilated state variables.

[0087] 5. Model state update and loop:

[0088] 5.1) State variable update. Update the WOFOST model state variables using the assimilated LAI values; adjust related variables such as biomass and developmental stage simultaneously; ensure that the model state is consistent with remote sensing observations;

[0089] 5.2) Growing Season Determination. Determine if the growing season has ended; if not, return to step 2 to continue daily simulation; if it has ended, output the final assimilation result.

[0090] 6. Effectiveness evaluation and feedback optimization:

[0091] 6.1) Assimilation effect evaluation: Calculate the assimilation performance index, MAE=45.74, RMSE=50.55, R²=0.875; generate an assimilation process report, including the effect statistics of each assimilation.

[0092] 6.2) Parameter optimization feedback: Analyze the sources and distribution characteristics of assimilation error; adjust key model parameters to improve the accuracy of subsequent simulations; optimize assimilation scheme parameters, such as the number of ensembles and observation error settings.

[0093] Technical effectiveness verification:

[0094] In practical application during the 2024 Xinjiang cotton growing season, this system performed excellently:

[0095] 1) Improved assimilation accuracy. Before assimilation, the simulated R² of LAI was 0.712; after assimilation, the simulated R² of LAI was 0.875. The accuracy improvement was 22.9%.

[0096] 2) Growth monitoring effects. It can accurately capture the transition of key growth stages; it can monitor the decline in LAI caused by drought stress; and it provides reliable crop growth status information for irrigation decisions.

[0097] This embodiment fully demonstrates the effectiveness and practicality of the technical solution of the present invention in crop growth monitoring. By assimilating and fusing the WOFOST model with remote sensing LAI, the accuracy and reliability of crop growth monitoring are significantly improved.

[0098] Example 2: Automatic irrigation decision system adapted to multiple crops (wheat, corn adapted)

[0099] The system structure of this embodiment is the same as that of Embodiment 1. The main difference between different crops lies in the configuration of the crop parameter library:

[0100] Load wheat growth parameters: effective accumulated temperature from emergence to flowering TSUM1=650℃·d, effective accumulated temperature from flowering to maturity TSUM2=1200℃·d, and adapt to wheat SLATB and FRTB tables;

[0101] Load maize growth parameters: effective accumulated temperature from emergence to flowering TSUM1=900℃·d, effective accumulated temperature from flowering to maturity TSUM2=1600℃·d, and adapt to maize SLATB and FRTB tables.

[0102] The system operation process is the same as in Example 1. After testing, the water-saving rate in the wheat planting area reached 25%, and the yield increased by 12%; the water-saving rate in the corn planting area reached 30%, and the yield increased by 14%, proving that the system has the ability to adapt to multiple crops.

[0103] Experiments show that this invention achieves four-dimensional data fusion of meteorology, soil, crops, and remote sensing, increasing data utilization to 92.3%, approximately 45% higher than traditional single-data-source methods; the R² after LAI assimilation reaches 0.875, 32% higher than the single WOFOST model; the R² for predicting water demand in the next 3–7 days is ≥0.875, and the hourly prediction R² reaches 0.986; the accuracy of irrigation timing reaches 95.2%, the irrigation amount control accuracy is ±8.7%, the number of ineffective irrigations is reduced by 67.3%, the overall water saving rate is 25%–35%, and the water use efficiency is improved by 2%. 8.6%; This invention has an automatic model parameter calibration function with a cycle of 7 to 15 days. After calibration, the prediction accuracy remains R²>0.85, adapting to different climate zones and soil types; manual management costs are reduced by 60% to 70%, comprehensive production costs are reduced by 20% to 25%, crop yield is increased by 12% to 18%, and crop quality indicators are improved; This invention realizes real-time continuous monitoring and forward-looking decision-making throughout the entire growing season: it supports continuous estimation of water demand at the hourly and field scale, can provide irrigation decision support 3 to 7 days in advance, and provides early warning of abnormal situations 24 to 48 hours in advance.

[0104] This invention, through multi-level technological innovation and system optimization, has achieved significant breakthroughs in the accuracy of crop water requirement prediction, the scientific nature of irrigation decisions, and the efficiency of water resource utilization, providing a complete and reliable technical solution for precision irrigation in modern agriculture. Practical application demonstrates that the system possesses good stability, adaptability, and economic efficiency, exhibiting significant promotional value and application prospects.

[0105] The above embodiments are merely preferred embodiments of the present invention and should not be considered as limiting the scope of the present invention. All equivalent variations and improvements made within the scope of the present invention should still fall within the patent coverage of the present invention.

Claims

1. An automatic irrigation decision-making system based on the WOFOST crop growth model, characterized in that... include: Data acquisition module: used to collect meteorological data, soil moisture data, crop growth data and remote sensing image data; WOFOST model processing module: used to receive output data from the data acquisition module, simulate crop growth process, invert and output crop LAI value; Remote sensing correction module: Used to receive the simulated LAI value from the WOFOST model processing module and the measured LAI value retrieved from remote sensing image inversion, and output the assimilated and corrected LAI value through the data assimilation algorithm; Water demand calculation module: Used to receive the corrected LAI value from the remote sensing correction module, and use the corrected Penman formula to calculate the actual water demand of crops as the irrigation target value; Machine learning prediction module: used to train models based on historical soil moisture data and meteorological data, use machine learning algorithms, input future meteorological data, and output predicted values ​​of crop water requirements for future periods; Irrigation control module: Based on the received water demand forecast and the effective water storage capacity of the soil, it determines the timing and amount of irrigation, automatically controls the start and stop of irrigation equipment, and achieves precision irrigation.

2. The automatic irrigation decision-making system based on the WOFOST crop growth model as described in claim 1, characterized in that... The meteorological data acquired by the data acquisition module includes solar radiation, air temperature, relative humidity, wind speed, and precipitation; soil moisture data is the volumetric water content at a depth of 0–60 cm in the crop root zone; and remote sensing image data uses Sentinel-2 multispectral images.

3. The automatic irrigation decision-making system based on the WOFOST crop growth model as described in claim 1, characterized in that... The WOFOST model processing module is deployed based on the PCSE framework and runs automatically daily, outputting simulated LAI values, biomass, and developmental stage data.

4. The automatic irrigation decision-making system based on the WOFOST crop growth model as described in claim 1, characterized in that... The data assimilation algorithm used in the remote sensing correction module is the ensemble Kalman filter algorithm.

5. The automatic irrigation decision-making system based on the WOFOST crop growth model as described in claim 1, characterized in that... The calculation formula of the water demand calculation module is as follows: , where f (Kc (LAI)) is the crop coefficient function driven by LAI, K_s is the water stress coefficient, and ET0 is the reference crop evapotranspiration.

6. The automatic irrigation decision-making system based on the WOFOST crop growth model as described in claim 1, characterized in that... The machine learning prediction module uses either the XGBoost algorithm or the LightGBM algorithm. The evaluation metrics for model training include mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²) to ensure prediction accuracy.

7. The automatic irrigation decision-making system based on the WOFOST crop growth model as described in claim 1, characterized in that... The machine learning prediction module can output crop water requirement predictions 3 to 7 days in advance, and issue early warnings 24 to 48 hours in advance in case of abnormal situations.

8. The automatic irrigation decision-making system based on the WOFOST crop growth model as described in claim 1, characterized in that... The irrigation control module sends control commands to the irrigation controller via a LoRa or 4G wireless communication network to control the opening and closing of the solenoid valve and the operating time of the water pump.

9. The automatic irrigation decision-making system based on the WOFOST crop growth model as described in claim 1, characterized in that... The system supports outputting crop water requirements on a daily or hourly basis to adapt to different irrigation strategies.

10. The automatic irrigation decision-making system based on the WOFOST crop growth model as described in claim 1, characterized in that... It also includes a crop parameter library, which stores growth parameters such as effective accumulated temperature, specific leaf area dynamic table, and assimilate allocation table for different crops, and supports adaptation to multiple crop types.

Images