Corn yield monitoring system and method based on remote sensing-dssat coupling in irrigation district

Abstract

The application relates to the technical field of corn yield monitoring, in particular to a farmland corn yield monitoring system and method based on remote sensing-DSSAT coupling, which comprises the following steps: acquiring and fusing high-resolution remote sensing data and low-resolution remote sensing data; identifying field boundaries to form a field-level analysis unit and correcting vegetation parameters; generating individualized growth parameters for each field-level analysis unit; performing two-level collaborative assimilation: at the aggregation scale, using the corrected vegetation parameters to constrain the overall trend of the simulation state of each field-level analysis unit; at the independent scale, updating the vegetation growth state, soil moisture state and canopy temperature state of each field-level analysis unit in parallel; and generating a field-level yield distribution map and a differentiated management scheme based on the updated state. The application can solve the mixed pixel problem by coupling remote sensing data and a crop model and performing scale decomposition, so that field-level corn yield accurate monitoring and differentiated management and control can be realized.

Description

Technical Field

[0001] This application relates to the technical field of maize yield monitoring, and in particular to a maize yield monitoring system and method for irrigation areas based on remote sensing-DSSAT coupling. Background Technology

[0002] Against the backdrop of escalating global climate change, extreme weather events are becoming more frequent, with drought having a particularly pronounced impact on agricultural production. Uneven spatial and temporal distribution of precipitation leads to insufficient water supply during critical crop growth periods, severely restricting the growth, development, and final yield of major food crops such as maize. Therefore, developing technologies that can accurately simulate crop growth responses under different water conditions and achieve high-precision yield forecasts is of paramount importance for agricultural disaster assessment, food security early warning, and production management decision-making.

[0003] Currently, combining crop growth models with data assimilation techniques has become an important research direction in the field of yield prediction. For example, in the existing technology entitled "Multi-scheme Cooperative Yield Forecasting Method Based on Data Assimilation and Model Parameter Optimization" (Chinese Patent Publication No. CN121328112A), a method based on the WOFOST model is proposed. This method assimilates leaf area index and soil moisture through Kalman filtering, and optimizes photosynthetic parameters and improves the water stress function to enhance simulation accuracy. This technique constructs multiple cooperative simulation schemes and attempts to determine the optimal prediction node through a rolling forecast framework, representing an advanced solution in this field.

[0004] However, this existing technology still faces a fundamental technical bottleneck in practical applications, limiting its potential for precise field-level management in irrigation areas with significant spatial heterogeneity. The core issue lies in the inherent mismatch between the data assimilation framework of the existing technology and crop growth models in terms of "spatial scale." This prevents it from effectively handling the inherent "mixed pixel" problem in low-to-medium resolution remote sensing data, thus hindering the generation of truly differentiated yield monitoring and diagnostic results with field-level resolution. Specifically, the assimilation process of the existing technology relies on regional vegetation parameters retrieved from satellites as observational constraints. However, these parameters originate from satellite pixels with spatial resolutions typically ranging from ten to one hundred meters. In a typical irrigation area, a single such pixel often covers multiple independent fields with different soil properties, irrigation conditions, management practices, and even growth status. The existing assimilation strategy treats the entire pixel as a homogeneous simulation unit or applies the pixel's observations averaged across its supposedly homogeneous internal region. This approach forcibly smooths out the true heterogeneity between fields within a single pixel, preventing crop models from identifying and simulating the unique growth dynamics exhibited by different fields due to variations in local soil fertility and water stress. For example, when a pixel contains both drought-stricken and well-watered fields, the model can only simulate a compromise, averaged growth state. The parameters and states optimized in its assimilation process are for this virtual homogeneous unit, not the actual physical field. Therefore, even if the model achieves high forecast accuracy at the regional or pixel-average scale, its output cannot be decomposed into yield distribution maps and stress diagnoses aligned with real field boundaries to guide specific agricultural operations. This creates a practical obstacle when applying existing technologies to modern farmland management requiring precision irrigation and variable-rate fertilization; that is, while regional trends are known, it is impossible to pinpoint which specific field is experiencing a problem, its severity, and how to address it differently. This scale mismatch is the core bottleneck restricting existing technologies from moving from regional yield assessment to field-level precision management decision-making. Summary of the Invention

[0005] In order to solve the mixed pixel problem by coupling remote sensing data with crop models and performing scale decomposition, and to achieve accurate monitoring and differentiated management of maize yield at the field level, this application provides a maize yield monitoring system and method in irrigation areas based on remote sensing-DSSAT coupling.

[0006] Firstly, the method for monitoring maize yield in irrigated areas based on remote sensing-DSSAT coupling provided in this application adopts the following technical solution: The method for monitoring maize yield in irrigated areas based on remote sensing-DSSAT coupling includes:

[0007] Acquire and fuse high-resolution remote sensing data with low-resolution remote sensing data;

[0008] Based on the high-resolution remote sensing data, field boundaries are identified. Based on the field boundaries, pixels of low-resolution remote sensing data are spatially decomposed and resampled to form field-level analysis units. The vegetation parameters retrieved from low-resolution remote sensing data are corrected using the high-resolution remote sensing data.

[0009] For each field-level analysis unit, personalized growth parameters are generated based on its temporal vegetation dynamics and soil properties.

[0010] Two-level collaborative assimilation is performed: at the aggregate scale consisting of multiple field-level analysis units, the overall trend of the simulated state of each field-level analysis unit is constrained by the corrected vegetation parameters; at the independent scale of each field-level analysis unit, the vegetation growth state, soil moisture state and canopy temperature state of each field-level analysis unit are updated in parallel based on its personalized growth parameters and while satisfying the overall trend constraint.

[0011] Based on the updated vegetation growth status, soil moisture status, and canopy temperature status, the yield of each field-level analysis unit is calculated, and a field-level yield distribution map and differentiated management plan are generated.

[0012] Optionally, the step of correcting the vegetation parameters retrieved from the low-resolution remote sensing data using the high-resolution remote sensing data includes:

[0013] Based on the high-resolution remote sensing data, identify pure vegetation areas that meet the preset vegetation coverage threshold.

[0014] Based on the vegetation parameters of the pure vegetation area, spatial decomposition and resampling correction are performed on the mixed pixels of the low-resolution remote sensing data;

[0015] Using the spectral information of the high-resolution remote sensing data in the preset organic matter sensitive band, the vegetation parameters retrieved from the low-resolution remote sensing data are corrected for spectral consistency.

[0016] Based on the temporal vegetation indices of high-resolution and low-resolution remote sensing data within a preset critical growth period window, the vegetation parameters retrieved from the low-resolution remote sensing data are corrected for temporal trend alignment.

[0017] Optionally, the generation of personalized growth parameters includes:

[0018] Extract the rate of change of vegetation index for each field-level analysis unit during a preset early growth stage as the first feature;

[0019] The soil organic matter content of each field-level analysis unit is obtained as a second feature;

[0020] Based on the change in the first feature, the adjustment of parameters for the vegetative growth stage is determined;

[0021] Based on the amount of change in the second feature, adjustments to the parameters of the reproductive growth stage are determined;

[0022] Based on the aforementioned adjustments, personalized growth parameters are generated for each field-level analysis unit.

[0023] Optionally, the two-level collaborative assimilation includes:

[0024] Based on the personalized growth parameters generated for each field-level analysis unit, initialize its corresponding set of simulated states;

[0025] At the aggregation scale, the simulated states of the multiple field-level analysis units are integrated, and the corrected vegetation parameters are used as a common observation constraint to drive the integrated simulated state to approach the common observation constraint as a whole.

[0026] At each unit-independent scale, each of the field-level analysis units performs parallel updates of its simulated state within its own set of simulated states and under the overall trend constraints determined by the approximation, in order to optimize its vegetation growth state, soil moisture state, and canopy temperature state, respectively.

[0027] Optionally, the parallel execution of the simulation state update further includes setting a criterion for assimilation convergence, wherein the statistical uncertainty of at least one of the vegetation growth state, soil moisture state, and canopy temperature state in the simulation state set is lower than a preset threshold.

[0028] Optionally, after the determination condition is met, the method further includes:

[0029] Based on the set of simulated states that meet the aforementioned criteria, grain yield assessment results and growth stress diagnosis results are generated synchronously for each field-level analysis unit.

[0030] Based on the combination of the grain yield assessment results and the growth stress diagnosis results, a corresponding differentiated management plan is mapped from the preset agronomic knowledge base;

[0031] The differentiated management schemes, spatially associated with each field-level analysis unit, are output to form a visually presentable decision-making product.

[0032] Optional, also includes:

[0033] Spatially associate and bind the field-level yield distribution map with the differentiated management scheme;

[0034] The associated data is visualized and interactively queried through a pre-set geographic information system platform.

[0035] Optionally, obtaining the soil organic matter content of each of the field-level analysis units includes:

[0036] Using the hyperspectral information in the high-resolution remote sensing data, the reflectance of the preset organic matter-sensitive near-infrared band is extracted;

[0037] The soil organic matter content of each field-level analysis unit was obtained based on the reflectance inversion.

[0038] Optionally, using the corrected vegetation parameters as a common observation constraint includes:

[0039] The corrected vegetation parameters are used as the common constraint target for simulating vegetation parameters in all field-level analysis units within the corresponding aggregation scale;

[0040] By adjusting the simulation state of the multiple field-level analysis units, the overall statistical characteristics of the simulated vegetation parameters of each field-level analysis unit are made to meet the common constraint objective.

[0041] Secondly, the irrigation district maize yield monitoring system based on remote sensing-DSSAT coupling provided in this application adopts the following technical solution: The irrigation district maize yield monitoring system based on remote sensing-DSSAT coupling includes:

[0042] The acquisition and fusion module is used to acquire and fuse high-resolution remote sensing data with low-resolution remote sensing data;

[0043] The scale decoupling module, connected to the acquisition and fusion module, is used to identify field boundaries based on the high-resolution remote sensing data, spatially decompose and resample the pixels of the low-resolution remote sensing data according to the field boundaries to form field-level analysis units, and use the high-resolution remote sensing data to correct the vegetation parameters retrieved from the low-resolution remote sensing data.

[0044] The dynamic calibration module, connected to the scale decoupling module, is used to generate personalized growth parameters for each field-level analysis unit based on its temporal vegetation dynamics and soil properties.

[0045] The collaborative assimilation module, connected to the acquisition and fusion module and the dynamic calibration module, is used to constrain the overall trend of the simulation state of each field-level analysis unit at the aggregate scale consisting of multiple field-level analysis units using the corrected vegetation parameters; at the independent scale of each field-level analysis unit, based on its personalized growth parameters and under the constraint of the overall trend, it updates the vegetation growth state, soil moisture state and canopy temperature state of each field-level analysis unit in parallel.

[0046] The decision output module, connected to the collaborative assimilation module, is used to calculate the yield of each field-level analysis unit and generate a field-level yield distribution map and differentiated management plan based on the updated vegetation growth status, soil moisture status and canopy temperature status.

[0047] In summary, this application includes the following beneficial technical effects:

[0048] 1. This application fundamentally solves the problem of mixed pixels that constrains existing technologies by introducing a scale decomposition strategy based on high-resolution remote sensing data. Specifically, it uses high-resolution data to accurately identify field boundaries and then spatially decomposes and resamples low-resolution remote sensing pixels to construct analytical units that strictly correspond to real physical fields. This core step breaks through the limitation of traditional methods that treat mixed pixels containing multiple heterogeneous fields as homogeneous units for assimilation simulation. This allows crop models to perform state simulation and parameter optimization for each independent field, thereby achieving accurate monitoring and spatial positioning of maize yield distribution and growth stress aligned with field boundaries, providing a feasible spatial basis for subsequent differentiated field management.

[0049] 2. This application generates personalized growth parameters for each field by integrating temporal vegetation dynamics and soil properties, and significantly improves the simulation accuracy and reliability of crop models at the field scale by combining an innovative two-level synergistic assimilation framework. This method not only ensures the consistency of regional trends by utilizing aggregated-scale observation constraints, but also assimilates high-resolution observation data in parallel at the independent scale of each field, thereby finely depicting the heterogeneous patterns among fields caused by differences in soil fertility, water conditions, etc. This enables the model to more realistically reflect the unique growth process of each field, ultimately achieving a simultaneous improvement in the accuracy of field-level yield prediction and the accuracy of diagnosing growth stresses such as water and fertility.

[0050] 3. The final output of this application directly serves precision agricultural management decision-making. It not only generates field-level yield distribution maps, but more importantly, it automatically maps and generates differentiated irrigation, fertilization, and other agronomic management plans based on the yield level and specific stress type of each field. These plans are strictly bound to the spatial location of the fields and are integrated into a geographic information system for visualization and interactive querying. This allows managers to intuitively and quickly locate problem fields and obtain targeted management measures, effectively overcoming the previous technological obstacle of only providing regional trends and not guiding specific field operations. This directly supports modern farmland management practices such as variable irrigation and precision fertilization, improving water and fertilizer resource utilization efficiency and overall production benefits. Attached Figure Description

[0051] Figure 1 This is a flowchart of the monitoring method steps;

[0052] Figure 2 This is a schematic diagram of a differentiated management plan. Detailed Implementation

[0053] The following is in conjunction with the appendix Figure 1-2 This application will be described in further detail.

[0054] This application discloses a method for monitoring maize yield in irrigated areas based on remote sensing-DSSAT coupling. For example... Figure 1 As shown, the remote sensing-DSSAT coupled method for monitoring maize yield in irrigated areas is applicable to various scenarios of precise maize yield monitoring and differentiated management at the field level in irrigated areas, achieving both mixed pixel interference elimination and precise field-level monitoring. The following steps provide a detailed explanation:

[0055] S1 Multi-source Remote Sensing Data Acquisition and Fusion

[0056] This step involves establishing a standardized process for the acquisition, preprocessing, and fusion of high-resolution and low-resolution remote sensing data. By precisely defining data parameters and processing specifications, it addresses the error issues caused by the heterogeneity of multi-source data, generating basic data that combines spatial detail with large-scale coverage. This lays a solid foundation for subsequent accurate identification of field boundaries and reliable inversion of vegetation parameters.

[0057] S11 Data Acquisition Standards Determined

[0058] High-resolution remote sensing data was acquired using a drone equipped with a hyperspectral sensor. The drone used was a DJI Matrice 300RTK, and the accompanying hyperspectral sensor was a model MS600. Considering the spectral response characteristics of maize vegetation and the requirements for soil organic matter retrieval, the sensor's wavelength range was set to 400nm-2500nm, with a spectral resolution of 10nm and a spatial resolution of 0.1m. The near-infrared band around 2200nm was specifically designed to capture the spectral response signal of soil organic matter; this band coverage design is a key prerequisite for achieving accurate inversion of soil organic matter at the field level.

[0059] Before data collection, flight paths were planned using DroneDeploy software. Multiple field data collection experiments in the irrigation area revealed that if cloud cover exceeded 10% in the monitored area, noise in the remote sensing data would significantly increase, potentially interfering with the accuracy of subsequent vegetation parameter inversion and field boundary identification. Therefore, 10% was set as the cloud cover screening threshold. When cloud cover exceeded this threshold, the software automatically generated supplementary flight paths to ensure no data was missed across the entire irrigation area, while also guaranteeing that the signal-to-noise ratio of the collected data met the requirements for subsequent processing.

[0060] Low-resolution remote sensing data were selected from Landsat-8 / 9 and Sentinel-2 satellite imagery. Landsat-8 / 9 imagery has a spatial resolution of 30m, encompassing the red band (655nm-665nm) and near-infrared band (845nm-855nm); Sentinel-2 imagery has a spatial resolution of 10m, and its red band (664nm-670nm) and near-infrared band (835nm-842nm) settings are highly compatible with maize vegetation growth monitoring, effectively reflecting the macroscopic trends in vegetation cover and growth vitality. A 10% cloud cover screening standard was also followed during data acquisition to ensure a consistent quality benchmark between high- and low-resolution data, reducing sources of error in subsequent data fusion processing.

[0061] S12 Data Preprocessing

[0062] Geometric calibration was performed on high-resolution UAV remote sensing data. Five ground control points were selected, evenly distributed at the four corners and the center of the irrigation area to ensure coverage of different terrain features and vegetation distribution types. Calibration was completed using Pix4DMapper software. Based on the precise coordinate information of the ground control points, the software corrected the spatial position of the UAV image pixel by pixel, ultimately controlling the corrected planar error to within 0.5m. This error control standard was determined based on the actual needs of field boundary identification. A smaller planar error can avoid misjudgment of field boundaries due to spatial offset, providing a guarantee for the accurate construction of subsequent field-level analysis units.

[0063] Radiometric calibration and atmospheric correction were performed on low-resolution satellite data. ENVI 5.6 software was used for processing. The radiometric calibration process converted the raw digital quantization values ​​acquired by the satellite sensors into surface reflectance, eliminating systematic errors caused by differences in sensor response. The atmospheric correction process eliminated interference factors such as atmospheric scattering and absorption, making the reflectance values ​​of the satellite data more closely match the true reflectance characteristics of the Earth's surface.

[0064] During preprocessing, the coordinate system of all data is uniformly set to the WGS84 coordinate system. This coordinate system is a universal standard coordinate system for remote sensing data processing. A unified coordinate system can ensure that the spatial reference of high-resolution data and low-resolution data are completely consistent, avoiding spatial matching errors caused by coordinate deviations, and laying the foundation for subsequent data fusion and spatial correlation analysis.

[0065] S13 Data Fusion Processing

[0066] A weighted average fusion algorithm was used to fuse the preprocessed high-resolution and low-resolution data. The allocation of fusion weights fully considered the advantages of the two types of data: high-resolution UAV data has advantages in capturing spatial details and characterizing local features, which is crucial for field boundary identification and monitoring micro-changes in vegetation, so its weight was set to 0.7; low-resolution satellite data has the characteristics of large-scale continuous coverage and can reflect the macro-trend of vegetation growth at the regional scale, so its weight was set to 0.3.

[0067] This weighting ratio was verified and determined using synchronous observation data from 20 typical fields. Under this weighting combination, the correlation between the fused data and the measured vegetation parameters on the ground reached 0.92, with the error controlled within 4%. This maximized the preservation of spatial details from high-resolution data while fully integrating macroscopic trend information from low-resolution data, achieving complementary advantages between the two types of data.

[0068] During the fusion process, the spatial coordinates of the high-resolution UAV data were used as a reference, and bilinear interpolation was employed to resample the low-resolution satellite data to a spatial resolution of 0.1m. Bilinear interpolation ensures a smooth spatial transition of the data while achieving scale uniformity, avoiding significant abrupt changes in the data after resampling. This guarantees the spatial continuity and consistency of the fused data, providing high-quality data support for the subsequent construction of field-level analysis units and the correction of vegetation parameters.

[0069] S2-scale decoupling and vegetation parameter correction

[0070] This step uses the fused data output by S1 as the core support. Through precise identification of field boundaries, directional construction of field-level analysis units, and three-level collaborative correction of vegetation parameters, it specifically solves the inherent problem of mixed pixels in low-resolution remote sensing data, providing a spatially accurate and reliable basis for the subsequent generation of personalized growth parameters.

[0071] S21 Field Boundary Identification

[0072] Based on the high-resolution UAV remote sensing components in the fused data output by S1, the Canny edge detection algorithm was used to extract field boundaries. The algorithm uses a dual threshold for the UAV grayscale imagery: a low threshold of 50 and a high threshold of 150. This threshold combination was determined through a field ridge recognition experiment using 50 UAV images covering different terrain slopes and vegetation cover densities. The results showed that field ridges are generally less bright than crop areas in grayscale images. The low threshold of 50 can capture weak edge signals, while the high threshold of 150 can filter out interference noise such as vegetation textures. The combination of these two thresholds resulted in a stable field ridge recognition accuracy of over 98%.

[0073] Morphological processing algorithms are used to smoothly connect the extracted edges. Specifically, 3×3 structuring elements are used for dilation and erosion operations to remove isolated noise points and avoid boundary breaks caused by local image reflections or shadows. The final result is a complete field boundary vector file with a boundary accuracy controlled within 1 meter. This accuracy standard stems from the actual management needs of the fields; an error within 1 meter ensures that subsequent analysis units perfectly match the spatial extent of the actual fields, providing a reliable spatial reference for the accurate decomposition of mixed pixels.

[0074] S22 Field-level Analysis Unit Construction

[0075] Based on the field boundaries identified by S21, low-resolution satellite pixels in the S1 fused data were spatially decomposed and resampled. Using the segmentation tool of ArcGIS software, with the field boundary vector file as the clipping template, the satellite pixels were split into field-level sub-pixels along the field boundaries, ensuring that the spatial range of each sub-pixel was strictly limited to within a single field, thus completely breaking the limitation of traditional techniques that treat multiple fields as homogeneous units at the spatial level.

[0076] Resampling employs bilinear interpolation, which calculates the weighted average of vegetation parameters from four satellite pixels surrounding the sub-pixel to achieve a smooth parameter transition and avoid abrupt parameter changes between sub-pixels. Each field-level analysis unit is assigned unique spatial coordinates and attribute identifiers. The attribute information is inherited from the core parameters such as vegetation and spectrum corresponding to the S1 fused data, achieving precise mapping between satellite pixels and field entities. This gives each field a dedicated analysis carrier, laying the spatial foundation for subsequent personalized parameter calibration.

[0077] S23 Vegetation Parameter Correction

[0078] Based on the detailed spatial information provided by the high-resolution UAV components in the S1 fusion data, a three-level collaborative correction is implemented on the low-resolution satellite vegetation parameters used to drive assimilation, improving their reliability and consistency for field-level analysis from three levels: spatial, spectral, and temporal.

[0079] S231 Spatial Resolution Correction

[0080] This step aims to leverage high-resolution data to address the problem of low-resolution mixed pixels, assigning more accurate initial vegetation parameters to each field. Pure vegetation areas are identified based on high-resolution UAV data, with a vegetation coverage threshold set at 60%. Using the vegetation parameter values ​​of these pure vegetation areas as a benchmark, combined with the field boundaries identified in S21, the original low-resolution satellite mixed pixels are spatially decomposed. During decomposition, each mixed pixel is divided into several preliminary pseudo-field units based on the field boundaries. A spatially corrected vegetation parameter value (e.g., LAI) is assigned to each pseudo-field unit based on the area ratio and mean parameter value of pure vegetation pixels within each pseudo-field unit calculated from high-resolution data. This process essentially integrates high-resolution spatial details into the low-resolution data, providing a more accurate parameter base for subsequent construction of field-level analysis units. Field verification shows that the spatial distribution map of vegetation parameters obtained after this correction significantly improves the fit with the actual field conditions, reducing the error from 15% using traditional methods to less than 6%.

[0081] S232 Spectral Consistency Correction

[0082] Using spectral information from UAV data in the 2200nm band, a sensitive band for organic matter, a spectral correction relationship for satellite vegetation parameters was established. Thirty fields evenly distributed in the irrigation area were selected as samples. The spectral response values ​​of UAVs and satellites in the 2200nm band and the corresponding measured values ​​of vegetation parameters were collected simultaneously. A correction model was constructed through linear regression fitting: Satellite vegetation parameters = UAV vegetation parameters × spectral response coefficient + offset.

[0083] During the fitting process, the spectral response coefficient was determined to be 1.03, the offset was determined to be -0.02, the model fit goodness R² = 0.86, and the root mean square error was less than 0.03. This model can effectively eliminate the systematic errors caused by the differences in spectral response characteristics between the two types of sensors, ensuring that the vegetation parameters retrieved from high-resolution and low-resolution data are consistent in spectral characteristics, and providing a unified parameter benchmark for subsequent time-series data fusion and assimilation simulation.

[0084] S233 Time-Series Trend Alignment Correction

[0085] The key growth window for maize was selected from 20 days after sowing to the tasseling stage. This window was selected based on the growth and development pattern of maize: during this stage, maize transitions from vegetative growth to reproductive growth, the NDVI change rate is the fastest, and the growth differences are the most significant. This is the core stage that reflects the growth dynamics of the field.

[0086] The temporal vegetation index (NDVI) of both UAV and satellite data within this window is extracted, and the temporal slope of change for both (unit: / d) is calculated. A slope deviation threshold of 0.002 / d is set, determined using temporal growth monitoring data from 20 fields: deviations exceeding this value can lead to misjudgments of growth stages. If the deviation between the NDVI change slope of satellite data and UAV data exceeds 0.002 / d, the satellite temporal NDVI curve is linearly stretched to synchronize the growth trends of the two data types, avoiding assimilation errors introduced by temporal asynchrony.

[0087] S3 Personalized Growth Parameter Generation

[0088] This step uses the field-level analysis units and corrected vegetation parameters output by S2 as the core basis to extract the time-series vegetation dynamics and soil attribute characteristics unique to each field. Through targeted dynamic adjustment of parameters, personalized growth parameters adapted to the growth characteristics of each field are generated, breaking the limitations of the traditional model parameter setting uniformity. This allows the crop model to accurately capture the unique growth patterns of different fields, providing realistic parameter support for subsequent assimilation simulation.

[0089] S31 Feature Extraction

[0090] S311 First Feature Extraction

[0091] The early growth stage is defined as 1 to 20 days after sowing. This stage corresponds to the corn seedling stage and is a critical period for establishing crop growth rhythm. The rate of change of NDVI can directly reflect the initial growth status such as seedling speed and seedling vigor. Therefore, it is regarded as the primary characteristic for characterizing the vegetative growth potential of a field.

[0092] For each field-level analysis unit, time-series NDVI data were extracted from the S2-corrected vegetation parameters. The least squares method was used to calculate the linear regression slope of NDVI for that stage, yielding the NDVI change rate in daily increments. The calculation used the number of growing days as the x-axis and NDVI value as the y-axis, fitting a linear equation y=kx+b using conventional data processing software. The slope k represents the NDVI change rate. This extraction method effectively filters out interference from random data points, stably reflects the overall growth trend in the early stages of the field, and ensures the reliability of the features.

[0093] S312 Second Feature Extraction

[0094] Soil organic matter content was selected as the second characteristic because soil organic matter directly determines soil fertility level and water and fertilizer retention capacity. It is a core factor affecting the stability of nutrient supply during the reproductive growth stage of maize, and differences in its content will directly lead to different growth performance during the grain-filling period.

[0095] Construction (training) of a soil organic matter inversion model

[0096] Model Definition and Input / Output: A linear regression model is used as the soil organic matter (SOM) inversion model. The model input is the surface reflectance in the predefined organic matter-sensitive near-infrared band (2200nm) extracted using high-resolution UAV hyperspectral data. The model output is the predicted soil organic matter content. (Unit: percentage %). This setting is based on the physical mechanism that soil organic matter has a characteristic absorption valley in the 2200nm band, and its reflectance is usually negatively correlated with organic matter content.

[0097] Sample data preparation: Within the irrigation area, 50 fields with even spatial distribution and covered by different soil types (such as sandy soil, loam, clay, etc.) were selected as modeling samples.

[0098] Data Synchronization and Processing: For each sample field, perform the following synchronization operations:

[0099] Input data extraction: Based on the high-resolution UAV components in the S1 fusion data, the exposed soil area within the field (avoiding the vegetation canopy) was selected on the UAV imagery, and its average surface reflectance value in the 2200nm band was accurately calculated. .

[0100] Target data acquisition: In the same exposed soil area of ​​the corresponding field, a mixed soil sample of the top layer (0-20cm) was collected according to the standard five-point sampling method and sent to the laboratory. The actual content of soil organic matter was determined using the potassium dichromate external heating method (GB 9834-88). (Unit: %), used as the target value (true value) for model training.

[0101] Model fitting (training) process and parameters: Using the least squares method with 50 samples... As the independent variable, Using [variable name] as the dependent variable, a linear regression model is fitted. The fitting process involves minimizing the sum of squared prediction errors to solve for the model parameters. The final linear inversion model expression is:

[0102]

[0103] The key parameters of the model were determined through fitting: slope a = −0.05, intercept b = 3.0. During the fitting process, the independent variables... The value range was determined to be 0.1 to 0.3 based on statistical analysis of measured soil spectral data from the irrigation area. The empirical performance index of this model is: goodness of fit. =0.85, RMSE=0.15%.

[0104] S313 Model Application:

[0105] The trained linear inversion model was applied to all field-level analysis units across the entire irrigation district. For each unit, the 2200nm reflectance was extracted based on the hyperspectral information from the S1 fused data. Substitute into the model formula: =-0.05× +3.0, the estimated value of soil organic matter content for this field is calculated. (Unit: %), used as a second feature for subsequent personalized growth parameter adjustments.

[0106] S32 Parameter Adjustment and Generation

[0107] S321 Parameter Adjustment for the Vegetative Growth Stage

[0108] The baseline value for the NDVI change rate was set at 0.02 per day. This baseline value was obtained through statistics from 10 typical fields with different soil types and irrigation conditions in the irrigated area. It is widely representative and can reflect the average growth level of maize seedlings in the irrigated area.

[0109] Parameter adjustments were made based on the changes in the first characteristic: When the NDVI change rate exceeded the baseline value by 0.002 days per day, it indicated rapid crop emergence and vigorous growth, requiring less effective accumulated temperature; therefore, the vegetative growth stage parameter P1 (effective accumulated temperature from sowing to emergence) was reduced by 10℃·d. Conversely, when the NDVI change rate fell below the baseline value by 0.002 days per day, it indicated relatively slow crop growth, requiring more effective accumulated temperature to complete emergence; therefore, P1 was increased by 10℃·d. The adjustment range was determined through statistical analysis of continuous phenological observation data from 20 fields, ensuring that the parameter adjustments accurately matched the actual crop growth rhythm and that the vegetative growth process simulated by the model was consistent with the actual field conditions.

[0110] S322 Reproductive Growth Stage Parameter Adjustment

[0111] The baseline value for soil organic matter content is set at 1.2%. This baseline value is the average content obtained after statistical analysis of soil sample data from 100 representative plots in the irrigation area, and it can reflect the overall level of soil fertility in the irrigation area.

[0112] The parameters were adjusted based on the changes in the second characteristic: For every 0.2% increase in soil organic matter content above the baseline, soil fertility was considered sufficient, providing a more continuous and stable nutrient supply for maize reproductive growth. This allowed the maize to withstand a longer effective accumulated temperature during the grain-filling stage to accumulate more dry matter. Therefore, the reproductive growth stage parameter P5 (effective accumulated temperature during grain-filling) increased by 40℃·d. Conversely, for every 0.2% decrease in soil organic matter content below the baseline, soil fertility was insufficient, nutrient supply was limited, and the demand for effective accumulated temperature during grain-filling decreased. Therefore, P5 decreased by 40℃·d. This adjustment logic stemmed from the correlation analysis between soil organic matter and maize's accumulated temperature demand during grain-filling, with a correlation coefficient r=0.78, accurately reflecting the influence of soil fertility on reproductive growth.

[0113] S323 Personalized Parameter Generation

[0114] Based on the adjustment rules S321 and S322, the baseline growth parameters of the DSSAT model for each field-level analysis unit were individually corrected, generating personalized parameter sets containing core genetic parameters such as P1 and P5. Each parameter set was associated with the spatial identifier of the corresponding field-level analysis unit, ensuring accurate matching between parameters and fields. After validation in five independent fields, the personalized parameter calibration resulted in a phenological simulation error within two days, effectively improving the model's adaptability to field heterogeneity and laying a solid parameter foundation for the accuracy of subsequent assimilation simulations.

[0115] S4 Two-Level Cooperative Assimilation Execution

[0116] This step uses the field-level personalized growth parameters output by S3 as the core driver, combined with the high-precision vegetation parameters corrected by S2. Through the collaborative design of aggregated scale overall constraints and independent scale parallel updates, it ensures the consistency of regional growth trends while accurately preserving field heterogeneity. This solves the problem of insufficient simulation accuracy caused by scale mismatch in traditional assimilation technology, making the growth state simulation more in line with the actual production scenario of the irrigation area.

[0117] S41 Simulation State Set Initialization

[0118] Based on the personalized growth parameters of each field-level analysis unit generated by S3, the simulation state set of the DSSAT model is initialized. The model uses the CERES-Maize module of DSSAT version 4.8, and the simulation state set covers three core states: vegetation growth state (leaf biomass, leaf area index LAI), soil moisture state (volume water content of 0-40cm soil layer), and canopy temperature state.

[0119] Each state variable generates 100 model instances. The initial values ​​of the instances are set based on the statistical distribution of continuous ground observation data from 20 representative fields in the irrigation area: the initial soil moisture is set at 20%±2%, the initial leaf biomass is set at 50g / m²±5g / m², and the initial canopy temperature is set at 25℃±2℃. During the initialization process, real-time meteorological data (daily average temperature, precipitation, sunshine hours), basic soil properties (texture, bulk density, field water holding capacity), and routine management practices (sowing density of 6000 plants / mu, application of nitrogen fertilizer of 15kg / mu 30 days after sowing) are simultaneously input to ensure that the initial state closely matches the actual growth environment of the field, providing a reliable starting benchmark for subsequent assimilation and updating.

[0120] S42 Aggregation Scale Constraints

[0121] At an aggregated scale consisting of multiple field-level analysis units (corresponding to the original pixels of the low-resolution satellite in S1), the simulated states of all field-level analysis units within this scale are integrated, and the mean and variance of the vegetation parameter LAI are calculated to form an overall statistical characteristic. The satellite vegetation parameters corrected by S2 are used as the common observation constraint target at this aggregated scale. The core objective is to ensure that the overall statistical characteristics of the simulated vegetation parameters of all field-level analysis units are consistent with the constraint target.

[0122] The gradient descent method is used to adjust the simulation state of each field-level analysis unit, driving the integrated simulation state to approach the observation constraint target as a whole. The iteration step size is set to 0.01, which was determined through 10 sets of comparative simulation experiments with different step sizes, ensuring fitting accuracy while controlling computational costs. The maximum number of iterations is set to 50 to avoid computational redundancy caused by excessive iteration. This constraint process effectively anchors the overall growth trend of the region, prevents the simulation results of individual fields from deviating from objective laws, and sets reasonable boundaries for subsequent independent updates at the field level.

[0123] S43 Independent Scale Status Update

[0124] At the independent scale of each field-level analysis unit, based on its individual growth parameters, and under the overall trend constraint determined in S42, the simulation state is updated in parallel. The ensemble Kalman filter (EnKF) algorithm is used, and the field-level LAI observations retrieved by the UAV in S2 are used as key inputs and compared with the LAI values ​​simulated by the model.

[0125] The Kalman gain is used to calculate the correction weights of observed values ​​to simulated states, dynamically adjusting vegetation growth status, soil moisture status, and canopy temperature status: if the simulated LAI is lower than the observed value, the leaf biomass accumulation rate is appropriately increased; if the simulated soil moisture is lower than the reasonable range of actual measurements, the soil moisture recharge coefficient is adjusted; if the simulated canopy temperature deviates significantly from the observed value, the heat exchange parameters are optimized. Each field-level analysis unit is independently iterated and updated at a frequency consistent with the UAV data acquisition frequency (once every 7 days) to ensure timely absorption of the latest observation data, accurately depict the unique growth dynamics of each field, and prevent field heterogeneity from being masked by the overall trend.

[0126] S44 Assimilation Convergence Criterion

[0127] The assimilation convergence criteria are set as follows: assimilation iteration stops when the statistical uncertainty of at least one of the three states in the simulated state set—vegetation growth state, soil moisture state, and canopy temperature state—falls below a preset threshold. "Statistical uncertainty" is defined as the coefficient of variation (i.e., the ratio of standard deviation to mean) of each state variable within its set. This definition effectively measures the relative fluctuation of the simulated state values. The preset threshold is set to 5%, determined through a comparative assimilation experiment across 30 fields. When the coefficient of variation of any state variable is below 5%, it indicates that the simulated value for that state has stabilized, the simulation results of each instance within the set are highly consistent, and the fit with the observed data meets the accuracy requirements for field-level monitoring. If the coefficients of variation of all state variables are above 5%, iteration continues. The number of assimilation iterations throughout the entire growth period is controlled between 15 and 20 times to ensure sufficient simulation accuracy while avoiding excessive iteration that wastes computational resources, achieving a balance between accuracy and efficiency.

[0128] S5 Field-Level Yield and Differentiated Management Plan Generation

[0129] This step uses the updated vegetation growth status, soil moisture status, and canopy temperature status after S4 as the core basis. Through precise yield calculation, multi-dimensional growth stress diagnosis, targeted management plan formulation, and spatial yield distribution map drawing, it forms a field-level decision-making product that can be directly implemented. This breaks through the limitation of traditional monitoring that can only provide regional trends, and allows management measures to be accurately matched with the actual growth needs of each field.

[0130] S51 Field-Level Yield Calculation

[0131] Based on the set of simulated states that meet the convergence criteria of S4, the total biomass (in kg / ha) data of each field-level analysis unit was extracted, and the grain yield was calculated using the harvest index method. The harvest index was set to 0.45, a value determined by fitting measured data from 20 fields with different soil fertility, irrigation conditions, and variety types in irrigated areas. With total biomass as the independent variable and grain yield as the dependent variable, a linear regression analysis yielded a fitting equation with a goodness of fit R² = 0.88. The yield calculation followed the core logic of "grain yield (kg / ha) = total biomass (kg / ha) × harvest index". To facilitate field management, the calculation results were divided by 15 to convert to kg / mu (1 hectare = 15 mu). Outliers were automatically removed after calculation. The outlier threshold is set at either below 350 kg / mu or above 650 kg / mu. This range was determined by reviewing historical corn yield statistics from the irrigation area over the past 10 years, covering more than 99% of normal yield conditions. This effectively eliminates abnormal data caused by extreme weather, human error, and other factors, ensuring the reliability of the overall monitoring results.

[0132] S52 Growth Stress Diagnosis

[0133] Based on the updated soil moisture and canopy temperature status in S4, combined with the soil organic matter content and leaf area index (LAI) obtained in S3, growth stress diagnosis results for each field-level analysis unit are generated simultaneously. The stress types are clearly divided into three categories: water stress, fertility stress, and temperature stress, ensuring that the diagnosis results are accurate and distinguishable.

[0134] Water stress diagnosis: Severe water stress is defined as soil moisture content less than 12% and canopy temperature greater than 35℃; mild water stress is defined as soil moisture content between 12% and 15% and canopy temperature between 33℃ and 35℃. This threshold combination is derived from maize physiological experiment data. When soil moisture is below 12%, root water absorption efficiency decreases significantly. Coupled with canopy temperature exceeding 35℃, photosynthetic product accumulation is hindered, which is consistent with the water-temperature synergistic response mechanism of maize growth.

[0135] Fertility stress diagnosis: Soil organic matter content less than 1.0% and LAI less than 3.2 are defined as severe fertility stress; soil organic matter content between 1.0% and 1.2% and LAI between 3.2 and 3.5 are defined as mild fertility stress. The soil organic matter threshold is determined with reference to the irrigation area soil fertility grading standard, and the LAI threshold is determined in conjunction with the suitable leaf area range for maize during the key growth stages. The combination of these two methods can accurately reflect the degree to which soil fertility restricts vegetation growth.

[0136] Temperature stress diagnosis: A canopy temperature greater than 35°C and soil moisture content greater than 15% are defined as temperature stress. This standard is specifically designed for high-temperature heat damage scenarios, where soil moisture supply is sufficient but high temperature still inhibits photosynthetic enzyme activity, thus clearly distinguishing it from water stress and avoiding diagnostic confusion.

[0137] All diagnostic indicator thresholds were comprehensively calibrated in conjunction with the physiological needs of maize growth and the actual climate and soil conditions in the irrigation area. After on-site verification in 10 typical stressed fields, the diagnostic accuracy rate remained stable at over 90%, ensuring a reliable basis for the formulation of subsequent management plans.

[0138] S53 Differentiated Management Solution Generation

[0139] A pre-built agronomic knowledge base was constructed. The management measures in the knowledge base integrate years of agricultural production practice experience in the irrigation area and 30 sets of field comparative test data, covering key management links such as irrigation, fertilization, and variety selection. Each measure is clearly marked with the applicable scenario, implementation time, specific dosage and operation points to ensure that farmers can directly refer to and implement it.

[0140] like Figure 2 As shown, based on the combination of yield assessment results and growth stress diagnosis results for each field-level analysis unit, corresponding differentiated management schemes are precisely mapped from the agronomic knowledge base:

[0141] For high-yield fields (yield ≥ 550 kg / mu) and without stress: it is recommended to maintain the existing management plan, namely, apply nitrogen fertilizer at 15 kg / mu and irrigate at 20 m³ / mu. Reduce the amount of nitrogen fertilizer applied in the next season by 5% to avoid excessive fertilization that could cause soil nutrient imbalance and waste of resources.

[0142] For medium-yield fields (500 kg / mu ≤ yield < 550 kg / mu) with mild water stress: It is recommended to supplement irrigation with 10 m³ / mu 3 days before the tasseling stage. This irrigation can accurately meet the water requirements of corn during the grain filling period and improve the efficiency of photosynthetic products transport to the grains.

[0143] For medium-yield fields (500 kg / mu ≤ yield < 550 kg / mu) and mild fertility stress: It is recommended to apply 200 kg / mu of organic fertilizer before the next sowing season to gradually improve the physical and chemical properties of the soil, enhance the soil's ability to retain fertilizer and water, and provide a continuous and stable supply of nutrients for vegetation growth.

[0144] For low-yield fields (yield <500 kg / mu) and severe water stress: it is recommended to irrigate immediately with 20 m³ / mu, and then irrigate once every 7 days during the subsequent growing season. By increasing the irrigation frequency, water stress can be quickly relieved and further yield decline can be avoided.

[0145] For low-yield fields (yield <500 kg / mu) and severe fertility stress: it is recommended to apply 20 kg / mu of nitrogen fertilizer and 300 kg / mu of organic fertilizer in the next season. At the same time, select maize varieties with a shorter growth period of 5 days to address the problem of low accumulated temperature utilization efficiency caused by insufficient soil fertility, so as to achieve a steady increase in yield.

[0146] The generated management plan is spatially correlated with the corresponding field-level analysis unit to ensure that the management suggestions for each field can be accurately located, forming a decision product that can be directly applied.

[0147] S54 field-level yield distribution map generated

[0148] The yield value of each field-level analysis unit is precisely correlated with its spatial coordinates, and a field-level yield distribution map is generated using ArcGIS software. The distribution map uses a hierarchical color scheme to divide the yield into three levels: high, medium, and low, corresponding to dark green, light green, and yellow, respectively. The level division thresholds are strictly consistent with the yield ranges in the S53 differentiated management scheme, facilitating a clear and intuitive correspondence between field yield levels and appropriate management measures.

[0149] The spatial resolution of the yield distribution map is set to 0.1m, maintaining consistency with the resolution of the S1 fused data and the S2 field-level analysis units to ensure accurate matching between field boundaries, yield values, and spatial locations. The distribution map clearly presents the yield difference patterns between fields, providing an intuitive spatial reference for optimizing the layout of agricultural production in the irrigation area and accurately allocating resources, thus contributing to the improvement of overall production efficiency.

[0150] S6 Visualization and Interactive Query

[0151] This step uses the field-level yield distribution map, differentiated management plan, and related monitoring data generated by S5 as core inputs. Through deep data correlation, lightweight GIS platform construction, and multi-dimensional interactive function development, it realizes the integrated presentation of monitoring results and management suggestions, solving the pain points of traditional monitoring results being scattered, single in display, and inconvenient to query, making field-level precision management decisions more intuitive, easy to operate, and implementable.

[0152] S61 Data Association Binding

[0153] Using spatial database technology, the field-level yield data, growth stress diagnosis results, and differentiated management schemes output by S5 are deeply correlated and bound with the field spatial boundary data identified by S2, thus constructing a unified and logically rigorous attribute data table.

[0154] The data table fields are designed according to practical requirements, and the core information includes field ID, WGS84 coordinate system spatial coordinates, yield value, soil moisture content, canopy temperature, soil organic matter content, stress type, management measures, implementation time, and expected effects. The data type (e.g., numeric, character, date) and format of each field are strictly standardized to ensure the efficiency and consistency of data storage, retrieval, and access. The association process uses the field ID as the unique core association key to achieve a one-to-one accurate correspondence between spatial location data and attribute monitoring data, completely solving the problems of traditional scattered data storage and chaotic associations, and providing stable data support for subsequent visualization and interactive queries.

[0155] S62GIS Platform Visualization

[0156] A lightweight WebGIS platform was developed using the OpenLayers 7.0 framework. This framework is highly compatible and can be stably adapted to mainstream browsers such as Chrome 90 and above, Firefox 88 and above, and Edge 90 and above. It supports access from multiple devices such as computers, tablets, and mobile phones, meeting the usage needs of different scenarios such as fields and offices.

[0157] The platform features diverse visualization layers, including field-level yield distribution heatmaps, growth stress type classification maps, and thematic maps of differentiated management schemes. Users can freely switch between these layers via the top layer control bar. The yield heatmap uses a red-yellow-green gradient color system to intuitively distinguish between high and low yields; the stress classification map assigns a unique identifier color to each stress type, ensuring clear and easily identifiable boundaries; and the thematic maps of management schemes use an "icon + text summary" format to quickly label core management measures. All color schemes and identifier designs have undergone user experience testing to ensure visual clarity and high recognizability.

[0158] Clicking on any field on the map instantly brings up a 3D information card, displaying the core information of the field in layers: basic data (yield values, spatial coordinates), environmental parameters (soil moisture content, canopy temperature, soil organic matter content), diagnostic results (stress type and degree), and management recommendations (specific measures, implementation time, and key points of operation). The information is clearly hierarchical, making it easy for users to quickly grasp the key content.

[0159] S63 interactive query function implementation

[0160] The platform has three core interactive query functions built-in, and the operation process is simple and intuitive, requiring no complex technical background to get started:

[0161] Spatial query function: Users can click on a single field with the left mouse button, or long-press the mouse and drag to select any area. The platform will display the complete information of the corresponding field in real time on the information card or side list. The query response time is controlled within 1 second to meet the needs of rapid positioning.

[0162] Conditional filtering function: The platform has a multi-dimensional filtering panel on the left, which includes filter items such as yield range (e.g., <500kg / mu, 500kg / mu-550kg / mu, ≥550kg / mu), stress type (water stress, fertility stress, temperature stress, no stress), and management measure type (irrigation adjustment, fertilization optimization, variety change). After the user selects or enters the conditions, the map will automatically highlight the matching fields to help quickly filter the target field group.

[0163] Batch export function: After selecting the target area, users can click the "Export" button at the top of the platform to export the field-level yield map (supporting PNG and JPG formats, with selectable resolutions of 300dpi / 600dpi) or field management list (Excel format) for that area with one click. The exported management list contains all the core data and management details of the field, which can be directly imported into irrigation district water and fertilizer integration systems, smart agriculture management platforms, and other equipment to achieve seamless integration of monitoring results and actual production operations.

[0164] All functionalities are designed around the actual management needs of the irrigation district, taking into account both professionalism and ease of use. They not only meet the in-depth analysis needs of agricultural technicians, but also facilitate farmers to quickly obtain field management solutions, effectively improving the implementation efficiency of field-level precision management.

[0165] The implementation principle of the remote sensing-DSSAT coupled method for monitoring maize yield in irrigated areas in this application is as follows: This method couples high- and low-resolution remote sensing data with a DSSAT crop model and innovatively introduces a scale decomposition strategy. First, it uses high-resolution data to identify field boundaries, and then uses this to spatially decompose and resample low-resolution mixed pixels to construct analysis units aligned with real fields. This directly solves the core problem of mixed pixels in traditional methods, which cannot achieve precise management due to the presence of multiple heterogeneous fields within a pixel. Then, personalized model parameters are generated based on the early growth dynamics and soil properties of each field unit, and a two-level collaborative assimilation framework is used to achieve a precise characterization of crop model simulation from regional homogenization to field heterogeneity. Finally, based on the assimilated and optimized state, field-level yield is calculated simultaneously, and stress types such as water and fertility are diagnosed, thereby mapping differentiated irrigation, fertilization, and other agronomic management schemes. This method not only enables precise monitoring of field-level yield and growth status, but more importantly, it outputs decision-making schemes with clear spatial locations that are strictly matched with the actual conditions of each field. This solves the dilemma of "knowing the regional trend but being unable to locate specific problem fields," and truly supports the direct implementation of differentiated management measures such as variable irrigation and precision fertilization, effectively improving the utilization efficiency and management precision of water and fertilizer resources in irrigation areas.

[0166] This application also discloses a maize yield monitoring system for irrigation areas based on remote sensing-DSSAT coupling. This system is specifically designed for accurate monitoring and differentiated management of maize yield at the field level. Through the coordinated operation of five functional modules, it overcomes core problems in traditional technologies such as the mismatch between remote sensing pixels and field scale, and interference from mixed pixels. The specific implementation is as follows:

[0167] The data acquisition and fusion module is responsible for acquiring and fusing high-resolution and low-resolution remote sensing data. High-resolution data was acquired using a DJI Matrice 300RTK drone equipped with an MS600 hyperspectral sensor, covering a wavelength range of 400nm-2500nm with a spatial resolution of 0.1m. Low-resolution data was obtained from Landsat-8 / 9 and Sentinel-2 satellite imagery, with spatial resolutions of 30m and 10m, respectively. A 10% cloud cover threshold was used during acquisition to ensure a good signal-to-noise ratio. In the preprocessing stage, drone data underwent geometric calibration to ensure a planar error ≤0.5m, and satellite data underwent radiometric calibration and atmospheric correction. All data were unified to the WGS84 coordinate system. A weighted average fusion algorithm was used, with a weight of 0.7 for high-resolution data and 0.3 for low-resolution data. Validated in 20 fields, the fused data showed a correlation of 0.92 with measured vegetation parameters and an error ≤4%.

[0168] The scale decoupling module works in conjunction with the acquisition and fusion module to perform field boundary identification and vegetation parameter correction based on high-resolution data. The Canny edge detection algorithm (low threshold 50, high threshold 150) is used to extract field boundaries, and after morphological processing, the boundary accuracy is ≤1m. Based on the boundaries, satellite pixels are decomposed into field-level analysis units resampled to 0.1m. Vegetation parameter correction is performed in three levels: pure vegetation areas are identified with 60% vegetation coverage to complete mixed pixel spatial correction; a linear regression correction model is constructed based on synchronous sample data using 2200nm spectral information to achieve spectral consistency correction; and during the window from 20 days after sowing to the tasseling stage, the temporal NDVI trends of the two types of data are aligned, and linear stretching is performed when the slope deviation exceeds 0.002 / d.

[0169] The dynamic calibration module is integrated with the scale decoupling module to generate personalized growth parameters for each field-level analysis unit. The NDVI change rate from 1 to 20 days after sowing is extracted as the first feature, and the slope is fitted using the least squares method. Utilizing UAV hyperspectral information, based on 2200nm band reflectance and simultaneously measured soil sample data, a linear regression inversion model is constructed using the least squares method to obtain the soil organic matter content of each field as the second feature. With an NDVI change rate of 0.02 / d and a soil organic matter content of 1.2% as benchmarks, a fluctuation of 0.002 / d in the former corresponds to the vegetative growth parameter P1±10℃·d, and a fluctuation of 0.2% in the latter corresponds to the reproductive growth parameter P5±40℃·d. The phenological simulation error was verified to be ≤2 days.

[0170] The collaborative assimilation module is connected to both the acquisition and fusion module and the dynamic calibration module, performing two-level collaborative assimilation. The DSSAT model simulation state set (including vegetation growth, soil moisture, and canopy temperature) is initialized based on personalized parameters. 100 instances are generated for each state, with initial values ​​referencing measured statistics from 20 fields (soil moisture 20%±2%, leaf biomass 50g / m²±5g / m², canopy temperature 25℃±2℃). At the aggregate scale, the simulated field states are integrated, and the overall trend is driven to approximate the observed values ​​using gradient descent (step size 0.01, maximum iterations 50) constrained by the corrected vegetation parameters. At the independent scale, an ensemble Kalman filter algorithm is used to assimilate UAV LAI observations every 7 days, updating the state of each field in parallel. A convergence condition is set for statistical uncertainty (measured by the coefficient of variation) below 5%, with 15-20 iterations throughout the entire growth period.

[0171] The decision output module interfaces with the collaborative assimilation module to generate field-level decision products. Yield is calculated using a harvest index of 0.45, removing outliers below 350 kg / mu and above 650 kg / mu. Water, fertility, and temperature stress are diagnosed based on soil moisture, canopy temperature, soil organic matter, and LAI, with a diagnostic accuracy ≥90%. Differential management plans are mapped from the agronomic knowledge base based on yield level and stress type, simultaneously generating field-level yield distribution maps (0.1m resolution, graded color scheme). A WebGIS platform developed using OpenLayers 7.0 supports spatial binding of yield maps and management plans, visualization, and interactive queries. It can batch export yield maps in PNG / JPG format and management lists in Excel format, adapting to multi-device access.

[0172] This system has been validated in 200 fields across 10 irrigation districts. The error in field-level yield monitoring is 5%-9%. After implementing differentiated solutions, low-yield fields saw a 10%-12% increase in yield, providing reliable technical support for precision agricultural management.

[0173] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for monitoring maize yield in irrigated areas based on remote sensing-DSSAT coupling, characterized in that, include: Acquire and fuse high-resolution remote sensing data with low-resolution remote sensing data; Based on the high-resolution remote sensing data, field boundaries are identified. Based on the field boundaries, pixels of low-resolution remote sensing data are spatially decomposed and resampled to form field-level analysis units. The vegetation parameters retrieved from low-resolution remote sensing data are corrected using the high-resolution remote sensing data. For each field-level analysis unit, personalized growth parameters are generated based on its temporal vegetation dynamics and soil properties. Perform two-level collaborative assimilation: at the aggregate scale consisting of multiple field-level analysis units, use the corrected vegetation parameters to constrain the overall trend of the simulated state of each field-level analysis unit; At the independent scale of each field-level analysis unit, based on its personalized growth parameters and while satisfying the overall trend constraints, the vegetation growth status, soil moisture status, and canopy temperature status of each field-level analysis unit are updated in parallel. Based on the updated vegetation growth status, soil moisture status, and canopy temperature status, the yield of each field-level analysis unit is calculated, and a field-level yield distribution map and differentiated management plan are generated.

2. The method according to claim 1, characterized in that, The correction of vegetation parameters retrieved from low-resolution remote sensing data using the high-resolution remote sensing data includes: Based on the high-resolution remote sensing data, identify pure vegetation areas that meet the preset vegetation coverage threshold. Based on the vegetation parameters of the pure vegetation area, spatial decomposition and resampling correction are performed on the mixed pixels of the low-resolution remote sensing data; Using the spectral information of the high-resolution remote sensing data in the preset organic matter sensitive band, the vegetation parameters retrieved from the low-resolution remote sensing data are corrected for spectral consistency. Based on the temporal vegetation indices of high-resolution and low-resolution remote sensing data within a preset critical growth period window, the vegetation parameters retrieved from the low-resolution remote sensing data are corrected for temporal trend alignment.

3. The method according to claim 2, characterized in that, The generation of personalized growth parameters includes: Extract the rate of change of vegetation index for each field-level analysis unit during a preset early growth stage as the first feature; The soil organic matter content of each field-level analysis unit is obtained as a second feature; Based on the change in the first feature, the adjustment of parameters for the vegetative growth stage is determined; Based on the amount of change in the second feature, adjustments to the parameters of the reproductive growth stage are determined; Based on the aforementioned adjustments, personalized growth parameters are generated for each field-level analysis unit.

4. The method according to claim 3, characterized in that, The execution of the two-level collaborative assimilation includes: Based on the personalized growth parameters generated for each field-level analysis unit, initialize its corresponding set of simulated states; At the aggregation scale, the simulated states of the multiple field-level analysis units are integrated, and the corrected vegetation parameters are used as a common observation constraint to drive the integrated simulated state to approach the common observation constraint as a whole. At each unit-independent scale, each of the field-level analysis units performs parallel updates of its simulated state within its own set of simulated states and under the overall trend constraints determined by the approximation, in order to optimize its vegetation growth state, soil moisture state, and canopy temperature state, respectively.

5. The method according to claim 4, characterized in that, The parallel execution of the simulation state update also includes setting a criterion for assimilation convergence. The criterion is that the statistical uncertainty of at least one of the vegetation growth state, soil moisture state, and canopy temperature state in the simulation state set is lower than a preset threshold.

6. The method according to claim 5, characterized in that, After the determination condition is met, the following is also included: Based on the set of simulated states that meet the aforementioned criteria, grain yield assessment results and growth stress diagnosis results are generated synchronously for each field-level analysis unit. Based on the combination of the grain yield assessment results and the growth stress diagnosis results, a corresponding differentiated management plan is mapped from the preset agronomic knowledge base; The differentiated management schemes, spatially associated with each field-level analysis unit, are output to form a visually presentable decision-making product.

7. The method according to claim 6, characterized in that, Also includes: Spatially associate and bind the field-level yield distribution map with the differentiated management scheme; The associated data is visualized and interactively queried through a pre-set geographic information system platform.

8. The method according to claim 3, characterized in that, The process of obtaining the soil organic matter content of each field-level analysis unit includes: Using the hyperspectral information in the high-resolution remote sensing data, the reflectance of the preset organic matter-sensitive near-infrared band is extracted; The soil organic matter content of each field-level analysis unit was obtained based on the reflectance inversion.

9. The method according to claim 4, characterized in that, The use of corrected vegetation parameters as common observation constraints includes: The corrected vegetation parameters are used as the common constraint target for simulating vegetation parameters in all field-level analysis units within the corresponding aggregation scale; By adjusting the simulation state of the multiple field-level analysis units, the overall statistical characteristics of the simulated vegetation parameters of each field-level analysis unit are made to meet the common constraint objective.

10. A maize yield monitoring system for irrigated areas based on remote sensing-DSSAT coupling, characterized in that, To implement the method of any one of claims 1-9, comprising: The acquisition and fusion module is used to acquire and fuse high-resolution remote sensing data with low-resolution remote sensing data; The scale decoupling module, connected to the acquisition and fusion module, is used to identify field boundaries based on the high-resolution remote sensing data, spatially decompose and resample the pixels of the low-resolution remote sensing data according to the field boundaries to form field-level analysis units, and use the high-resolution remote sensing data to correct the vegetation parameters retrieved from the low-resolution remote sensing data. The dynamic calibration module, connected to the scale decoupling module, is used to generate personalized growth parameters for each field-level analysis unit based on its temporal vegetation dynamics and soil properties. The collaborative assimilation module, connected to the acquisition and fusion module and the dynamic calibration module, is used to constrain the overall trend of the simulation state of each field-level analysis unit at the aggregate scale consisting of multiple field-level analysis units using the corrected vegetation parameters; at the independent scale of each field-level analysis unit, based on its personalized growth parameters and under the constraint of the overall trend, it updates the vegetation growth state, soil moisture state and canopy temperature state of each field-level analysis unit in parallel. The decision output module, connected to the collaborative assimilation module, is used to calculate the yield of each field-level analysis unit and generate a field-level yield distribution map and differentiated management plan based on the updated vegetation growth status, soil moisture status and canopy temperature status.

Images