Satellite precipitation error correction and fusion method based on elevation gradient under complex terrain

Abstract

This application provides a satellite precipitation error correction and fusion method based on elevation gradient under complex terrain, belonging to the field of meteorological data processing technology. The method includes: acquiring multi-source precipitation observation data of complex terrain areas, obtaining gridded precipitation data and inputting it into an elevation-error correlation model, obtaining an error intensity distribution map and identifying grid areas with error intensity values ​​greater than or equal to a preset error threshold as uncertain areas and performing feature extraction, obtaining regional feature vectors and inputting them into an error correction strategy selection model, determining the correction strategy type and applicability of the uncertain areas, obtaining the correction areas that need to be deeply fused from the uncertain areas and obtaining ground verification point data, calculating the local precipitation true field and error spatial structure parameters; using a correction algorithm to correct the satellite precipitation data of the correction areas, fusing the satellite precipitation data of the correction areas with those of the uncorrected areas, and generating precipitation products and uncertainty assessment reports.

Description

Technical Field

[0001] This application relates to the field of meteorological data processing technology, and in particular to a method for satellite precipitation error correction and fusion based on elevation gradient under complex terrain. Background Technology

[0002] In obtaining accurate precipitation information in complex terrain areas, one approach is on-site measurement using ground-based meteorological stations. This method directly collects and measures precipitation data, but it is limited by the number and distribution of stations and cannot comprehensively represent the overall precipitation characteristics of complex terrain areas. Another approach utilizes satellite remote sensing technology to acquire precipitation data. This involves receiving and analyzing precipitation-related electromagnetic signals through specific sensors to retrieve precipitation information. However, this method is affected by complex terrain, leading to errors in the precipitation data. Because complex terrain causes complex changes in atmospheric motion and water vapor distribution, existing methods cannot fully consider the impact of terrain factors on satellite precipitation data errors. This makes it difficult to accurately determine the intensity and distribution of errors during error correction, and prevents precise correction of areas with large errors, ultimately affecting the accuracy of the final precipitation data.

[0003] Therefore, there is an urgent need for a satellite precipitation error correction and fusion method based on elevation gradient under complex terrain. Summary of the Invention

[0004] To address the aforementioned technical issues, this application provides a satellite precipitation error correction and fusion method based on elevation gradient under complex terrain.

[0005] A first aspect of this application provides a method for satellite precipitation error correction and fusion based on elevation gradient under complex terrain, including: Acquire multi-source precipitation observation data of the target complex terrain area, perform spatiotemporal matching and quality control on the multi-source precipitation observation data, and obtain standardized grid precipitation data; The standardized grid precipitation data is input into a preset elevation-error correlation model to obtain an error intensity distribution map of the potential error level of satellite precipitation data in each grid unit of the target complex terrain area; Based on the error intensity distribution map, grid regions with error intensity values ​​greater than or equal to a preset error threshold are identified as uncertainty regions. Multi-dimensional feature extraction is performed on the uncertain region to obtain the region feature vector; The region feature vector is input into a preset error correction strategy selection model to determine the type of correction strategy and the corresponding applicability of the strategy for the uncertain region. Based on the correction strategy type and the corresponding strategy applicability, the correction region that needs to be deeply fused from multiple sources is obtained from the uncertainty region; Obtain ground verification point data for the calibration area, and calculate the local precipitation true field and error spatial structure parameters based on the ground verification point data; Based on the correction strategy type and the error spatial structure parameters, the satellite precipitation data of the correction area is corrected using a correction algorithm, and the satellite precipitation data of the correction area is spatially fused with the satellite precipitation data of the uncorrected area to generate the final gridded fused precipitation product and the corresponding uncertainty assessment report.

[0006] A second aspect of this application provides a satellite precipitation error correction and fusion system based on elevation gradient under complex terrain, comprising: The data acquisition module is used to acquire multi-source precipitation observation data of the target complex terrain area, perform spatiotemporal matching and quality control on the multi-source precipitation observation data, and obtain standardized grid precipitation data. The error diagnosis module is used to input the standardized grid precipitation data into a preset elevation-error correlation model to obtain an error intensity distribution map of the potential error level of satellite precipitation data in each grid unit of the target complex terrain area; The region identification module is used to identify grid regions with error intensity values ​​greater than or equal to a preset error threshold, based on the error intensity distribution map, as uncertain regions; The feature extraction module is used to extract multi-dimensional features from the uncertain region to obtain a region feature vector; The strategy determination module is used to input the regional feature vector into a preset error correction strategy selection model to determine the type of correction strategy and the corresponding applicability of the strategy for the uncertain region. The region correction module is used to obtain the correction region that needs to be deeply fused from the uncertain region based on the correction strategy type and the corresponding strategy applicability. The data calculation module is used to acquire ground verification point data in the correction area and calculate the local precipitation true field and error spatial structure parameters based on the ground verification point data. The calibration and fusion module is used to calibrate the satellite precipitation data of the calibration area based on the calibration strategy type and the error spatial structure parameters, and to spatially fuse the satellite precipitation data of the calibration area with the satellite precipitation data of the uncalibrated area to generate the final gridded fused precipitation product and the corresponding uncertainty assessment report.

[0007] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the above-described method for satellite precipitation error correction and fusion based on elevation gradient under complex terrain.

[0008] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method for satellite precipitation error correction and fusion based on elevation gradient under complex terrain.

[0009] The beneficial effects of the satellite precipitation error correction and fusion method based on elevation gradient under complex terrain provided in this application are as follows: This application obtains standardized gridded precipitation data by acquiring and processing multi-source precipitation observation data, which can standardize the data format and quality; it uses an elevation-error correlation model to obtain an error intensity distribution map, which can identify potential errors in satellite precipitation data; it determines the uncertainty area and performs multi-dimensional feature extraction and correction strategy selection, which can formulate appropriate correction strategies for different situations; it obtains the correction area and calculates relevant parameters, providing a basis for correction; it uses the correction algorithm to correct satellite precipitation data and fuses it with data from uncorrected areas, which can generate gridded fused precipitation products and corresponding uncertainty assessment reports, thereby improving the accuracy and reliability of satellite precipitation data under complex terrain. Attached Figure Description

[0010] Figure 1 A flowchart illustrating a satellite precipitation error correction and fusion method based on elevation gradient under complex terrain provided in an embodiment of this application; Figure 2 A structural block diagram of a satellite precipitation error correction and fusion system based on elevation gradient under complex terrain provided in an embodiment of this application; Figure 3 This is a schematic block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0011] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0012] To make the purpose, technical solution, and advantages of this application clearer, the following will be described in conjunction with the appendix. Figure 1-3 The following is an explanation using specific examples.

[0013] Please refer to Figure 1 , Figure 1 A flowchart illustrating a satellite precipitation error correction and fusion method based on elevation gradient under complex terrain, provided in an embodiment of this application, is shown below. The method includes: S101: Acquire multi-source precipitation observation data for the target complex terrain area, perform spatiotemporal matching and quality control on the multi-source precipitation observation data, and obtain standardized grid precipitation data.

[0014] In this embodiment, the target complex terrain area refers to the geographical region corresponding to precipitation correction, characterized by complex spatial structures such as elevation variations, mountains, hills, plateaus, and canyons, where topography has a strong influence on precipitation. Multi-source precipitation observation data comprises precipitation observation data from different observation methods, platforms, and spatiotemporal resolutions, including: satellite-retrieved precipitation products (such as GPM and TRMM series satellite precipitation), ground rain gauge observation data, radar-estimated precipitation data, and reanalysis precipitation data.

[0015] In this embodiment, spatiotemporal matching unifies precipitation data from different sources, with different temporal resolutions, and different spatial grids, to the same time step and the same spatial grid system, ensuring that each grid has comparable multi-source precipitation values ​​at the same time. Quality control involves processing the raw observation data, including outlier detection, removal, missing data repair, and consistency checks, to eliminate obvious errors, missing data, jumps, and unreasonable values, making the data usable and reliable. Standardized gridded precipitation data, after spatiotemporal unification, quality control, unit unification, and range normalization, forms a gridded precipitation product with regular grids, a unified format, and a unified spatiotemporal reference, which can be directly input into subsequent models for error correction and fusion.

[0016] S102: Input standardized grid precipitation data into a preset elevation-error correlation model to obtain an error intensity distribution map of the potential error level of satellite precipitation data in each grid unit of the target complex terrain area.

[0017] In this embodiment, the standardized gridded precipitation data, after spatiotemporal matching and quality control processing, possesses a unified spatial grid, unified time reference, unified data format, unified unit (mm / h or mm / d), and is reliable and anomaly-free. It serves as the standardized data source for the elevation-error correlation model, encompassing all grids in the target complex terrain area, with each grid corresponding to a unique precipitation observation. The preset elevation-error correlation model is a prediction model designed, pre-trained, and deployed for complex terrain scenarios. Its function is to establish a mapping relationship between standardized gridded precipitation data, elevation gradient features, and potential errors in satellite precipitation. Based on the input data, it can output the error level of satellite precipitation in each grid, adapting to the error distribution patterns of complex terrains such as mountains and plateaus.

[0018] Specifically, the pre-defined elevation-error correlation model is constructed as follows: First, based on standardized grid precipitation data from historical periods, the systematic error between satellite remote sensing inversion precipitation data and station interpolated precipitation fields within each grid cell is calculated; second, the elevation gradient features corresponding to each grid cell are extracted, including the elevation of the grid center point, the elevation variation coefficient within the grid, and the mean elevation difference between the grid and adjacent grids; finally, with the systematic error as the dependent variable and the elevation gradient features as the independent variable, a geographic weighted regression model is used for fitting to establish a relationship model between error and terrain features, which serves as the elevation-error correlation model.

[0019] The elevation-error correlation model employs a four-layer structure: a feature input layer, an elevation feature embedding layer, a correlation fitting layer, and an error intensity output layer. The feature input layer receives standardized gridded precipitation data along with corresponding elevation and topographic gradient data. The elevation feature embedding layer discretizes and encodes the elevation information, highlighting the impact of complex terrain on satellite precipitation errors. The correlation fitting layer establishes the correlation between elevation, topographic relief, and satellite precipitation errors through multi-layer nonlinear mapping. The error intensity output layer outputs the normalized error intensity value. These modules are sequentially connected forward to form an end-to-end fitting structure from elevation and precipitation data to error intensity.

[0020] In this embodiment, a grid cell is the smallest spatial unit into which the target complex terrain area is divided according to a preset spatial resolution (e.g., 1km×1km, 5km×5km). It is the basic carrier of gridded data. Each grid cell corresponds to unique geographic coordinates, elevation information, and precipitation data. The prediction and output of error intensity are both based on grid cells. Satellite precipitation data is precipitation observation data obtained through inversion from meteorological satellites, such as the GPM and TRMM series satellites. It is the object of this error correction. Due to the influence of elevation obstruction and airflow disturbances caused by complex terrain, observation errors will occur, and it is necessary to predict the potential degree of error through the model.

[0021] In this embodiment, the potential error level refers to the probability and magnitude of deviation between satellite precipitation data and actual ground precipitation data. It has not been verified in the field and is predicted by an elevation-error correlation model based on factors such as elevation gradient and precipitation characteristics. It is represented by a value in the range of 0-1, with larger values ​​indicating larger errors. The error intensity distribution map is a thematic map that uses a spatial grid of the target complex terrain area as its carrier to spatially visualize the potential error level of each grid unit. It intuitively represents the spatial distribution pattern of satellite precipitation errors within the area. Areas with large numerical values ​​(high error intensity) correspond to low satellite precipitation reliability, while areas with small numerical values ​​correspond to high reliability, providing a direct basis for identifying areas of uncertainty.

[0022] S103: Based on the error intensity distribution map, identify grid areas where the error intensity value is greater than or equal to a preset error threshold as uncertain areas.

[0023] In this embodiment, the error intensity value is the numerical value corresponding to each grid cell in the error intensity distribution map, used to characterize the potential error magnitude of satellite precipitation data at that grid, and is predicted by the elevation-error correlation model. The preset error threshold is a critical value determined in advance through experiments, statistical analysis, and terrain features, used to distinguish between acceptable and unacceptable error areas, and serves as the basis for determining whether to enter an uncertainty area.

[0024] Specifically, the method for determining the preset error threshold is as follows: First, through actual measurement and verification, when the error intensity value is greater than or equal to 0.6, the satellite precipitation error is greater than 10% (exceeding the tolerance range of business applications); second, data statistics are conducted to analyze the precipitation error distribution of the target area over the past 5 years, and 95% of the error grids have an error intensity greater than or equal to 0.6; finally, standard sample testing shows that when the threshold is 0.6, the recall rate for identifying uncertain areas is greater than or equal to 0.9, and the precision rate is greater than or equal to 0.88. Therefore, the preset error threshold is finally set to 0.6.

[0025] In this embodiment, a grid region is a spatial range composed of one or more spatially adjacent or discrete grid cells. An uncertainty region is a region formed by grids whose error intensity value is greater than or equal to a preset error threshold. Satellite precipitation data in this region has large errors and low reliability, and requires subsequent key feature extraction, strategy judgment, and error correction.

[0026] S104: Perform multi-dimensional feature extraction on the uncertain region to obtain the region feature vector.

[0027] In this embodiment, the uncertainty region is a set of grid areas where the error intensity value is greater than or equal to a preset error threshold. Satellite precipitation data within this region has large errors and low reliability, making it a key area for error correction. Spatially, it is characterized by contiguous distribution (e.g., mountain clusters) or discrete distribution (e.g., isolated high-altitude grids). Multi-dimensional feature extraction extracts features from multiple independent dimensions (topography, climate, weather) related to satellite precipitation errors within the uncertainty region. These features characterize the region and represent the factors influencing the error, capturing the comprehensive impact of topographic relief, climate background, and weather conditions on satellite precipitation errors, providing data support for the selection of correction strategies. The regional feature vector is a single high-dimensional quantized vector formed by standardizing and weighted fusion of the various features extracted from multiple dimensions (topography, climate, weather features). It is a condensed representation of the features of the uncertainty region, with each uncertainty region corresponding to a unique regional feature vector, which can be directly input into the error correction strategy selection model for inference.

[0028] S105: Input the regional feature vector into the preset error correction strategy selection model to determine the type of correction strategy for the uncertain region and the corresponding applicability of the strategy.

[0029] In this embodiment, the preset error correction strategy selection model is a pre-trained and deployed classification prediction model designed for satellite precipitation error scenarios in complex terrain. Its function is to establish the mapping relationship between regional feature vectors and error correction strategies, and to output correction strategies and their applicability that are suitable for the current uncertain regions.

[0030] Specifically, the pre-defined error correction strategy selection model is constructed as follows: First, a training sample set including various typical complex terrain scenarios is constructed. Each sample includes a regional feature vector of a historically uncertain area and an optimal correction strategy label determined through expert knowledge and data experiments. Second, a random forest or gradient boosting decision tree algorithm is used, with the regional feature vector as the input feature and the optimal correction strategy label as the output target, to train the error correction strategy selection model. Finally, during the training process, the importance score of each input feature for strategy classification and the prediction probability of the error correction strategy selection model for each strategy type are output simultaneously. The prediction probability is used as a measure of the applicability of the strategy when applied.

[0031] Specifically, the error correction strategy selection model adopts a multi-layer feature classification architecture, consisting of a feature input layer, a feature encoding layer, a strategy classification layer, an applicability regression layer, and an output layer. The feature input layer receives regional feature vectors; the feature encoding layer performs weighted fusion and dimensionality normalization on multi-dimensional features such as terrain, climate, and error spatial structure; the strategy classification layer uses a fully connected classifier to output the correction strategy type corresponding to the uncertain region; the applicability regression layer is connected in parallel with the classification layer to simultaneously predict the applicability of each strategy; the output layer performs constraint matching on the classification and regression results to ensure a one-to-one correspondence between strategy type and strategy applicability. All modules are connected in a forward propagation manner, forming an end-to-end strategy selection network.

[0032] In this embodiment, the uncertainty region is a set of grid areas where the error intensity value is greater than or equal to a preset error threshold. Satellite precipitation data in these areas has large errors and low reliability, making them the target of this correction strategy selection. The correction strategy type is a preset, directly applicable error correction method category for satellite precipitation errors in complex terrain. Common types, based on the characteristics of complex terrain, include: interpolation correction strategy, multi-source data fusion correction strategy, elevation gradient correction strategy, ensemble Kalman filter correction strategy, and local bias correction strategy. Strategy applicability is a value output by the error correction strategy selection model, representing the degree to which a certain correction strategy adapts to the current uncertainty region. The value ranges from 0 to 1; the closer the value is to 1, the better the error correction effect and the stronger the adaptability of the strategy for the current region; the smaller the value, the worse the adaptability of the strategy, and the correction effect cannot be guaranteed.

[0033] S106: Based on the type of correction strategy and the corresponding applicability of the strategy, obtain the correction area that needs to be deeply fused from the uncertain area.

[0034] In this embodiment, the correction strategy type is the output of the error correction strategy selection model. It represents the specific error correction method category adapted to each uncertainty region. Different types correspond to different correction logic and applicable scenarios. Among them, the multi-source data fusion correction strategy is the target type selected in this study. The strategy applicability characterizes the degree to which a certain type of correction strategy is adapted to the corresponding uncertainty region. It takes a value of 0-1. The larger the value, the better the strategy applicability and correction effect; the smaller the value, the worse the applicability, and it is necessary to determine whether a better multi-source data deep fusion method should be adopted.

[0035] In this embodiment, deep multi-source data fusion, compared to ordinary correction strategies, integrates precipitation data from multiple sources (satellite, ground rain gauge, radar, reanalysis data, etc.) and achieves error cancellation through spatial correlation, error complementarity, and weight allocation. It is suitable for areas with insufficient strategy applicability and complex errors, and serves as a correction method for selecting correction areas. The correction areas requiring deep multi-source data fusion are grid areas identified from uncertain regions through strategy type matching, applicability screening, and spatial clustering optimization. These areas are characterized by inadequate strategy applicability, complex errors, and poor performance of single correction strategies.

[0036] S107: Obtain ground verification point data for the calibration area, and calculate the local precipitation true field and error spatial structure parameters based on the ground verification point data.

[0037] In this embodiment, the correction area is selected from areas of uncertainty. It is a grid area where satellite precipitation errors are large and the effect of a single strategy is insufficient, requiring deep fusion of multi-source data. The ground verification point data is precipitation data from ground rain gauges, automatic weather stations, rain sensors, and other field observations within or around the correction area. It includes information such as observation location, observation time, measured precipitation, and observation accuracy, and is used to represent the actual precipitation situation.

[0038] In this embodiment, the local precipitation truth field is a highly reliable and approximately realistic spatial distribution field of precipitation generated across the entire correction area grid using measured data from ground verification points as a benchmark, through spatial interpolation and fitting methods. This field serves as the reference truth for satellite precipitation error correction. Error spatial structure parameters describe the spatial distribution and correlation characteristics of satellite precipitation errors, including: nugget value, sill value, range, and the nugget / sill value ratio, which are used in subsequent steps such as Kalman filtering, spatial fusion, and error covariance matrix setting.

[0039] S108: Based on the correction strategy type and error spatial structure parameters, the satellite precipitation data in the correction area is corrected using the correction algorithm, and the satellite precipitation data in the correction area is spatially fused with the satellite precipitation data in the uncorrected area to generate the final gridded fused precipitation product and the corresponding uncertainty assessment report.

[0040] In this embodiment, the correction algorithm is a specific calculation method used to eliminate satellite precipitation errors in the correction area. This embodiment employs an ensemble Kalman filter algorithm, which can be adjusted according to the correction strategy type and error spatial structure parameters. Its function is to correct the bias and noise of satellite precipitation data based on the local precipitation truth field. The satellite precipitation data in the correction area is determined through standardization, error prediction, and screening. It is satellite precipitation grid data that needs to be corrected through deep fusion of multi-source data and the correction algorithm. After correction, the accuracy is improved and the reliability meets the standards.

[0041] Specifically, the ensemble Kalman filter algorithm adopts a three-layer architecture of input layer, processing layer, and output layer. The modules at each layer are closely related and have clear functions, making it suitable for satellite precipitation correction scenarios with complex terrain. The input layer provides input data for the algorithm, including a data preprocessing module responsible for standardizing, outlier removal, and dimension matching of the input satellite precipitation data and ground verification point data, ensuring the integrity and consistency of the input data. The processing layer is the algorithm operation layer, including a set initialization module, an error covariance matrix construction module, a Kalman gain calculation module, and an analysis update module. The set initialization module is responsible for generating the initial set members. The error covariance matrix construction module generates the background field error covariance matrix (B matrix) and the observation error covariance matrix (R matrix) respectively. The Kalman gain calculation module calculates the Kalman gain (K) based on the B matrix and the adjusted R matrix. The analysis update module completes error correction by fusing background field and observation field information. The output layer includes a result verification module and a data output module. The result verification module performs quality control and accuracy verification on the analyzed and updated precipitation field. The data output module outputs the corrected data that has passed verification in a formatted manner, providing input for the subsequent spatial fusion step. The modules at each level are linked through a data interface, enabling the algorithm to run efficiently and stably.

[0042] In this embodiment, the satellite precipitation data in the uncorrected region is grid data outside the identified uncertainty area (error intensity less than a preset error threshold), or regional data in the uncertainty area where a single correction strategy can meet the accuracy requirements. No multi-source fusion correction is needed; only the standardized original satellite precipitation data (with acceptable error) is retained. Spatial fusion is used to avoid discontinuities or abrupt changes in the connection between corrected and uncorrected region data. A smooth transition algorithm is employed, based on terrain gradient, error distribution, and data reliability, to weight and connect the precipitation data of the two types of regions, achieving spatial continuity and consistency of precipitation data across the entire region.

[0043] In this embodiment, the final gridded fused precipitation product is a spatially continuous, accurate, and formatted gridded precipitation data product formed after correction algorithms and spatial fusion processing. It encompasses the entire target complex terrain area, possesses a unified spatiotemporal reference, unit, and grid resolution, and can be directly used for operational applications such as hydrological simulation and weather forecasting. The uncertainty assessment report, generated alongside the gridded fused precipitation product, represents the uncertainty (error magnitude, reliability) of precipitation estimates for each grid unit in the final precipitation product. It includes comprehensive uncertainty indicators, data quality levels, error source analysis, and applicable scenario suggestions, ensuring product traceability and applicability.

[0044] As can be seen from the above, this application obtains standardized gridded precipitation data by acquiring and processing multi-source precipitation observation data, which can standardize data format and quality; it uses an elevation-error correlation model to obtain an error intensity distribution map, which can identify potential errors in satellite precipitation data; it identifies uncertainty areas and performs multi-dimensional feature extraction and correction strategy selection, which can formulate appropriate correction strategies for different situations; it obtains correction areas and calculates relevant parameters, providing a basis for correction; and it uses correction algorithms to correct satellite precipitation data and fuses it with data from uncorrected areas, which can generate gridded fused precipitation products and corresponding uncertainty assessment reports, improving the accuracy and reliability of satellite precipitation data under complex terrain.

[0045] In one embodiment of this application, multi-dimensional feature extraction is performed on the uncertain region to obtain a region feature vector, including: Extracting terrain feature sub-vectors from high-resolution digital elevation model data in uncertain regions; Based on the long-term climate statistical grid dataset and climate zoning map of the target complex terrain region, extract climate background feature sub-vectors; Weather feature sub-vectors are extracted from atmospheric reanalysis grid data that are spatiotemporally matched with uncertain regions. The topographic feature vector, climate background feature vector, and weather feature vector are weighted and fused to form the regional feature vector.

[0046] In this embodiment, high-resolution digital elevation model (DEM) data is high-precision gridded data, such as 30m resolution, used to characterize the topographic relief features of the target area. It includes information such as elevation, slope, and aspect of each grid cell and serves as the data source for extracting topographic features, adapting to the detailed capture of complex terrain. The topographic feature sub-vector is a set of features extracted from the high-resolution DEM data that characterize the topographic properties of uncertain areas, such as elevation and slope. After standardization, it forms a sub-vector, for example, 80-dimensional, representing the impact of topographic relief on satellite precipitation errors.

[0047] In this embodiment, the long-term climate statistical grid dataset is a dataset formed by statistical analysis and gridding of historical precipitation, temperature, and other meteorological observation data of the target area over the past 30 years (or longer). It is typically 1km × 1km resolution and includes climate indicators such as annual precipitation and precipitation variability. The climate zoning map is a thematic map that divides the target area into different climate zones based on its climate type and precipitation distribution patterns. It is used to extract the characteristics of the climate type of uncertain areas and to help capture the impact of long-term climate background on precipitation errors. The climate background feature sub-vector is a set of features extracted from the long-term climate statistical grid dataset and the climate zoning map that characterize the long-term climate characteristics of uncertain areas. These features include average annual precipitation and climate zoning type. After standardization, these sub-vectors are typically 96-dimensional and represent the potential impact of long-term climate conditions on satellite precipitation errors. The atmospheric reanalysis grid data is gridded meteorological data formed by assimilating multi-source data from meteorological satellites, ground observation stations, and radar. It typically has a 1-hour temporal resolution and a 1km spatial resolution and includes real-time / near-real-time weather indicators such as air pressure, humidity, and wind speed.

[0048] In this embodiment, spatiotemporal matching ensures that the temporal range, spatial range, and grid resolution of the atmospheric reanalysis grid data completely correspond to the uncertain region (synchronized in time, overlapping in space, and consistent in grid size), enabling the extracted weather features to match the real-time weather conditions of the uncertain region. The weather feature sub-vector is a set of features extracted from the spatiotemporally matched atmospheric reanalysis grid data, representing the real-time / near-real-time weather conditions of the uncertain region, such as precipitation and relative humidity over the past 24 hours. These features, after standardization, are typically 80-dimensional and represent the direct impact of short-term weather conditions on satellite precipitation errors.

[0049] In this embodiment, the weighted fusion process is to assign different fusion weights to the three types of feature sub-vectors—terrain, climate background, and weather—based on their confidence levels (i.e., the reliability of the features and the strength of their correlation with precipitation errors). The process then splices and fuses the three types of sub-vectors into a single regional feature vector. This highlights features with strong correlation and high reliability, thereby improving the accuracy of subsequent model inference.

[0050] As can be seen from the above, this embodiment can represent the topographic features of an uncertain region by extracting topographic feature sub-vectors from high-resolution digital elevation model data; it can reflect the regional climate background by extracting climate background feature sub-vectors from long-term climate statistical grid datasets and climate zoning maps of the target complex topographic region; it can obtain regional weather features by extracting weather feature sub-vectors from atmospheric reanalysis grid data that is spatiotemporally matched with the uncertain region; and it can construct a regional feature vector by weighting and fusing these sub-vectors, providing a more comprehensive and accurate input basis for determining the type and applicability of the correction strategy.

[0051] In one embodiment of this application, the terrain feature sub-vector, climate background feature sub-vector, and weather feature sub-vector are weighted and fused to form a regional feature vector, including: The terrain feature subvectors are evaluated based on the terrain complexity index of the region corresponding to the terrain feature subvector and the data source resolution to generate terrain feature confidence scores. The climate feature confidence scores are generated by evaluating the historical climate data on which the climate background feature subvectors depend, based on the temporal length and spatial representativeness of the data. The weather feature subvectors are evaluated based on the spatiotemporal resolution of the atmospheric reanalysis data on which they are based and the density of assimilated observations, and a weather feature confidence score is generated. Based on the confidence scores of terrain features, climate features, and weather features, the weight allocation of terrain feature sub-vectors, climate background feature sub-vectors, and weather feature sub-vectors in the fusion process is determined. Based on the assigned weights, the terrain feature sub-vector, climate background feature sub-vector, and weather feature sub-vector are weighted and fused to generate a regional feature vector.

[0052] In this embodiment, the terrain feature sub-vector is a feature vector extracted from high-resolution digital elevation model data of uncertain regions, representing terrain information such as topographic relief, slope, and aspect, indicating the impact of terrain on satellite precipitation errors. The climate background feature sub-vector is a feature vector extracted from long-term climate statistical grid data and climate zoning maps, representing long-term regional climate patterns such as precipitation, temperature, and climate type. The weather feature sub-vector is a feature vector extracted from atmospheric reanalysis grid data, representing short-term weather conditions such as current precipitation, humidity, and wind field.

[0053] In this embodiment, the terrain complexity index represents the degree of undulation and fragmentation of the regional terrain. A higher index indicates more complex terrain and more important terrain features. Data source resolution refers to the spatial precision of terrain, climate, and weather data, such as 30m or 1km. Higher resolution results in more refined data and higher confidence levels. Terrain feature confidence is a score (0-1) of the reliability and importance of terrain feature sub-vectors, determined jointly by terrain complexity and data source resolution.

[0054] In this embodiment, the historical climate data duration refers to the observation period used to construct the climate statistical grid data, such as 30 years. The longer the duration, the more stable the climate background and the higher the confidence level. Spatial representativeness refers to the extent to which climate data can cover the entire uncertain region and represent the true climate background of the region. Climate feature confidence is a score (0-1) for the reliability and representativeness of the climate background feature sub-vector. The spatiotemporal resolution of atmospheric reanalysis data is the time interval (e.g., 1 hour) and spatial grid size (e.g., 1 km) of the reanalysis data; the higher the resolution, the higher the confidence level. Assimilation observation density is the density of ground / satellite observation points used in the assimilation process of the reanalysis data; the higher the density, the higher the data quality. Weather feature confidence is a score (0-1) for the reliability and timeliness of the weather feature sub-vector. Weight allocation is based on the three confidence levels, assigning a weight to the topography, climate, and weather sub-vectors during fusion; the higher the confidence level, the higher the weight. The regional feature vector is a single high-dimensional vector obtained after weighted fusion, which serves as the input to the error correction strategy selection model.

[0055] As can be seen from the above, this embodiment obtains standardized grid precipitation data by acquiring multi-source precipitation observation data of the target complex terrain area and processing it. Then, it obtains an error intensity distribution map through the elevation-error correlation model, identifies the uncertainty area, and extracts feature sub-vectors for the area from the aspects of terrain, climate background, and weather. Based on the evaluation factors corresponding to each feature sub-vector, corresponding confidence scores are generated. Then, the fusion weight allocation is determined and the weighted fusion is used to generate a regional feature vector. This can comprehensively consider the reliability and importance of different features to construct a more accurate and reasonable regional feature vector, providing a more reliable basis for determining the type and applicability of correction strategies, and helping to improve the accuracy and effectiveness of satellite precipitation error correction and fusion under complex terrain.

[0056] In one embodiment of this application, based on the correction strategy type and the corresponding strategy applicability, a correction region requiring deep fusion of multi-source data is obtained from the uncertainty region, including: Match the correction strategy types with the preset strategy type level library to determine the basic screening threshold corresponding to each strategy type; Compare the applicability of the strategy with the basic screening threshold for the corresponding strategy type; For any region of uncertainty, if the applicability of the strategy is less than the basic screening threshold of the corresponding strategy type, the region of uncertainty is marked as a candidate correction region. Based on the spatial distribution characteristics of the candidate correction regions, a region clustering operation is performed to merge candidate correction regions whose spatial distance is less than the preset adjacency threshold and whose correction strategy types are the same or compatible into contiguous regions. Based on the area and geographical integrity of the merged contiguous areas, boundary optimization is performed on the contiguous areas that meet the preset area threshold conditions to generate the final correction area that needs to be deeply fused from multiple sources.

[0057] In this embodiment, the preset strategy type level library is a relational database pre-built based on historical correction data and operational accuracy requirements of the target complex terrain area. It stores the mapping relationship between various correction strategy types and corresponding basic screening thresholds, used to determine whether a single correction strategy can meet the accuracy requirements. The basic screening threshold is a critical value corresponding to each type of correction strategy, ranging from 0 to 1. It is determined by historical correction effect statistics and operational accuracy requirements (e.g., precipitation error greater than or equal to 0.1 mm / h). It is the standard for judging whether the applicability of a strategy meets the standard. If it is less than this threshold, it means that a single strategy cannot meet the correction accuracy and multi-source fusion is required.

[0058] In this embodiment, the candidate correction region is an area within an uncertain region where the applicability of the strategy is less than the basic screening threshold for the corresponding strategy type. This is initially determined to be an area where a single strategy's correction effect is insufficient, requiring deep fusion of multi-source data. Spatially, it consists of discrete grids or small-scale, scattered, contiguous grids. Spatial distribution characteristics include the spatial location of the candidate correction region within the target complex terrain area, grid spacing, distribution pattern (discrete / scattered / contiguous), and correspondence with terrain features. These characteristics are used to determine whether candidate regions are suitable for merging into contiguous regions, avoiding fragmented correction. Region clustering is a process based on the spatial coordinates of the candidate correction regions. By calculating the spatial distance between grids, discrete candidate regions that meet the conditions of spatial proximity and strategy compatibility are merged into contiguous regions. This aims to improve the efficiency of multi-source fusion correction and ensure the spatial continuity of the corrected data. The preset adjacency threshold is a critical distance pre-determined based on the average elevation and terrain undulation of the target complex terrain area. It is used to determine whether two candidate correction regions are spatially adjacent. If the spatial distance between the grid centers of two regions is less than the preset adjacency threshold, they are determined to be adjacent regions that can be merged.

[0059] In this embodiment, "same or compatible correction strategy type" means that the correction strategy types corresponding to the candidate correction regions are completely identical, or that different strategy types can be adapted to the same set of multi-source data deep fusion correction logic. For example, ensemble Kalman filter correction and local bias correction are both compatible with multi-source fusion logic, allowing the merged contiguous regions to adopt a unified multi-source fusion correction scheme. The contiguous region is a spatially continuous grid region without obvious discontinuities formed by merging discrete candidate correction regions through region clustering operations. Compared with discrete candidate regions, it is more convenient for subsequent ground verification point data matching, error parameter calculation, and multi-source fusion correction.

[0060] In this embodiment, geographical integrity refers to the boundary shape and spatial extent of contiguous areas, whether they conform to the natural geographical features of the target complex terrain, such as the boundaries of mountains and hills, and whether there are obvious grid gaps or fragmentation faults, ensuring the geographical rationality of the correction area and avoiding invalid corrections across terrain units. The preset area threshold is a critical area determined in advance based on grid resolution (e.g., 1km × 1km), business correction costs, and accuracy requirements. It is used to screen effective contiguous areas, eliminate fragmented contiguous areas with too small an area, avoid excessive correction costs and lack of practical business value, and retain contiguous areas with correction significance.

[0061] Specifically, the target area has a unified grid resolution; in this embodiment, the grid is 1km × 1km, with a single grid of 1... To avoid fragmented correction, a constraint coefficient of 5-8 is set. In this embodiment, the constraint coefficient is set to 5, corresponding to a minimum of 5 effective grids. The formula is: Preset area threshold = area of ​​a single grid × constraint coefficient. In this embodiment, 5 is obtained. Further adjustments were made based on the complex terrain features of the target area, ultimately determining the threshold condition as a contiguous area with an area greater than or equal to 5. .

[0062] In this embodiment, boundary optimization involves morphological processing (dilation followed by erosion) of contiguous areas that meet a preset area threshold to fill small holes and smooth irregular boundaries. This process also aligns the area boundaries with the peak position of the terrain gradient, improving the spatial rationality and data continuity of the correction area and adapting to the correction requirements of complex terrain.

[0063] As can be seen from the above, this embodiment can screen out candidate correction regions suitable for deep fusion correction of multi-source data by matching the correction strategy type with the preset strategy type level library to determine the basic screening threshold and comparing it with the strategy applicability. Performing regional clustering operation on the candidate correction regions and merging regions with small spatial distances and the same or compatible correction strategy types into contiguous regions can improve the efficiency and accuracy of the correction operation. Optimizing the boundaries of contiguous regions that meet the preset area threshold conditions can make the correction regions more reasonable, thereby more accurately obtaining the correction regions that need to be deeply fused with multi-source data, and providing a more suitable processing range for subsequent satellite precipitation data correction and fusion.

[0064] In one embodiment of this application, the method for determining the preset adjacency threshold includes: The basic value of the adjacency threshold is determined based on the average elevation and topographic relief of the target complex terrain area; Obtain the climate type and precipitation variability of the candidate correction area, and correct the base value of the adjacency threshold based on climate stability to obtain the corrected adjacency threshold. Based on the requirements of the correction strategy type for data spatial continuity, the corrected adjacency threshold is adjusted a second time to obtain the final preset adjacency threshold.

[0065] In this embodiment, the preset adjacency threshold is a critical distance (unit: km) used to determine whether candidate correction areas are spatially adjacent. It is a parameter for the region clustering operation, and the final value is obtained through three steps: base value determination, climate correction, and strategy adjustment, adapting to the spatial distribution characteristics of complex terrain. The average elevation is the arithmetic mean (unit: m) of the elevation values ​​of all grid cells within the target complex terrain area, representing the overall altitude level of the area and directly affecting the spatial distribution density of candidate correction areas. The higher the altitude, the more complex the terrain, and the more dispersed the candidate areas, requiring corresponding adjustments to the base value of the adjacency threshold. The terrain undulation is the maximum elevation difference between any grid cell and its surrounding grid cells within the target complex terrain area (or the difference between the highest and lowest elevations within the area, unit: m), indicating the fragmentation and steepness of the terrain. The greater the undulation, the more dispersed the candidate areas, requiring a larger base value. The base value of the adjacency threshold is an initial value (unit: km) determined solely based on the average elevation and terrain undulation of the target complex terrain area, without considering climate or strategy factors. It serves as the basis for correction and adjustment, conforming to the spatial characteristics of the terrain itself.

[0066] In this embodiment, the candidate correction region is a region whose strategy applicability is less than the corresponding basic screening threshold and which is initially determined to require deep fusion of multi-source data. The climate and precipitation characteristics of its location serve as the basis for adjacency threshold correction. Climate type is the long-term climate classification of the candidate correction region (e.g., plateau mountain climate, temperate continental climate), determined by the region's long-term precipitation and temperature patterns. It directly affects the spatial continuity of precipitation distribution; the more stable the climate, the more uniform the spatial distribution of precipitation, the more concentrated the candidate regions, and the lower the adjacency threshold can be. Precipitation variability is the degree of dispersion (dimensionless) of precipitation observations over recent years (e.g., 30 years) within the candidate correction region, representing the stability of precipitation distribution. The greater the variability, the greater the spatial difference in precipitation, the more dispersed the candidate regions, and the higher the adjacency threshold needs to be. Climate stability is jointly determined by climate type and precipitation variability, characterizing the spatial continuity and stability of precipitation distribution in the candidate correction region. The more stable the climate (e.g., low precipitation variability), the more concentrated the spatial distribution of candidate regions, and the lower the adjacency threshold can be.

[0067] In this embodiment, the corrected adjacency threshold is an intermediate value (unit: km) obtained by correcting the base adjacency threshold for climate stability. This compensates for the shortcomings of only considering topography and not the spatial distribution characteristics of precipitation, making the preset adjacency threshold more closely match the actual distribution of the candidate area. Data spatial continuity is a requirement during the execution of the correction strategy, demanding that satellite precipitation data be spatially coherent, without discontinuities or abrupt changes. Some strategies (such as multi-source fusion correction) have strict continuity requirements, necessitating a reduction in the adjacency threshold to make the merged contiguous areas more concentrated; other strategies (such as Kalman filtering) have less stringent requirements and can be appropriately increased. The secondary adjustment is a final fine-tuning of the corrected adjacency threshold based on the spatial continuity requirements of the correction strategy type. This ensures that the adjacency threshold not only conforms to topographic and climatic characteristics but also adapts to the execution requirements of subsequent correction strategies, forming the final preset adjacency threshold.

[0068] As can be seen from the above, this embodiment determines the basic value of the adjacency threshold by using the average elevation and topographic relief of the target complex terrain area, which can initially consider the impact of terrain factors on data continuity; it obtains the climate type and precipitation variability of the candidate correction area and corrects the basic value based on climate stability, making the adjacency threshold more in line with the actual climate conditions; it makes a secondary adjustment based on the requirements of data spatial continuity according to the correction strategy type, further optimizing the adjacency threshold, so that the determined preset adjacency threshold is more scientific and reasonable, and can better meet the requirements of data spatial continuity for multi-source data deep fusion correction area clustering operation under complex terrain, improving the accuracy and rationality of area merging.

[0069] In one embodiment of this application, based on the area and geographical integrity of the merged contiguous areas, boundary optimization is performed on the contiguous areas that meet the preset area threshold conditions, including: Based on morphological image processing methods, a dilation-erosion closing operation is performed on the boundary of contiguous regions to fill small holes inside the regions and smooth irregular boundaries, thus obtaining the basic optimized boundary. Extract digital elevation model data of the contiguous area and calculate the terrain gradient change; The basic optimization boundary is compared with the location where the terrain gradient changes drastically. If the distance between the basic optimization boundary and the peak location of the terrain gradient is less than a preset distance threshold, the basic optimization boundary is adjusted to the peak location of the terrain gradient. The boundary of the correction area is aligned with the terrain features to obtain the adjusted boundary. Based on the adjusted boundary, a smoothing optimization process is performed to complete the boundary optimization.

[0070] In this embodiment, the morphological image processing method borrows from morphological operations in digital image processing. It treats the gridded contiguous regions as binary images (a grid belonging to the correction region is 1, otherwise 0), and corrects the region morphology through operations such as dilation and erosion to address boundary issues at the grid level. The dilation operation is the first step of the morphological closing operation, expanding the boundary of the contiguous region outward by one (or more) grid units, filling individual discrete holes (missing grids) within the region, and connecting adjacent small fragments, thus resolving the issues of concave boundaries and holes. The erosion operation is the second step of the morphological closing operation, shrinking the boundary of the contiguous region inward by one (or more) grid units after the dilation operation, restoring the region's extent, and simultaneously smoothing outward convexities and irregularities caused by the dilation.

[0071] In this embodiment, the closing operation is a combination of dilation and erosion morphological operations. Its function is to fill small holes within the region and smooth irregular boundaries without changing the overall area or core extent of the region; it only optimizes the boundary morphology. The basic optimized boundary is the initial optimized boundary of the contiguous region obtained after morphological closing processing. This solves the problems of holes and irregularities at the mesh level, but does not consider terrain feature adaptability. The digital elevation model data is high-precision terrain data (e.g., 30m resolution) of the location of the contiguous region, including information such as elevation, slope, and aspect. It is the data source for calculating terrain gradient changes. The terrain gradient change is the ratio of the elevation difference between adjacent grid cells within the contiguous region to the horizontal distance (unit: ° or m / km). It represents the steepness and undulation of the terrain; the larger the gradient value, the more drastic the terrain change, and it is an indicator of terrain characteristics.

[0072] In this embodiment, the peak position of the terrain gradient is the grid position where the maximum value of the terrain gradient change is located within a contiguous area, such as the transition line between mountain slopes, hills, and plains. It is the natural boundary of complex terrain and also a reference for aligning the boundary of the correction area. The preset distance threshold is a critical distance used to determine whether the basic optimization boundary needs to be adjusted towards the peak position of the terrain gradient. If the distance between the basic boundary and the peak position is less than the preset distance threshold, it indicates that the boundary is out of sync with the terrain features and needs to be adjusted for alignment.

[0073] Specifically, the method for determining the preset distance threshold is as follows: First, when the distance is less than 500m, the boundary and the peak of the terrain gradient are not aligned, which will cause the correction area to cross terrain units; second, the average spacing of the peak of the terrain gradient in the target area is 800m, and 500m is the minimum effective alignment distance; finally, the standard sample test shows that when the preset distance threshold is 500m, the matching degree between the boundary and the terrain features is greater than or equal to 0.9, so it is finally set to 500m.

[0074] In this embodiment, terrain feature alignment involves adjusting the basic optimization boundary to the peak position of the terrain gradient, ensuring that the boundary of the correction area conforms to the contours of natural terrains such as mountains and hills. This avoids the correction area crossing terrain units and ensures the geographical rationality of subsequent corrections. The adjusted boundary is obtained after terrain gradient comparison and terrain alignment, which solves the shape problem at the mesh level and adapts to terrain features, but still requires final smoothing. The smoothing optimization process involves curve fitting to the adjusted boundary, such as cubic spline interpolation, to eliminate jagged or broken-line irregularities, resulting in a smooth and continuous boundary finish, ultimately completing the boundary optimization.

[0075] As can be seen from the above, this embodiment can fill small holes inside the region and smooth irregular boundaries by performing a closing operation on the boundary of the contiguous area. Calculating the terrain gradient change and comparing and adjusting the basic optimized boundary with the peak position of the terrain gradient can make the boundary of the correction area aligned with the terrain features. Finally, smoothing optimization can further improve the boundary, thereby improving the geographical integrity and boundary quality of the correction area and enhancing the accuracy and reliability of the fused precipitation products.

[0076] In one embodiment of this application, the correction algorithm employs an ensemble Kalman filter algorithm; before correcting the satellite precipitation data of the correction area using the correction algorithm based on the correction strategy type and error spatial structure parameters, the method further includes: The local observation noise level in the correction area is determined based on the correction strategy type and the ratio of nugget value to sill value in the error space structure parameters. Adjust the observation error covariance matrix of the ensemble Kalman filter algorithm according to the local observation noise level; If the ratio of nugget value to sill value is greater than the preset ratio threshold, it is determined that the local observation noise level is high, and the value of the diagonal element in the observation error covariance matrix is ​​increased based on the first coefficient. If the ratio of nugget value to sill value is less than a preset ratio threshold, the local observation noise level is determined to be low, and the value of the diagonal element in the observation error covariance matrix is ​​reduced based on the second coefficient.

[0077] In this embodiment, the ratio of nugget value to sill value is one of the calculated error spatial structure parameters, ranging from 0 to 1. It represents the proportion of random noise in the satellite precipitation error; a larger ratio indicates more random observation noise in the error, while a smaller ratio indicates stronger spatial correlation and less random noise. The local observation noise level is the intensity of random, irregular noise in the satellite precipitation observation data within the correction area. It is determined by both the correction strategy type and the nugget value / sill value ratio and serves as the basis for adjusting the observation error covariance matrix. The observation error covariance matrix (R matrix) is the parameter matrix of the Kalman filter algorithm, characterizing the error distribution features of the observation data (ground verification point precipitation data). Larger diagonal elements indicate greater observation noise and lower algorithm confidence in the observation data. The preset ratio threshold is a critical value determined in advance based on historical correction data. It is used to determine the magnitude of the nugget value / sill value ratio, thereby distinguishing between large and small local observation noise levels, and serves as the critical standard for noise level determination.

[0078] Specifically, the method for determining the preset ratio threshold is as follows: First, when the nugget value / sill value is greater than 0.5, observation noise dominates the error, and the observation error covariance matrix needs to be increased; after analyzing 100 sets of topographic precipitation error data, when the ratio is greater than 0.5, the proportion of observation noise is greater than 60%; finally, the standard sample test shows that when the preset ratio threshold is 0.5, the accuracy of noise level judgment is greater than or equal to 0.92, and it is finally set to 0.5.

[0079] In this embodiment, the first coefficient is an adjustment factor used to increase the diagonal elements of the observation error covariance matrix when the local observation noise level is determined to be high. Its value is greater than 1, for example, 1.2-1.5, reducing the algorithm's dependence on high-noise observation data. The second coefficient is an adjustment factor used to decrease the diagonal elements of the observation error covariance matrix when the local observation noise level is determined to be low. Its value is less than 1, for example, 0.7-0.9, increasing the algorithm's confidence in low-noise observation data. The diagonal elements are the elements on the main diagonal of the observation error covariance matrix, directly corresponding to the error variance of each observation point (ground verification point). The larger the element value, the greater the noise weight of that observation point, and the smaller the correction magnitude during algorithm fusion.

[0080] As can be seen from the above, this embodiment determines the local observation noise level by adjusting the ratio of nugget value to sill value in the error space structure parameters of the correction strategy type, thereby understanding the observation noise situation in the correction area; adjusting the observation error covariance matrix of the ensemble Kalman filter algorithm according to the local observation noise level makes the algorithm more adaptable to different noise environments; adjusting the values ​​of the diagonal elements in the observation error covariance matrix according to the relationship between the ratio of nugget value to sill value and the preset ratio threshold enables flexible adjustment of algorithm parameters according to the noise level, thereby improving the correction accuracy of satellite precipitation data in the correction area.

[0081] In one embodiment of this application, satellite precipitation data from the corrected area and satellite precipitation data from the uncorrected area are spatially fused to generate a final gridded fused precipitation product and a corresponding uncertainty assessment report, including: A fusion buffer zone centered on the boundary of the correction area is established. The width of the buffer zone is adaptively determined based on the difference in terrain gradient change rate and correction strategy type between adjacent correction and non-correction areas. Within the fusion buffer, the normalized distance between each grid point and the boundary of the nearest correction region is calculated. Based on the policy applicability ratio of the regions on both sides of the boundary of the nearest correction region, the fusion weight of the corresponding grid point using the correction data is calculated through a preset weight function. For grid points adjacent to multiple correction areas, the comprehensive weight distribution of data from different correction areas is calculated based on the terrain accessibility distance to the boundary of each correction area, the policy applicability of each area, and the policy compatibility matrix. Multi-source data fusion is then performed to obtain precipitation data after global fusion. Based on the precipitation data after global fusion, the comprehensive uncertainty index of the final precipitation estimate for each grid point is calculated through error propagation analysis, taking into account the original error intensity, the applicability of the correction strategy, the fusion weight coefficient, and the spatial distribution density of the ground validation data. The comprehensive uncertainty index is mapped to a preset data quality level, and a corresponding uncertainty assessment report is generated.

[0082] In this embodiment, the fusion buffer is a transitional region (strip-shaped) extending from the boundary of the correction region into the correction region and to both sides of the uncorrected region. Its function is to smoothly connect the data from the two types of regions, avoiding abrupt data changes at the boundary, and its width is adaptively adjusted (to fit terrain and strategy differences). The terrain gradient change rate is the magnitude of change in terrain elevation per unit distance (e.g., 1 km) (unit: m / km), characterizing the steepness of the terrain at the boundary between the correction and uncorrected regions. The greater the change rate, the wider the buffer needs to be to avoid fusion discontinuities caused by terrain differences. The degree of difference in correction strategy types is the degree of difference (dimensionless, 0-1) between the correction strategies (or no correction strategy) used in adjacent correction and uncorrected regions. The greater the difference, the wider the buffer needs to be to accommodate the differences in data after correction using different strategies.

[0083] In this embodiment, the adaptive determination involves a variable buffer width, calculated based on the rate of change of terrain gradient at the boundary and the degree of difference in strategy types, rather than a preset fixed value, adapting to the spatial heterogeneity of complex terrain. A grid point is the smallest unit (1km × 1km) of the global gridded precipitation data, serving as the smallest object for spatial fusion, weight calculation, and uncertainty assessment. Normalized distance is the actual distance between a grid point and the boundary of the nearest correction area, converted into a dimensionless value in the range of 0-1 (the closer the actual distance, the closer the normalized distance is to 0; the farther away, the closer it is to 1), used to represent the spatial correlation between the grid point and the correction area. The strategy applicability ratio is the ratio of the strategy applicability of the correction area to that of the adjacent non-correction area, representing the difference in reliability between the two types of regional data, used to allocate fusion weights. The preset weight function is a pre-constructed mathematical function that calculates fusion weights based on the normalized distance and the strategy applicability ratio, such as a linear weight function; the closer the grid point is to the correction area and the greater the applicability of the correction strategy, the greater the weight of the correction data.

[0084] In this embodiment, the fusion weight is the weight (0-1) assigned to the data of each grid point in the correction area, with a total weight of 1. The weight of the corrected data + the weight of the uncorrected data = 1, determining the composition ratio of the final precipitation data of the grid point. The terrain accessibility distance is the actual reachable distance between the grid point and the boundaries of each adjacent correction area, taking into account terrain undulations (e.g., mountains, canyons) (distinct from straight-line distance). It represents the impact of terrain on data fusion; this distance is more consistent with real-world scenarios in complex terrain. The strategy compatibility matrix is ​​a pre-constructed matrix (element values ​​0-1) characterizing the degree of adaptation between different correction strategy types. The closer the element value is to 1, the stronger the compatibility between the two strategies, and the smoother the data connection during fusion. The comprehensive weight distribution, when a grid point is adjacent to multiple correction areas, calculates the weight ratio of each correction area data at that grid point based on the terrain accessibility distance, strategy applicability, and strategy compatibility matrix, ensuring the rationality of multi-source correction data fusion.

[0085] In this embodiment, the fused precipitation data is obtained after buffer fusion and multi-source correction data fusion. It includes the entire target complex terrain area, fault-free, and spatially continuous precipitation grid data, which forms the basis for generating the final product. The original error intensity is the calculated error quantization value (0-1) of the initial satellite precipitation data for each grid point, and is the basic parameter for uncertainty assessment. The spatial distribution density of the ground validation data is the spatial density of ground validation points (automatic rain gauges, etc.) within the target area (unit: points / ). The higher the density, the more accurate the uncertainty assessment, serving as an important reference for error propagation analysis. Error propagation analysis, based on the original error, correction effect, fusion weight, and validation point density of each grid point, analyzes the transmission and superposition of errors throughout the correction-fusion process, representing the uncertainty of the final precipitation estimate. The comprehensive uncertainty index, calculated through error propagation analysis, is a dimensionless value (0-1) representing the uncertainty of the final precipitation estimate for each grid point. The closer the value is to 0, the more reliable the precipitation estimate and the lower the uncertainty. The preset data quality levels are pre-defined based on the comprehensive uncertainty index, such as levels 1-5, with level 1 (uncertainty 0-0.2) being the best and level 5 (uncertainty 0.8-1.0) being the worst, used to visually present the reliability of the grid data.

[0086] As can be seen from the above, this embodiment adaptively determines the width of the fusion buffer, making the buffer setting more closely match the actual terrain and strategy differences, and better transitioning between corrected and uncorrected areas; calculating the fusion weight based on the ratio of normalized distance and strategy applicability, and performing multi-source data fusion by calculating the comprehensive weight distribution of adjacent grid points in multiple corrected areas, can improve the accuracy and rationality of data fusion; calculating the comprehensive uncertainty index through error propagation analysis and mapping it to the data quality level can generate a more scientific and accurate uncertainty assessment report, improving the quality and reliability of the final gridded fused precipitation product.

[0087] In one embodiment of this application, after generating the final gridded fused precipitation product, the method further includes: Obtain ground station observation data during the validation period, independent of the model training and calibration process; The ground station observation data during the verification period were compared point by point with the final gridded fused precipitation products of the corresponding spatiotemporal locations; Calculate the evaluation indicators at the watershed and station scales. The evaluation indicators should include at least the root mean square error, mean absolute error, correlation coefficient, and bias. Based on the results of the evaluation indicators, a final accuracy verification report for the gridded fused precipitation product is generated.

[0088] In this embodiment, the ground station observation data during the validation period (hereinafter referred to as validation period ground data), which is independent of the model training and calibration process, is measured precipitation data from ground rain gauges and automatic weather stations specifically used for product accuracy validation. Its characteristics include independence and lack of correlation, ensuring the objectivity and impartiality of the validation results and avoiding validation bias caused by data overfitting. The validation period is a pre-defined time range used for product accuracy validation (not overlapping with the model training / calibration time range), for example, the training / calibration period is the last 29 years, and the validation period is the 30th year, ensuring the independence and representativeness of the validation. The ground station observation data is measured precipitation data from ground automatic rain gauges and regional meteorological observation stations, including station latitude and longitude, observation time, and measured precipitation (unit: mm / h), etc., and is the gold standard for evaluating product accuracy (based on the true value). The corresponding spatiotemporal location is the time + space of the ground station during the validation period, perfectly matching the grid cells of the final gridded precipitation product. Spatially, the grid cell where the station is located is found; temporally, the observation time is consistent with the product's temporal resolution (e.g., 1 hour), achieving point-to-point and time-to-time comparison. Point-by-point comparison involves comparing the measured precipitation data from each ground station during the verification period with the gridded fused precipitation product data of the corresponding time in the grid where the station is located, one by one, to obtain the deviation data for each station / grid, which is the basis for the calculation of subsequent evaluation indicators.

[0089] In this embodiment, the watershed scale uses the complete geographical boundary of the target complex terrain region (e.g., a watershed in the western Sichuan Plateau) as the scope, summarizing and calculating the comparative data of all verification stations within the entire watershed to evaluate the overall accuracy performance of the product across the entire watershed. The station scale uses a single ground station during the verification period as the unit, calculating evaluation indicators separately for each station's time-series and grid-by-grid comparative data to evaluate the local accuracy performance of the product at a single station location (enhancing the product's spatial heterogeneity adaptability). The evaluation indicators are used to represent the degree of deviation and correlation between the gridded fused precipitation product and the ground-measured data, objectively representing the product's accuracy.

[0090] In this embodiment, the root mean square error (RMSE) is an indicator (unit: mm / h) for evaluating the degree of deviation between the product data and the measured data. It is calculated by taking the square root of the sum of squares of the deviations and the average value. The smaller the value, the smaller the deviation between the product data and the measured data, and the greater the accuracy. The mean absolute error (MAE) is an indicator (unit: mm / h) for evaluating the average deviation between the product data and the measured data. It is calculated by averaging the absolute values ​​of the deviations. The smaller the value, the smaller the average deviation of the product, and the better the stability. The correlation coefficient is an indicator (dimensionless, ranging from 0 to 1) for evaluating the linear correlation between the product data and the measured data. The closer the value is to 1, the stronger the linear correlation between the product data and the measured data, and the better the product can represent the true precipitation distribution pattern. The deviation is an indicator (unit: mm / h, can be positive or negative) for evaluating the systematic deviation of the product. It is calculated by the difference between the average value of the product data and the average value of the measured data. A positive value indicates that the product data is generally too large, and a negative value indicates that the data is generally too small. The smaller the absolute value, the smaller the systematic deviation. The accuracy verification report is a supporting report generated based on the calculation results of the evaluation indicators. It is used to represent the accuracy level, deviation characteristics, and applicable scope of the final gridded fused precipitation product. It serves as proof of the product's reliability and needs to be output to the product users simultaneously.

[0091] As can be seen from the above, this embodiment can effectively verify the accuracy of the final gridded fused precipitation product by comparing the ground station observation data during the verification period with the final gridded fused precipitation product point by point, calculating the watershed and station-scale evaluation indicators and generating an accuracy verification report.

[0092] In one embodiment of this application, a correction algorithm is used to correct satellite precipitation data in the correction area based on the correction strategy type and error spatial structure parameters, including: The satellite remote sensing inversion precipitation data within the correction area is used as the background field for the ensemble Kalman filter algorithm; The local precipitation true field calculated based on ground verification point data is used as the observation field; Using the adjusted observation error covariance matrix and the background field error covariance matrix determined based on the spatial autocorrelation of satellite precipitation data, an ensemble Kalman filter analysis is performed to update the data. The output is the updated precipitation field, which serves as the corrected satellite precipitation data for the correction area.

[0093] In this embodiment, the satellite remote sensing inverted precipitation data is precipitation grid data obtained by satellite sensors and calculated using an inversion algorithm. It is the original source of satellite precipitation data for the correction area. The background field of the ensemble Kalman filter algorithm is the prior estimate data to be corrected. In this embodiment, it refers to the original satellite remote sensing inverted precipitation data within the correction area, representing the precipitation estimate before algorithm correction, including systematic bias and random noise. The local precipitation ground truth field (observation field) is high-precision precipitation grid data (with the same grid resolution as the satellite precipitation data, 1km×1km) obtained by spatial interpolation based on ground verification point data within the correction area. It represents the true distribution of precipitation within the correction area. The adjusted observation error covariance matrix (adjusted R matrix) is calculated based on the algorithm parameters adjusted according to the local observation noise level (determined by the ratio of nugget value to sill value). It represents the error distribution of the observation field (local precipitation ground truth field), with the diagonal elements corresponding to the error variance of each observation point to adapt to the noise characteristics of the correction area.

[0094] In this embodiment, the spatial autocorrelation of satellite precipitation data is the degree of spatial correlation (dimensionless, 0-1) between satellite precipitation data, i.e., the similarity of precipitation data at adjacent grid points. The greater the similarity, the stronger the spatial autocorrelation. This is characterized by the range in the error spatial structure parameters; the greater the range, the wider the autocorrelation range. The background field error covariance matrix represents the error distribution characteristics of the background field (original satellite precipitation data) (e.g., spatial correlation and variance of errors), determined based on the spatial autocorrelation of satellite precipitation data (according to the error spatial structure parameters), enabling the algorithm to adapt to the error spatial distribution of complex terrain. The ensemble Kalman filter analysis update is based on the background field, the observation field, the adjusted observation error covariance matrix, and the background field error covariance matrix. Through Kalman gain calculation, the true information of the observation field is integrated into the background field, correcting the errors in the background field to obtain posterior estimated data (corrected precipitation data), achieving error correction and data fusion. The updated precipitation field is the posterior estimated precipitation data output after analysis and update, i.e., the result of error correction of satellite precipitation data within the correction area, eliminating most systematic biases and random noise. The corrected satellite precipitation data is based on the analysis of the updated precipitation field. After quality control (removing outliers and ensuring spatial continuity), the resulting standardized satellite precipitation grid data covers the entire correction area and meets the required accuracy.

[0095] As can be seen from the above, this embodiment uses the satellite remote sensing inversion precipitation data within the correction area as the background field of the ensemble Kalman filter algorithm, and the local precipitation true field calculated based on ground verification point data as the observation field. By using the adjusted observation error covariance matrix and the background field error covariance matrix determined according to the spatial autocorrelation of satellite precipitation data to perform ensemble Kalman filter analysis and update, the corrected satellite precipitation data can be output, improving the accuracy of satellite precipitation data. Correction is performed according to the correction strategy type and error spatial structure parameters, which can better adapt to precipitation conditions under complex terrain, and generate more accurate gridded fused precipitation products and corresponding uncertainty assessment reports.

[0096] Corresponding to the satellite precipitation error correction and fusion method based on elevation gradient under complex terrain in the above embodiment, Figure 2 This is a structural block diagram of a satellite precipitation error correction and fusion system based on elevation gradient under complex terrain, provided as an embodiment of this application. For ease of explanation, only the parts relevant to the embodiment of this application are shown. References Figure 2 The satellite precipitation error correction and fusion system 20 based on elevation gradient under complex terrain includes: a data acquisition module 21, an error diagnosis module 22, a region identification module 23, a feature extraction module 24, a strategy determination module 25, a region correction module 26, a data calculation module 27, and a correction fusion module 28.

[0097] Among them, the data acquisition module 21 is used to acquire multi-source precipitation observation data of the target complex terrain area, perform spatiotemporal matching and quality control on the multi-source precipitation observation data, and obtain standardized grid precipitation data. Error diagnosis module 22 is used to input standardized grid precipitation data into a preset elevation-error correlation model to obtain an error intensity distribution map of the potential error level of satellite precipitation data in each grid unit of the target complex terrain area; The region identification module 23 is used to identify grid regions with error intensity values ​​greater than or equal to a preset error threshold based on the error intensity distribution map, as uncertain regions; Feature extraction module 24 is used to extract multi-dimensional features from uncertain regions to obtain regional feature vectors; The strategy determination module 25 is used to input the regional feature vector into the preset error correction strategy selection model to determine the type of correction strategy for the uncertain region and the corresponding strategy applicability. The region correction module 26 is used to obtain the correction region that needs to be deeply fused from the uncertain region based on the correction strategy type and the corresponding strategy applicability. The data calculation module 27 is used to acquire ground verification point data in the calibration area and calculate the local precipitation true field and error spatial structure parameters based on the ground verification point data. The calibration and fusion module 28 is used to correct satellite precipitation data in the calibration area based on the calibration strategy type and error spatial structure parameters, and to spatially fuse the satellite precipitation data in the calibration area with the satellite precipitation data in the uncalibrated area to generate the final gridded fused precipitation product and the corresponding uncertainty assessment report.

[0098] See Figure 3 , Figure 3 This is a schematic block diagram of an electronic device provided according to an embodiment of this application. Figure 3 The electronic device 300 in this embodiment may include one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memories 304 store computer programs, including program instructions. The processors 301 execute the program instructions stored in the memories 304. Specifically, the processors 301 are configured to invoke the program instructions to perform the functions of the modules in the aforementioned device embodiments, for example... Figure 2 The functions of the data acquisition module 21, error diagnosis module 22, region identification module 23, feature extraction module 24, strategy determination module 25, region correction module 26, data calculation module 27, and correction fusion module 28 are shown.

[0099] It should be understood that, in the embodiments of this application, the processor 301 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0100] Input device 302 may include a touchpad, a fingerprint sensor (for collecting the user's fingerprint information and fingerprint orientation information), a microphone, etc., and output device 303 may include a display (LCD, etc.), a speaker, etc.

[0101] The memory 304 may include read-only memory and random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include non-volatile random access memory. For example, the memory 304 may also store device type information.

[0102] In specific implementations, the processor 301, input device 302, and output device 303 described in the embodiments of this application can execute the implementation methods described in any embodiment of the satellite precipitation error correction and fusion method based on elevation gradient under complex terrain provided in the embodiments of this application, or they can execute the implementation methods of the electronic devices described in the embodiments of this application, which will not be repeated here.

[0103] In another embodiment of this application, a computer-readable storage medium is provided. This computer-readable storage medium stores a computer program, which includes program instructions. When executed by a processor, the program instructions implement all or part of the processes in the methods described above. Alternatively, the computer program can instruct related hardware to complete the process. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0104] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the foregoing embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the electronic device. Furthermore, the computer-readable storage medium can include both internal and external storage units of the electronic device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0105] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0106] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the electronic devices and units described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0107] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces or units, or it may be an electrical, mechanical, or other form of connection.

[0108] The units described as separate components may or may not be physically separate. 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 units can be selected to achieve the purpose of the embodiments of this application, depending on actual needs.

[0109] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0110] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for satellite precipitation error correction and fusion based on elevation gradient under complex terrain, characterized in that, include: Acquire multi-source precipitation observation data of the target complex terrain area, perform spatiotemporal matching and quality control on the multi-source precipitation observation data, and obtain standardized grid precipitation data; The standardized grid precipitation data is input into a preset elevation-error correlation model to obtain an error intensity distribution map of the potential error level of satellite precipitation data in each grid unit of the target complex terrain area; Based on the error intensity distribution map, grid regions with error intensity values ​​greater than or equal to a preset error threshold are identified as uncertainty regions. Multi-dimensional feature extraction is performed on the uncertain region to obtain the region feature vector; The feature vector of the region is input into a preset error correction strategy selection model to determine the type of correction strategy and the corresponding applicability of the strategy for the uncertain region. Based on the correction strategy type and the corresponding strategy applicability, the correction region that needs to be deeply fused from multiple sources is obtained from the uncertainty region; Obtain ground verification point data for the calibration area, and calculate the local precipitation true field and error spatial structure parameters based on the ground verification point data; Based on the correction strategy type and the error spatial structure parameters, the satellite precipitation data of the correction area is corrected using a correction algorithm, and the satellite precipitation data of the correction area is spatially fused with the satellite precipitation data of the uncorrected area to generate the final gridded fused precipitation product and the corresponding uncertainty assessment report.

2. The satellite precipitation error correction and fusion method based on elevation gradient under complex terrain as described in claim 1, characterized in that, The step of extracting multi-dimensional features from the uncertain region to obtain a region feature vector includes: Based on the high-resolution digital elevation model data of the aforementioned uncertain region, extract terrain feature sub-vectors; Based on the long-term climate statistical grid dataset and climate zoning map of the target complex terrain region, extract climate background feature sub-vectors; Weather feature sub-vectors are extracted from atmospheric reanalysis grid data that are spatiotemporally matched with the uncertain region. The terrain feature sub-vector, climate background feature sub-vector, and weather feature sub-vector are weighted and fused to form the regional feature vector.

3. The method for satellite precipitation error correction and fusion based on elevation gradient under complex terrain as described in claim 2, characterized in that, The step of weightedly fusing the terrain feature sub-vector, climate background feature sub-vector, and weather feature sub-vector to form the regional feature vector includes: The terrain feature subvectors are evaluated based on the terrain complexity index of the region corresponding to the terrain feature subvectors and the data source resolution to generate terrain feature confidence scores. The climate feature subvectors are evaluated based on the temporal length and spatial representativeness of the historical climate data upon which they depend, and climate feature confidence scores are generated. The weather feature subvectors are evaluated based on the spatiotemporal resolution of the atmospheric reanalysis data and the density of assimilated observations to generate weather feature confidence scores. Based on the confidence scores of terrain features, climate features, and weather features, the weight allocation of the terrain feature sub-vectors, climate background feature sub-vectors, and weather feature sub-vectors in the fusion process is determined. Based on the assigned weights, the terrain feature sub-vector, climate background feature sub-vector, and weather feature sub-vector are weighted and fused to generate the regional feature vector.

4. The method for satellite precipitation error correction and fusion based on elevation gradient under complex terrain as described in claim 1, characterized in that, The step of obtaining the correction region requiring deep multi-source data fusion from the uncertainty region based on the correction strategy type and corresponding strategy applicability includes: The correction strategy type is matched with a preset strategy type level library to determine the basic screening threshold corresponding to each strategy type. Compare the applicability of the strategy with the basic screening threshold for the corresponding strategy type; For any uncertain region, if the applicability of the strategy is less than the basic screening threshold of the corresponding strategy type, then the uncertain region is marked as a candidate correction region. Based on the spatial distribution characteristics of the candidate correction regions, a region clustering operation is performed to merge candidate correction regions whose spatial distance is less than a preset adjacency threshold and whose correction strategy types are the same or compatible into contiguous regions. Based on the area and geographical integrity of the merged contiguous areas, boundary optimization is performed on the contiguous areas that meet the preset area threshold conditions to generate the final correction area that requires deep fusion of multi-source data.

5. The satellite precipitation error correction and fusion method based on elevation gradient under complex terrain as described in claim 4, characterized in that, The method for determining the preset adjacency threshold includes: The basic value of the adjacency threshold is determined based on the average elevation and topographic relief of the target complex terrain area; Obtain the climate type and precipitation variability of the candidate correction area, and correct the base value of the adjacency threshold based on climate stability to obtain the corrected adjacency threshold. Based on the requirements of the correction strategy type for data spatial continuity, the corrected adjacency threshold is adjusted a second time to obtain the final preset adjacency threshold.

6. The satellite precipitation error correction and fusion method based on elevation gradient under complex terrain as described in claim 4, characterized in that, The step of optimizing the boundaries of contiguous areas that meet a preset area threshold condition based on the area and geographical integrity of the merged contiguous areas includes: Based on morphological image processing methods, a dilation-erosion closing operation is performed on the boundary of the contiguous region to fill the small holes inside the region and smooth the irregular boundary, thus obtaining the basic optimized boundary. Extract digital elevation model data of the location of the contiguous area and calculate the terrain gradient change; The basic optimization boundary is compared with the location where the terrain gradient changes drastically. If the distance between the basic optimization boundary and the peak location of the terrain gradient is less than a preset distance threshold, the basic optimization boundary is adjusted to the peak location of the terrain gradient. The boundary of the correction area is aligned with the terrain features to obtain the adjusted boundary. Based on the adjusted boundary, a smoothing optimization process is performed to complete the boundary optimization.

7. The method for satellite precipitation error correction and fusion based on elevation gradient under complex terrain as described in claim 1, characterized in that, The correction algorithm employs an ensemble Kalman filter algorithm; before correcting the satellite precipitation data of the correction area using the correction algorithm based on the correction strategy type and the error spatial structure parameters, the method further includes: The local observation noise level of the correction region is determined based on the correction strategy type and the ratio of nugget value to sill value in the error space structure parameters. The observation error covariance matrix of the ensemble Kalman filter algorithm is adjusted based on the local observation noise level. If the ratio of the nugget value to the sill value is greater than a preset ratio threshold, it is determined that the local observation noise level is high, and the value of the diagonal element in the observation error covariance matrix is ​​increased based on the first coefficient. If the ratio of the nugget value to the sill value is less than a preset ratio threshold, the local observation noise level is determined to be low, and the value of the diagonal element in the observation error covariance matrix is ​​reduced based on the second coefficient.

8. The method for satellite precipitation error correction and fusion based on elevation gradient under complex terrain as described in claim 1, characterized in that, The process of spatially fusing satellite precipitation data from the corrected area with that from the uncorrected area to generate a final gridded fused precipitation product and a corresponding uncertainty assessment report includes: A fusion buffer is established centered on the boundary of the correction region. The width of the buffer is adaptively determined based on the terrain gradient change rate between adjacent correction and non-correction regions and the degree of difference between the correction strategy types. Within the fusion buffer, the normalized distance between each grid point and the boundary of the nearest correction region is calculated, and based on the policy applicability ratio of the regions on both sides of the boundary of the nearest correction region, the fusion weight of the corresponding grid point using the correction data is calculated through a preset weight function. For grid points adjacent to multiple correction areas, the comprehensive weight distribution of data from different correction areas is calculated based on the terrain accessibility distance to the boundary of each correction area, the policy applicability of each area, and the policy compatibility matrix. Multi-source data fusion is then performed to obtain precipitation data after global fusion. Based on the precipitation data after the whole domain fusion, the comprehensive uncertainty index of the final precipitation estimate of each grid point is calculated by error propagation analysis according to the original error intensity, the applicability of the correction strategy, the fusion weight coefficient and the spatial distribution density of the ground verification data. The comprehensive uncertainty index is mapped to a preset data quality level, and a corresponding uncertainty assessment report is generated.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images