Remote sensing estimation method and system for per unit area yield of crops

Abstract

The invention provides a remote sensing estimation method and system for crop yield per unit area. The method comprises the following steps: acquiring target remote sensing data corresponding to a crop area; inputting the target remote sensing data into a crop yield per unit estimation model to obtain crop yield per unit prediction data of the crop region output by the crop yield per unit estimation model; wherein the crop per unit yield estimation model is obtained by training on the basis of crop per unit yield simulation data and atmosphere top reflectivity simulation data corresponding to the crop region; the crop per unit area yield simulation data and the atmosphere top layer reflectivity simulation data are generated by a crop per unit area yield agricultural remote sensing joint simulation model, and the crop per unit area yield agricultural remote sensing joint simulation model is constructed based on a crop growth model and an atmosphere radiation transmission model. The accuracy of the crop per unit yield prediction result is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural production, and particularly relates to a method and system for remotely sensing and estimating the single crop yield. Background Art

[0002] The existing estimation of crop yields mainly relies on field sampling techniques, which require analyzing crop samples to estimate and extrapolate the crop yields of the entire planting area. However, the method of destructive sampling requires a large amount of human and material resources and is affected by the inherent differences in geographical and biological diversities.

[0003] Remote sensing data, with its advantages of being macroscopic, dynamic, and fast, enables surface observations at multiple scales (regions, countries, fields, etc.) and realizes more accurate and reliable quantitative estimation of soil properties and canopy state variables. More and more high spatio-temporal resolution optical remote sensing data (such as Sentinel-2, Landsat 8, SPOT-6, and Gaofen-1, etc.), synthetic aperture radar data that is not affected by weather (such as Sentinel-1, ALOS, and RADARSAT-2, etc.), and unmanned aerial vehicle data that can provide fine changes of field crops provide strong data support for the estimation of single crop yields.

[0004] However, the existing empirical statistical models strongly rely on the data quality of a large number of training samples and the representativeness of the models for the accuracy of yield prediction. Although remote sensing data provides certain data support for the construction of yield estimation models, it is often difficult to obtain sufficient high-quality images in practical applications. Moreover, since the acquisition of ground truth samples requires a large amount of human and material resources, the problem of sample scarcity often limits the accuracy and spatio-temporal generalization ability of yield estimation models. In addition, due to the lack of agronomic knowledge in empirical statistical models, the improvement of the estimation accuracy of single crop yields is always limited. Therefore, there is an urgent need for a method and system for remotely sensing and estimating single crop yields to solve the above problems. Summary of the Invention

[0005] In view of the problems existing in the prior art, the present invention provides a method and system for remotely sensing and estimating single crop yields.

[0006] The present invention provides a method for remotely sensing and estimating single crop yields, including: Obtaining target remote sensing data corresponding to a crop area; Inputting the target remote sensing data into a single crop yield estimation model to obtain the single crop yield prediction data of the crop area output by the single crop yield estimation model; Among them, the crop yield estimation model is trained based on the crop yield simulation data and the top-of-atmosphere reflectance simulation data corresponding to the crop area; the crop yield simulation data and the top-of-atmosphere reflectance simulation data are generated by a combined crop yield and agronomic remote sensing simulation model, and the combined crop yield and agronomic remote sensing simulation model is constructed based on a crop growth model and an atmospheric radiative transfer model.

[0007] According to a crop yield remote sensing estimation method provided by the present invention, the combined crop yield and agronomic remote sensing simulation model is specifically used for: Obtain the historical meteorological data, historical soil data, historical crop variety data, and historical crop management measure data of the crop area at a historical moment; Input the historical meteorological data, the historical soil data, the historical crop variety data, and the historical crop management measure data into the crop growth model to obtain the crop growth process simulation data corresponding to the crop area output by the crop growth model, where the crop growth process simulation data includes the crop yield simulation data, leaf area index simulation data, leaf dry weight simulation data, and simulation date; Obtain the historical chlorophyll data and historical incident longwave radiation data of the crop area at a historical moment; Input the historical chlorophyll data, the historical incident longwave radiation data, the leaf area index simulation data, the leaf dry weight simulation data, specific leaf area input data, the simulation date, and longitude and latitude information into the canopy radiative transfer model to obtain the canopy top reflectance simulation data output by the canopy radiative transfer model, where the specific leaf area input data and the longitude and latitude information are input parameters obtained based on the crop growth model; Obtain the historical ozone concentration data and historical aerosol optical depth data of the crop area at a historical moment; Input the historical ozone concentration data, the historical aerosol optical depth data, and the canopy top reflectance simulation data into the atmospheric radiative transfer model to obtain the top-of-atmosphere reflectance simulation data output by the atmospheric radiative transfer model.

[0008] According to a crop yield remote sensing estimation method provided by the present invention, the crop growth model is the WOFOST model, the canopy radiative transfer model is the SCOPE model, and the atmospheric radiative transfer model is the SMAC model.

[0009] A method for remotely estimating the single crop yield provided by the present invention, wherein inputting the chlorophyll historical data, the incident long-wave radiation historical data, the leaf area index simulation data, the leaf dry weight simulation data, the specific leaf area input data, the simulation date and the longitude and latitude information into a canopy radiation transfer model to obtain the canopy top reflectance simulation data output by the canopy radiation transfer model includes: Constructing leaf optical parameters according to the chlorophyll historical data, the leaf dry weight simulation data and the specific leaf area input data; Obtaining the plant height data, the leaf inclination data, the leaf inclination change data and the leaf width data of the crop area at a historical moment, and constructing canopy structure parameters according to the leaf area index simulation data, the plant height data, the leaf inclination data, the leaf inclination change data and the leaf width data; Obtaining the solar zenith angle data, the observation zenith angle data and the relative azimuth angle data of the crop area at a historical moment, and constructing time and angle parameters according to the simulation date, the longitude and latitude information, the solar zenith angle data, the observation zenith angle data and the relative azimuth angle data; Constructing target meteorological parameters according to the incident long-wave radiation historical data, the meteorological historical data and preset meteorological data; Obtaining the preset aerodynamic parameters and the preset soil reflectance corresponding to the canopy radiation transfer model; Based on the canopy radiation transfer model, simulating reflectance data for the leaf optical parameters, the canopy structure parameters, the time and angle parameters, the target meteorological parameters, the preset aerodynamic parameters and the preset soil reflectance to obtain the canopy top reflectance simulation data.

[0010] A method for remotely estimating the single crop yield provided by the present invention, wherein the crop single yield estimation model is trained through the following steps: Performing a sensitivity analysis on the crop variety historical data and the canopy structure parameters, and constructing a sensitivity coupling parameter set according to the results of the sensitivity analysis and different preset soil and management measure conditions, wherein the preset soil and management measure conditions are constructed according to the soil historical parameters and the crop management measure historical data; Inputting the sensitivity coupling parameter set into the crop single yield agronomic remote sensing joint simulation model to obtain the crop single yield simulation data, the canopy top reflectance simulation data and the top-of-atmosphere reflectance simulation data; Use the top-of-atmosphere reflectance simulation data as sample crop remote sensing data, and generate crop yield labels corresponding to the sample crop remote sensing data according to the crop yield simulation data, so as to construct a training sample set; Based on the training sample set, train a recurrent neural network model to obtain the crop yield estimation model.

[0011] According to a crop yield remote sensing estimation method provided by the present invention, perform a sensitivity analysis on the crop variety historical data and the canopy structure parameters, and construct a sensitivity coupling parameter set according to the results of the sensitivity analysis and different preset soil and management measure conditions, including: According to the crop variety historical data, obtain the initial crop total dry weight historical data, specific leaf area historical data, total dry matter distribution ratio historical data, first accumulated temperature historical data, leaf life historical data, and development stage historical data of the crop area, wherein the total dry matter distribution ratio historical data includes the ratio historical data of the total dry matter distributed to the leaves, the ratio historical data of the total dry matter distributed to the storage organs, and the ratio historical data of the total dry matter distributed to the stems; the first accumulated temperature historical data represents the accumulated temperature historical data of the crop from emergence to flowering; the leaf life historical data represents the life data of the crop leaves at 35 degrees Celsius; the development stage historical data includes the state variables of the crop from the initial development stage to the harvest development stage; Obtain the historical data of the leaf equivalent water layer thickness; Construct the sensitivity coupling parameter set corresponding to different preset soil and management measure conditions according to the initial crop total dry weight historical data, the specific leaf area historical data, the total dry matter distribution ratio historical data, the first accumulated temperature historical data, the leaf life historical data, the development stage historical data, the leaf inclination data, and the leaf equivalent water layer thickness historical data.

[0012] The present invention also provides a crop yield remote sensing estimation system, including: A crop remote sensing data acquisition module, configured to acquire target remote sensing data corresponding to a crop area; A crop yield estimation module, configured to input the target remote sensing data into a crop yield estimation model to obtain crop yield prediction data of the crop area output by the crop yield estimation model; Among them, the crop yield estimation model is trained based on the crop yield simulation data and the top-of-atmosphere reflectance simulation data corresponding to the crop area; the crop yield simulation data and the top-of-atmosphere reflectance simulation data are generated by a combined crop yield and agronomic remote sensing simulation model, and the combined crop yield and agronomic remote sensing simulation model is constructed based on a crop growth model and an atmospheric radiative transfer model.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the method for remotely estimating crop yield as described in any one of the above is implemented.

[0014] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method for remotely estimating crop yield as described in any one of the above is implemented.

[0015] The present invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, the method for remotely estimating crop yield as described in any one of the above is implemented.

[0016] The method and system for remotely estimating crop yield provided by the present invention generate simulation data to train a crop yield estimation model by constructing a combined crop yield and agronomic remote sensing simulation model, which improves the construction quality of the crop sample dataset, and further improves the accuracy of the crop yield prediction result, providing more reliable decision-making support for agricultural production and management. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 It is a schematic flowchart of the method for remotely estimating crop yield provided by the present invention; Figure 2 It is a schematic overall diagram of the process for remotely estimating winter wheat yield provided by the present invention; Figure 3 It is a schematic architecture diagram of the combined crop yield and agronomic remote sensing simulation model provided by the present invention; Figure 4 It is a schematic diagram of the sensitivity analysis result of the WOFOST model parameters to the leaf area index provided by the present invention; Figure 5Schematic diagram of the sensitivity analysis results of WOFOST model parameters provided by the present invention for spike dry weight accumulation; Figure 6 Schematic diagram of the sensitivity analysis results of the coupled model provided by the present invention for the top-of-atmosphere reflectance; Figure 7 Schematic diagram of the distribution of meteorological data in a certain area provided by the present invention; Figure 8 Schematic diagram of the simulated yield of a certain climate annual type provided by the present invention; Figure 9 Schematic diagram of the simulated reflectance during the flowering period of winter wheat of a certain climate annual type provided by the present invention; Figure 10 Schematic diagram of the single-point scale verification results provided by the present invention; Figure 11 Schematic diagram of the municipal scale verification results provided by the present invention; Figure 12 Schematic diagram of the structure of the remote sensing estimation system for crop yield provided by the present invention; Figure 13 Schematic diagram of the structure of the electronic device provided by the present invention. Detailed implementation manners

[0019] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the present invention. Obviously, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without creative efforts shall fall within the protection scope of the present invention.

[0020] Traditional crop yield estimation models are mainly data-driven. By using remotely sensed characteristic factors, such as Leaf Area Index (LAI for short) and Fraction of Absorbed Photosynthetically Active Radiation (FAPAR for short), an empirical statistical model is constructed with crop yield to achieve yield estimation.

[0021] To further improve the estimation accuracy of crop yield per unit area, in existing research, multi-source remote sensing data are often used to obtain different types of characteristic parameters for yield estimation modeling. Compared with traditional statistical models, machine learning algorithms have a stronger ability to handle non-linear tasks among a large amount of input data, providing powerful and flexible technical support for data-driven approaches. Some machine learning models in the context of data-driven, such as Extreme Gradient Boosting (abbreviated as XGBoost), Random Forest (abbreviated as RF), Support Vector Machine (abbreviated as SVM), and Artificial Neural Network (abbreviated as ANN), have been widely applied to remote sensing estimation of crop yield. However, the accuracy of empirical statistical models for yield prediction depends on the data quality of a large number of training samples and the representativeness of the models. Although remote sensing data provide certain data support for the construction of yield estimation models, it is often difficult to obtain sufficient high-quality images in practical applications. In addition, the acquisition of ground-measured samples requires a large amount of manpower and material resources, and the problem of sample scarcity often limits the accuracy and spatio-temporal generalization ability of yield estimation models. Moreover, due to the lack of agronomic knowledge in existing empirical statistical models, the improvement of crop yield estimation accuracy has always been limited.

[0022] With the continuous in-depth understanding of the crop growth process, crop growth models driven by agronomic mechanism knowledge have gradually developed. Crop growth models can predict the evolution process of crops from sowing to harvesting by simulating photosynthesis, gas exchange between the canopy and the atmosphere, phenology, soil moisture and temperature changes, biomass accumulation, and grain formation, further strengthening the agronomic foundation and improving the estimation accuracy. Considering that the driving factors and the underlying mechanism knowledge of different crop growth models are different, coupling different types of models can not only avoid the shortcomings of the original models, give full play to the advantages of different models, but also improve the simulation accuracy and the stability of the modeling system, and reduce the operation cost of the models. However, in the actual popularization and application of existing crop growth models, they have to face the problem that a large number of parameters cannot be accurately obtained during the model parameter calibration process.

[0023] To improve the applicability of crop growth models at the regional scale, a data assimilation method that combines crop physiological variables inverted from remote sensing data with crop growth models has gradually emerged. On the one hand, data assimilation can, while maintaining the mechanistic meaning, reduce the difficulty of obtaining model parameters in large-area studies and quantify the growth status of crops at the regional scale. On the other hand, it can correct the biases in the model simulation process and improve the accuracy of yield estimation. However, due to the generally cumbersome assimilation algorithms and large computational requirements, knowledge-driven crop growth models still have difficulty getting rid of the situation where the fixed-point simulation accuracy is strong but the extrapolation ability outside the region is limited when estimating single crop yields.

[0024] Aiming at the problems existing in the above-mentioned prior art, the present invention focuses on the process of single crop yield formation, takes the photosynthesis mechanism of the vegetation canopy intercepting and absorbing the electromagnetic spectrum as the starting point, studies the coupled model constructed by the crop growth model and the vegetation canopy radiation transfer model, and further couples the atmospheric radiation transfer model, thereby establishing a joint simulation model of agricultural remote sensing for single crop yields.

[0025] Furthermore, based on the joint simulation model of agricultural remote sensing for single crop yields established above, the present invention takes winter wheat as an example. For the study area, it comprehensively and systematically simulates the growth and single crop yield formation process of winter wheat under various agricultural production scenarios such as different climate conditions, soil characteristics, management measures, and crop varieties by using historical data, obtains the association rule set between the single crop yield elements of winter wheat and spectral characteristics, realizes the quantitative description of the single crop yield formation process of winter wheat, constructs a multi-scenario winter wheat growth simulation data set, and provides a sample simulation data set for the single crop yield estimation modeling of winter wheat. And, in the present invention, based on the obtained winter wheat growth simulation data set, key index factors affecting single crop yield elements are analyzed and screened, the importance ranking and optimal combination of characteristic factors are realized, and an association analysis model between the single crop yield of winter wheat and characteristic factors is carried out based on deep learning methods, so as to realize the model-based description of the single crop yield formation process of winter wheat.

[0026] Finally, based on the winter wheat single crop yield estimation model, combined with high-resolution remote sensing data, characteristic factors required by the model are obtained considering their band characteristics, and a demonstration application at the regional scale is carried out. For winter wheat in the target area, a long-term spatial distribution map of winter wheat single crop yields is made, and the accuracy evaluation and error analysis of the single crop yield estimation results are carried out at different spatial scales (such as provinces, cities, and sample plots), and the effectiveness of the hybrid model in practical applications is analyzed.

[0027] Figure 1 The flow chart of the agricultural remote sensing estimation method for single crop yields provided by the present invention is as Figure 1 shown. The present invention provides an agricultural remote sensing estimation method for single crop yields, including: Step 101, obtaining target remote sensing data corresponding to the crop area.

[0028] In the present invention, multispectral images of crop areas can be captured by satellites such as Landsat 5 or Landsat 8. These images contain rich surface information and are crucial for analyzing the growth status of crops. The present invention utilizes satellites such as Landsat 5 or Landsat 8 to observe the crop areas according to a predetermined orbit and time interval to obtain remote sensing image data.

[0029] Furthermore, according to the research requirements of crops, remote sensing image processing techniques such as band combination, image enhancement, and classification are used to identify and extract the crop areas from the preprocessed remote sensing images. By methods such as visual interpretation or automatic classification, the crop areas are separated from the background ground objects to provide an accurate target area for subsequent yield prediction, that is, to obtain the target remote sensing data.

[0030] Step 102: Input the target remote sensing data into the crop yield estimation model to obtain the crop yield prediction data of the crop area output by the crop yield estimation model. Among them, the crop yield estimation model is trained based on the crop yield simulation data and the top-of-atmosphere reflectance simulation data corresponding to the crop area; the crop yield simulation data and the top-of-atmosphere reflectance simulation data are generated by a combined crop yield and agronomic remote sensing simulation model, and the combined crop yield and agronomic remote sensing simulation model is constructed based on a crop growth model and an atmospheric radiation transfer model.

[0031] Figure 2 For the overall schematic diagram of the winter wheat yield remote sensing estimation process provided by the present invention, reference can be made to Figure 2 As shown, in the present invention, the target remote sensing data (for example, image data from Landsat 5 / 8) is input into a deep learning model (i.e., the crop yield estimation model). The crop yield estimation model processes and analyzes these data and finally outputs the crop yield prediction data of the crop area. These crop yield prediction data can be used to evaluate the yield potential of crops and provide a basis for agricultural management and decision-making.

[0032] In the present invention, the crop yield estimation model is trained based on a series of simulation data, including crop yield simulation data and top-of-atmosphere reflectance simulation data. These simulation data provide rich training samples for the model and help the deep learning model learn the complex relationship between crop yield and remote sensing data.

[0033] In the present invention, the simulated data of crop yield per unit area and the simulated data of top-of-atmosphere reflectance are generated by a combined crop yield and agronomic remote sensing simulation model (coupled model). This combined simulation model is constructed based on a crop growth model, a canopy radiation transfer model, and an atmospheric radiation transfer model. The three sub-models are respectively responsible for simulating the growth process of crops, the radiation transfer characteristics of the canopy, and the influence of the atmosphere on radiation. Specifically, the crop growth model is used to simulate the growth and development process of crops, including growth rate, biomass accumulation, and yield formation, etc., providing a basis for the simulation of crop yield per unit area. The canopy radiation transfer model is used to simulate the transfer process of solar radiation in the crop canopy, including absorption, reflection, and transmission, etc., providing a basis for the simulation of the reflectance at the top of the canopy. The atmospheric radiation transfer model is used to simulate the influence of the atmosphere on solar radiation and surface radiation, including atmospheric absorption, scattering, etc., providing support for the simulation of top-of-atmosphere reflectance.

[0034] In the present invention, for the simulated data output by the combined crop yield and agronomic remote sensing simulation model, further screening of the simulated data is performed according to different growth stages of winter wheat (such as greening, flowering, and maturity) to obtain characteristic data that can better reflect the prediction of winter wheat yield per unit area, thereby providing comprehensive training samples for the crop yield per unit area estimation model. These simulated data help the crop yield per unit area estimation model learn the mapping relationship between remote sensing data and crop yield per unit area, thereby improving the prediction accuracy of the model.

[0035] The crop yield per unit area remote sensing estimation method provided by the present invention generates simulated data to train the crop yield per unit area estimation model by constructing a combined crop yield and agronomic remote sensing simulation model, which improves the construction quality of the crop sample dataset, and further improves the accuracy of the crop yield per unit area prediction result, providing more reliable decision-making support for agricultural production and management.

[0036] Based on the above embodiments, the combined crop yield and agronomic remote sensing simulation model is specifically used for: Obtaining the historical meteorological data, historical soil data, historical crop variety data, and historical crop management measure data of the crop area at a historical moment; Inputting the historical meteorological data, the historical soil data, the historical crop variety data, and the historical crop management measure data into the crop growth model to obtain the simulated data of the crop growth process corresponding to the crop area output by the crop growth model, wherein the simulated data of the crop growth process includes the simulated data of crop yield per unit area, the simulated data of leaf area index, the simulated data of leaf dry weight, and the simulated date; Obtaining the historical chlorophyll data and the historical incident long-wave radiation data of the crop area at a historical moment; Input the chlorophyll historical data, the incident longwave radiation historical data, the leaf area index simulation data, the leaf dry weight simulation data, the specific leaf area input data, the simulation date, and the longitude and latitude information into the canopy radiative transfer model to obtain the canopy top reflectance simulation data output by the canopy radiative transfer model, where the specific leaf area input data and the longitude and latitude information are input parameters obtained based on the crop growth model; Obtain the ozone concentration historical data and the aerosol optical depth historical data of the crop area at historical times; Input the ozone concentration historical data, the aerosol optical depth historical data, and the canopy top reflectance simulation data into the atmospheric radiative transfer model to obtain the top of atmosphere reflectance simulation data output by the atmospheric radiative transfer model.

[0037] Figure 3 It is a schematic diagram of the architecture of the crop yield agronomic remote sensing joint simulation model provided by the present invention. For reference, see Figure 3 As shown, in the present invention, the crop growth model is the WOFOST (World Food Studies) model, the canopy radiative transfer model is the SCOPE (Soil Canopy Observation, Photochemistry and Energy fluxes) model, and the atmospheric radiative transfer model is the SMAC (Synchronous Monitoring Atmospheric Corrector) model.

[0038] In the present invention, the winter wheat yield is estimated by coupling the crop growth model WOFOST, the canopy radiative transfer model SCOPE, and the atmospheric radiative transfer model SMAC. First, a WOFOST-SCOPE-SMAC coupling model is constructed in a series connection manner, that is, a crop yield agronomic remote sensing joint simulation model. While simulating the winter wheat yield formation process, considering the top of atmosphere (TOA) reflectance simulation, and further setting multiple scenarios for the input parameters of the coupling model to construct a simulation dataset of the winter wheat yield formation process under multiple scenarios, providing a data basis for learning the association rules between the winter wheat yield and the TOA reflectance at different growth stages. In the present invention, the WOFOST-SCOPE-SMAC coupling model mainly realizes the pairwise connection between models by passing the input and simulation output results of the previous model to the next model.

[0039] In the present invention, historical data of the crop area is first obtained. These historical data include at least meteorological historical data, soil historical data, crop variety historical data, and crop management measure historical data. Among them, meteorological historical data: includes records of meteorological elements such as temperature, humidity, precipitation, wind speed, sunshine hours, etc. at historical times. Soil historical data: includes data of soil characteristics such as soil type, soil humidity, and soil nutrient content at historical times. Crop variety historical data: records information of crop varieties, such as variety name, growth cycle, etc. Crop management measure historical data: includes records of agricultural management activities such as sowing date, fertilization amount, irrigation method, and pest and disease control measures at historical times.

[0040] In the present invention, the simulation of the WOFOST model can be realized based on the PCSE environment. The input parameters of the model include the four parameters of meteorology, soil, crop variety, and management measures in the above embodiments. Since it is difficult to obtain field experiment data, the present invention collects historical data of the existing WOFOST model for simulating wheat growth, and combines the collected multi-source data products to carry out the localization calibration work of the WOFOST model, so that the model parameter localization work is completely independent of ground samples and has strong generalizability.

[0041] Specifically, meteorological historical data is a key parameter for driving the simulation of the WOFOST model and has an important impact on the crop growth and development process and the formation of unit yield. In the present invention, the meteorological parameters required by the WOFOST model mainly include 8 types, namely date (DAY), incident short-wave radiation (kJ m -2 d -1 ), daily minimum temperature (°C), daily maximum temperature (°C), daily average water vapor pressure (kPa), daily average wind speed at 2 meters above the ground surface (m s -1 ), daily precipitation (mm), and snow depth (cm). The above meteorological parameters can all be provided by meteorological station data.

[0042] Soil historical data mainly includes soil wilting point water content (SMW), field capacity (SMFCF), saturated water content (SM0), and saturated soil hydraulic conductivity (K0). In the present invention, according to the existing relevant soil data sets, the main soil type in the crop area is loam, and it is further divided into three types: sandy loam, light loam, and medium loam. In this article, the parameter settings of the three soil types are carried out with reference to the property data provided by the existing research and soil data sets in the crop area. The specific parameter settings of the three soil types can be referred to Table 1, Table 2, and Table 3.

[0043] Table 1 Parameter settings for sandy loam soil

[0044] Table 2 Parameter settings for light loam soil

[0045] Loam soil parameter settings in Table 3

[0046] It should be noted that in the present invention, a sufficient simulation dataset is constructed based on the coupling model for supporting the training of the machine learning model, rather than determining the global optimal solution of the model parameters. The main purpose of setting the model parameters is to adapt to various agricultural production scenarios during the growth period of winter wheat. Therefore, the present invention does not use ground observation data and remotely sensed inversion parameters, etc. to calibrate the model parameters.

[0047] Furthermore, in the present invention, the values of the historical data of crop varieties are based on the variety parameters of Winter_wheat_105 provided by the WOFOST model and are set according to existing research literature. Specifically, reference can be made to the content of Tables 4a, 4b, and 4c.

[0048] Initial settings of crop variety parameters in Table 4a

[0049] Initial settings of crop variety parameters in Table 4b

[0050] Initial settings of crop variety parameters in Table 4c

[0051] In the present invention, based on Tables 4a, 4b and 4c, the crop emergence date (IDEM), the accumulated temperature from emergence to flowering (TSUM1), the accumulated temperature from flowering to maturity (TSUM2), the initial total crop dry weight (TDWI), the leaf lifespan at 35 degrees Celsius (SPAN), the specific leaf area (SLATB), the maximum leaf CO2 assimilation rate (AMAXTB), the efficiency of assimilated biomass conversion to storage organs (CVO), the proportion of total dry matter allocated to leaves (FLTB), the proportion of total dry matter allocated to storage organs (FOTB), and the proportion of total dry matter allocated to stems (FSTB) are taken as important parameters during the growth and development process of winter wheat. Preferably, considering that the values of some parameters are presented in list form and change continuously at different growth stages of winter wheat. For the convenience of unified parameter setting and analysis, the present invention establishes 10 single mapping variables (β_IDEM, α_TSUM1, α_TSUM2, α_TDWI, α_SPAN, αSLATB, α_AMAXTB, α_CVO, α_v and β_DVS) to establish a mapping relationship with the above 11 parameters to control the parameter values, where α_v corresponds to the comparison of the allocation of three types of total dry matter, namely the proportion of total dry matter allocated to leaves (FLTB), the proportion of total dry matter allocated to storage organs (FOTB), and the proportion of total dry matter allocated to stems (FSTB). The optimized setting of crop variety-related parameters is shown in Table 5.

[0052] Table 5 Optimized Setting of Crop Variety Parameters

[0053] For the historical data of crop management measures, the WOFOST model in the present invention is based on the simulation of crop physiological and ecological processes such as assimilation, respiration, transpiration and dry matter allocation, mainly including the simulation of crop growth under potential growth conditions, water-limited conditions and nutrient-limited conditions. Since the winter wheat planting in the crop area generally does not suffer from nutrient stress, the present invention does not consider the growth simulation of winter wheat under nutrient-limited conditions, and mainly considers the simulation under two planting modes: rain-fed (Wofost72_WLP_CWB) and irrigation (Wofost72_PP). According to the field survey results, the sowing date of the Winter_wheat_105 winter wheat variety is initially set as October 7th.

[0054] Further, in the present invention, the historical data of meteorology, soil, crop variety, and management measures obtained in the above embodiments are input into a crop growth model. The crop growth model simulates the growth process of the crop at historical moments according to the input data and outputs the simulated data of the crop growth process. The simulated data includes simulated data of crop yield per unit area (predicted yield), leaf area index simulated data (the degree of leaf coverage on the ground), leaf dry weight simulated data (dry matter weight of the leaves), and simulated date, etc.

[0055] Since the WOFOST model runs in a Python environment and the canopy radiation transfer model SCOPE runs in a Matlab environment, the present invention couples the two models based on the MATLAB 2019b Engine for Python to achieve data transmission. Subsequent analysis of the coupled model is mainly implemented based on Jupyter Notebook of the Python interpreter.

[0056] Further, by learning the input and output parameters of the WOFOST model and the SCOPE model, the input and output results of the WOFOST model are used to be passed into the SCOPE model, so as to dynamically assign values to the parameters of the SCOPE model. In the present invention, the parameters of the SCOPE model mainly include: leaf optical parameters, canopy structure parameters, time and angle parameters, meteorological parameters, aerodynamic parameters, and soil parameters.

[0057] Based on the above embodiments, inputting the chlorophyll historical data, the incident long-wave radiation historical data, the leaf area index simulated data, the leaf dry weight simulated data, specific leaf area input data, the simulated date, and the longitude and latitude information into the canopy radiation transfer model to obtain the simulated data of the canopy top reflectance output by the canopy radiation transfer model includes: Constructing leaf optical parameters according to the chlorophyll historical data, the leaf dry weight simulated data, and the specific leaf area input data; Obtaining the plant height data, leaf inclination data, leaf inclination change data, and leaf width data of the crop area at historical moments, and constructing canopy structure parameters according to the leaf area index simulated data, the plant height data, the leaf inclination data, the leaf inclination change data, and the leaf width data; Obtaining the solar zenith angle data, observation zenith angle data, and relative azimuth angle data of the crop area at historical moments, and constructing time and angle parameters according to the simulated date, the longitude and latitude information, the solar zenith angle data, the observation zenith angle data, and the relative azimuth angle data; Construct a target meteorological parameter based on the incident long-wave radiation historical data, the meteorological historical data, and the preset meteorological data; Obtain the preset aerodynamic parameter and the preset soil reflectivity corresponding to the canopy radiation transfer model; Based on the canopy radiation transfer model, perform reflectivity data simulation on the leaf optical parameter, the canopy structure parameter, the time and angle parameter, the target meteorological parameter, the preset aerodynamic parameter, and the preset soil reflectivity to obtain the simulated data of the canopy top reflectivity.

[0058] In the present invention, the chlorophyll historical data, the simulated data of leaf dry weight, and the specific leaf area input data reflect the photosynthesis ability, the substance content, and the structural characteristics of the leaves. By comprehensively analyzing these data, the optical parameters of the leaves can be constructed, which describe the absorption, reflection, and transmission characteristics of the leaves to light. Among them, the specific leaf area input data and the longitude and latitude information can be obtained based on the relevant parameters input into the WOFOST model.

[0059] The plant height, the leaf inclination angle, the change of the leaf inclination angle, and the leaf width jointly describe the three-dimensional structure of the canopy. The leaf area index reflects the ratio of the total area of the leaves in the canopy to the ground area. Through these data, the canopy structure parameter can be constructed.

[0060] The solar zenith angle and the observation zenith angle describe the angles of the sun and the observation point relative to the ground, and the relative azimuth angle describes the direction difference between the sun and the observation point. Combining the simulation date and the longitude and latitude information, the solar position and the observation angle at a specific time and location can be calculated, so as to construct the time and angle parameter.

[0061] The incident long-wave radiation reflects the radiation heating effect of the atmosphere on the ground, the meteorological historical data provides the past meteorological conditions, and the preset meteorological data is set according to the required weather conditions for prediction. By comprehensively analyzing these data, the target meteorological parameter can be constructed, which describes the meteorological conditions during simulation, such as temperature, humidity, wind speed, etc.

[0062] The preset aerodynamic parameter and the preset soil reflectivity are fixed inputs in the canopy radiation transfer model, which are used to describe the air flow characteristics in the canopy and the light reflection characteristics of the soil.

[0063] Finally, by inputting all the above parameters into the canopy radiation transfer model, the reflectivity data of the canopy top can be simulated, which is used to describe the light reflection characteristics of the canopy under specific conditions.

[0064] Specifically, in the present invention, the leaf optical parameters of the SCOPE model are set as shown in Table 6.

[0065] Table 6 Leaf optical parameters

[0066] In Table 6, the chlorophyll content (Cab) is assigned by obtaining the leaf chlorophyll data product (LCC). The carotenoid content (Cca) is set to one-fourth of the Cab content according to the initial setting of the SCOPE model parameters. The leaf equivalent water layer thickness (Cw) can be set to 0.01 with reference to relevant literature studies. The slope of the Ball-Berry stomatal conductance model (BallBerrySlope) is set to 9 with reference to existing studies. The leaf photochemical pathway (Type) and the proportion of photons allocated to PSII (beta) are set with reference to the initial values set by the SCOPE model for C3 crops.

[0067] In order to seamlessly integrate the simulation results of the WOFOST model with the SCOPE model, the present invention performs corresponding conversions on some simulation results. Based on the leaf dry weight (WLV) and specific leaf area SLATB simulated by the WOFOST model, the leaf dry matter (Cdm) and leaf mesophyll structure (N) in the SCOPE model are assigned. Among them, WLV is expressed as the number of kilograms of leaves per hectare of ground area and needs to be converted to the number of grams of leaf dry weight per square centimeter of leaf area. The formula is: ; Formula (1) The value of N can be obtained through the empirical relationship estimation formula with SLA: ; Formula (2) Since the maximum light energy utilization rate of leaves at 25°C (Vcmax25) determines the maximum ability of Rubisco carboxylation and is a basic parameter determining the photosynthesis ability of plant leaves, it has a very important impact on photosynthesis, energy balance, and reflectance simulation. Therefore, accurately estimating Vcmax25 is crucial for improving the accuracy of canopy reflectance simulation. The present invention uses LCC data and based on the ratio of the maximum electron transfer rate (Jmax) to Vcmax of plant leaves at 25°C ( ), realizes the dynamic estimation of Vcmax25. The formula is: ; Formula (3) Among them, is the quantum yield of photosynthetic electron transfer at 25°C (mol mol -1 ), which can be further expressed as Formula (4): ; Formula (4) Among them, is the leaf absorbance, which is the output variable of the RTMo sub-module in the SCOPE model; 0.5 means that half of the absorbed light reaches photosystem II; is the photochemical quantum yield of photosystem II, which can be calculated from the air temperature (Ta; °C) using the formula: ; Equation (5) Furthermore, I is the absorbed photosynthetic photon flux density , which can be calculated from the incident shortwave radiation (Rin; W m -2 ) input parameter of the SCOPE model using the formula: ; Equation (6) can be further expressed as Equation (7): ; Equation (7) (Pa) is the intercellular CO2 concentration; (Pa) represents the CO2 compensation point without mitochondrial respiration; (Pa) is the intercellular oxygen concentration, which is equivalent to the O2 concentration in the atmosphere according to relevant research; (Pa) and (Pa) are the Michaelis-Menten coefficients of RuBisCO for CO2 and O2 activities, respectively. , and in Equation (7) can be represented by the atmospheric pressure (P; hPa) input parameter of the SCOPE model, where: ; Equation (8) ; Equation (9) ; Equation (10) can be further expressed as a function of the CO2 partial pressure in the atmosphere (pca; Pa), and pca can be obtained by multiplying the atmospheric CO2 concentration (Ca; ppm) and P input parameters of the SCOPE model: ; Equation (11) ; Equation (12) where, (unitless) represents the optimal ratio of to pca based on the minimum cost theory at 25 °C: ; Equation (13) where, is defined as at 25 °C The sensitivity to vapor pressure deficit (D; Pa) can be further expressed by Equation (14): ; Equation (14) where (unitless) is the ratio of the photosynthesis cost factor to the transpiration cost factor, which is set to 146 in the present invention; D in Equation (13) can be calculated from relative humidity (RH; %) and Ta.

[0068] ; Equation (15) Finally, in Equation (3) can be calculated from its empirical relationship with the average temperature during the growing season (Tg; °C), which can be expressed by Equation (16): ; Equation (16) It should be noted that for other parameters of the leaf optical parameters not mentioned in the present invention, the values are all from the initial values provided by the SCOPE model Furthermore, the canopy structure parameters required by the SCOPE model mainly include LAI, plant height (hc), leaf inclination angle (LIDFa), leaf inclination angle variation (LIDFb), and leaf width (leafwidth), and the values are shown in Table 7.

[0069] Table 7 Canopy structure parameters

[0070] Among them, LAI is transmitted from the daily simulation results of the WOFOST model to the SCOPE model to achieve the connection and dynamic simulation between the two models. The plant height and leaf width are obtained from ground survey data and are set to 0.72 m and 0.0138 m respectively. Since LIDFa has a great influence on the simulated reflectance and chlorophyll fluorescence, while LIDFb has a small influence on the simulated reflectance and chlorophyll fluorescence. In the present invention, the parameter change of LIDFa is mainly considered, and the initial simulation value is set to -0.35, and LIDFb is set to the default value of -0.15.

[0071] Furthermore, the time series data required by the SCOPE model can be referred to Table 8.

[0072] Table 8 Time and angle parameters

[0073] In Table 8, the time series data includes the dates for starting and stopping the simulation, latitude, longitude, east longitude, and is mainly set through the input parameters and simulation results of the WOFOST model. In addition, the SCOPE model also requires the provision of angular information, including the solar zenith angle, the viewing zenith angle, and the relative azimuth angle, and these angular parameters are obtained from satellite metadata.

[0074] In the present invention, the meteorological parameters required by the SCOPE model and their values are shown in Table 9.

[0075] Table 9 Meteorological Parameters

[0076] In the present invention, these meteorological parameters in Table 9 are mainly obtained through the input meteorological parameters of the WOFOST model, the historical data of the original meteorological stations, and the obtained incident longwave radiation data products. In addition to the meteorological parameters required in the original SCOPE model, the present invention adds the growing season average temperature (Tg) and the 2-meter relative humidity (RH) parameters to dynamically set the value of Vcmax25, and the values of the two are provided by the data of the original meteorological stations.

[0077] Furthermore, the preset aerodynamic parameters required by the SCOPE model are mainly the canopy momentum roughness length (zo), displacement height (d), leaf drag coefficient (Cd), leaf edge resistance (rb), drag coefficient of an isolated tree (CR), fitting parameter (CD1), roughness layer correction (Psicor), soil drag coefficient (CSSOIL), soil boundary layer resistance (rbs), and canopy internal resistance (rwc). These parameters are used to describe the momentum, heat, and moisture exchange processes between the canopy and the atmosphere, and the values are derived from the initial parameter settings provided by the SCOPE model. For details, please refer to Table 10.

[0078] Table 10 Aerodynamic Parameters

[0079] Furthermore, the preset soil reflectance can be obtained by simulating the soil spectrum in two ways provided by SCOPE: (1) Using the BSM model, the soil reflectance is simulated by setting soil parameters; (2) The SCOPE model data file contains 3 original soil spectra for selection. The present invention simulates the soil reflectance based on the BSM model. The soil parameters required by the BSM model are shown in Table 11, including the resistance of the soil to pore space evaporation (rss), the broadband reflectance of the soil in the thermal infrared range (ss_thermal), the specific heat capacity of the soil (cs), the soil specific gravity (rhos), the soil thermal conductivity (lambdas), the soil water content in the root zone (smc), the soil brightness (BSMBrightness), and two model parameters lat and lon. The values of all parameters are only the standard parameter values provided by the SCOPE model.

[0080] Table 11 Soil Parameters of the SCOPE Model

[0081] In the present invention, the historical chlorophyll data (the content of chlorophyll in the leaves) and the historical incident long-wave radiation data (the intensity of the long-wave part of solar radiation) of the crop area at a historical moment are obtained. Then, the historical chlorophyll data, the historical incident long-wave radiation data, and the leaf area index, leaf dry weight, specific leaf area, simulation date, and longitude and latitude information simulated and output by the crop growth model are input into the canopy radiation transfer model. The canopy radiation transfer model simulates the absorption, reflection, and transmission processes of radiation by the crop canopy according to the input data, and outputs the simulated data of the reflectance at the top of the canopy.

[0082] Furthermore, the input parameters required by the SMAC model are shown in Table 12, including the observation geometry parameters (tts, tto, and psi), the aerosol optical depth at 550 nm (AOT 550 ), the ozone content (U O3 ), the water vapor content (U H2O ), and the atmospheric pressure (P), and these data can be the historical data of the crop area. Among them, the atmospheric pressure data can be provided by the ground meteorological station data. AOT 550 and U O3 are respectively assigned values by the collected aerosol optical depth data product MOD04_L2 and the ozone concentration data product Sentinel-5P NRTI O3. U H2O is obtained by converting the average relative humidity (RH), atmospheric pressure (P), and surface average temperature (Ta) parameters: ; Formula (17) where 0.622 is the ratio constant of water vapor to dry air, is the saturated water vapor pressure (hPa), which can be expressed according to the Tetens formula as function of: ; Equation (18) Table 12 Input parameters of the SMAC model

[0083] In the present invention, historical data of ozone concentration and aerosol optical depth at historical moments in the crop area are obtained. Then, the historical data of ozone concentration, the historical data of aerosol optical depth, and the simulated data of canopy top reflectance output by the canopy radiative transfer model are input into the atmospheric radiative transfer model SMAC. The atmospheric radiative transfer model SMAC simulates the processes of absorption, scattering, and reflection of radiation by the atmosphere according to the input data, and outputs the simulated data of top-of-atmosphere reflectance.

[0084] for reference Figure 3 As shown, in the present invention, the SCOPE model and the SMAC model are selected to simulate the TOA reflectance during the growth period of winter wheat. First, the canopy top (Top of Canopy, abbreviated as TOC) reflectance during the growth period of winter wheat is simulated based on the SCOPE model, and then the TOC reflectance is combined with the SMAC model. The four TOC reflectances simulated by the SCOPE model (i.e., , , and ) are converted into TOA reflectance. However, since the crop growth process is complex and continuous, and the SCOPE model and the SMAC model are simulated based on a single-point time step and cannot achieve continuous simulation of the reflectance during the entire growth period of winter wheat, the present invention takes into account that the WOFOST model can dynamically simulate the growth and development process of winter wheat every day during the growth period, further couples the WOFOST model with the SCOPE and SMAC models, and uses the dynamic growth and development parameters provided by the WOFOST model to provide necessary inputs for the radiative transfer model, thereby achieving continuous simulation of the reflectance during the growth period of winter wheat.

[0085] Based on the above embodiments, the crop yield estimation model is trained through the following steps: Perform a sensitivity analysis on the historical data of the crop variety and the canopy structure parameters, and construct a sensitivity coupling parameter set according to the results of the sensitivity analysis and different preset soil and management measure conditions, where the preset soil and management measure conditions are constructed based on the historical soil parameters and the historical data of the crop management measures; Input the set of sensitivity coupling parameters into the combined agronomic remote sensing simulation model of crop yield to obtain the simulated crop yield data, the simulated canopy top reflectance data, and the simulated top-of-atmosphere reflectance data; Use the simulated top-of-atmosphere reflectance data as sample crop remote sensing data, and generate a crop yield label corresponding to the sample crop remote sensing data according to the simulated crop yield data to construct a training sample set; Train a recurrent neural network model based on the training sample set to obtain the crop yield estimation model.

[0086] In the present invention, the gated recurrent unit (GRU) in the deep learning model is used to perform intelligent modeling on the winter wheat yield. The GRU model is a type of recurrent neural network, mainly used to solve problems such as long-term memory and gradients in backpropagation. GRU combines the gate structures of LSTM, improving the complex unit structure of the previous recurrent neural network, which can increase the training speed of the network and ensure that the training accuracy of the model does not decrease. Compared with the "three-gate" structure of LSTM, GRU only has two gate structures, the update gate and the reset gate, to control the flow of long-term information, simplifying the network structure and improving the training speed of the model. In the present invention, the calculation process of the GRU model is as follows: where sigmoid() is the S-shaped activation function; tanh() represents the hyperbolic tangent activation function; represents t the output of the reset gate at time represents t the output of the update gate at time represents the weight matrix of the reset gate; represents the weight matrix of the update gate; represents the weight matrix of the candidate hidden state; represents the input of the GRU model at t time represents the output of the hidden layer state of the GRU model at t time represents the candidate hidden state of the current input.

[0087] During the model construction process, the present invention separately calculated the reflectance characteristics of different bands of winter wheat from the green-reverting stage to the flowering stage (vegetative growth stage) and from the flowering stage to the maturity stage (reproductive growth stage) as the temporal input characteristics of the model, and took the yield per unit area of winter wheat as the output characteristic of the model. The simulated dataset was divided into training samples and test samples according to a ratio of 9:1. The model optimizer was Adam (Adaptive Moment Estimation), and the loss function was mean squared error. The network was trained multiple times to find the optimal values of parameters such as the number of GRU layers, the number of neurons units, epochs, batch size, and dropout.

[0088] In the present invention, to explore the best prediction factors of the model, three vegetation indices of RVI, DVI, and NDVI for any two-band combinations in different growth stages were constructed respectively. The reflectance of a single band and the average value, maximum value, median value, and cumulative value of these vegetation indices during the growth stage were calculated respectively. These calculated metrics were used as the characteristic factors for prediction, and their correlations with the final yield of winter wheat were analyzed. The combination of characteristic factors with the highest correlation was selected as the prediction characteristics of the model.

[0089] In one embodiment, based on the existing winter wheat growth phenology dataset, the green-reverting and maturity times of winter wheat in different regions and different years were obtained respectively, and the accumulated temperature required for winter wheat to flower (DVS = 1) in the winter wheat growth simulation dataset of the corresponding partition was calculated. Based on the daily average air temperature data at a height of 2 meters above the ground provided by the ERA5-Land Daily Aggregated - ECMWF Climate Reanalysis data, the time for winter wheat to reach flowering was calculated pixel by pixel, so as to obtain the corresponding remote sensing image data of winter wheat from the green-reverting stage to the flowering stage (vegetative growth stage) and from the flowering stage to the maturity stage (reproductive growth stage) for calculating the prediction characteristics required by the GRU model. Among them, the estimation of winter wheat yield per unit area from 2000 to 2012 mainly used Landsat-5 satellite image data, and the estimation from 2013 to 2024 mainly used Landsat-8 satellite image data.

[0090] In the present invention, the verification of the estimation accuracy of winter wheat yield per unit area includes two aspects: verification with measured sample points and verification with statistical data. Among them, the verification data of measured sample points comes from the ground survey data from 2017 to 2024, and the main content collected is the yield per unit area of winter wheat. The statistical data comes from the statistical yearbooks of crop regions in each city from 2000 to 2023, and the yield per unit area of winter wheat in each city is mainly obtained. The accuracy verification indicators include the coefficient of determination R 2(Coefficient of Determination), Root Mean Square Error (RMSE), and Mean Relative Error (MRE). R 2 The higher it is, the smaller the RMSE and MRE are, indicating that the performance of the crop yield estimation model is better. The calculation formulas are as follows: ; Formula (23) ; Formula (24) ; Formula (25) Among them, and represent the actual yield and the estimated yield respectively, is the mean value of the actual yield.

[0091] Based on the above embodiments, perform a sensitivity analysis on the historical data of the crop variety and the canopy structure parameters, and construct a sensitivity coupling parameter set according to the results of the sensitivity analysis and different preset soil and management measure conditions, including: According to the historical data of the crop variety, obtain the historical data of the initial total crop dry weight, specific leaf area, total dry matter distribution ratio, first accumulated temperature, leaf life, and development stage in the crop area. Among them, the historical data of the total dry matter distribution ratio includes the historical data of the proportion of total dry matter allocated to leaves, the historical data of the proportion of total dry matter allocated to storage organs, and the historical data of the proportion of total dry matter allocated to stems; the historical data of the first accumulated temperature represents the accumulated temperature historical data of the crop from emergence to flowering; the historical data of the leaf life represents the leaf life data of the crop at 35 degrees Celsius; the historical data of the development stage includes the state variables of the crop from the initial development stage to the harvest development stage; Obtain the historical data of the leaf equal water layer thickness; Construct the sensitivity coupling parameter sets corresponding to different preset soil and management measure conditions according to the historical data of the initial total crop dry weight, the specific leaf area, the total dry matter distribution ratio, the first accumulated temperature, the leaf life, the development stage, the leaf inclination data, and the historical data of the leaf equal water layer thickness.

[0092] Since the impacts of model parameters on reflectance and crop growth simulation vary significantly at different time periods and development stages, different parameters or parameter combinations have different degrees of influence on the simulation results at different growth stages. Therefore, in order to visualize the contributions and sensitivities of each parameter to the results throughout the simulation process, the present invention conducts a global sensitivity analysis on the constructed WOFOST-SCOPE-SMAC agro-remote sensing joint simulation model in the time domain and spectral domain of crop growth simulation. For different preset soil and management measure conditions, key parameters that have significant impacts on the model output at different growth stages under different soil parameters and management measures are identified, providing a basis for subsequent model analysis. The global sensitivity analysis in the present invention is based on the Fourier Amplitude Sensitivity Test (FAST) method.

[0093] Regarding the WOFOST model parameters, the present invention selects 11 crop parameters shown in Table 5, and calculates their value ranges by setting the initial values and distribution states of their mapping variables. Referring to existing studies, it is assumed that the prior information of the 10 constructed mapping variables (β_IDEM, α_TSUM1, α_TSUM2, α_TDWI, α_SPAN, αSLATB, α_AMAXTB, α_CVO, α_v and β_DVS) all follow a normal distribution. For the SCOPE model, it is analyzed that the model input parameters Cab, Cw, LIDFa and Vcmax25 are crucial for the simulation of crop canopy reflectance. Since the sensitivity analysis results are mainly used to support the construction of multi-scenario simulation datasets, Cab has been assigned values through the input data products in the present invention, and dynamic assignment has been implemented for Vcmax25. Therefore, when conducting parameter sensitivity analysis, only the impacts of Cw and LIDFa on reflectance simulation are considered. Further, the present invention assumes their prior information to be uniformly distributed. Since the SMAC model only requires additional data products and meteorological station data as model inputs and is not considered additionally during sensitivity analysis. In summary, the prior information of the coupled model parameters is shown in Table 13.

[0094] Table 13 Prior Information of Coupled Model Parameters

[0095] By obtaining the prior distribution information of different parameters, more than 5,000 sets of crop growth simulation results with different initial parameter sets were generated. First, taking the LAI and the dry weight of the ear (TWSO) obtained by the WOFOST model as the target variables, a sensitivity analysis of the model parameters was carried out to determine the model parameters that are sensitive to leaf growth and development and the accumulation of ear dry weight. Furthermore, taking Landsat-8 as the target sensor and the TOA reflectance of each band of Landsat-8 simulated by the WOFOST-SCOPE-SMAC coupled model as the target variable, a further analysis of the sensitive parameters of the coupled model was carried out. To ensure that the cumulative effect of all parameters at each time step is equal to 1, the sensitivity index was normalized, and the influence of model parameters on different target variables was visualized with the growth stage of winter wheat (DVS) as the independent variable.

[0096] Figure 4 Schematic diagram of the sensitivity analysis results of the WOFOST model parameters provided by the present invention for the leaf area index, for reference Figure 4 As shown, taking the LAI simulated by the WOFOST model as the target variable, the sensitivity analysis results of the WOFOST model parameters are shown. It can be seen from the results that before the heading of winter wheat (DVS < 0.75), the model parameters sensitive to the change of winter wheat LAI are mainly α_TDWI, α_SLATB and α_v; after the heading of winter wheat (DVS > 0.75), the model parameters sensitive to the change of winter wheat LAI are mainly α_TSUM1, α_SPAN, α_SLATB, β_DVS, α_v.

[0097] Figure 5 Schematic diagram of the sensitivity analysis results of the WOFOST model parameters provided by the present invention for the accumulation of ear dry weight, for reference Figure 5 As shown, taking the simulated ear dry weight of the WOFOST model as the target variable, a sensitivity analysis of the WOFOST model parameters was carried out. The model parameters sensitive to the change of winter wheat ear dry weight are mainly α_TSUM1 and β_DVS, which mainly determine the time from emergence to flowering of winter wheat and the biomass allocation ratio of each organ in the later growth stage.

[0098] Figure 6 Schematic diagram of the sensitivity analysis results of the coupled model provided by the present invention for the top-of-atmosphere reflectance, for reference Figure 6 As shown, taking the TOA reflectance simulated by the coupled model as the target variable, a sensitivity analysis of the coupled model parameters was carried out. The value of the leaf inclination angle (LIDFa) is highly sensitive to all nine bands of Landsat-8. Since Cw is related to the leaf water content, it is also relatively sensitive in the short-wave infrared bands (band 6 and band 7).

[0099] The present invention mainly considers different crop varieties by dynamically setting the above sensitive parameters, and constructs a sensitivity coupling parameter set by combining meteorological data of different annual types, different types of soil data, and different management measures, so as to provide a sufficient simulation data set for the agronomic knowledge base of the winter wheat yield formation process under various agricultural production scenarios.

[0100] In one embodiment, based on the climate zone division data, for example, a certain climate region contains nine secondary climate zones. Considering the winter wheat planting area in this region, the climate zones with a small proportion of the climate zone area and a small winter wheat planting area are merged, and finally the climate zoning of this region is realized. For details, please refer to Table 14.

[0101] Table 14 Secondary climate zones covered by a certain climate zone

[0102] Considering that the climate differences in different years in the study area will affect the growth status and final yield of winter wheat, based on the meteorological station data in the study area, the average temperature and cumulative precipitation of each growth period of winter wheat are calculated, and the average value and standard deviation of the average temperature and cumulative precipitation from 2000 to 2020 are calculated, and the climate annual types are divided by the average value ± standard deviation.

[0103] Furthermore, simulation data sets are constructed for each annual type in each ecological zone. The construction of the simulation data set is mainly realized by constructing multiple scenarios for the parameters of the WOFOST-SCOPE-SMAC model. For the WOFOST model, multiple scenario settings of four model input parameters, namely meteorological parameters, soil type, crop variety, and management measures, are considered. First, the average value and standard deviation of each index of the multi-year meteorological data from 2000 to 2021 of the meteorological stations included in each ecological zone are calculated respectively, and three meteorological data are provided for each ecological zone as model inputs with the average value ± standard deviation; for the soil type, three soil types, namely sandy loam, light loam, and medium loam, are considered for parameter setting; for the crop variety, according to the above model parameter sensitivity analysis results, different crop varieties are constructed by equidistant sampling within the 95% confidence interval of the sensitive parameter values; for the management measures, the simulations of rain-fed and irrigated planting modes are realized by using Wofost72_WLP_CWB and Wofost72_PP respectively, and the simulations of different sowing dates are realized by setting the β_IDEM parameter. For the SCOPE model, according to the above model parameter sensitivity analysis results, different values of the sensitive parameters are obtained by equidistant sampling within the value range of the sensitive parameters.

[0104] Based on the idea of "lookup table", the present invention permutes and combines the above parameter inputs to generate different input coupling models for growth scenarios, so as to simulate different winter wheat growth scenarios. For each annual type in each ecological region, more than 350,000 (2×3 11 ) simulation data are generated. A simulation result of the winter wheat growth scenario can be referred to Table 15: Table 15 Simulation of Winter Wheat Growth Scenarios

[0105] In one embodiment, Figure 7 is a schematic diagram of the distribution of meteorological data in a certain area provided by the present invention, which can be referred to Figure 7 as shown. The shaded areas respectively represent the average temperature ± standard deviation and the cumulative precipitation ± standard deviation. The present invention divides the climate annual types by using the years within the intersection area as normal annual types. The division results of the annual types in three climate regions (such as Region A, Region B, and Region C) are shown in Table 16, Table 17, and Table 18. For Region A, a total of five annual types are divided, namely normal year, low temperature year, high temperature year, rainy year, and dry year. Among them, there are 8 years of normal annual type, 3 years of low temperature annual type, 5 years of high temperature annual type, 3 years of rainy annual type, and 2 years of dry annual type; for Region B, a total of six annual types are divided, namely normal year, low temperature year, high temperature year, high temperature and rainy year, rainy year, and dry year. Among them, there are 11 years of normal annual type, 3 years of low temperature annual type, 1 year of high temperature annual type, 2 years of high temperature and rainy annual type, 2 years of rainy annual type, and 2 years of dry annual type; for Region C, a total of six annual types are divided, namely normal year, low temperature year, high temperature year, high temperature and rainy year, rainy year, and dry year. Among them, there are 10 years of normal annual type, 2 years of low temperature annual type, 3 years of high temperature annual type, 1 year of high temperature and rainy annual type, 2 years of rainy annual type, and 2 years of dry annual type.

[0106] Table 16 Division of Annual Types in Region A

[0107] Table 17 Division of Annual Types in Region B

[0108] Table 18 Division of Annual Types in Region C

[0109] Furthermore, in order to verify whether the constructed agronomic knowledge base for the yield formation process conforms to the actual growth situation of winter wheat, the present invention first statistically analyzed the simulation of the final yield of winter wheat under different climate regions and annual types. Taking the yield simulation results of different annual types in Region A as an example, Figure 8 is a schematic diagram of the yield simulation in a certain climate annual type provided by the present invention, and the statistical situation is as Figure 8As shown. The results show that with the change of climate, the yield of winter wheat also changes. The climate conditions of high temperature and less rainfall have a greater impact on the yield of winter wheat and are more likely to cause low yield To further verify the simulation of TOA reflectance, referring to the existing winter wheat phenology dataset and combining with the winter wheat planting distribution data, 10 sampling points with large phenological differences and uniform distribution were selected for different years respectively, and the TOA reflectance data of Landsat-8 / 5 images at three time nodes of winter wheat greening, flowering and maturity were extracted to verify the simulated reflectance. Taking the normal year in Area A as an example, Figure 9 is a schematic diagram of the simulated reflectance of winter wheat at the flowering stage in a certain climate year type provided by the present invention. The simulation results can be referred to Figure 9 As shown. Among them, the black curve represents the average simulated reflectance of the simulated data at different growth stages of winter wheat, and the shaded area represents one standard deviation of the simulated data. The results show that the simulated dataset can accurately simulate the TOA reflectance under different climate years and different growth stages of winter wheat while ensuring simulation diversity.

[0110] Table 19 shows the screening results of characteristic factors for different regions and years. Taking the correlation analysis of characteristic factors in the normal year type of Area A as an example, finally, NDVI(2,5)max and NDVI(6,7)max are selected as the characteristics of the vegetative growth stage and reproductive growth stage respectively for model training and prediction.

[0111] Table 19 Optimal Selection Results of Characteristic Factors

[0112] Furthermore, the ground measured yield data from 2017 to 2024 are used to verify the accuracy of the model prediction results at the single-point scale. Figure 10 is a schematic diagram of the single-point scale verification result provided by the present invention. The accuracy verification result can be referred to Figure 10 As shown. It can be seen from the results that the prediction model achieves a relatively accurate estimation of the winter wheat yield at the single-point scale, and the overall prediction accuracy is higher than 90%.

[0113] The statistical data of winter wheat yield at the municipal level from 2000 to 2023 are used to verify the accuracy of the yield estimation results of the model at the regional scale respectively. Figure 11 is a schematic diagram of the verification result at the municipal scale provided by the present invention. The verification result is as Figure 11 shown. The coefficient of determination is 0.67, the root mean square error is 839.79 kg / ha, and the average relative error is 13.07%.

[0114] Figure 12 is a schematic diagram of the spatial distribution of winter wheat yield in a certain area in 2024 provided by the present invention. It can be referred to Figure 12As shown, the model prediction results exhibit good spatio-temporal continuity. The pixel-scale yield estimation not only obtains yield information at a finer scale but also shows high accuracy at the municipal scale. Additionally, the method provided by the present invention can achieve dynamic and flexible yield estimation according to the available data situation, demonstrating great potential in large-scale winter wheat yield estimation.

[0115] The crop yield remote sensing estimation system provided by the present invention will be described below. The crop yield remote sensing estimation system described below can be mutually referred to and corresponding to the crop yield remote sensing estimation method described above.

[0116] Figure 13 It is a schematic structural diagram of the crop yield remote sensing estimation system provided by the present invention, as Figure 13 shown, the present invention provides a crop yield remote sensing estimation system, including a crop remote sensing data acquisition module 1301 and a crop yield estimation module 1302. Among them, the crop remote sensing data acquisition module 1301 is used to obtain target remote sensing data corresponding to the crop area; the crop yield estimation module 1302 is used to input the target remote sensing data into the crop yield estimation model to obtain the crop yield prediction data of the crop area output by the crop yield estimation model; among them, the crop yield estimation model is trained based on the crop yield simulation data and the top-of-atmosphere reflectance simulation data corresponding to the crop area; the crop yield simulation data and the top-of-atmosphere reflectance simulation data are generated by a crop yield agronomic remote sensing joint simulation model, and the crop yield agronomic remote sensing joint simulation model is constructed based on a crop growth model and an atmospheric radiation transfer model.

[0117] The crop yield remote sensing estimation system provided by the present invention constructs a crop yield agronomic remote sensing joint simulation model to generate simulation data for training the crop yield estimation model, which improves the construction quality of the crop sample dataset and further improves the accuracy of the crop yield prediction results, providing more reliable decision-making support for agricultural production and management.

[0118] The system provided in the embodiments of the present invention is used to execute the above method embodiments. For the specific process and detailed content, please refer to the above embodiments and will not be elaborated here.

[0119] FIG. 14 is a schematic structural diagram of an electronic device provided by the present invention. As shown in FIG. 14, the electronic device may include: a processor 1401, a communications interface 1402, a memory 1403, and a communication bus 1404. Among them, the processor 1401, the communications interface 1402, and the memory 1403 complete mutual communication through the communication bus 1404. The processor 1401 may call logic instructions in the memory 1403 to execute a method for remotely estimating the single crop yield. The method includes: obtaining target remote sensing data corresponding to a crop area; inputting the target remote sensing data into a single crop yield estimation model to obtain single crop yield prediction data of the crop area output by the single crop yield estimation model; wherein, the single crop yield estimation model is trained based on single crop yield simulation data and top-of-atmosphere reflectance simulation data corresponding to the crop area; the single crop yield simulation data and the top-of-atmosphere reflectance simulation data are generated by a combined single crop yield and agricultural remote sensing simulation model, and the combined single crop yield and agricultural remote sensing simulation model is constructed based on a crop growth model and an atmospheric radiative transfer model.

[0120] In addition, when the logic instructions in the foregoing memory 1403 can be implemented in the form of a software functional unit and sold or used as an independent product, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk, or an optical disc that can store program codes.

[0121] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the crop yield remote sensing estimation method provided by each of the above methods. The method includes: obtaining target remote sensing data corresponding to a crop area; inputting the target remote sensing data into a crop yield estimation model to obtain crop yield prediction data of the crop area output by the crop yield estimation model; wherein, the crop yield estimation model is trained based on crop yield simulation data and top-of-atmosphere reflectance simulation data corresponding to the crop area; the crop yield simulation data and the top-of-atmosphere reflectance simulation data are generated by a crop yield agro-remote sensing joint simulation model, and the crop yield agro-remote sensing joint simulation model is constructed based on a crop growth model and an atmospheric radiative transfer model.

[0122] In another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is configured to execute the crop yield remote sensing estimation method provided by each of the above embodiments. The method includes: obtaining target remote sensing data corresponding to a crop area; inputting the target remote sensing data into a crop yield estimation model to obtain crop yield prediction data of the crop area output by the crop yield estimation model; wherein, the crop yield estimation model is trained based on crop yield simulation data and top-of-atmosphere reflectance simulation data corresponding to the crop area; the crop yield simulation data and the top-of-atmosphere reflectance simulation data are generated by a crop yield agro-remote sensing joint simulation model, and the crop yield agro-remote sensing joint simulation model is constructed based on a crop growth model and an atmospheric radiative transfer model.

[0123] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative labor.

[0124] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0125] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for remotely sensing and estimating the single yield of crops, characterized in that, Including: Obtaining target remote sensing data corresponding to a crop area; Inputting the target remote sensing data into a crop yield estimation model to obtain crop yield prediction data of the crop area output by the crop yield estimation model; Wherein, the crop yield estimation model is trained based on crop yield simulation data and top-of-atmosphere reflectance simulation data corresponding to the crop area; the crop yield simulation data and the top-of-atmosphere reflectance simulation data are generated by a combined crop yield and agronomic remote sensing simulation model, and the combined crop yield and agronomic remote sensing simulation model is constructed based on a crop growth model and an atmospheric radiative transfer model.

2. The remote sensing estimation method for crop yield per unit area according to claim 1, wherein The combined crop yield and agronomic remote sensing simulation model is specifically used for: Obtaining historical meteorological data, historical soil data, historical crop variety data, and historical crop management measure data of the crop area at a historical moment; Inputting the historical meteorological data, the historical soil data, the historical crop variety data, and the historical crop management measure data into the crop growth model to obtain crop growth process simulation data corresponding to the crop area output by the crop growth model, wherein the crop growth process simulation data includes the crop yield simulation data, leaf area index simulation data, leaf dry weight simulation data, and simulation date; Obtaining historical chlorophyll data and historical incident longwave radiation data of the crop area at a historical moment; Inputting the historical chlorophyll data, the historical incident longwave radiation data, the leaf area index simulation data, the leaf dry weight simulation data, specific leaf area input data, the simulation date, and longitude and latitude information into a canopy radiative transfer model to obtain canopy top reflectance simulation data output by the canopy radiative transfer model, wherein the specific leaf area input data and the longitude and latitude information are input parameters obtained based on the crop growth model; Obtaining historical ozone concentration data and historical aerosol optical depth data of the crop area at a historical moment; Inputting the historical ozone concentration data, the historical aerosol optical depth data, and the canopy top reflectance simulation data into the atmospheric radiative transfer model to obtain the top-of-atmosphere reflectance simulation data output by the atmospheric radiative transfer model.

3. The remote sensing estimation method for crop yield per unit area according to claim 2, characterized in that, The crop growth model is the WOFOST model, the canopy radiative transfer model is the SCOPE model, and the atmospheric radiative transfer model is the SMAC model.

4. The remote sensing estimation method for the single crop yield according to claim 2, wherein The step of inputting the historical chlorophyll data, the historical incident longwave radiation data, the leaf area index simulation data, the leaf dry weight simulation data, specific leaf area input data, the simulation date, and longitude and latitude information into a canopy radiative transfer model to obtain canopy top reflectance simulation data output by the canopy radiative transfer model includes: Constructing leaf optical parameters according to the historical chlorophyll data, the leaf dry weight simulation data, and the specific leaf area input data; Obtain the plant height data, leaf inclination data, leaf inclination change data, and leaf width data of the crop area at a historical moment, and construct canopy structure parameters based on the leaf area index simulation data, the plant height data, the leaf inclination data, the leaf inclination change data, and the leaf width data. Obtain the solar zenith angle data, observation zenith angle data, and relative azimuth angle data of the crop area at a historical moment, and construct time and angle parameters based on the simulation date, the longitude and latitude information, the solar zenith angle data, the observation zenith angle data, and the relative azimuth angle data. Construct target meteorological parameters based on the incident longwave radiation historical data, the meteorological historical data, and the preset meteorological data. Obtain the preset aerodynamic parameters and the preset soil reflectivity corresponding to the canopy radiation transfer model. Based on the canopy radiation transfer model, perform reflectivity data simulation on the leaf optical parameters, the canopy structure parameters, the time and angle parameters, the target meteorological parameters, the preset aerodynamic parameters, and the preset soil reflectivity to obtain the canopy top reflectivity simulation data.

5. The remote sensing estimation method for crop yield per unit area according to claim 4, characterized in that, The crop yield estimation model is trained through the following steps: Conduct a sensitivity analysis on the crop variety historical data and the canopy structure parameters, and construct a sensitivity coupling parameter set according to the results of the sensitivity analysis and different preset soil and management measure conditions, where the preset soil and management measure conditions are constructed based on the soil historical parameters and the crop management measure historical data. Input the sensitivity coupling parameter set into the crop yield agro-remote sensing joint simulation model to obtain the crop yield simulation data, the canopy top reflectivity simulation data, and the top-of-atmosphere reflectivity simulation data. Use the top-of-atmosphere reflectivity simulation data as the sample crop remote sensing data, and generate a crop yield label corresponding to the sample crop remote sensing data according to the crop yield simulation data to construct a training sample set. Based on the training sample set, train a recurrent neural network model to obtain the crop yield estimation model.

6. The remote sensing estimation method for crop yield per unit area according to claim 5, wherein The sensitivity analysis of the crop variety historical data and the canopy structure parameters, and the construction of the sensitivity coupling parameter set according to the results of the sensitivity analysis and different preset soil and management measure conditions include: According to the historical data of the crop variety, obtain the historical data of the initial total dry weight of crops in the crop area, the specific leaf area historical data, the historical data of the total dry matter distribution ratio, the historical data of the first accumulated temperature, the historical data of the leaf life, and the historical data of the development stage. Among them, the historical data of the total dry matter distribution ratio includes the historical data of the ratio of the total dry matter distributed to the leaves, the historical data of the ratio of the total dry matter distributed to the storage organs, and the historical data of the ratio of the total dry matter distributed to the stems; the historical data of the first accumulated temperature represents the historical data of the accumulated temperature of the crops from emergence to flowering; the historical data of the leaf life represents the life data of the leaves of the crops at 35 degrees Celsius; the historical data of the development stage includes the state variables of the crops from the initial development stage to the harvest development stage. Obtain the historical data of the leaf equal water layer thickness. According to the historical data of the initial total dry weight of crops, the specific leaf area historical data, the historical data of the total dry matter distribution ratio, the historical data of the first accumulated temperature, the historical data of the leaf life, the historical data of the development stage, the leaf inclination data, and the historical data of the leaf equal water layer thickness, construct the set of sensitivity coupling parameters corresponding to different preset soil and management measure conditions.

7. A remote sensing estimation system for the single crop yield, characterized in that, Including: A crop remote sensing data acquisition module for obtaining target remote sensing data corresponding to the crop area; A crop yield estimation module for inputting the target remote sensing data into a crop yield estimation model to obtain the crop yield prediction data of the crop area output by the crop yield estimation model; Among them, the crop yield estimation model is trained based on the crop yield simulation data corresponding to the crop area and the top-of-atmosphere reflectance simulation data; the crop yield simulation data and the top-of-atmosphere reflectance simulation data are generated by a crop yield agronomic remote sensing joint simulation model, and the crop yield agronomic remote sensing joint simulation model is constructed based on a crop growth model and an atmospheric radiation transfer model.

8. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the crop yield remote sensing estimation method according to any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the crop yield remote sensing estimation method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the crop yield remote sensing estimation method according to any one of claims 1 to 6.

Images