High-precision surface precipitation deviation correction method based on space-time correlation

Abstract

The invention discloses a high-precision surface rainfall deviation correction method based on space-time association. The method comprises the steps of evaluating satellite remote sensing rainfall data estimation precision and modeling a space-time association relationship; and space-time association relationship modeling: analyzing space-time error distribution of the long-time-sequence satellite remote sensing rainfall data, performing uncertainty modeling on a satellite rainfall estimation result by using a probability density distribution function, and providing data uncertainty information for rainfall deviation correction. And space-time association relationship modeling: researching a modeling method of the space-time association relationship of the satellite-ground data, and constructing space-time association models between the ground observation sites and the corresponding remote sensing pixels and between the pixels. Meanwhile, environment variables, precipitation deviation time sequence change characteristics and uncertainty information are introduced, and the surface precipitation deviation correction precision is improved. Through the research, the estimation error of the satellite remote sensing surface rainfall data can be effectively reduced, and the accuracy of deviation correction is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of precipitation prediction, and in particular relates to a high-precision surface precipitation deviation correction method based on spatiotemporal correlation. Background Art

[0002] With the development of sky and ground observation technology, precipitation monitoring has entered a new stage of multi-source data fusion. In high-altitude and complex terrain areas, precipitation prediction faces the problems of data sparsity and bias correction. The current bias correction method relies on ground observation data and statistical regression relationships, but its effect is limited in a changing environment. In order to improve the accuracy of precipitation estimation, researchers combine remote sensing images with meteorological data and use satellite-ground fusion algorithms to improve data accuracy. Machine learning and spatiotemporal clustering methods have also been introduced to mine the spatial information of precipitation characteristics and improve the effect of bias correction. Although some progress has been made, there is still a need to improve the existing time series clustering algorithm to better deal with the spatial and temporal heterogeneity of precipitation. Summary of the invention

[0003] In order to solve the problem that the traditional satellite-ground fusion surface precipitation deviation correction method in the background technology relies on the spatial layout density of ground in-situ sensor meteorological observation stations and lacks consideration of the spatiotemporal changes of precipitation and terrain distribution characteristics, the present invention focuses on the current situation that the ground stations in the river basins in complex terrain are sparsely distributed and the daily scale estimation error of satellite remote sensing surface precipitation data is large. The purpose is to provide a high-precision surface precipitation deviation correction method based on spatiotemporal correlation, and quantify the spatiotemporal correlation relationship of multi-source data in the consistent area of ​​surface precipitation feature space to achieve high-precision surface precipitation deviation correction. It mainly includes: studying the estimation accuracy evaluation and spatiotemporal uncertainty analysis of multi-source satellite remote sensing surface precipitation data, analyzing the spatiotemporal distribution of errors in long-time series satellite remote sensing surface precipitation data, and studying the use of probability density distribution function to achieve uncertainty modeling of satellite remote sensing precipitation estimation results, providing data uncertainty information for subsequent precipitation bias correction; studying the modeling method of the spatiotemporal correlation between satellite and ground data in areas with consistent precipitation characteristics, constructing a spatiotemporal correlation model between ground observation sites and remote sensing pixels at corresponding spatial positions (sites and pixels) and between pixels, and introducing environmental variables, information on the temporal changes in precipitation biases, and uncertainty information, to further achieve high-precision bias correction of surface precipitation.

[0004] In order to solve the technical problem, the technical solution of the present invention is:

[0005] A high-precision surface precipitation deviation correction method based on spatiotemporal correlation, the method comprising:

[0006] S1: Construct a satellite-ground multi-source dataset for the study area, including precipitation data from ground meteorological observation sites and multi-source satellite remote sensing precipitation data, and perform data preprocessing. Then, extract the pixel values ​​of remote sensing images of ground meteorological observation sites and their corresponding spatial locations.

[0007] S2: Based on the principal component analysis method, quantitative calculation is performed on the contribution of natural changes and human activity-related factors affecting long-term precipitation, and the factors with high contribution and significant correlation are screened;

[0008] S3: Based on the screened high-contribution influencing factors, a consistent precipitation feature space is constructed that considers the spatial attributes of surface precipitation, trend changes of non-spatial attributes, and values ​​of non-spatial attributes;

[0009] S4: Use the spatiotemporal correlation relationship to carry out surface precipitation deviation correction of co-located, spatially correlated and spatially adjacent pixels in areas with consistent precipitation characteristics, so as to achieve high-precision deviation correction of surface precipitation.

[0010] Further, the step S1 comprises:

[0011] Preprocessing the precipitation data of the ground meteorological observation station includes outlier processing, missing value processing and standardization processing; reprojecting the multi-source satellite remote sensing precipitation data, extracting by mask and resampling; and then extracting the pixel values ​​of the remote sensing images of the ground meteorological observation station and its corresponding spatial position.

[0012] Furthermore, based on the remote sensing image pixel values ​​extracted in step S1 and the preprocessed data, the principal component analysis method is used to quantitatively calculate the contribution of natural changes and human activity-related factors that affect long-term precipitation, and screen out influencing factors with high contribution and significant correlation.

[0013] Further, the step S3 specifically includes:

[0014] Step S301: construct an initial precipitation feature space that considers the similarity of spatial attributes; use the fuzzy C-means spatial clustering method to identify the spatial distribution characteristics of precipitation and its influencing factors, and perform spatial clustering on pixels with similar characteristics. In the spatial clustering process, the model input variables are precipitation and precipitation influencing factors with higher contribution after principal component analysis screening. After spatial clustering, if the pixels are not in the same clustering space, they do not belong to the space with consistent precipitation change characteristics; if the pixels are in the same clustering space, mark the clustering space index to which the pixel belongs;

[0015] Step S302: Construct a precipitation feature space that considers the consistency of non-spatial attribute trend changes; based on the spatial clustering results constructed in step S301, in each clustering space, the remote sensing image grid pixel corresponding to the ground meteorological observation station, i.e., the station pixel, is used as the starting node C 0 The Sen slope and Mann-Kendall trend analysis methods were used to analyze the historical rainfall trend changes of the station pixel and other pixels, and the pixels with the station pixel C were screened. 0 Image pixels with consistent rainfall trend changes;

[0016] Step S303: Consider the correlation of non-spatial attribute values ​​and construct an initial spatial correlation path; construct a pixel spatial correlation path, each of which has a starting node, a process node, and an end node. In the precipitation trend change consistent space constructed in step S302, C 0 The pixel is the starting node, and the pixels within the neighborhood space of the pixel are the initial ending nodes C (1，i) , connect the starting node and the ending node to obtain the initial path starting from the station pixel;

[0017] Step S304: Initial end node C (1) As the new starting node, continue to search for spatial neighboring points related to precipitation statistics within the 8-neighborhood range, and this neighboring point is the new ending node C. (2) , the determination of new nodes must meet the significant correlation test conditions. Among the candidate pixels, the Pearson correlation coefficient PCC and the corresponding significance test p value are used to realize the determination of significantly correlated adjacent points within the 8-neighborhood range;

[0018] The PCC calculation formula is as follows:

[0019]

[0020] Where n is the number of sample pairs in the dataset, here is the number of historical multi-temporal precipitation estimates of remote sensing images at the station pixel, and x i and i Represent the pixel values ​​of the station pixel and the other pixels at the i-th moment, respectively. is the sample mean of the data set x: is the sample mean of the data set y:

[0021] The criteria for determining significantly correlated adjacent pixels are shown in formula (2). The critical values ​​of PCC and p are set to 0.5 and 0.05, respectively. When the absolute value of the correlation coefficient is greater than or equal to 0.5 and the p value is less than 0.05, the two sets of data are statistically significantly correlated.

[0022] |PCC|≥0.5and p<0.05 (2)

[0023] Assuming that each terminal node has n significantly correlated adjacent pixels within the 8-neighborhood range, the adjacent pixel with the largest correlation coefficient PCC value is added to the spatial correlation path as the next starting node C (2) ; After that, continue to look for the next adjacent pixel C with the largest correlation coefficient (3) , repeat this step until there are no new nodes added to the spatial correlation path, and finally establish a rainfall-significantly correlated spatial connectivity path;

[0024] Step S305: The spatial area occupied by the remote sensing image pixels on the spatially connected path significantly related to rainfall is the spatially consistent area of ​​precipitation characteristics that integrates spatial attributes, non-spatial attribute trend changes, and non-spatial attribute values.

[0025] Further, the step S4 specifically includes:

[0026] Step S401: combining variational mode decomposition (VMD) and multivariate random forest regression algorithm (MRFR) to correct precipitation deviations in co-located pixels;

[0027] Firstly, the precipitation deviation between the true value of precipitation at the ground meteorological observation station and the raster pixel of the remote sensing image at the corresponding spatial position is calculated to obtain the precipitation deviation time series data set; then, the precipitation deviation time series is decomposed by the variational mode decomposition algorithm to obtain K deviation modal components; multivariate random forest regression modeling is performed on the K deviation modal components respectively to obtain the quantitative correlation between the deviation value and the satellite remote sensing image precipitation estimation value, where the input variables of the multivariate random forest regression model are the precipitation deviation modal components and the relevant driving factors affecting precipitation; the deviation output results of the multivariate random forest regression of all modal components are integrated to reconstruct the precipitation deviation between the satellite remote sensing image and the ground observation station; the precipitation deviation result is added to the satellite remote sensing image estimation value to obtain the surface precipitation deviation correction result of the same-position associated pixels;

[0028] Step S402: spatial correlation pixel surface precipitation deviation correction;

[0029] Based on the result of the consistent spatial construction of the spatiotemporal variation characteristics of precipitation and its influencing factors in step S3, except for the station pixels, the remaining pixels in the feature space are all spatially correlated pixels to be corrected. According to the order of nodes on the spatially correlated path, the multivariate random forest regression algorithm is used to establish mathematical relationships in sequence to realize the correction of the surface precipitation deviation of spatially correlated pixels. The multivariate random forest regression surface precipitation deviation correction is shown in formulas (3) and (4):

[0030]

[0031]

[0032] Among them, C (i-1) and C (i) They represent the historical values ​​of the remote sensing image pixels before deviation correction at the i-1th and i-th nodes on the spatial correlation path, and represents the deviation correction result value at the i-1th and ith nodes, C (i-1) and C (i)The precipitation driving factors at the i-1th and i-th nodes include environmental variables such as temperature, evapotranspiration, elevation, and precipitation time series variation characteristics and uncertainty information. i represents the decision tree established between the historical pixel values ​​of the remote sensing image before the deviation correction at the current node and the previous node on the spatial correlation path, and n is the number of decision trees; wherein the station pixel is used as the starting node of the spatial path, and the precipitation deviation correction result at the starting node is the same-position correlation deviation correction result in step S401;

[0033] Step S403: Correction of precipitation deviation of spatially adjacent pixels;

[0034] In the area where rainfall characteristics are spatially inconsistent, the remaining pixels are spatially neighboring pixels. In the process of correcting the surface precipitation deviation of spatially neighboring pixels, the inverse distance weighted interpolation method IDW is used. Based on the surface precipitation deviation correction results of the co-located association in step S401 and the spatially correlated pixels in step S402, the surface precipitation deviation correction of the remaining pixels is realized.

[0035] Compared with the prior art, the advantages of the present invention are:

[0036] The method for high-precision correction of surface precipitation deviation proposed in the present invention firstly, spatial attributes, trend changes of non-spatial attributes and characteristic information of non-spatial attribute values ​​shown by spatiotemporal clustering of precipitation affected by natural changes and human activities are used to divide the spatially consistent regions of precipitation characteristics; secondly, within the spatially consistent regions of precipitation characteristics, spatiotemporal correlation models of precipitation between ground observation sites and image pixels, and between image pixels are constructed from three levels of co-location association, spatial correlation and spatial proximity, thus breaking through the limitations of the surface precipitation deviation correction method based on a small number of ground observation sites and a large number of image pixels, and realizing high-precision deviation correction of satellite remote sensing surface precipitation products. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 , implementation process framework diagram;

[0038] Figure 2 , a flowchart of the construction of a space with consistent precipitation characteristics;

[0039] Figure 3 , rainfall significant correlation spatial connectivity path map;

[0040] Figure 4 , Schematic diagram of the correction process of precipitation deviations in co-located pixels;

[0041] Figure 5 , flow chart for correction of spatially correlated pixel surface precipitation deviations. DETAILED DESCRIPTION

[0042] The specific implementation mode of the present invention is described below in conjunction with embodiments:

[0043] It should be noted that the structures, proportions, sizes, etc. shown in this specification are only used to match the contents disclosed in the specification so that people familiar with this technology can understand and read them, and are not used to limit the conditions under which the present invention can be implemented. Any structural modification, change in proportional relationship or adjustment of size should still fall within the scope of the technical content disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0044] At the same time, the terms such as "upper", "lower", "left", "right", "middle" and "one" cited in this specification are only for the convenience of description and are not used to limit the scope of implementation of the present invention. Changes or adjustments to their relative relationships should be regarded as the scope of implementation of the present invention without substantially changing the technical content.

[0045] Embodiment 1:

[0046] The present invention provides a method for high-precision correction of surface precipitation deviation in a spatially consistent region of precipitation characteristics based on spatiotemporal association. First, the satellite-ground multi-source data set of the study area is collected and sorted and preprocessed, and the correlation analysis and screening of long-term precipitation influencing factors based on principal component analysis are performed. Then, a surface precipitation characteristic consistent space considering the surface precipitation spatial attributes, non-spatial attribute trend changes and non-spatial attribute values ​​is constructed. Finally, surface precipitation deviation correction of co-localized, spatially correlated and spatially adjacent pixels is carried out in the spatially consistent region of precipitation characteristics to achieve high-precision deviation correction of surface precipitation. The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0047] like Figure 1 As shown, the process of the embodiment of the present invention includes the following steps:

[0048] Step 1: Construct a satellite-ground multi-source dataset for the study area, mainly including ground meteorological observation site data (such as precipitation, temperature, evapotranspiration, etc.), satellite remote sensing data (such as precipitation data CHIRPS, MSWEP, PERSIANN-CDR, GSMaP, etc., soil moisture SMAP, vegetation data MODIS) and terrain environment data (such as land use type data, DEM data, etc.), and perform data preprocessing before model input.

[0049] Step 1.1, ground meteorological observation site data preprocessing, including outlier processing, missing value processing and standardization processing.

[0050] Firstly, the box plot method is used to identify and delete outliers in the time series data set of ground meteorological observation stations;

[0051] Secondly, interpolation methods are used to fill in missing values ​​in the time series dataset of ground meteorological observation sites;

[0052] Finally, the time series data set of ground meteorological observation stations is converted into a standard normal distribution (with a mean of 0 and a standard deviation of 1) through standardization.

[0053] Step 1.2, remote sensing image reprojection, mask extraction, and resampling.

[0054] First, all satellite remote sensing data were reprojected to the 2000 National Geodetic Coordinate System;

[0055] Secondly, all satellite remote sensing data were extracted to the size of the study area according to the mask;

[0056] Finally, all satellite remote sensing data were resampled to a spatial resolution of 0.1°.

[0057] Step 1.3, extract pixel values ​​of remote sensing images of ground meteorological observation stations and their corresponding spatial locations.

[0058] Step 2: Based on the principal component analysis method, quantitative calculation is performed on the contribution of factors related to natural changes (such as temperature, evapotranspiration, soil moisture, vegetation index, etc.) and human activities (such as land use type, etc.) that affect long-term precipitation, and the factors with high contribution and significant correlation are screened;

[0059] Step 3, such as Figure 2 As shown, a consistent precipitation feature space considering multiple attribute characteristics is constructed;

[0060] Step 3.1, construct the initial precipitation feature space considering the similarity of spatial attributes. The fuzzy C-means spatial clustering method is used to identify the spatial distribution characteristics of precipitation and its influencing factors, and spatially cluster pixels with similar characteristics. In the spatial clustering process, the model input variables are precipitation and precipitation influencing factors with higher contribution after principal component analysis screening. After spatial clustering, if the pixels are not in the same clustering space, they do not belong to the space with consistent precipitation change characteristics; if the pixels are in the same clustering space, mark the clustering space index to which the pixel belongs.

[0061] Step 3.2: Construct a precipitation feature space that considers the consistency of non-spatial attribute trend changes. Based on the spatial clustering results constructed in step 3.1, in each clustering space, the remote sensing image raster pixel corresponding to the ground meteorological observation station (hereinafter referred to as the station pixel) is used as the starting node C. 0 , the Sen slope and Mann-Kendall trend analysis methods are used to analyze the historical rainfall trend changes of the station pixel and other pixels, and the non-spatial attribute trend change characteristic information of the pixel is marked at the significance level. If the trend is rising, it is marked as 1; if the trend is falling, it is marked as -1; if there is no trend change, it is marked as 0. Screening and station pixel C 0 Image pixels with consistent changes in rainfall trends.

[0062] Step 3.3: Consider the correlation of non-spatial attribute values ​​and construct the initial spatial correlation path. Since there is statistical correlation between pixels and pixels, the pixel spatial correlation path can be constructed. Each path has a starting node, a process node, and an end node. In the consistent precipitation trend change space constructed in step 3.2, C 0 The pixel is the starting node, and the pixels within the neighborhood space of the pixel are the initial ending nodes C (1,i) . Connect the start node and the end node to obtain the initial path starting from the station pixel.

[0063] Step 3.4: Change the initial terminal node C (1) As the new starting node, continue to search for spatial neighboring points related to precipitation statistics within the 8-neighborhood range, and this neighboring point is the new ending node C. (2) , the determination of new nodes must meet the significant correlation test conditions. In the candidate pixels, the determination of significantly correlated adjacent points within the 8-neighborhood range is realized according to formula (2). Each terminal node has n significantly correlated adjacent pixels within the 8-neighborhood range. The adjacent pixel with the largest correlation coefficient PCC value is added to the spatial correlation path as the next starting node C. (2) After that, continue to look for the next adjacent pixel C with the largest correlation coefficient (3) Repeat this step until there are no new nodes added to the spatial correlation path, and finally establish a rainfall-significantly correlated spatial connectivity path, such as Figure 3 shown.

[0064] Step 3.5: The spatial area occupied by the remote sensing image pixels on the spatially connected path with significant rainfall correlation is the spatially consistent area of ​​precipitation characteristics that integrates spatial attributes, trend changes of non-spatial attributes, and non-spatial attribute values.

[0065] Step 4: high-precision correction of surface precipitation deviation in spatially consistent areas of precipitation characteristics based on spatiotemporal correlation;

[0066] Step 4.1, the precipitation bias correction of co-located pixels is performed by combining variational mode decomposition and multivariate random forest regression algorithm (VMD-MRFR), as follows: Figure 4 As shown;

[0067] Firstly, the precipitation deviation between the true value of the ground meteorological observation station and the remote sensing image raster pixel at the corresponding spatial position is calculated to obtain the precipitation deviation time series dataset.

[0068] Then, the precipitation deviation time series is decomposed by the variational mode decomposition algorithm to obtain K deviation modal components.

[0069] Then, multivariate random forest regression modeling was performed on the K deviation modal components to obtain the quantitative correlation between the deviation value and the estimated value of precipitation from satellite remote sensing images. The input variables of the multivariate random forest regression model are the precipitation deviation modal components and the relevant driving factors affecting precipitation (such as temperature, evapotranspiration, elevation, longitude and latitude, and slope, etc.).

[0070] Then, the deviation output results of multivariate random forest regression of all modal components are integrated to reconstruct the precipitation deviation between satellite remote sensing images and ground observation sites.

[0071] Then, the precipitation deviation result is added to the estimated value of the satellite remote sensing image to obtain the surface precipitation deviation correction result of the same-position associated pixels.

[0072] Step 4.2, spatially correlated pixel surface precipitation bias correction;

[0073] Based on the results of the consistent spatial construction of the spatiotemporal variation characteristics of precipitation and its influencing factors in step 3, except for the station pixels, the remaining pixels in the feature space are all spatially related pixels to be corrected. According to the order of nodes on the spatially related path, the multivariate random forest regression algorithm is used to establish mathematical relationships in sequence to realize the correction of precipitation deviation on the spatially related pixel surface, such as Figure 5 shown.

[0074] Step 4.3, correction of precipitation deviation of spatially adjacent pixels;

[0075] In the area where rainfall characteristics are spatially inconsistent, the remaining pixels are spatially adjacent pixels. In the process of correcting the surface precipitation deviation of spatially adjacent pixels, the inverse distance weighted interpolation method is used to correct the surface precipitation deviation of the remaining pixels based on the correction results of the surface precipitation deviation of the co-located pixels in step 4.1 and the spatially correlated pixels in step 4.2.

[0076] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0077] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0078] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0079] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0080] The preferred embodiments of the present invention are described in detail above, but the present invention is not limited to the above embodiments. Various changes can be made within the knowledge of ordinary technicians in the field without departing from the purpose of the present invention. Many other changes and modifications can be made without departing from the concept and scope of the present invention. It should be understood that the present invention is not limited to specific embodiments, and the scope of the present invention is defined by the appended claims.

Claims

1. A high-precision surface precipitation deviation correction method based on spatiotemporal correlation, characterized in that: The method comprises: S1: Construct a satellite-ground multi-source dataset for the study area, including precipitation data from ground meteorological observation sites and multi-source satellite remote sensing precipitation data, and perform data preprocessing. Then, extract the pixel values ​​of remote sensing images of ground meteorological observation sites and their corresponding spatial locations. S2: Based on the principal component analysis method, quantitative calculation is performed on the contribution of natural changes and human activity-related factors affecting long-term precipitation, and the factors with high contribution and significant correlation are screened; S3: Based on the screened high-contribution influencing factors, a consistent precipitation feature space is constructed that considers the spatial attributes of surface precipitation, trend changes of non-spatial attributes, and values ​​of non-spatial attributes; S4: Use the spatiotemporal correlation relationship to carry out surface precipitation deviation correction of co-located, spatially correlated and spatially adjacent pixels in areas with consistent precipitation characteristics, so as to achieve high-precision deviation correction of surface precipitation.

2. The high-precision surface precipitation deviation correction method based on spatiotemporal correlation according to claim 1 is characterized in that: The step S1 comprises: Preprocessing the precipitation data of the ground meteorological observation station includes outlier processing, missing value processing and standardization processing; reprojecting the multi-source satellite remote sensing precipitation data, extracting by mask and resampling; and then extracting the pixel values ​​of the remote sensing images of the ground meteorological observation station and its corresponding spatial position.

3. The high-precision surface precipitation deviation correction method based on spatiotemporal correlation according to claim 1 is characterized in that: Based on the remote sensing image pixel values ​​extracted in step S1 and the preprocessed data, the principal component analysis method is used to quantitatively calculate the contribution of natural changes and human activity-related factors that affect long-term precipitation, and screen out influencing factors with high contribution and significant correlation.

4. The high-precision surface precipitation deviation correction method based on spatiotemporal correlation according to claim 1 is characterized in that: The step S3 specifically includes: Step S301: construct an initial precipitation feature space that considers the similarity of spatial attributes; use the fuzzy C-means spatial clustering method to identify the spatial distribution characteristics of precipitation and its influencing factors, and perform spatial clustering on pixels with similar characteristics. In the spatial clustering process, the model input variables are precipitation and precipitation influencing factors with higher contribution after principal component analysis screening. After spatial clustering, if the pixels are not in the same clustering space, they do not belong to the space with consistent precipitation change characteristics; if the pixels are in the same clustering space, mark the clustering space index to which the pixel belongs; Step S302: construct a precipitation feature space that considers the consistency of non-spatial attribute trend changes; based on the spatial clustering results constructed in step S301, in each clustering space, the remote sensing image grid pixel corresponding to the ground meteorological observation station, that is, the station pixel, is used as the starting node C0, and the Sen slope and Mann-Kendall trend analysis methods are used to analyze the historical rainfall trend changes of the station pixel and the remaining pixels, and the image pixels that are consistent with the rainfall trend changes of the station pixel C0 are selected; Step S303: Consider the correlation of non-spatial attribute values ​​and construct an initial spatial correlation path; construct a pixel spatial correlation path, each path has a starting node, a process node and an end node. In the precipitation trend change consistent space constructed in step S302, the C0 pixel is taken as the starting node, and the pixels within the neighborhood space of the pixel are taken as the initial end node C0. (1,i) , connect the starting node and the ending node to obtain the initial path starting from the station pixel; Step S304: Initial end node C (1) As the new starting node, continue to search for spatial neighboring points related to precipitation statistics within the 8-neighborhood range, and this neighboring point is the new ending node C. (2) , the determination of new nodes must meet the significant correlation test conditions. Among the candidate pixels, the Pearson correlation coefficient PCC and the corresponding significance test p value are used to realize the determination of significantly correlated adjacent points within the 8-neighborhood range; The PCC calculation formula is as follows: Where n is the number of sample pairs in the dataset, here is the number of historical multi-temporal precipitation estimates of remote sensing images at the station pixel, and x i and i Represent the pixel values ​​of the station pixel and the other pixels at the i-th moment, respectively. is the sample mean of the data set x: is the sample mean of the data set y: The criteria for determining significantly correlated adjacent pixels are shown in formula (2). The critical values ​​of PCC and p are set to 0.5 and 0.05, respectively. When the absolute value of the correlation coefficient is greater than or equal to 0.5 and the p value is less than 0.05, the two sets of data are statistically significantly correlated. |PCC|≥0.5 and p<0.05 (2) Assuming that each terminal node has n significantly correlated adjacent pixels within the 8-neighborhood range, the adjacent pixel with the largest correlation coefficient PCC value is added to the spatial correlation path as the next starting node C (2) ; After that, continue to look for the next adjacent pixel C with the largest correlation coefficient (3) , repeat this step until there are no new nodes added to the spatial correlation path, and finally establish a rainfall-significantly correlated spatial connectivity path; Step S305: The spatial area occupied by the remote sensing image pixels on the spatially connected path significantly related to rainfall is the spatially consistent area of ​​precipitation characteristics that integrates spatial attributes, non-spatial attribute trend changes, and non-spatial attribute values.

5. The high-precision surface precipitation deviation correction method based on spatiotemporal correlation according to claim 1 is characterized in that: The step S4 specifically includes: Step S401: combining variational mode decomposition (VMD) and multivariate random forest regression algorithm (MRFR) to correct precipitation deviations in co-located pixels; Firstly, the precipitation deviation between the true value of precipitation at the ground meteorological observation station and the raster pixel of the remote sensing image at the corresponding spatial position is calculated to obtain the precipitation deviation time series data set; then, the precipitation deviation time series is decomposed by the variational mode decomposition algorithm to obtain K deviation modal components; multivariate random forest regression modeling is performed on the K deviation modal components respectively to obtain the quantitative correlation between the deviation value and the satellite remote sensing image precipitation estimation value, where the input variables of the multivariate random forest regression model are the precipitation deviation modal components and the relevant driving factors affecting precipitation; the deviation output results of the multivariate random forest regression of all modal components are integrated to reconstruct the precipitation deviation between the satellite remote sensing image and the ground observation station; the precipitation deviation result is added to the satellite remote sensing image estimation value to obtain the surface precipitation deviation correction result of the same-position associated pixels; Step S402: spatial correlation pixel surface precipitation deviation correction; Based on the result of the consistent spatial construction of the spatiotemporal variation characteristics of precipitation and its influencing factors in step S3, except for the station pixels, the remaining pixels in the feature space are all spatially correlated pixels to be corrected. According to the order of nodes on the spatially correlated path, the multivariate random forest regression algorithm is used to establish mathematical relationships in sequence to realize the correction of the surface precipitation deviation of spatially correlated pixels. The multivariate random forest regression surface precipitation deviation correction is shown in formulas (3) and (4): Among them, C (i-1) and C (i) They represent the historical values ​​of the remote sensing image pixels before deviation correction at the i-1th and i-th nodes on the spatial correlation path, respectively. and represents the deviation correction result value at the i-1th and i-th nodes, G (i-1) and G (i) The precipitation driving factors at the i-1th and i-th nodes include environmental variables such as temperature, evapotranspiration, elevation, and precipitation time series variation characteristics and uncertainty information. i represents the decision tree established between the historical pixel values ​​of the remote sensing image before the deviation correction at the current node and the previous node on the spatial correlation path, and n is the number of decision trees; wherein the station pixel is used as the starting node of the spatial path, and the precipitation deviation correction result at the starting node is the same-position correlation deviation correction result in step S401; Step S403: Correction of precipitation deviation of spatially adjacent pixels; In the area where rainfall characteristics are spatially inconsistent, the remaining pixels are spatially neighboring pixels. In the process of correcting the surface precipitation deviation of spatially neighboring pixels, the inverse distance weighted interpolation method IDW is used. Based on the surface precipitation deviation correction results of the co-located association in step S401 and the spatially correlated pixels in step S402, the surface precipitation deviation correction of the remaining pixels is realized.

Images