A high-precision evapotranspiration fusion correction method and system based on multi-source data

Abstract

The application relates to the technical field of hydrological remote sensing and water resource management, and discloses a high-precision evapotranspiration fusion correction method and system based on multi-source data, which comprises the following steps: acquiring remote sensing and reanalysis evapotranspiration products and unifying spatial and temporal resolutions and grids; calculating reference evapotranspiration based on meteorological station data, extracting multi-source product values at the station, and constructing a support vector regression model to realize nonlinear fusion; carrying out basin partitioning by combining hydrological data such as flow stations, rainfall stations and reservoir pits, and predicting storage capacity by using a random forest; calculating evapotranspiration of each partitioning area by water balance, and obtaining a deviation correction coefficient to correct the fusion result. Under the dual actions of station constraints and water balance, the method carries out fusion correction on evapotranspiration products, generates high-precision evapotranspiration fusion products, and effectively improves the evapotranspiration estimation precision, physical consistency and response ability to human activities.

Description

Technical Field

[0001] This application relates to the fields of hydrological remote sensing and water resources management technology, and in particular to a high-precision evapotranspiration fusion correction method and system based on multi-source data. Background Technology

[0002] Evapotranspiration (ET) is the combined process by which surface water enters the atmosphere through soil evaporation and plant transpiration, and it is a key component of the land surface water cycle and energy balance. Accurate estimation of ET is not only fundamental to understanding global and regional water cycle mechanisms and responding to climate change, but also a core basis for agricultural irrigation management, drought disaster monitoring, watershed water resource planning, and ecosystem assessment. However, because ET involves complex soil-vegetation-atmosphere continuum (SPAC) interactions, its direct and accurate observation and estimation have always been a major challenge in the fields of hydrology and remote sensing applications.

[0003] Traditionally, evapotranspiration (ET) data acquisition has relied on site observations, such as eddy covariance systems (EC) and lysimeters. While these methods offer high accuracy at the point scale, their high instrument costs, complex maintenance requirements, and limited spatial representativeness hinder their widespread application at the regional scale, particularly in small and medium-sized watersheds with complex topography and highly heterogeneous underlying surfaces. To overcome this limitation, with the development of remote sensing technology and land surface process models, various global-scale evapotranspiration products have emerged. These primarily include remote sensing inversion products based on physical models (such as MOD16 and GLEAM) and land surface model products based on data assimilation and reanalysis (such as GLDAS and ERA5-Land). These products offer the advantage of wide-area, long-term series coverage, providing a valuable data foundation for global and regional water cycle research.

[0004] Despite significant progress in existing evapotranspiration products, their application in small-scale areas (such as small and medium-sized watersheds and irrigation districts) still faces significant limitations and uncertainties, mainly in the following aspects: First, there are significant systematic differences and uncertainties among multi-source products. Different ET products differ in their algorithm mechanisms, input data, and parameterization schemes, leading to substantial discrepancies in their estimation results in terms of magnitude and spatiotemporal dynamics. For example, remote sensing products (such as MOD16) are susceptible to cloud cover and vegetation dynamics, and may underestimate evapotranspiration in humid areas; while reanalysis products (such as ERA5-Land, GLDAS), although possessing better physical consistency, generally have coarser spatial resolution (typically greater than 0.1°), making it difficult to capture detailed evapotranspiration features caused by complex land use (such as urban and farmland mosaics), and they also have insufficient response in areas significantly affected by human activities such as irrigation.

[0005] Second, existing fusion methods have shortcomings in small-scale applications. To reduce the uncertainty of individual products, researchers have developed multi-source flux tower fusion methods. Early methods, such as arithmetic mean, Bayesian model averaging (BMA), and triangular hat method (TC), while integrating information from various products to some extent, are mostly linear or based on specific statistical assumptions, making it difficult to capture complex nonlinear relationships between products. In recent years, machine learning methods (such as random forests, support vector regression, and neural networks) have been introduced into fusion research due to their powerful nonlinear fitting capabilities, significantly improving fitting accuracy. However, these purely data-driven methods heavily rely on training samples (usually sparse flux tower data) and lack explicit physical process constraints, resulting in limited extrapolation capabilities and insufficient stability in regions with scarce samples or under extreme climatic conditions. The generated fusion results may also violate the basic laws of the water cycle.

[0006] Third, the application of physical constraints for regional water balance faces challenges. Incorporating watershed water balance (ET) into the watershed water balance framework for verification or constraint is an effective way to improve its physical rationality. Traditionally, researchers have used precipitation, runoff, and water storage changes retrieved from GRACE satellite gravity data to estimate regional ET. However, the spatial resolution of GRACE data is too low (approximately 0.5°), completely failing to meet the needs of small watershed studies. Simultaneously, in areas with intense human activity, processes such as irrigation diversion, reservoir scheduling, and inter-basin water transfer have significantly altered the natural water cycle, and most ET products and verification methods do not fully consider these factors, leading to a serious discrepancy between the estimated results and the actual water consumption in the region. Furthermore, the storage capacity of widely distributed small reservoirs, ponds, and other water bodies within the region is a key factor in water balance calculations, but due to a lack of monitoring data, their dynamic changes are difficult to quantify accurately, becoming a major bottleneck for high-precision water balance calculations.

[0007] In summary, there is an urgent need to develop a small-scale evapotranspiration estimation method that can balance high spatiotemporal resolution, nonlinear fusion capabilities, and consistency with physical processes. An ideal method should effectively integrate complementary information from multi-source ET products, incorporate atmospheric driving constraints provided by dense meteorological station observations, and closely integrate regional water balance physical constraints that consider human activities. This would generate high-precision, high-reliability evapotranspiration products that reflect both fine spatial patterns and the actual watershed water budget. Summary of the Invention

[0008] The purpose of this application is to provide a high-precision evapotranspiration fusion correction method and system based on multi-source data, so as to solve the technical problems of insufficient resolution, weak physical consistency and limited response to human activities in existing evapotranspiration products in small and medium watersheds.

[0009] According to a first aspect of the embodiments of this application, a high-precision evapotranspiration fusion correction method based on multi-source data is provided, comprising the following steps: Download the required remote sensing and atmospheric reanalysis evapotranspiration products, preprocess the evapotranspiration products to ensure consistent spatiotemporal resolution and perform raster alignment; Collect meteorological data from weather stations and calculate reference evapotranspiration; The preprocessed evapotranspiration products are used to extract the evapotranspiration values ​​corresponding to each evapotranspiration product at the meteorological station location. The evapotranspiration values ​​are used as input features and the reference evapotranspiration values ​​are used as target variables to construct a support vector regression model and perform nonlinear fusion to generate evapotranspiration fusion products. Hydrological data of the study area are collected, and the study area is divided into watersheds according to flow stations. The hydrological data includes flow data from flow stations, precipitation data from rain gauges, non-agricultural water use data, water diversion data from water diversion projects, and monitoring data of reservoirs and ponds. Using the monitoring data of the reservoirs and ponds, a random forest prediction model for the water storage capacity of typical small reservoirs was established to predict the water storage variable data of all reservoirs and ponds in the study area. Based on the hydrological data and storage variable data, the regional water balance is calculated, and combined with the station reference evapotranspiration values, the water balance evapotranspiration data of each watershed zone is calculated. The deviation correction coefficients for each watershed zone are calculated using the water balance evapotranspiration data and the evapotranspiration fusion product. The deviation correction coefficient is used to correct the deviation of the evaporation fusion product to obtain the final evaporation fusion corrected product.

[0010] Optionally, the meteorological data can be used as input to calculate a reference evapotranspiration value using the FAO-modified Penman–Monteith formula.

[0011] Optionally, when performing nonlinear fusion, a radial basis function (RBF) is used as the kernel function type to capture the high-dimensional relationship between nonlinear features. The optimal hyperparameter combination is determined by grid search and five-fold cross-validation, and the hyperparameter combination includes a penalty coefficient and kernel parameters.

[0012] Optionally, using the monitoring data of the reservoirs and ponds, a random forest prediction model for the water storage capacity of typical small reservoirs is established to predict the water storage variable data of all reservoirs and ponds in the study area, including: Collect monitoring data of the reservoirs and ponds, including daily water level, reservoir capacity curve, catchment area, reservoir area and total capacity of typical small reservoirs; The water storage capacity of the typical small reservoir is calculated based on the daily water level and reservoir capacity curves. The precipitation data collected by the rain gauges and the evapotranspiration fusion product are used to calculate the unit rainfall and unit evapotranspiration data of reservoirs and ponds. Using the catchment area, total storage capacity, reservoir area, unit rainfall, and unit evapotranspiration of the typical small reservoir as input features, and the water storage capacity of the typical small reservoir as a label, a random forest prediction model is established based on random forest RF machine learning. The random forest prediction model described above is used to predict the water storage capacity of all small reservoirs and ponds. Statistics on the storage capacity of known large and medium-sized reservoirs and predicted small reservoirs and ponds in each watershed are compiled according to watershed division.

[0013] Optionally, based on the hydrological data and storage variable data, regional water balance is calculated, and combined with station reference evapotranspiration values, water balance evapotranspiration data for each watershed zone is calculated, including: Based on the flow data, precipitation data, non-agricultural water use data, and water transfer data in the hydrological data, the inflow, outflow, unit precipitation, water transfer, and non-agricultural water use of each watershed zone are calculated through allocation. Based on the calculated inflow, outflow, unit precipitation, water transfer, non-agricultural water use, and storage variable data of each watershed sub-region, the regional water balance is calculated to obtain the evapotranspiration of each watershed sub-region with complete data. Based on the water balance equivalence method, it is assumed that the evapotranspiration ratio calculated based on water balance between adjacent watersheds is consistent with the evapotranspiration ratio based on station reference. Through equivalent derivation, the water balance evapotranspiration data of all watershed partitions are obtained based on water balance calculation.

[0014] Optionally, the deviation correction coefficients for each watershed zone are calculated using the water balance evapotranspiration data and the evapotranspiration fusion product, including: Calculate the evapotranspiration value of the evapotranspiration fusion product in each watershed region; Evapotranspiration is calculated based on the evapotranspiration data of water balance in each watershed and the evapotranspiration fusion product. Calculate the ratio of the evapotranspiration of the water balance evapotranspiration data to the evapotranspiration of the evapotranspiration fusion product for each watershed partition. This ratio is the deviation correction coefficient for that watershed partition.

[0015] Optionally, the deviation correction coefficient is used to correct the deviation of the evaporation fusion product to obtain the final evaporation fusion corrected product, including: The evapotranspiration product of each watershed zone is multiplied by the corresponding deviation correction coefficient to obtain the final evapotranspiration product after fusion correction.

[0016] According to a second aspect of the embodiments of this application, a high-precision evapotranspiration fusion correction system based on multi-source data is provided, comprising: The preprocessing module is used to download the required evapotranspiration products from remote sensing and atmospheric reanalysis, preprocess the evapotranspiration products to ensure consistent spatiotemporal resolution and perform raster alignment. The data collection and calculation module is used to collect meteorological data from meteorological stations and calculate reference evapotranspiration. The fusion module is used to extract the evapotranspiration values ​​corresponding to each evapotranspiration product at the meteorological station location using the preprocessed evapotranspiration products, and to construct a support vector regression model using the evapotranspiration values ​​as input features and the reference evapotranspiration values ​​as target variables, and to perform nonlinear fusion to generate evapotranspiration fusion products. The zoning module is used to collect hydrological data of the study area and divide the study area into watersheds according to flow stations. The hydrological data includes flow data from flow stations, precipitation data from rain gauges, non-agricultural water use data, water diversion data from water diversion projects, and monitoring data from reservoirs and ponds. The prediction module is used to establish a random forest prediction model for the water storage capacity of typical small reservoirs using the monitoring data of the reservoirs and ponds, and to predict the water storage variable data of all reservoirs and ponds in the study area. The calculation module is used to calculate the regional water balance based on the hydrological data and storage variable data, and to calculate the water balance evapotranspiration data of each watershed zone by combining the station reference evapotranspiration values. The coefficient calculation module is used to calculate the deviation correction coefficient for each watershed zone using the water balance evapotranspiration data and the evapotranspiration fusion product. The correction module is used to correct the deviation of the evaporation fusion product using the deviation correction coefficient, so as to obtain the final evaporation product after fusion correction.

[0017] According to a third aspect of the embodiments of this application, an electronic device is provided, comprising: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors perform the method as described in the first aspect.

[0018] According to a fourth aspect of the embodiments of this application, a computer-readable storage medium is provided that stores computer instructions thereon, which, when executed by a processor, implement the steps of the method as described in the first aspect.

[0019] The technical solutions provided by the embodiments of this application may include the following beneficial effects: As can be seen from the above embodiments, this application achieves unified preprocessing and grid alignment of multi-source evapotranspiration products, nonlinear fusion based on support vector regression of meteorological station reference evapotranspiration, and combines water balance calculation of watershed zones with random forest water storage prediction. Furthermore, it corrects the fusion results by constructing a partition bias correction coefficient through water balance evapotranspiration. This overcomes the problems of insufficient resolution of single data sources, poor extrapolation ability of pure data-driven models, and lack of spatial detail and difficulty in reflecting the impact of human activities in traditional water balance methods in existing technologies. Thus, it achieves accurate evapotranspiration estimation that takes into account both high spatiotemporal resolution and physical consistency in small-scale areas, significantly improving the accuracy, stability and responsiveness to actual hydrological processes of the results, and providing reliable data support for the refined management and scheduling of watershed water resources.

[0020] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0021] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0022] Figure 1 This is a flowchart illustrating the small-scale multi-source evapotranspiration fusion method of the present invention based on dual constraints of meteorological and water balance, according to an exemplary embodiment.

[0023] Figure 2 It is calculated on the user site according to an exemplary embodiment. A comparison of evapotranspiration at meteorological stations on an 8-day scale with four different types of evapotranspiration.

[0024] Figure 3 This is the accuracy of the training set (left) and validation set (right) of the evapotranspiration SVR fusion model illustrated according to an exemplary embodiment; Figure 4 This is an example diagram of a partial study area showing the annual unit evapotranspiration of SVR_ET after ET fusion, according to an exemplary embodiment.

[0025] Figure 5 The diagram illustrates, according to an exemplary embodiment, the accuracy of a random forest model for predicting the water storage capacity of small reservoirs and ponds during the training period (left) and the accuracy during the validation period (right), respectively.

[0026] Figure 6 This is an example diagram of a partial study area showing the annual unit evapotranspiration of SVR_ET adjusted by BAF according to an exemplary embodiment.

[0027] Figure 7This is a comparison of the accuracy of the original evaporation and the evaporation after fusion correction, as shown in an exemplary embodiment.

[0028] Figure 8 This is a block diagram illustrating a high-precision evapotranspiration fusion correction system based on multi-source data, according to an exemplary embodiment.

[0029] Figure 9 This is a structural diagram of an electronic device according to an exemplary embodiment. Detailed Implementation

[0030] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of systems and methods consistent with some aspects of this application.

[0031] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used herein are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0032] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0033] Explanation of terms: Typical small reservoirs: Individual small reservoirs with water level monitoring and reservoir capacity curves that are evenly distributed within the study area.

[0034] Large and medium-sized reservoirs: These are large and medium-sized reservoirs with detailed water level monitoring.

[0035] Watershed zoning: Closed catchment units are defined based on flow stations, and water volume changes can be determined based on flow station data.

[0036] Deviation correction factor: represents the deviation between the target product and the actual product, and is used to calibrate the target product so that its total amount matches the actual product.

[0037] Figure 1 This is a flowchart illustrating a high-precision evapotranspiration fusion correction method based on multi-source data, according to an exemplary embodiment. Figure 1 As shown, the method may include the following steps: S1: Download the required remote sensing and atmospheric reanalysis evapotranspiration products, preprocess the evapotranspiration products to ensure consistent spatiotemporal resolution and perform raster alignment; Specifically, in this embodiment, the evapotranspiration products from remote sensing and atmospheric reanalysis are selected from MOD16, GLDAS, GLEAM, and ERA5-Land. To address the localized voids in MOD16, bilinear interpolation is used to fill them, ensuring spatial continuity. For the other three types of products, MOD16 is used as a unified benchmark. First, an 8-day cumulative processing is performed on a temporal scale. Then, spatial resampling is used to unify the resolution to 500 m, followed by raster alignment. Finally, all data are cropped according to the study area to obtain an evapotranspiration dataset with consistent spatiotemporal resolution for subsequent fusion analysis. However, this invention is not limited to these four types of remote sensing and atmospheric reanalysis evapotranspiration data.

[0038] Four mainstream evapotranspiration data sources were used: MODIS remote sensing evapotranspiration product MOD16 (500m, 8 days), GLDAS NOAH model land surface model product (0.25°, 3 hours), reanalysis data product ERA5-Land (0.1°, 1 hour), and micrometeorological model remote sensing product GLEAM (0.1°, 1 day). Local voids in MOD16 were filled using bilinear interpolation. The other three ET products were accumulated at 8-day resolution based on the spatiotemporal resolution of MOD16, then resampled to 500m, and finally raster-aligned with the MOD16 raster image. The processed image was cropped to the study area to obtain preprocessed ET data with uniform spatiotemporal resolution for subsequent fusion.

[0039] The above processing flow is adopted because different data sources have significant differences in spatiotemporal resolution and data structure, and direct fusion would introduce large errors. By using MOD16 as the benchmark for unified processing, we can fully utilize its high spatial resolution advantage, and ensure the comparability of multi-source data at the same scale through temporal accumulation and spatial resampling. At the same time, interpolation imputation avoids the interference of missing values ​​on model training. This design effectively improves the consistency and integrity of the data, provides reliable input for subsequent machine learning fusion, and thus improves the accuracy and stability of evapotranspiration estimation results.

[0040] S2: Collect meteorological data from weather stations and calculate reference evapotranspiration; Specifically, the meteorological data comes from the National Meteorological Information Center (https: / / data.cma.cn / ), and consists of daily meteorological data from six stations. The daily meteorological data includes average relative humidity, sunshine hours, average temperature, maximum temperature, minimum temperature, and 2m average wind speed. Using this daily meteorological data as input, the reference evapotranspiration value is calculated using the FAO-corrected Penman–Monteith formula. The FAO-corrected Penman–Monteith formula is: (1) in, For reference evaporation rate (mm / day); The slope of the saturated water vapor pressure curve ( ); Net radiation to crop surface ( ); Soil heat flux ( (0 can be taken on a day-scale). The daily average temperature (°C); Wind speed at a height of 2 m ( ); The saturated vapor pressure is (kPa). This represents the actual water vapor pressure (kPa). Specific parameter calculations are shown in Table 1. Table 1. Calculation table of parameters for the Penman–Monteith formula

[0041] Saturated vapor pressure Calculate and average the values ​​based on the highest and lowest temperatures. Actual vapor pressure Based on relative humidity calculations, only the daily average relative humidity is considered. ,use and Approximation. Slope of the saturated water vapor pressure curve. according to The net radiation is obtained from the formula for calculating T. Net shortwave radiation and net longwave radiation The difference consists of two components, both of which can be estimated using empirical methods. This represents the crop albedo (0.23 is used as a reference for grassland). The sunshine duration n is estimated using the Angström formula, where... , These are empirical coefficients (FAO recommends 0.25 and 0.50 respectively), where N is the astronomical probability of sunshine hours. extraterrestrial radiation, It can be obtained from latitude and solar eclipse angle using astronomical formulas provided by the FAO. Net longwave radiation It is then calculated using the Stefan-Boltzmann formula, which is corrected based on the average temperature, actual water vapor pressure, and cloud cover coefficient.

[0042] By substituting the daily average relative humidity, sunshine hours, average temperature, maximum temperature, minimum temperature, and 2 m wind speed into the above formulas, the parameters can be obtained daily and then substituted into the Penman–Monteith formula for calculation. .

[0043] Figure 2 It uses the calculation on the site. A comparison of evapotranspiration at 8-day scales with four different meteorological stations is presented. The results show that all products effectively reflect the intra-annual variation of evapotranspiration, exhibiting an overall seasonal trend of increasing with rising temperature and precipitation. However, differences exist in numerical amplitude and temporal response among the different products, revealing significant systematic bias. Because the evapotranspiration at this station cannot represent accurate evapotranspiration information, joint analysis with regional evapotranspiration calculated based on water balance is necessary, with emphasis on spatial information references at the station level. Overall, GLEAM, GLDAS, MOD16, and ERA5-Land all demonstrate good applicability in the study area and can serve as important data sources for subsequent evapotranspiration fusion correction.

[0044] The reason for using daily data from national meteorological stations and calculating reference evapotranspiration based on the FAO-modified Penman–Monteith formula is that this method comprehensively considers key meteorological factors such as radiation, air temperature, humidity, and wind speed, possessing a clear physical mechanism and international applicability. Using data from multiple stations effectively reflects regional climate differences, enhancing the representativeness of the results. This design not only overcomes the problem of insufficient accuracy of single meteorological elements or empirical models but also provides a highly reliable "benchmark truth" for subsequent evapotranspiration fusion, thereby improving the model training accuracy and the physical rationality of the results.

[0045] S3: Extract the evapotranspiration values ​​corresponding to each evapotranspiration product at the meteorological station location using the preprocessed evapotranspiration products, and construct a support vector regression (SVR) model using the evapotranspiration values ​​as input features and the reference evapotranspiration values ​​as target variables, and perform nonlinear fusion to generate evapotranspiration fusion products; Specifically, in the model construction process, the 8-day evapotranspiration sequences of four evapotranspiration products are used as input variables, and the reference evapotranspiration calculated by the Penman-Monteith formula for each site is used as the target output. For nonlinear fusion, a radial basis function (RBF) is used as the kernel function type to capture the high-dimensional relationships between nonlinear features. The optimal hyperparameter combination is determined through grid search and 5-fold cross-validation, and this combination includes the penalty coefficient C and the kernel parameter γ. The dataset is divided into training and validation sets in a 6:4 ratio, and finally, the fused evapotranspiration product SVR_ET is generated.

[0046] After determining the Support Vector Regression (SVR) model structure, this invention constructs an evapotranspiration fusion correction model using multi-source evapotranspiration products as input and site-measured reference evapotranspiration (ET0) as constraint variables. The input feature matrix consists of 8-day evapotranspiration values ​​for four types of products, namely: (2) The output target is the reference evapotranspiration value for the corresponding time period. .

[0047] During the model training phase, the optimal approximation of the nonlinear mapping relationship is achieved by minimizing the balance between prediction error and model complexity. Its prediction function can be expressed as: (3) in, To predict evapotranspiration, For Lagrange multipliers, This is the kernel function.

[0048] Figure 3 The left and right images show the accuracy of the evapotranspiration SVR fusion model. SVR demonstrates high accuracy and stability in multi-source evapotranspiration fusion correction. The training set results show that the model exhibits high fitting ability; the validation set results also show good generalization performance.

[0049] Figure 4 This is a partial example map of the annual evapotranspiration per unit area of ​​SVR_ET after ET fusion. The spatial distribution characteristics of evapotranspiration SVR_ET after fusion differ significantly from those of topographic and hydrological zoning, and are closer to the comprehensive spatial distribution characteristics of multi-source evapotranspiration products, and are greatly affected by low-resolution products.

[0050] After SVR fusion, the model effectively integrates complementary information from multiple data sources, reduces random and systematic errors, and makes the estimated values ​​closer to the measured reference in both magnitude and time series. Overall, the stability and consistency of SVR fusion products have been significantly enhanced across different sites and climate zones, enabling them to more accurately reflect the spatiotemporal variations of regional evapotranspiration and providing more reliable data support for subsequent water balance analysis and water resource management.

[0051] The reason for extracting multi-source evapotranspiration values ​​from meteorological stations and constructing a support vector regression model using reference evapotranspiration as the target is that different evapotranspiration products exhibit systematic biases and their interrelationships possess significant nonlinear characteristics, making direct weighting insufficient to fully exploit their complementary information. The SVR model employing the RBF kernel function effectively characterizes high-dimensional nonlinear relationships, and parameter optimization through grid search and cross-validation enhances the model's generalization ability. This design not only overcomes the limited accuracy of traditional linear fusion methods but also maintains stability even with a small sample size, thereby generating more accurate and robust fused evapotranspiration products.

[0052] S4: Collect hydrological data of the study area and divide the study area into watersheds according to flow stations. The hydrological data includes flow data from flow stations, precipitation data from rain gauges, non-agricultural water use data, water diversion data from water diversion projects, and monitoring data of reservoirs and ponds. Specifically, each zone is divided into flow stations that control the inflow and outflow of water in that area. Each zone is a completely closed catchment area, and water volume changes between zones are recorded by the flow data of the flow stations controlling those zones.

[0053] The hydrological data includes flow data from flow stations, precipitation data from rain gauge stations, non-agricultural water use data (composed of industrial and domestic water use data from the water resources bulletin), water diversion data from water transfer projects, and monitoring data from reservoirs and ponds (including daily water level, reservoir capacity curves, catchment area, reservoir area, and total capacity). Rainfall information is obtained from measured daily precipitation data at 52 rain gauge stations in and around the study area. Voronoi polygons were constructed for each rain gauge station, and spatial intersections were performed with the study areas to calculate area weights and average precipitation per unit area for each area. Total water consumption statistics are primarily derived from the water resources bulletin for the study area, including water consumption for forestry, animal husbandry, fisheries, industry, urban public services, residential use, and ecological environment from 2018 to 2022, collectively constituting the non-agricultural water consumption of the study area. Water diversion information is obtained from inter-county water diversion tables or water diversion volume tables of water diversion projects in the study area, yielding the water balance inflow and outflow volumes. The watershed is divided into hydrological zones based on flow stations. Each zone is a completely closed catchment area, and water volume changes between zones are recorded by controlling the flow data of the flow stations in those zones.

[0054] The reason for integrating multi-source hydrological data and dividing the watershed based on flow stations is that single data sources cannot fully reflect the regional water balance process. By incorporating information on precipitation, water use, and water transfer, and combining this with Thiessen polygons to calculate spatial precipitation distribution, the spatial representativeness of water elements can be improved. Dividing the watershed into closed units based on flow stations allows for accurate constraints on water quantity changes. This design effectively enhances the completeness and accuracy of water balance calculations, providing reliable regional-scale constraints for evapotranspiration estimation.

[0055] S5: Using the aforementioned hydrological data, establish a random forest prediction model for the water storage capacity of typical small reservoirs to predict the water storage variable data of all reservoirs and ponds within the study area; this step includes the following sub-steps: S51: Collect daily water level, reservoir capacity curves, catchment area, reservoir area and total capacity of typical small reservoirs in the study area; Specifically, the daily water level, reservoir capacity curves, catchment area, reservoir area and total capacity of 12 large and medium-sized reservoirs and 47 typical small reservoirs within the hydrological zone were collected.

[0056] S52: Calculate the water storage capacity of the typical small reservoir based on the daily water level and reservoir capacity curves; Specifically, based on the daily water level and storage capacity curves of 12 large and medium-sized reservoirs and 47 typical small reservoirs within the collected hydrological zones, the daily water storage of each reservoir is calculated.

[0057] S53: Based on the precipitation data collected by the rain gauges and the evapotranspiration fusion product, the unit rainfall and unit evapotranspiration data of reservoirs and ponds are statistically analyzed. Specifically, using the precipitation data collected by the rain gauges and the daily-scale resolution ERA5-Land evapotranspiration data as data sources, the unit precipitation and unit evapotranspiration of all reservoirs and ponds are statistically analyzed.

[0058] S54: Using the catchment area, total storage capacity, reservoir area, unit rainfall and unit evapotranspiration of the typical small reservoir as input features, and the water storage capacity of the typical small reservoir as a label, a random forest prediction model is established based on random forest RF machine learning. Specifically, Figure 5This invention utilizes information from typical small reservoirs to predict the water storage capacity of small reservoirs and ponds using a random forest model. The model's training period accuracy (left) and validation period accuracy (right) are presented. This invention collects daily water levels, storage capacity curves, catchment areas, reservoir areas, and total storage capacity data from 12 large and medium-sized reservoirs and 47 typical small reservoirs within a hydrological zone. For the more numerous small water bodies (including small reservoirs and ponds) within the region, a generalized model capable of estimating water storage capacity from available feature variables is established. Based on observational data from a typical sample library, the relationships between water storage capacity and catchment areas, total storage capacity, reservoir area, and meteorological and topographic indicators such as unit rainfall and unit evapotranspiration are constructed. Considering the complex characteristics of nonlinearity, multi-scale, and multivariate interactions among the variables, traditional linear regression is insufficient to effectively characterize these relationships. Therefore, this embodiment employs a random forest (RF) machine learning model for predicting the water storage capacity of small water bodies.

[0059] S55: Predict the water storage capacity of all small reservoirs and ponds using the aforementioned random forest prediction model; Specifically, the trained random forest prediction model is used to predict the water storage capacity of small reservoirs and ponds in the study area that have no measured water level, based on their catchment area, total storage capacity, reservoir area, unit rainfall, and unit evapotranspiration.

[0060] S56: Statistics on the storage capacity of known large and medium-sized reservoirs and predicted small reservoirs and ponds in each region according to watershed division; Specifically, the statistical results of the annual water storage changes in each zone of the study area are shown in Table 2: Table 2. Annual Water Storage Changes in Each River Basin Sub-region (Unit: 10,000 cubic meters)

[0061] The storage variable data of all reservoirs and ponds were collected, and the daily storage variable was calculated. This resulted in daily storage variable data for 12 large and medium-sized reservoirs, 417 small reservoirs, and 2332 ponds within the study area. The total storage capacity of these three types of water bodies accounted for 62.7%, 31.4%, and 5.9%, respectively. The storage variable data for large and medium-sized reservoirs were mainly supported by measured data, while the results for small reservoirs and ponds were obtained through random forest model inversion. The errors in both methods were controlled within a reasonable range, ensuring the reliability and representativeness of the regional water balance calculation results. Then, based on the geographical location of each reservoir and pond, the total storage variable data for each watershed sub-region was statistically analyzed.

[0062] Random forest (RF) models are ensemble learning methods composed of numerous decision trees. They utilize a bootstrap method for sample resampling and randomly select feature subsets during the splitting of each tree, thereby enhancing the model's generalization ability and stability. Their main advantages include: capturing nonlinear relationships between variables; insensitivity to the scale and distribution of multi-source input variables; strong noise resistance; and unbiased estimation of model performance using out-of-bag (OOB) error. Furthermore, RF can output feature importance, helping to identify key factors influencing water storage dynamics. Given the complex relationships between topographic, meteorological, and structural variables involved in this study, and the lack of a unified water level-storage capacity relationship across different small water bodies, the RF model can more effectively characterize the spatial heterogeneity of water storage, making it a suitable method for predicting water storage in small water bodies.

[0063] S6: Based on the hydrological data and storage variable data, calculate the regional water balance, and combine it with the station reference evapotranspiration values ​​to calculate the water balance evapotranspiration data for each watershed zone; this step includes the following sub-steps: S61: Based on the flow data, precipitation data, non-agricultural water use data and water transfer data in the hydrological data, the inflow, outflow, unit precipitation, water transfer and non-agricultural water use of each watershed zone are calculated through allocation. Specifically, based on the flow data, precipitation data, non-agricultural water use data, and water transfer data in the hydrological data, the average unit precipitation for each region is calculated from the precipitation data; the non-agricultural water use for each zone is calculated from the water consumption data of forestry, animal husbandry, fishery, industry, urban public use, residential use, and ecological environment in the water resources bulletin; the water transfer volume for each zone is calculated from the inter-county water transfer information table or the water transfer volume table of water transfer projects; and the inflow and outflow of each zone are calculated from the measured flow data of the flow stations in the control watershed zone.

[0064] S62: Based on the calculated inflow, outflow, unit precipitation, water transfer, non-agricultural water use, and storage variable data of each watershed sub-region, calculate the regional water balance to obtain the evapotranspiration of each watershed sub-region with complete data. Specifically, Table 3 shows the annual unit ET (WBET) calculation results for each zone of the study area based on water balance calculation.

[0065] Table 3. Annual water balance calculations (unit: mm) for each watershed sub-region

[0066] The traditional water balance formula (Formula 4) provides a relatively simple understanding of regional water volume changes, using only precipitation, runoff, and total water storage data to estimate ET.

[0067] (4) Where ET is the actual evapotranspiration, P is the precipitation, and Q is the runoff. This represents the change in surface and groundwater reserves.

[0068] To more effectively characterize local water volume changes, water storage variables are added. This invention incorporates the water volume changes of water bodies such as reservoirs and ponds as... At the same time, the specific changes in the water balance process have been improved by adding variables such as water diversion, water regulation, and water consumption. This has enriched the content of the water balance formula and made the general water balance formula more consistent with the local water cycle process, thereby improving the calculation accuracy of regional water balance ET.

[0069] The regional water balance formula can be expressed as: (5) (6) in: This refers to precipitation. , These refer to the inflow and outflow of water in the basin, respectively. , These refer to the water volume transferred into and out of the basin, respectively. For water consumption in the basin, To store variables for the watershed, Regional evapotranspiration Indicates industrial water consumption. This indicates the amount of water consumed in daily life.

[0070] The evapotranspiration of each watershed region, calculated based on water balance, is obtained by substituting the inflow, outflow, unit precipitation, water transfer, and non-agricultural water use (comprising industrial and domestic water use) into the formula, along with the reservoir and pond storage variables of each region.

[0071] Traditional water balance formulas have a relatively simplistic understanding of regional water volume changes, relying solely on precipitation, runoff, and total water storage data to estimate evapotranspiration. Most studies use GRACE data to calculate water storage changes, but this has significant limitations due to its large scale, particularly in small-scale regional studies. This study incorporates water volume changes from reservoirs and ponds as ∆W. Furthermore, it refines the specific changes in the water balance process by including variables related to water diversion, regulation, and consumption, thus enriching the water balance formula and making it more consistent with local water cycle processes, thereby improving the accuracy of regional water balance evapotranspiration calculations.

[0072] S63: Based on the water balance equivalent method, it is assumed that the evapotranspiration ratio calculated based on water balance between adjacent watersheds is consistent with the evapotranspiration ratio based on station reference. The water balance evapotranspiration data of all watershed partitions are obtained through equivalent derivation. Specifically, for situations where accurate estimation of regional water balance evapotranspiration is impossible due to monitoring gaps and the lack of typical large and medium-sized reservoirs, the Water Balance Equivalence Method (WABE) is used to derive the water balance evapotranspiration of the region in question by combining evapotranspiration calculated based on accurate regional water balance (WBET) with evapotranspiration calculated based on meteorological stations (FCET). The WABE method assumes that the ratio of ET in watershed A to ET in watershed B (both adjacent and sharing the same hydroclimate with no substantial difference in major ET drivers such as net radiation) calculated using watershed water balance is equivalent to the ratio calculated using FCET. The WBET of the region in question can be calculated based on the known WBET and the FCET of the two regions. The calculation formula is as follows: (7) (8) in Evapotranspiration is calculated for the water balance of the region to be determined. Evapotranspiration calculated for a known regional water balance; Evapotranspiration for the region in question, calculated based on the site. Evapotranspiration calculated based on the site for a known area.

[0073] Then, evapotranspiration data based on water balance calculations for all regions and all times are obtained (WBET).

[0074] S7: Calculate the deviation correction coefficient for each watershed zone using the aforementioned water balance evapotranspiration data and evapotranspiration fusion product; this step includes the following sub-steps: S71: Calculate the evapotranspiration value of the evapotranspiration fusion product in each watershed zone; Specifically, the unit evapotranspiration value of the evapotranspiration fusion product is statistically analyzed in each watershed zone.

[0075] S72: The evapotranspiration values ​​of the water balance evapotranspiration data and the evapotranspiration fusion product of each watershed are statistically analyzed by year to calculate the annual evapotranspiration. Specifically, the unit evapotranspiration value of the evapotranspiration fusion product in each zone and the unit evapotranspiration value based on water balance in each zone are statistically analyzed and summarized into the total annual unit evapotranspiration value.

[0076] S73: Calculate the ratio of the annual evapotranspiration data of the water balance to the evapotranspiration of the evapotranspiration fusion product for each watershed partition, which is the deviation correction coefficient corresponding to that watershed partition. Specifically, the annual regional total evapotranspiration is used to calculate BAF to avoid interference from extreme rainfall years on normal years.

[0077] (9) Where i represents the corresponding year. This represents the deviation correction factor for the region in the corresponding year. This represents the unit evapotranspiration for the corresponding year in this region, calculated based on water balance. The unit evapotranspiration of the fusion product for the corresponding year in this region is given. The deviation correction factor (BAF) for each zone is calculated and shown in Table 4. Table 4. Annual Deviation Correction Factors (BAF) for Each Watershed Sub-region

[0078] It should be noted that this embodiment uses a year as the time scale, but it can also use other time scales, such as months or multi-year averages.

[0079] S8: The deviation correction coefficient is used to correct the deviation of the evaporation fusion product to obtain the final evaporation product after fusion correction.

[0080] Specifically, deviation correction is performed based on the evaporation fusion product and the calculated annual deviation correction coefficient for each region. After deviation correction, the evaporation fusion product after fusion correction (hereinafter referred to as BAET) can be obtained.

[0081] (10) Where i represents the corresponding year, and j represents the corresponding 8D resolution. This represents the deviation correction factor for the region in the corresponding year. This represents the evapotranspiration of the region for the corresponding year, fused using site-constrained machine learning. This represents the evapotranspiration value of the final product for the corresponding year and 8d resolution in this region.

[0082] Figure 6 This is a partial example map of the annual unit evapotranspiration of SVR_ET after BAF adjustment. The corrected BAET has the same regional evapotranspiration value as WBET, which preserves the regional spatial distribution of SVR_ET and meets the specific water volume change characteristics of the local hydrology.

[0083] The evapotranspiration product of each watershed zone is multiplied by the corresponding deviation correction coefficient to obtain the final evapotranspiration product after fusion correction.

[0084] Figure 7This comparison examines the accuracy of evapotranspiration between the original data and the evapotranspiration data after fusion correction. The SVR_ET data, after site fusion, exhibits the highest accuracy across all metrics, though its accuracy may be excessively high due to inherent data limitations. The BAET data, after correction for physical constraints in the water balance region, shows improvements not only in the correlation coefficient (r), RMSE, and MAE compared to the original products, but also in r (which increased by 0.05 from the average correlation coefficient of 0.87 to 0.92) and RMSE (which decreased by 1.36 mm / 8d from the average of 8.71 mm / 8d to 5.35 mm / 8d). Furthermore, its evapotranspiration data accurately reflects the specific water body evapotranspiration variations under local hydrological conditions. Improvements are observed at both the site and regional levels.

[0085] As can be seen from the above embodiments, this application achieves unified preprocessing and grid alignment of multi-source evapotranspiration products, nonlinear fusion based on support vector regression of meteorological station reference evapotranspiration, and combines water balance calculation of watershed zones with random forest water storage prediction. Furthermore, it corrects the fusion results by constructing a partition bias correction coefficient through water balance evapotranspiration. This overcomes the problems of insufficient resolution of single data sources, poor extrapolation ability of pure data-driven models, and lack of spatial detail and difficulty in reflecting the impact of human activities in traditional water balance methods in existing technologies. Thus, it achieves accurate evapotranspiration estimation that takes into account both high spatiotemporal resolution and physical consistency in small-scale areas, significantly improving the accuracy, stability and responsiveness to actual hydrological processes of the results, and providing reliable data support for the refined management and scheduling of watershed water resources.

[0086] The final product BAET obtained by this application after site and water balance evapotranspiration constraints not only meets the spatial evapotranspiration distribution but also the specific changes in local water bodies, providing a more refined and effective long-term estimate of small-area evapotranspiration that is more in line with the regional water cycle process.

[0087] Corresponding to the aforementioned embodiments of the high-precision evapotranspiration fusion correction method based on multi-source data, this application also provides embodiments of a high-precision evapotranspiration fusion correction system based on multi-source data.

[0088] Figure 8 This is a block diagram illustrating a high-precision evapotranspiration fusion correction system based on multi-source data, according to an exemplary embodiment. (Refer to...) Figure 8 The system includes: Preprocessing module 1 is used to download the required evapotranspiration products from remote sensing and atmospheric reanalysis, preprocess the evapotranspiration products to ensure consistent spatiotemporal resolution and perform raster alignment; The data collection and calculation module 2 is used to collect meteorological data from meteorological stations and calculate reference evapotranspiration. Fusion module 3 is used to extract the evapotranspiration values ​​corresponding to each evapotranspiration product at the meteorological station location using the preprocessed evapotranspiration products, and to construct a support vector regression model using the evapotranspiration values ​​as input features and the reference evapotranspiration values ​​as target variables, and to perform nonlinear fusion to generate evapotranspiration fusion products. The zoning module 4 is used to collect hydrological data of the study area and divide the study area into watersheds according to the flow station. The hydrological data includes flow data from the flow station, precipitation data from the rain gauge station, non-agricultural water use data, water diversion data from water diversion projects, and monitoring data of reservoirs and ponds. Prediction module 5 is used to establish a random forest prediction model for the water storage capacity of typical small reservoirs using the monitoring data of the reservoirs and ponds, and to predict the water storage variable data of all reservoirs and ponds in the study area. Calculation module 6 is used to calculate the regional water balance based on the hydrological data and storage variable data, and to calculate the water balance evapotranspiration data of each watershed zone by combining the station reference evapotranspiration values. The coefficient calculation module 7 is used to calculate the deviation correction coefficient for each watershed zone using the water balance evapotranspiration data and the evapotranspiration fusion product. The correction module 8 is used to correct the deviation of the evaporation fusion product using the deviation correction coefficient, so as to obtain the final evaporation product after fusion correction.

[0089] Regarding the system in the above embodiments, the specific ways in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0090] For the system embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0091] Accordingly, this application also provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; and, when the one or more programs are executed by the one or more processors, causing the one or more processors to implement the high-precision evapotranspiration fusion correction method based on multi-source data as described above. Figure 9The diagram shown is a hardware structure diagram of any device with data processing capabilities, which is part of a high-precision evapotranspiration fusion correction system based on multi-source data provided in an embodiment of the present invention. (Except for...) Figure 9 In addition to the processor, memory, DMA controller, disk, and non-volatile memory shown, any data processing device in which the system is located in the embodiment may also include other hardware depending on the actual function of the data processing device, which will not be described in detail here.

[0092] Accordingly, this application also provides a computer-readable storage medium storing computer instructions thereon, which, when executed by a processor, implement the high-precision evaporation fusion correction method based on multi-source data as described above. The computer-readable storage medium can be an internal storage unit of any data-processing device as described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium can also be an external storage device, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc., equipped on the device. Furthermore, the computer-readable storage medium can include both internal storage units of any data-processing device and external storage devices. The computer-readable storage medium is used to store the computer program and other programs and data required by the data-processing device, and can also be used to temporarily store data that has been output or will be output.

[0093] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only.

[0094] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A high-precision evapotranspiration fusion correction method based on multi-source data, characterized in that, include: Download the required remote sensing and atmospheric reanalysis evapotranspiration products, preprocess the evapotranspiration products to ensure consistent spatiotemporal resolution and perform raster alignment; Collect meteorological data from weather stations and calculate reference evapotranspiration; The preprocessed evapotranspiration products are used to extract the evapotranspiration values ​​corresponding to each evapotranspiration product at the meteorological station location. The evapotranspiration values ​​are used as input features and the reference evapotranspiration is used as the target variable to construct a support vector regression model and perform nonlinear fusion to generate evapotranspiration fusion products. Hydrological data of the study area are collected, and the study area is divided into watersheds according to flow stations. The hydrological data includes flow data from flow stations, precipitation data from rain gauges, non-agricultural water use data, water diversion data from water diversion projects, and monitoring data of reservoirs and ponds. Using the monitoring data of the reservoirs and ponds, a random forest prediction model for the water storage capacity of typical small reservoirs was established to predict the water storage variable data of all reservoirs and ponds in the study area. Based on the hydrological data and storage variable data, the regional water balance is calculated, and combined with the station reference evapotranspiration values, the water balance evapotranspiration data of each watershed zone is calculated. The deviation correction coefficients for each watershed zone are calculated using the water balance evapotranspiration data and the evapotranspiration fusion product. The deviation correction coefficient is used to correct the deviation of the evaporation fusion product to obtain the final evaporation fusion corrected product.

2. The high-precision evapotranspiration fusion correction method based on multi-source data as described in claim 1, characterized in that, Using the meteorological data as input, the reference evapotranspiration value is calculated using the FAO-modified Penman–Monteith formula.

3. The high-precision evapotranspiration fusion correction method based on multi-source data as described in claim 1, characterized in that, When performing nonlinear fusion, the radial basis function (RBF) is used as the kernel function type to capture the high-dimensional relationship between nonlinear features. The optimal hyperparameter combination is determined by grid search and five-fold cross-validation. The hyperparameter combination includes the penalty coefficient and the kernel parameter.

4. The high-precision evapotranspiration fusion correction method based on multi-source data as described in claim 1, characterized in that, Using the monitoring data of the reservoirs and ponds, a random forest prediction model for the water storage capacity of typical small reservoirs was established to predict the water storage variables of all reservoirs and ponds in the study area, including: Collect monitoring data of the reservoirs and ponds, including daily water level, reservoir capacity curve, catchment area, reservoir area and total capacity of typical small reservoirs; The water storage capacity of the typical small reservoir is calculated based on the daily water level and reservoir capacity curves. The precipitation data collected by the rain gauges and the evapotranspiration fusion product are used to calculate the unit rainfall and unit evapotranspiration data of reservoirs and ponds. Using the catchment area, total storage capacity, reservoir area, unit rainfall, and unit evapotranspiration of the typical small reservoir as input features, and the water storage capacity of the typical small reservoir as a label, a random forest prediction model is established based on random forest RF machine learning. The random forest prediction model described above is used to predict the water storage capacity of all small reservoirs and ponds. Statistics on the storage capacity of known large and medium-sized reservoirs and predicted small reservoirs and ponds in each watershed are compiled according to watershed division.

5. The high-precision evapotranspiration fusion correction method based on multi-source data as described in claim 1, characterized in that, Based on the aforementioned hydrological data and storage variable data, regional water balance is calculated. Combined with station reference evapotranspiration values, water balance evapotranspiration data for each watershed region is calculated, including: Based on the flow data, precipitation data, non-agricultural water use data, and water transfer data in the hydrological data, the inflow, outflow, unit precipitation, water transfer, and non-agricultural water use of each watershed zone are calculated through allocation. Based on the calculated inflow, outflow, unit precipitation, water transfer, non-agricultural water use, and storage variable data of each watershed sub-region, the regional water balance is calculated to obtain the evapotranspiration of each watershed sub-region with complete data. Based on the water balance equivalence method, it is assumed that the evapotranspiration ratio calculated based on water balance between adjacent watersheds is consistent with the evapotranspiration ratio based on station reference. Through equivalent derivation, the water balance evapotranspiration data of all watershed partitions are obtained based on water balance calculation.

6. The high-precision evapotranspiration fusion correction method based on multi-source data as described in claim 1, characterized in that, The deviation correction coefficients for each watershed zone are calculated using the aforementioned water balance evapotranspiration data and evapotranspiration fusion product, including: Calculate the evapotranspiration value of the evapotranspiration fusion product in each watershed region; Evapotranspiration is calculated based on the evapotranspiration data of water balance in each watershed and the evapotranspiration fusion product. Calculate the ratio of the evapotranspiration of the water balance evapotranspiration data to the evapotranspiration of the evapotranspiration fusion product for each watershed partition. This ratio is the deviation correction coefficient for that watershed partition.

7. The high-precision evapotranspiration fusion correction method based on multi-source data as described in claim 1, characterized in that, The deviation correction coefficient is used to correct the deviation of the evaporation fusion product, resulting in the final fusion-corrected evaporation product, including: The evapotranspiration product of each watershed zone is multiplied by the corresponding deviation correction coefficient to obtain the final evapotranspiration product after fusion correction.

8. A high-precision evapotranspiration fusion correction system based on multi-source data, characterized in that, include: The preprocessing module is used to download the required evapotranspiration products from remote sensing and atmospheric reanalysis, preprocess the evapotranspiration products to ensure consistent spatiotemporal resolution and perform raster alignment. The data collection and calculation module is used to collect meteorological data from meteorological stations and calculate reference evapotranspiration. The fusion module is used to extract the evapotranspiration values ​​corresponding to each evapotranspiration product at the meteorological station location using the preprocessed evapotranspiration products, and to construct a support vector regression model using the evapotranspiration values ​​as input features and the reference evapotranspiration as the target variable, and to perform nonlinear fusion to generate evapotranspiration fusion products. The zoning module is used to collect hydrological data of the study area and divide the study area into watersheds according to flow stations. The hydrological data includes flow data from flow stations, precipitation data from rain gauges, non-agricultural water use data, water diversion data from water diversion projects, and monitoring data from reservoirs and ponds. The prediction module is used to establish a random forest prediction model for the water storage capacity of typical small reservoirs using the monitoring data of the reservoirs and ponds, and to predict the water storage variable data of all reservoirs and ponds in the study area. The calculation module is used to calculate the regional water balance based on the hydrological data and storage variable data, and to calculate the water balance evapotranspiration data of each watershed zone by combining the station reference evapotranspiration values. The coefficient calculation module is used to calculate the deviation correction coefficient for each watershed zone using the water balance evapotranspiration data and the evapotranspiration fusion product. The correction module is used to correct the deviation of the evaporation fusion product using the deviation correction coefficient, so as to obtain the final evaporation product after fusion correction.

9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-7.

10. A computer-readable storage medium storing computer instructions thereon, characterized in that, When executed by the processor, this instruction implements the steps of the method as described in any one of claims 1-7.

Images