Multi-scheme collaborative yield prediction method based on data assimilation and model parameter optimization

Abstract

The application provides a multi-scheme cooperative yield prediction method based on data assimilation and model parameter optimization, and belongs to the field of agricultural information technology.The method comprises the following steps: obtaining historical and real-time data of a target region and a target crop growing season; constructing multiple combination simulation schemes of a WOFOST model; using an ensemble Kalman filter (EnKF) method combined with a Gaussian disturbance strategy to perform data assimilation on a leaf area index and soil humidity; performing sensitivity analysis and optimization on photosynthetic parameters of the WOFOST model, determining an optimal photosynthetic parameter combination, and running the model; improving a water stress function; constructing a rolling update yield prediction framework, dynamically optimizing yield prediction results; and dynamically selecting an optimal simulation strategy to perform yield simulation and prediction.The application significantly improves the yield simulation accuracy and prediction stability of the crop model under different water year types, and provides a reference for developing a new framework of crop model meteorological year type self-adaptive dynamic simulation.

Description

Technical Field

[0001] This invention relates to the field of agricultural information technology, and in particular to a multi-scheme collaborative yield forecasting method based on data assimilation and model parameter optimization. Background Technology

[0002] In recent years, drought has become particularly prominent due to the impact of global climate change. Uneven spatial and temporal distribution of precipitation and insufficient water supply during key growth periods severely inhibit the growth of crops such as corn, thereby affecting yields. Therefore, accurately simulating crop growth and yield changes under different water conditions is of great significance for disaster impact assessment and food security.

[0003] Crop growth models are important digital tools that have been widely used in crop management and yield prediction, and have made significant progress in their long-term development. Among the many models, the WOFOST model is a general-purpose crop growth mechanism model. This model simulates the phenological development of crops by setting different crop parameters and using a daily time step, covering key physiological and ecological processes such as leaf growth and light interception, CO2 assimilation, respiration, transpiration, and dry matter accumulation and distribution. Due to its strong mechanistic logic and relatively mature development, the WOFOST model has been widely used for simulating the growth and yield of major food crops such as maize. Furthermore, to facilitate the use of the WOFOST model in a Python environment, the PCSE platform provides flexible interfaces and a modular design, making crop growth simulation, parameter setting, and result output more convenient.

[0004] Although the WOFOST model is capable of simulating crop growth under water-limited conditions, it still exhibits significant biases in simulating crop growth dynamics and yield under water stress. The current WOFOST model uses a linear water stress response function (RFWS), assuming that soil moisture reduction is proportional to transpiration rate inhibition. However, extensive experimental evidence shows that in actual drought conditions, some crops are insensitive to water stress in the early stages, with transpiration rates only declining rapidly when water levels approach a critical threshold. This linear assumption leads to underestimation of yield in mild drought and overestimation in moderate to severe drought. Furthermore, model parameters (such as maximum photosynthetic rate AMAXTB) lack localized optimization, making it difficult to adapt to the physiological characteristics of different crops in different regions. Single data assimilation schemes (such as assimilating only leaf area index or soil moisture) cannot meet the needs of both normal and drought years, and bivariate assimilation is prone to reduced accuracy due to information redundancy. In addition, existing models lack clear criteria for stable yield prediction nodes and drought thresholds, hindering dynamic adaptive simulation. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-scheme collaborative yield forecasting method based on data assimilation and model parameter optimization. It selects an appropriate assimilation strategy based on climate conditions to achieve collaborative assimilation of multi-source data, thereby improving the simulation accuracy of crop yield. Through data assimilation, parameter optimization, function improvement and rolling forecast integration, it constructs a differentiated simulation strategy, which significantly improves the adaptability and simulation accuracy of the WOFOST model under different moisture conditions.

[0006] To achieve the above objectives, this invention proposes a multi-scheme collaborative yield forecasting method based on data assimilation and model parameter optimization, comprising the following steps:

[0007] Step S1: Obtain historical and real-time data on the target area and the growth period of the target crop, including meteorological data, crop parameters, soil parameters, and agricultural management data;

[0008] Step S2: By judging the data assimilation status, photosynthetic parameter optimization status, water stress function improvement status, and cumulative precipitation threshold, different situations are combined to construct a combined simulation scheme for the WOFOST model;

[0009] Step S3: Using the ensemble Kalman filter (EnKF) method combined with a Gaussian perturbation strategy, the leaf area index is... With soil moisture Perform data assimilation;

[0010] Step S4: Use the EFAST method to perform sensitivity analysis on the WOFOST model parameters to determine the key parameters affecting crop yield, optimize the key parameters of the WOFOST model based on the optimization algorithm, and determine the optimal combination of photosynthetic parameters;

[0011] Step S5: Improve the water stress function into a piecewise function;

[0012] Step S6: Construct a rolling yield forecasting framework, identify the forecast stability window based on the difference between the predicted yield and the actual yield, define the time period closest to the crop maturity as the optimal node, record the optimal node, and dynamically optimize the yield forecast results.

[0013] Step S7: Evaluate the combined simulation scheme using the mean absolute relative error (MAER) and mean absolute error (MAE), and dynamically select the optimal simulation strategy for production simulation and forecasting based on the evaluation results.

[0014] Preferably, in step S1, the growth period includes the three-leaf stage, seven-leaf stage, jointing stage, tasseling stage, milk stage, and maturity stage; the meteorological data includes daily maximum temperature, daily minimum temperature, wind speed, precipitation, sunshine duration, water vapor pressure, and solar radiation; the crop parameters include leaf area index, leaf biomass, stem biomass, and fruit biomass; and the soil parameters include soil moisture, wilting moisture, field capacity, and saturated water content.

[0015] Preferably, in step S2, the combined simulation scheme includes: the original simulation scheme without data assimilation, Data assimilation simulation scheme Data assimilation and photosynthetic parameter optimization collaborative simulation scheme Data assimilation simulation scheme Data assimilation and photosynthetic parameter optimization collaborative simulation scheme and Bivariate data assimilation simulation scheme and A collaborative simulation scheme for bivariate data assimilation and photosynthetic parameter optimization. A collaborative simulation scheme for data assimilation, photosynthetic parameter optimization, and improvement of water stress function. A collaborative simulation scheme combining data assimilation, photosynthetic parameter optimization, and water stress function improvement; and a simulation scheme considering the cumulative precipitation threshold.

[0016] Preferably, in the simulation scheme considering the cumulative precipitation threshold, combined with the improved strategy of the water stress function, a parameterization scheme for the precipitation-triggered assimilation strategy based on the cumulative precipitation during the key growth period of crops is designed, the critical value for drought occurrence is defined, the daily canopy evapotranspiration is calculated, and the drought threshold is obtained. The formula for calculating the daily canopy evapotranspiration is as follows:

[0017] ;

[0018] in, This represents daily canopy evapotranspiration. Field holding capacity The point of wilting. For crop coefficients, For reference crop evapotranspiration.

[0019] Preferably, step S3 includes the following steps:

[0020] Step S31: Obtain and ,right and Apply a Gaussian perturbation to generate the perturbed result. and Constructing an observation set ;

[0021] Step S32: Filter sensitive model parameters, apply uncertainty perturbations to the sensitive model parameters, generate parameter combinations, and construct a parameter set. ;

[0022] Step S33: Set the quantitative gradient of parameter combinations, calculate the model simulation accuracy of different quantitative gradient parameter combinations, and select the quantitative gradient with the highest model simulation accuracy, which is recorded as the optimal set size;

[0023] Step S34: Input the parameter combinations from the parameter set into the WOFOST model, run the WOFOST model, obtain the simulated values ​​corresponding to the parameter combinations, and construct the prediction set. ;

[0024] Step S35: Find the perturbation-containing and At any given moment, the Ensemble Kalman Filter (EnKF) method is used to fuse the prediction set and the observation set, and update... and And output the assimilated data and Until the next moment.

[0025] Preferably, in step S4, during the parameter optimization process, the phenological process is kept unchanged, and the yield difference before and after parameter optimization and the growth simulation difference are evaluated. The yield difference is evaluated by calculating the absolute and relative errors between the predicted yield and the actual yield before and after model optimization, and the growth simulation difference is evaluated by comparing the deviation between the simulated values ​​of crop parameters and the actual crop parameters before and after model optimization.

[0026] Preferably, in step S5, the improved water stress function is:

[0027] ;

[0028] in, It is a water stress reduction factor. For continuity coefficients, This represents the current soil moisture content. The soil moisture content at the wilting point. This is the critical water content.

[0029] Preferably, step S6 includes the following steps:

[0030] Step S61: Construct a rolling yield forecast framework. From the crop jointing stage to the crop maturity stage, set the time interval for rolling simulation of the model, acquire and import meteorological data to drive the model to perform rolling simulation, and output the predicted yield results.

[0031] Step S62: Design a prediction stable window identification method, obtain the predicted yield results, calculate the relative error between the predicted yield results and the measured yield, set a relative error threshold, record the number of times the relative error is less than the relative error threshold, determine and record the prediction stable window, and retain the stable window closest to the maturity period.

[0032] Step S63: Obtain the stable window closest to maturity, and record the earliest predicted time point that meets the stability conditions within the stable window closest to maturity as the optimal node, and dynamically optimize the yield prediction results.

[0033] Preferably, in step S61, during the rolling simulation process, the EnKF algorithm is used to assimilate and update the measured state variables in real time, and to dynamically correct the predicted crop growth parameters in the model.

[0034] Preferably, in step S7, the formula for calculating the Mean Absolute Relative Error (MAER) is:

[0035] ;

[0036] in, The mean absolute relative error, For simulating quantities, For the first One simulated value, For the first One observation value;

[0037] The formula for calculating the Mean Absolute Error (MAE) is:

[0038] ;

[0039] in, This represents the mean absolute error.

[0040] Therefore, this invention proposes a multi-scheme collaborative yield forecasting method based on data assimilation and model parameter optimization, the beneficial effects of which are as follows:

[0041] (1) This invention significantly reduces the simulation error of the model in different water year types by using a differentiated combination simulation scheme, effectively solves the defect of the original WOFOST model in the inaccurate response to water stress, and provides a quantitative tool for drought impact assessment.

[0042] (2) This invention innovatively proposes a dynamic simulation strategy based on precipitation threshold, which can automatically select the optimal assimilation scheme according to real-time moisture conditions. At the same time, by determining the best node for yield prediction, it takes into account both the lead time and stability of the forecast, and realizes the leap from static simulation to dynamic adaptive forecast.

[0043] (3) This invention optimizes photosynthetic parameters to accurately reflect the photosynthetic characteristics of crops during their growth period. At the same time, it changes the water stress function from linear to piecewise nonlinear, which more accurately describes the nonlinear response of crops to drought and enhances the rationality and universality of the model.

[0044] (4) The present invention integrates real-time data to dynamically correct predictions through the constructed rolling forecast framework, thereby reducing the randomness error of a single simulation and taking into account both the prediction accuracy and continuity of the model. Attached Figure Description

[0045] Figure 1 A flowchart of a multi-scheme collaborative yield forecasting method based on data assimilation and model parameter optimization;

[0046] Figure 2 Schematic diagram of soil relative humidity during the corn growing season in different years;

[0047] Figure 3 This is a schematic diagram comparing the water stress function before and after the improvement.

[0048] Figure 4 A schematic diagram comparing the simulation results with measured values ​​of the combined simulation scheme of the constructed WOFOST model;

[0049] Figure 5 A schematic diagram comparing the simulation results with measured values ​​for three sowing periods of the combined simulation scheme of the constructed WOFOST model;

[0050] Figure 6 Comparison of simulated and measured yields of maize rolling forecasts for sowing period II;

[0051] Figure 7 A schematic diagram comparing the predicted and actual production values ​​of the combined simulation scheme of the constructed WOFOST model;

[0052] Figure 8 A schematic diagram comparing the combined simulation scheme of the WOFOST model constructed under sowing period III with the measured values ​​of TAGP;

[0053] Figure 9 For the second half of the broadcast period A schematic diagram of the rolling forecast results of maize yield using the data assimilation simulation scheme. Detailed Implementation

[0054] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0055] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0056] Example 1

[0057] like Figure 1 As shown, this invention provides a multi-scheme collaborative yield forecasting method based on data assimilation and model parameter optimization, comprising the following steps:

[0058] Step S1: Obtain historical and real-time data for the target area and the target crop's growing season, including meteorological data, crop parameters, soil parameters, and agricultural management data. The growing season includes the three-leaf stage, seven-leaf stage, jointing stage, tasseling stage, milk stage, and maturity stage. Meteorological data includes daily maximum temperature, daily minimum temperature, wind speed, precipitation, sunshine duration, water vapor pressure, and solar radiation. Crop parameters include leaf area index, leaf biomass, stem biomass, and fruit biomass. Soil parameters include soil moisture, wilting moisture, field capacity, and saturated water content.

[0059] Under rainfed conditions, the maize varieties observed were Danyu 39 and Danyu 405, which belong to the same series and have basically the same genetic parameters. Three sowing periods were set, namely sowing period I, sowing period II and sowing period III.

[0060] To reflect soil moisture conditions during maize growth, the soil relative humidity was measured at 10 cm intervals in the 0–50 cm soil layer using the soil auger method. The statistical results are as follows: Figure 2 Years with soil relative humidity greater than or equal to 60% are defined as water-sufficient years, and years with soil relative humidity less than 60% are defined as drought years. The collected data covers both water-sufficient and drought years, providing reliable data for analyzing the simulation performance of the WOFOST model under different moisture conditions.

[0061] Step S2: By assessing data assimilation, photosynthetic parameter optimization, water stress function improvement, and cumulative precipitation threshold, different scenarios are combined to construct a combined simulation scheme for the WOFOST model, as shown in Table 1. The combined simulation scheme includes: the original simulation scheme without data assimilation, ... Data assimilation simulation scheme Data assimilation and photosynthetic parameter optimization collaborative simulation scheme Data assimilation simulation scheme Data assimilation and photosynthetic parameter optimization collaborative simulation scheme and Bivariate data assimilation simulation scheme and A collaborative simulation scheme for bivariate data assimilation and photosynthetic parameter optimization. A collaborative simulation scheme for data assimilation, photosynthetic parameter optimization, and improvement of water stress function. A collaborative simulation scheme for data assimilation, photosynthetic parameter optimization, and improved water stress function; and a simulation scheme that considers the cumulative precipitation threshold.

[0062] Table 1. Multiple combined simulation schemes

[0063]

[0064] In years with ample water availability, data assimilation based on leaf area index significantly improves yield simulation accuracy, while in years with limited water availability, soil moisture assimilation is more effective. Temperature and precipitation are the main meteorological factors affecting maize's final yield, and maize growth and development are typically controlled by the cumulative effect of meteorological factors. Therefore, combining an improved water stress function strategy, a precipitation-triggered assimilation strategy parameterization scheme based on cumulative precipitation during key maize growth stages was designed to improve the model's adaptability and prediction accuracy under different water conditions.

[0065] Defining a soil relative humidity of 70% as the critical value for drought occurrence, we obtain the expression for daily canopy evapotranspiration:

[0066] ;

[0067] in, This represents daily canopy evapotranspiration. Field holding capacity The point of wilting. For crop coefficients, For reference crop evapotranspiration.

[0068] During the critical growth period of maize, from the jointing stage to the tasseling stage, if factors such as irrigation, groundwater recharge, and runoff are not considered, the cumulative precipitation and cumulative evapotranspiration during this period can be approximated as balanced. Therefore, the cumulative evapotranspiration during this period can be used as the threshold precipitation for drought assessment. To avoid interference from extreme years, daily meteorological data from years with normal precipitation were selected, and the cumulative evapotranspiration during each of the three sowing periods (Sowing Period I, Sowing Period II, and Sowing Period III) was calculated. The average value was then used as the drought threshold.

[0069] In rolling forecast simulations, the cumulative precipitation for each stage changes continuously as daily meteorological data is updated. Correspondingly, the simulation strategy based on threshold judgments is dynamically adjusted: when the cumulative precipitation for a stage is higher than the drought threshold, the S3 scheme is used; when the cumulative precipitation is lower than or equal to the threshold, it is determined to be water-limited, and the S9 scheme is used. As the rolling process progresses, when the cumulative precipitation for a stage is fully known, the simulation strategy remains fixed, thus ensuring the determinism and repeatability of the final simulation scheme.

[0070] Step S3: Using the ensemble Kalman filter (EnKF) method combined with a Gaussian perturbation strategy, the leaf area index is... With soil moisture Data assimilation includes the following steps:

[0071] Step S31: Obtain and ,right and Apply a Gaussian perturbation to generate the perturbed result. and Constructing an observation set ;

[0072] Step S32: Filter sensitive model parameters, apply uncertainty perturbations to the sensitive model parameters, generate parameter combinations, and construct a parameter set. ;

[0073] Step S33: Set five parameter combinations for quantitative gradients: 10, 30, 50, 70, and 90. Calculate the model simulation accuracy for different quantitative gradient parameter combinations. Select the quantitative gradient with the highest model simulation accuracy and denote it as the optimal set size.

[0074] Step S34: When observation data exists at time t, input the parameter combinations from the parameter set into the WOFOST model, run the WOFOST model, obtain the simulated values ​​corresponding to the parameter combinations, and construct the prediction set. When there is no observation data at time t, the model maintains its current state and runs until the next time step.

[0075] Step S35: Find the perturbation-containing and At any given moment, the Ensemble Kalman Filter (EnKF) method is used to fuse the prediction set and the observation set, and update... and And output the assimilated data and At the next moment, the subsequent simulation will be conducted.

[0076] During the data assimilation process, all set members run independently, and the final output is the mean of the simulation results of the set members.

[0077] Step S4: Before running the WOFOST model, sensitivity analysis and key parameter optimization are performed to improve model performance. The EFAST method is used to perform sensitivity analysis on the WOFOST model parameters to determine the key parameter AMAXTB that affects crop yield. Based on trial and error, the optimization is performed within a reasonable range. The key parameter AMAXTB of the WOFOST model was optimized and adjusted to determine the optimal combination of photosynthetic parameters:

[0078] ;

[0079] ;

[0080] ;

[0081] ;

[0082] ;

[0083] This parameter combination reflects the gradual decrease in photosynthetic potential of maize during the transition from vegetative to reproductive growth. During parameter optimization, the phenological process is kept constant, and the differences in yield and growth simulation before and after parameter optimization are evaluated. The yield difference is evaluated by calculating the absolute and relative errors between the crop yield before and after optimization and the measured crop yield. The difference in growth simulation is evaluated by comparing the deviations between the simulated values ​​and the measured values ​​of crop growth indicators before and after optimization.

[0084] Step S5: In crop models, transpiration is a key link connecting soil moisture and crop photosynthetic rate, significantly impacting yield formation. The WOFOST model calculates actual transpiration under different environmental conditions by multiplying the canopy's potential transpiration rate by water and oxygen stress reduction factors, as shown in the following formula:

[0085] ;

[0086] ;

[0087] ;

[0088] in, This is the actual transpiration rate. For the maximum potential transpiration rate, The water stress reduction factor has a value range of [value range missing]. , The oxygen stress reduction factor has a value range of [value range missing]. The default value here is 1. To reference the potential evapotranspiration of crops, The light interception coefficient is the canopy light. Leaf area index, This represents the current soil moisture content. The soil moisture content at the wilting point. This is the critical water content;

[0089] Most crops exhibit nonlinear responses to different water stress conditions; that is, crops are not sensitive to water stress in the early stages of drought, but their sensitivity increases as drought intensifies. However, in the WOFOST model... The function takes linear form. This means that once the relative soil moisture content falls below field capacity, the transpiration rate immediately decreases linearly. This treatment makes the model overly sensitive to water stress during mild drought stages, failing to reflect the buffering capacity that crops typically possess in the early stages of drought.

[0090] To address this deficiency, a power function form is used in the DSSAT-CERES model. describe This method delays the appearance of water stress effects under mild drought conditions, which is more consistent with the nonlinear response characteristics of crops to water conditions. However, a single power function may underestimate the stress intensity during severe drought and is difficult to take into account the response characteristics of different soil moisture ranges.

[0091] Therefore, this study proposes a piecewise function improvement scheme: when the relative soil moisture content is ≤0.6, a linear function is used to describe the water stress effect to ensure the sensitivity of the model under moderate to severe drought conditions; when the relative soil moisture content is >0.6, an exponential function (power of 0.3) is used to reduce the premature suppression of transpiration rate under mild drought. Simultaneously, a continuity coefficient A is used to ensure smooth connection of the function at the segmentation points. This piecewise function improvement scheme is used to refine the water stress function. The improved water stress function is as follows:

[0092] ;

[0093] in, This is the continuity coefficient.

[0094] like Figure 3 As shown, compared with the original linear function and power function schemes, the piecewise function can reflect the buffering effect under mild drought and maintain the rapid response under severe drought, thus improving the model's adaptability and universality to different soil moisture conditions.

[0095] Step S6: Construct a rolling yield forecasting framework. Based on the difference between predicted and actual yields, identify the forecast stability window, define the time period closest to crop maturity as the optimal node, record the optimal node, and dynamically optimize the yield forecast results, including the following steps:

[0096] Step S61: Construct a rolling yield forecast framework. From the corn jointing stage to the corn maturity stage, set the time interval for rolling model simulation to 10 days. Acquire and import meteorological data to drive the model for rolling simulation, and output the predicted yield results. During the rolling simulation, the EnKF algorithm is used for real-time assimilation and updating. and For periods when there is no actual meteorological data in the future, the daily average meteorological data of the meteorological station over the past 30 years are spliced ​​together to ensure the continuity of the simulation and reduce the impact of future meteorological uncertainties on the prediction results. As the model simulation progresses, the yield prediction results are dynamically corrected. As the real driving data is gradually improved and the measured state variables are continuously assimilated, the model prediction error should gradually decrease and tend to stabilize.

[0097] Step S62: Design a prediction stable window identification method, obtain the predicted yield result, calculate the relative error between the predicted yield result and the measured yield, set a relative error threshold, record the number of times the relative error is less than the relative error threshold, determine and record the prediction stable window, and retain the stable window closest to the maturity period. Specifically, if the relative error is less than 15% and the number of occurrences is not less than 5, it is determined to be a stable window.

[0098] Step S63: Obtain the stable window closest to maturity. Define the earliest predicted time point in the stable window closest to maturity as the optimal node, record the optimal node, and dynamically optimize the yield prediction results. Considering that multiple stable windows may occur during the maize growth period, this paper backtracks from maturity and only retains the time period closest to maturity. The earliest predicted time in this time period that meets the above conditions is recorded as the optimal node, and the yield prediction results are dynamically optimized to ensure that the prediction has entered the final stable stage and avoid interference from fluctuations after a brief period of stability.

[0099] Step S7: Evaluate the combined simulation scheme using the Mean Absolute Relative Error (MAER) and the Mean Absolute Error (MAE). Based on the evaluation results, dynamically select the optimal simulation strategy for yield simulation and forecasting. The formula for calculating the Mean Absolute Relative Error (MAER) is as follows:

[0100] ;

[0101] in, The mean absolute relative error, For simulating quantities, For the first One simulated value, For the first One observation value;

[0102] The formula for calculating the Mean Absolute Error (MAE) is:

[0103] ;

[0104] in, This represents the mean absolute error.

[0105] The MARE value reflects the relative deviation between the simulation results and the observed data; the smaller the value, the higher the accuracy of the model simulation. The MAE value, on the other hand, directly reflects the deviation between the simulation results and the observed values ​​in absolute terms.

[0106] The results were analyzed, and the findings are as follows:

[0107] (1) Simulation results of maize yield under different climatic conditions under different data assimilation schemes and whether or not parameter optimization is performed are as follows: Figure 4 As shown: In years with sufficient moisture, the MARE of the predicted yield for scheme S2 is only 12.1%, significantly better than 20.6% for scheme S4 ​​and 14.7% for scheme S6, indicating that... This approach can more effectively characterize crop growth status and improve yield simulation accuracy. In drought years, Scheme S4 ​​predicts a yield MARE of 39%, which is better than Scheme S2's 49.3% and Scheme S6's 45.1%, more accurately reflecting the limiting effect of soil moisture on yield formation. Bivariate assimilation did not show optimal accuracy in either type of year, indicating... and Redundancy or cross-interference between information sources may weaken the information fusion effect. Compared with the original simulation, all three assimilation schemes significantly reduced the systematic underestimation bias in all years, indicating the key role of data assimilation in improving the yield simulation accuracy of the WOFOST model. After optimizing the key photosynthetic parameter AMAXTB, the simulated yields of each assimilation scheme are closer to the measured values: in years with sufficient moisture, the MARE of the predicted yield by scheme S3 is 11.5%, which is better than the 12.1% of scheme S2; in years with drought, the MARE of the predicted yield by scheme S5 is 38.4%, which is better than the 39% of scheme S4. However, the MARE in drought years is still higher than that in normal years, and the yield is still significantly underestimated under extreme drought conditions, which is presumably related to the excessive sensitivity of the original water stress function RFWS to soil moisture.

[0108] (2) After improving the water stress function, the differences between the predicted and measured yields under different simulation schemes for the three sowing periods are as follows: Figure 5 As shown, the model accuracy is significantly improved after introducing the improved RFWS in drought years: the MARE of predicted yield in scheme S9 is 34.1%, which is better than 47.4% in scheme S8, and the minimum MARE of scheme S9 is 14.5%. In years with normal water availability, the RFWS improvement has minimal impact on the simulation results because there is no water limitation for crops. However, in extreme drought conditions, scheme S9 still underestimates the yield, exposing the limitations of the model in responding to extreme drought.

[0109] (3) After rolling forecasts and confirmation of the optimal node, the results are as follows: Figure 6As shown, rolling forecasts were conducted using S3, the best-performing scheme under sowing period II, as the benchmark. The results show that the model converges quickly in normal years; in drought years, it takes until 60 days after the jointing stage to stabilize, and the predicted yield values ​​are in high agreement with the measured values ​​after stabilization. After comprehensive analysis, approximately 60 days after the jointing stage is determined to be the universally optimal node, which can balance forecast stability and lead time.

[0110] (4) A simulation strategy considering the cumulative precipitation threshold is introduced. Taking the optimal cumulative precipitation of 149 mm as the drought threshold, a differentiated strategy S10 is constructed: when the cumulative precipitation > 149 mm, scheme S3 is adopted; when the cumulative precipitation ≤ 149 mm, scheme S9 is adopted. The results are as follows: Figure 7 As shown, scheme S10 can effectively adapt to different water scenarios: in years with sufficient water, the predicted MAE is 1145.5 kg / ha, and in years with drought, the predicted MAE is 642.7 kg / ha. In the dynamic simulation of total aboveground biomass (TAGP) at sowing date III, scheme S10 also outperforms other schemes, as shown in the comparison results. Figure 8 As shown.

[0111] (5) Under the rolling forecast verification, the rolling forecast results for sowing date II are as follows: Figure 9 As shown, the rolling forecast for sowing period II indicates that because the cumulative precipitation after the jointing stage is higher than the drought threshold, the system adopts scheme S2. The optimal node is still 60 days after the jointing stage. After that, the predicted value deviates little from the measured value, which verifies the universality of the optimal node and also proves the reliability of the model simulation accuracy.

[0112] It is worth noting that all contents not described in detail in this invention are existing technologies and are well known to those skilled in the art.

[0113] Therefore, this invention provides a multi-scheme collaborative yield forecasting method based on data assimilation and model parameter optimization. Based on the WOFOST model, this method constructs differentiated simulation schemes through data assimilation, photosynthetic parameter optimization, water stress function improvement, and rolling forecast integration. At the same time, it proposes a dynamic simulation strategy selection mechanism based on the cumulative precipitation threshold during the key growth period. By determining the optimal node for yield forecast stabilization through rolling forecasts, it achieves dynamic and accurate simulation of crop yield under different water year types, improves the stability of model prediction, and provides reliable technical support for crop yield forecasting, drought impact assessment, and agricultural management decision-making.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A multi-scheme collaborative yield forecasting method based on data assimilation and model parameter optimization, characterized in that, Includes the following steps: Step S1: Obtain historical and real-time data on the target area and the growth period of the target crop, including meteorological data, crop parameters, soil parameters, and agricultural management data; Step S2: By assessing data assimilation, photosynthetic parameter optimization, water stress function improvement, and cumulative precipitation threshold, different scenarios are combined to construct a combined simulation scheme for the WOFOST model. The combined simulation scheme includes: the original simulation scheme without data assimilation, ... Data assimilation simulation scheme Data assimilation and photosynthetic parameter optimization collaborative simulation scheme Data assimilation simulation scheme Data assimilation and photosynthetic parameter optimization collaborative simulation scheme and Bivariate data assimilation simulation scheme and A collaborative simulation scheme for bivariate data assimilation and photosynthetic parameter optimization. A collaborative simulation scheme for data assimilation, photosynthetic parameter optimization, and improvement of water stress function. A collaborative simulation scheme for data assimilation, photosynthetic parameter optimization, and improved water stress function; and a simulation scheme that considers the cumulative precipitation threshold. In the simulation scheme considering the cumulative precipitation threshold, combined with the improved strategy of the water stress function, a parameterization scheme for the precipitation-triggered assimilation strategy based on the cumulative precipitation during the key growth period of crops is designed. A critical value for drought occurrence is defined, daily canopy evapotranspiration is calculated, and the drought threshold is obtained. The formula for calculating daily canopy evapotranspiration is as follows: ； in, This represents daily canopy evapotranspiration. Field holding capacity The point of wilting. For crop coefficients, For reference crop evapotranspiration; Step S3: Using the ensemble Kalman filter (EnKF) method combined with a Gaussian perturbation strategy, the leaf area index is... With soil moisture Data assimilation includes the following steps: Step S31: Obtain and ,right and Apply a Gaussian perturbation to generate the perturbed result. and Constructing an observation set ; Step S32: Filter sensitive model parameters, apply uncertainty perturbations to the sensitive model parameters, generate parameter combinations, and construct a parameter set. ; Step S33: Set the quantitative gradient of parameter combinations, calculate the model simulation accuracy of different quantitative gradient parameter combinations, and select the quantitative gradient with the highest model simulation accuracy, which is recorded as the optimal set size; Step S34: Input the parameter combinations from the parameter set into the WOFOST model, run the WOFOST model, obtain the simulated values ​​corresponding to the parameter combinations, and construct the prediction set. ; Step S35: Find the perturbation-containing and At any given moment, the Ensemble Kalman Filter (EnKF) method is used to fuse the prediction set and the observation set, and update... and And output the assimilated data and Until the next moment; Step S4: Use the EFAST method to perform sensitivity analysis on the WOFOST model parameters to determine the key parameters affecting crop yield, optimize the key parameters of the WOFOST model based on the optimization algorithm, and determine the optimal combination of photosynthetic parameters; Step S5: Improve the water stress function into a piecewise function; Step S6: Construct a rolling yield forecasting framework. Based on the difference between predicted and actual yields, identify the forecast stability window, define the time period closest to crop maturity as the optimal node, record the optimal node, and dynamically optimize the yield forecast results, including the following steps: Step S61: Construct a rolling yield forecast framework. From the crop jointing stage to the crop maturity stage, set the time interval for rolling simulation of the model, acquire and import meteorological data to drive the model to perform rolling simulation, and output the predicted yield results. Step S62: Design a prediction stable window identification method, obtain the predicted yield results, calculate the relative error between the predicted yield results and the measured yield, set a relative error threshold, record the number of times the relative error is less than the relative error threshold, determine and record the prediction stable window, and retain the stable window closest to the maturity period. Step S63: Obtain the stable window closest to maturity, and obtain the earliest prediction time point in the stable window closest to maturity that meets the stability conditions. Record it as the best node and dynamically optimize the yield prediction results. Step S7: Evaluate the combined simulation scheme using the mean absolute relative error (MARE) and mean absolute error (MAE), and dynamically select the optimal simulation strategy for production simulation and forecasting based on the evaluation results.

2. The multi-scheme collaborative yield forecasting method based on data assimilation and model parameter optimization according to claim 1, characterized in that, In step S1, the growth period includes the three-leaf stage, seven-leaf stage, jointing stage, tasseling stage, milk stage, and maturity stage. Meteorological data include daily maximum temperature, daily minimum temperature, wind speed, precipitation, sunshine duration, water vapor pressure, and solar radiation. Crop parameters include leaf area index, leaf biomass, stem biomass, and fruit biomass. Soil parameters include soil moisture, wilting moisture, field holding capacity, and saturated water content.

3. The multi-scheme collaborative yield forecasting method based on data assimilation and model parameter optimization according to claim 1, characterized in that, In step S4, during parameter optimization, the phenological process is kept unchanged, and the yield difference before and after parameter optimization and the growth simulation difference are evaluated. The yield difference is evaluated by calculating the absolute and relative errors between the predicted yield and the actual yield before and after model optimization, and the growth simulation difference is evaluated by comparing the deviation between the simulated values ​​of crop parameters and the actual crop parameters before and after model optimization.

4. The multi-scheme collaborative yield forecasting method based on data assimilation and model parameter optimization according to claim 1, characterized in that, In step S5, the improved water stress function is: ； in, It is a water stress reduction factor. For continuity coefficients, This represents the current soil moisture content. The soil moisture content at the wilting point. This is the critical water content.

5. The multi-scheme collaborative yield forecasting method based on data assimilation and model parameter optimization according to claim 1, characterized in that, In step S61, during the rolling simulation, the EnKF algorithm is used to assimilate and update the measured state variables in real time, and to dynamically correct the predicted crop growth parameters in the model.

6. The multi-scheme collaborative yield forecasting method based on data assimilation and model parameter optimization according to claim 1, characterized in that, In step S7, the formula for calculating the mean absolute relative error (MARE) is: ； in, The mean absolute relative error, For simulating quantities, For the first One simulated value, For the first One observation value; The formula for calculating the Mean Absolute Error (MAE) is: ； in, This represents the mean absolute error.

Images