A back-estimation method for satellite precipitation data in mountainous areas

Abstract

A method for back-calculating satellite precipitation data in mountainous areas includes: reading APHRODITE and IMERG precipitation data and calculating daily precipitation statistics at observation stations; building a model using a random forest algorithm at a 0.25° scale to simulate the long-term trend of the two types of precipitation data; generating spatial disaggregation weights for precipitation data at a 0.1° scale using a 3×3 sliding window; and combining the temporal and spatial scales to obtain the 0.1° IMERG precipitation data before 2001, which is the NIMERG data. This method utilizes the temporal and spatial relationships between APHRODITE reanalysis precipitation products and IMERG satellite precipitation products over the same timeframe to reversely back-calculate and extend the IMERG satellite precipitation data. This method can provide a set of high-quality, long-term precipitation products for research related to mountain hydrology, ecology, or geological hazards.

Description

Technical Field

[0001] The invention belongs to the technical field of hydrology and meteorology, and in particular relates to a back-calculation method for satellite precipitation data in mountainous areas. Background Art

[0002] Precipitation is a crucial component of the macro-water cycle, significantly influencing climate regulation, plant and animal survival, and topographic changes. Studying precipitation is crucial for regional water cycles, water resource management, economic development, and ecological and environmental governance. In recent decades, under the backdrop of global warming, the frequency and intensity of extreme precipitation events have increased. Mountainous areas are characterized by complex and relatively rugged terrain, and the distribution of meteorological stations at high altitudes is limited and uneven, making the study of the spatial distribution of precipitation in mountainous areas challenging. Furthermore, precipitation in mountainous areas exhibits significant temporal and spatial variability, making meteorological station data inadequate for exploring patterns and trends. Furthermore, this temporal and spatial variability in precipitation is a direct contributing factor to natural disasters such as droughts and floods. Monitoring and forecasting precipitation in mountainous watersheds based on satellite precipitation data has become a key application area for meteorological and radar satellites.

[0003] Currently, there are a variety of precipitation monitoring products based on remote sensing satellites, including the Fengyun meteorological satellites, the precipitation product developed by the U.S. Weather Prediction Center (CMORPH), the IMERG precipitation satellite product, and the APHRODITE precipitation product. These precipitation products provide continuous spatiotemporal precipitation information, providing strong data support for regional precipitation research and hydrological modeling. Among these precipitation products, the IMERG series not only provides strong precipitation visualization but also has potential for drought monitoring. Furthermore, IMERG features a large observation range, long observation time, and continuous dynamic changes, enabling effective observation and prediction of extreme precipitation events and floods. The root mean square error of the APHRODITE precipitation product is not significantly affected by elevation and latitude, indicating that this precipitation product is more suitable for applications at higher latitudes. Therefore, we selected the APHRODITE precipitation product for verification with the IMERG series to jointly explore precipitation patterns in mountainous areas.

[0004] Remote sensing satellites, as a new method for measuring regional precipitation, offer significant advantages, including being unaffected by topography. Currently, many researchers use remote sensing satellite products to predict future precipitation over long time series and short time periods (e.g., one hour) using forecasting techniques. This has been instrumental in climate forecasting and early warning. Building on previous research, we analyzed the patterns of precipitation products and satellite precipitation products over the same timeframe to reverse-derive previously unavailable satellite precipitation data for mountainous areas. This not only extends the temporal scale of satellite precipitation products but also increases their spatial scale. Furthermore, reverse-deriving satellite precipitation data is crucial for exploring the altitudinal distribution of precipitation in high-altitude mountainous areas and analyzing geological hazards triggered by extreme precipitation. Therefore, reverse-deriving satellite precipitation data for mountainous areas is essential for understanding the patterns and connections between geological hazard environments and extreme precipitation. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, such as the short time series and low spatial resolution of existing satellite precipitation products when applied in mountainous areas with complex terrain and sparse observation sites, a method for back-calculating satellite precipitation data in mountainous areas based on the APHRODITE reanalysis precipitation product and the IMERG satellite precipitation product was proposed. This method organically integrates the APHRODITE and IMERG precipitation products, extending the time scale of the IMERG satellite precipitation product and effectively improving its accuracy, providing an alternative for analysis related to hydrology, meteorology, and geological hazards in mountainous areas.

[0006] To achieve the above object, the technical solution adopted by the present invention is:

[0007] A method for back-calculating satellite precipitation data in mountainous areas, comprising the following specific steps, characterized in that:

[0008] Step 1: Read the APHRODITE and IMERG precipitation data and calculate the daily precipitation at the observation station;

[0009] Step 2: Using the random forest algorithm to build a model at a scale of 0.25°, simulate the long-term trend of the two types of precipitation data;

[0010] Step 3: Generate spatial disaggregation weights of precipitation data using a 3×3 sliding window at a 0.1° scale;

[0011] Step 4: Combine the temporal and spatial scales to obtain the 0.1° IMERG precipitation data before 2001, which is the NIMERG data.

[0012] As a further improvement of the present invention, the step 1 specifically includes the following steps:

[0013] Step 11: First, convert the APHRODITE and IMERG precipitation products into tif format, and then extract the precipitation bands.

[0014] Step 12: Obtain the spatial coordinates of the upper left, upper right, lower left, and lower right vertices of the spatial range based on the vector boundary of the watershed along the Sichuan-Tibet Railway.

[0015] Step 13: Based on the rectangular boundary determined by the four vertices, in order to fully utilize the precipitation data of the observation stations and better reflect the influence of terrain on the precipitation distribution in mountainous areas, a buffer zone is established by extending 0.25° outward along the boundary of the study area. The APHRODITE and IMERG precipitation information within the buffer zone is read respectively, and the APHRODITE and IMERG precipitation data of the study area with a time interval of one day and a spatial resolution of 0.25° and 0.1° are obtained.

[0016] Step 14: Count the extracted satellite precipitation product information pixel by pixel to obtain the daily APHRODITE and IMERG precipitation data for each pixel;

[0017] Step 15: Organize the daily precipitation monitored by observation stations in the buffer study area.

[0018] As a further improvement of the present invention, the step 2 specifically includes the following steps:

[0019] Step 21, calculate the daily average precipitation value of APHRODITE and IMERG data for 15 years;

[0020] Step 22: resample the IMERG daily average precipitation data to the same spatial resolution as the APHRODITE precipitation data, changing its spatial resolution from 0.1° to 0.25°;

[0021] Step 23, the elevation data is obtained by splicing and clipping multiple ASTER GDEM V2 data covering the study area to obtain the DEM data of the study area;

[0022] Step 24, extracting the longitude, latitude, slope, aspect, openness, roughness and undulation raster data by calculating the DEM data of the study area;

[0023] Step 25: resample the Longitude, Latitude, Slope, Aspect, Openness, Roughness, and Undulation raster data to the same spatial resolution as the APHRODITE precipitation data, changing its spatial resolution from 90 m to 0.25°;

[0024] Step 26: Based on the APHRODITE and IMERG precipitation data and the surface parameters, a nonlinear machine learning model based on random forest is constructed to establish the relationship equation between the IMERG precipitation data and the surface parameters and the APHRODITE precipitation data:

[0025] PER = f RF (C)+ε (1)

[0026] Where PER represents the average daily precipitation value of IMERG during the training phase; C is the input vector, which represents the input variables, including Longitude, Latitude, Slope, Aspect, Openness, Roughness, Undulation, and APHRODITE daily precipitation data; f RF It is a nonlinear function that establishes a relationship between the input variables and the output PER;

[0027] Step 27, using a decision tree to predict the IMERG precipitation data with a spatial resolution of 0.25°, substituting the precipitation data of a certain day of APHRODITE into the model to obtain the IMERG precipitation data A of the corresponding day;

[0028] In the regression task, several decision trees are first established during the training period. Each decision tree is established by Bootstrap samples, where the training input data accounts for about 2 / 3 of the total samples, and the remaining 1 / 3 of the samples are used to verify each tree;

[0029] The final model prediction value is generated by taking the arithmetic average of the results of many independent regression trees. The final model result is expressed as:

[0030]

[0031] Where P(PER|C) is the final prediction result, m is the number of regression trees, and P i (PER|C) represents the prediction result of the i-th tree;

[0032] Step 28: Use the average out-of-bag data (OOB) of each decision tree to get the prediction accuracy. The calculation principle is as follows:

[0033]

[0034] Where n is the number of OOB samples, Y^(X i ) is based on the given sample X i , the output data of the RF model; Y i is the IMERG daily average precipitation data;

[0035] Step 29: Extract the APHRODITE and IMERG precipitation values for each grid cell, analyze the distribution patterns of the two precipitation data on a temporal scale, use a polynomial calculation method to adjust the parameters to be as close as possible to the precipitation data, and obtain the IMERG precipitation data B for a grid cell on the corresponding date.

[0036] In step 210, the IMERG precipitation data of a grid cell on the corresponding date is taken as the average of the precipitation data obtained in steps 27 and 29. This idea is used to correct the precipitation values of all grid cells of the IMERG precipitation data at a scale of 0.25° in the entire study area to generate the NIMERG precipitation data for the corresponding date.

[0037] As a further improvement of the present invention, the step 3 specifically includes the following steps:

[0038] Step 31, calculate the daily average precipitation value of APHRODITE and IMERG data for 15 years;

[0039] Step 32: Select a 3×3 moving window size, and calculate the ratio of the daily rainfall of the central grid, i.e., the average APHRODITE precipitation data of a certain day, to the daily average rainfall of the corresponding 3×3 window, i.e., the average IMERG precipitation data of a certain day, to obtain a daily spatial decomposition weight of 0.1° based on IMERG.

[0040] As a further improvement of the present invention, the step 4 specifically includes the following steps:

[0041] Step 41: Multiply the daily spatial decomposition weight value of IMERG at the 0.1° scale obtained in step 32 by the NIMERG precipitation data at the 0.25° scale obtained in step 210 to obtain the NIMERG daily precipitation data at 0.1° on the corresponding date. In this way, the NIMERG daily precipitation data between the corresponding years are obtained.

[0042] Step 42, based on the precipitation data of the observation station, the accuracy verification analysis of the NIMERG daily precipitation data is performed using the correlation coefficient R and the root mean square error RMSE related indicators;

[0043] R represents the correlation between the two. The closer the value is to 1, the better the correlation is, and the higher the accuracy of the downscaling model is. RMSE is used to measure the deviation between the precipitation obtained by the downscaling model and the measured precipitation data, reflecting the overall level of assessment error. The smaller the value, the closer the two are. The calculation formulas for each indicator are as follows:

[0044]

[0045]

[0046] Where: P c 、P o They represent the NIMERG daily precipitation value and the precipitation value at the observation station, respectively, and n is the number of observation stations.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] The present invention provides a method for back-calculating satellite precipitation data in mountainous areas. This method includes: reading APHRODITE and IMERG precipitation data and calculating daily precipitation statistics at observation stations; establishing a model using a random forest algorithm at a 0.25° scale to simulate the long-term trend of the two types of precipitation data; generating spatial disaggregation weights for precipitation data at a 0.1° scale using a 3×3 sliding window; and combining the temporal and spatial scales to obtain 0.1° IMERG precipitation data before 2001, namely the NIMERG data. The present invention utilizes the relationship between APHRODITE reanalyzed precipitation products and IMERG satellite precipitation products at the same temporal and spatial scales to reversely calculate and extend the IMERG satellite precipitation data. This method aims to explore a scientifically feasible method for assimilating satellite remote sensing precipitation products. This method is also of great significance for ecological, hydrological, and geological disaster research and the comprehensive utilization of water resources in complex mountainous areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 Schematic diagram of the research process of the present invention.

[0050] Figure 2 This is a schematic diagram of the longitude spatial distribution of the study area of the present invention.

[0051] Figure 3 This is a schematic diagram of the latitudinal spatial distribution of the study area of the present invention.

[0052] Figure 4 This is a schematic diagram of the spatial distribution of slope in the study area of the present invention.

[0053] Figure 5 This is a schematic diagram of the spatial distribution of slope aspects in the study area of the present invention.

[0054] Figure 6This is a schematic diagram of the spatial distribution of terrain relief in the study area of the present invention.

[0055] Figure 7 This is a schematic diagram of the spatial distribution of terrain openness in the study area of the present invention.

[0056] Figure 8 Schematic diagram of the spatial distribution of surface roughness in the study area of the present invention.

[0057] Figure 9 This is a schematic diagram of the spatial distribution of APHRODITE precipitation data for the study area of the present invention on August 9, 2000.

[0058] Figure 10 This is a schematic diagram of the spatial distribution of NIMERG precipitation data for the study area of the present invention on August 9, 2000.

[0059] Figure 11 This is a comparison chart of the NIMERG data and the site precipitation data for the study area of the present invention on August 9, 2000. DETAILED DESCRIPTION

[0060] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0061] The study area is located in the heart of the Asian continent, centered along the Sichuan-Tibet Railway. The 1,838-kilometer-long Sichuan-Tibet Railway runs from Chengdu, Sichuan Province, to Lhasa, Tibet Autonomous Region, traversing the major river basins of the Yarlung Zangbo, Nujiang, Lancangjiang, Jinshajiang, Yalongjiang, Dadujiang, and Minjiang Rivers. The study area exhibits significant elevation differences (ranging from 116 to 7,019 meters above sea level) and a wide longitude span (25°–33°N, 85°–105°E). Solar radiation increases from east to west, and the terrain gradually rises. Regional climatic conditions vary significantly with altitude and topographic location. The climate varies from subtropical monsoon to plateau mountain climate, to temperate plateau monsoon semi-humid, to semi-arid, to subarctic plateau monsoon semi-arid, and finally to arid and arid plateau arctic monsoon arid. The study area boasts a complex terrain, with alternating high mountains and valleys, and a highly undulating topography. The main landforms include hills, basins, river valley plains, alpine valleys, and glaciers. Temperatures are relatively low, especially on the plateau, where the average annual temperature remains below 0°C. For example, in Lhasa, Shigatse, and Nyingchi, the average temperature in the coldest month ranges from -4°C to -2°C. Precipitation is low and extremely unevenly distributed. Bashika, located in the lower reaches of the Yarlung Zangbo River, receives 4,495 mm of annual precipitation, making it one of the precipitation centers in my country. 80%-90% of the annual precipitation is concentrated between May and September. The number of rainy days decreases from southeast to northwest, with dry winters and rainy summers, and less spring rain than autumn.

[0062] like Figure 1As shown in the figure, the present invention provides a method for back-calculating satellite precipitation data in mountainous areas, wherein the boundary of the study area, the distribution of meteorological stations and the DEM diagram of the present invention, and the longitude spatial distribution diagram of the study area extracted by the present invention are shown in the figure. Figure 2 As shown in the figure, the spatial distribution diagram of the latitude of the research area extracted by the present invention is as follows Figure 3 As shown in the figure, the spatial distribution diagram of the slope of the study area extracted by the present invention is as follows Figure 4 As shown in the figure, the spatial distribution diagram of the slope of the study area extracted by this invention is as follows: Figure 5 As shown in the figure, the spatial distribution diagram of the terrain relief of the study area extracted by the present invention is as follows: Figure 6 As shown in the figure, the spatial distribution diagram of the terrain openness of the study area extracted by the present invention is as follows Figure 7 As shown in the figure, the spatial distribution diagram of the surface roughness of the study area extracted by this invention is as follows Figure 8 As shown in FIG, the spatial distribution diagram of the APHRODITE precipitation data extracted by the present invention in the study area on August 9, 2000 is shown in FIG. Figure 9 As shown in FIG, the spatial distribution diagram of the NIMERG precipitation data extracted by the present invention in the study area on August 9, 2000 is shown in FIG. Figure 10 As shown in the figure, the comparison of the NIMERG data and the station precipitation data of the study area on August 9, 2000 extracted by the present invention is shown in the figure Figure 11 As shown, the following steps are included:

[0063] Step 1: Read the APHRODITE and IMERG precipitation data and calculate the daily precipitation at the observation station. Specifically:

[0064] Step 11: First, convert the APHRODITE and IMERG precipitation products to tif format, extract the precipitation bands, use the Make NetCDF Raster Layer tool, select precipitation as the variable, and then export to tif format.

[0065] Step 12: Obtain the spatial coordinates of the upper left, upper right, lower left, and lower right vertices of the spatial range based on the vector boundary of the watershed along the Sichuan-Tibet Railway.

[0066] Step 13: Based on the rectangular boundary determined by the four vertices, in order to make full use of the precipitation data of the observation stations and better reflect the influence of terrain on the precipitation distribution in mountainous areas, a buffer zone is established by extending 0.25° outward along the boundary of the study area. The APHRODITE and IMERG precipitation information within the buffer zone are read respectively. The APHRODITE precipitation data with a time interval of one day and a spatial resolution of 0.25° from 1951 to 2015 and the IMERG precipitation data with a time interval of one day and a spatial resolution of 0.1° from 2001 to 2020 in the study area are obtained.

[0067] Step 14: Count the extracted satellite precipitation product information pixel by pixel to obtain the daily APHRODITE and IMERG precipitation data for each pixel;

[0068] Step 15, collating the daily precipitation monitored by observation stations in the buffer study area;

[0069] Step 2: Use the random forest algorithm to build a model at a scale of 0.25° to simulate the long-term trend of the two types of precipitation data series. Specifically:

[0070] Step 21: Use the Raster Calculator tool to calculate the average daily precipitation values for the APHRODITE and IMERG data for a total of 15 years (2001-2015).

[0071] Step 22: Use the Data Management tool to resample the IMERG daily average precipitation data from step 21 to the same spatial resolution as the APHRODITE precipitation data, changing its spatial resolution from 0.1° to 0.25°.

[0072] Step 23: Elevation data is obtained by mosaicing (Mosaic to new raster) and clipping (Extract by mask) multiple ASTER GDEM V2 data covering the study area to obtain the DEM data of the study area;

[0073] Step 24, extracting raster data such as longitude, latitude, slope, aspect, openness, roughness, and undulation by calculating the DEM data of the study area;

[0074] Step 25: Use the Data Management tool to resample the Longitude, Latitude, Slope, Aspect, Openness, Roughness, and Undulation raster data from Step 24 to match the spatial resolution of the APHRODITE precipitation data, changing its spatial resolution from 90 m to 0.25°.

[0075] Step 26: Based on the APHRODITE and IMERG precipitation data and the surface parameters, a nonlinear machine learning model based on random forest is constructed to establish the relationship equation between the IMERG precipitation data and the surface parameters and the APHRODITE precipitation data:

[0076] PER = f RF (C)+ε (1)

[0077] Where PER represents the average daily precipitation value of IMERG during the training phase; C is the input vector, which represents the input variables, including Longitude, Latitude, Slope, Aspect, Openness, Roughness, Undulation, and APHRODITE daily precipitation data; f RF It is a nonlinear function that establishes a relationship between the input variables and the output PER;

[0078] Step 27, using a decision tree to predict the IMERG precipitation data with a spatial resolution of 0.25°, substituting the precipitation data of a certain day of APHRODITE (such as August 9, 2000) into the model to obtain the IMERG precipitation data A on August 9, 2000;

[0079] In the regression task, several decision trees are first built during the training period. Each decision tree is built by Bootstrap samples, where the training input data accounts for about 2 / 3 of the total samples, and the remaining (1 / 3) samples are used to validate each tree. In order to further improve the generalization ability of the random forest model, the final model prediction value is generated by taking the arithmetic average of the results of many independent regression trees. The final model result is expressed as:

[0080]

[0081] Where P(PER|C) is the final prediction result, m is the number of regression trees, and P i (PER|C) represents the prediction result of the i-th tree.

[0082] In step 28, the average out-of-bag (OOB) value of each decision tree is used to obtain the prediction accuracy. The calculation principle is as follows:

[0083]

[0084] Where n is the number of OOB samples, Y^(X i ) is based on the given sample X i , output data of RF model; Y i It is the average daily precipitation data of IMERG;

[0085] Step 29: Extract the APHRODITE and IMERG precipitation values for each grid cell at a 0.25° scale, analyze the temporal distribution of the two precipitation data, and use a polynomial calculation method to adjust the parameters to be as close to the precipitation data as possible. This will yield IMERG precipitation data B for a particular grid cell on August 9, 2000.

[0086] Step 210: At a scale of 0.25°, the IMERG precipitation data of a certain grid cell on August 9, 2000, is taken as the average of the precipitation data obtained in steps 27 and 29. This idea is used to correct the precipitation values of all grid cells of the IMERG precipitation data at a scale of 0.25° for the entire study area to generate the NIMERG precipitation data for August 9, 2000.

[0087] Step 3: Generate the spatial disaggregation weights of precipitation data using a 3×3 sliding window at a 0.1° scale. Specifically:

[0088] Step 31, use the Raster Calculator tool to calculate the average daily precipitation values for the APHRODITE and IMERG data for a total of 15 years (2001-2015);

[0089] Step 32: When the spatial resolution of the APHRODITE data is 0.25°, a 3×3 moving window size is selected, and the daily spatial decomposition weight (0.1°) based on IMERG is obtained by calculating the ratio of the daily rainfall of the central grid (the average APHRODITE precipitation data of a certain day) to the daily average rainfall of the corresponding 3×3 window (the average IMERG precipitation data of a certain day);

[0090] Step 4: Combine the temporal and spatial scales to obtain the 0.1° IMERG precipitation data before 2001, which is the NIMERG data. Specifically:

[0091] Step 41: Multiply the daily spatial decomposition weight of IMERG at the 0.1° scale obtained in step 32 by the NIMERG precipitation data at the 0.25° scale obtained in step 210, and fill all blank grid cells with precipitation values at the 0.1° scale to obtain the NIMERG daily precipitation data at 0.1° on August 9, 2000. In this way, the NIMERG daily precipitation data from January 1951 to December 2000 can be obtained.

[0092] In step 42, based on the precipitation data collected and organized at the observation stations in step 15, the NIMERG daily precipitation data is verified and analyzed for accuracy using relevant indicators such as the correlation coefficient (R) and the root mean square error (RMSE). R represents the correlation between the two. The closer the value is to 1, the better the correlation and the higher the accuracy of the downscaling model. RMSE is used to measure the deviation between the precipitation obtained by the downscaling model and the measured precipitation data, reflecting the overall level of assessment error. The smaller the value, the closer the two are. The calculation formulas for each indicator are as follows:

[0093]

[0094]

[0095] Where: P c 、P o represent the daily precipitation values of NIMERG and the precipitation values of the observation stations, respectively, and n is the number of observation stations;

[0096] Step 43: The correlation coefficient R between the NIMERG daily precipitation data at 0.1° on August 9, 2000, and the precipitation data at the observation station is 0.6685, and the root mean square error RMSE is 9.7861 mm / day, which passes the test with a confidence level of 0.01. This indicates that there is a significant linear correlation between the two and the error is small, indicating that this method has a good effect.

[0097] The above description is merely a preferred embodiment of the present invention and does not constitute any other form of limitation to the present invention. Any modification or equivalent variation based on the technical essence of the present invention shall still fall within the scope of protection claimed by the present invention.

Claims

1. A method for back-calculating satellite precipitation data in mountainous areas, comprising the following steps, characterized in that: Step 1: Read the APHRODITE and IMERG precipitation data and calculate the daily precipitation at the observation station; Step 2: Using the random forest algorithm to build a model at a scale of 0.25°, simulate the long-term trend of the two types of precipitation data; The step 2 specifically includes the following steps: Step 21, calculate the daily average precipitation value of APHRODITE and IMERG data for 15 years; Step 22: resample the IMERG daily average precipitation data to the same spatial resolution as the APHRODITE precipitation data, changing its spatial resolution from 0.1° to 0.25°; Step 23, the elevation data is obtained by splicing and clipping multiple ASTER GDEM V2 data covering the study area to obtain the DEM data of the study area; Step 24, extracting the longitude, latitude, slope, aspect, openness, roughness and undulation raster data by calculating the DEM data of the study area; Step 25: resample the Longitude, Latitude, Slope, Aspect, Openness, Roughness, and Undulation raster data to the same spatial resolution as the APHRODITE precipitation data, changing its spatial resolution from 90 m to 0.25°; Step 26: Based on the APHRODITE and IMERG precipitation data and the surface parameters, a nonlinear machine learning model based on random forest is constructed to establish the relationship equation between the IMERG precipitation data and the surface parameters and the APHRODITE precipitation data: (1) Where PER represents the average daily precipitation value of IMERG during the training phase; C is the input vector, which represents the input variables, including Longitude, Latitude, Slope, Aspect, Openness, Roughness, Undulation, and APHRODITE daily precipitation data; f RF It is a nonlinear function that establishes a relationship between the input variables and the output PER; Step 27, using a decision tree to predict the IMERG precipitation data with a spatial resolution of 0.25°, substituting the precipitation data of a certain day of APHRODITE into the model to obtain the IMERG precipitation data A of the corresponding day; In the regression task, several decision trees are first established during the training period. Each decision tree is established by Bootstrap samples, where the training input data accounts for about 2 / 3 of the total samples, and the remaining 1 / 3 of the samples are used to verify each tree; The final model prediction value is generated by taking the arithmetic average of the results of many independent regression trees. The final model result is expressed as: (2) Where, is the final prediction result, m is the number of regression trees, Represents the prediction result of the i-th tree; Step 28: Use the average out-of-bag data (OOB) of each decision tree to get the prediction accuracy. The calculation principle is as follows: (3) Where n is the number of OOB samples, According to the given sample , output data of RF model; is the IMERG daily average precipitation data; Step 29: Extract the APHRODITE and IMERG precipitation values for each grid cell, analyze the distribution patterns of the two precipitation data on a temporal scale, use a polynomial calculation method to adjust the parameters to be as close as possible to the precipitation data, and obtain the IMERG precipitation data B for a grid cell on the corresponding date. Step 210: The IMERG precipitation data of a grid cell on the corresponding date is taken as the average of the precipitation data obtained in steps 27 and 29. The precipitation values of all grid cells of the IMERG precipitation data at a scale of 0.25° in the entire study area are corrected using the above steps to generate the NIMERG precipitation data for the corresponding date. Step 3: Generate spatial disaggregation weights of precipitation data using a 3×3 sliding window at a 0.1° scale; Step 4: Combine the temporal and spatial scales to obtain the IMERG precipitation data before 2001 on a 0.1° scale, which is the NIMERG data.

2. The method for back-calculating satellite precipitation data in mountainous areas according to claim 1, wherein: The step 1 specifically includes the following steps: Step 11: First, convert the APHRODITE and IMERG precipitation products into tif format, and then extract the precipitation bands. Step 12: Obtain the spatial coordinates of the upper left, upper right, lower left, and lower right vertices of the spatial range based on the vector boundary of the watershed along the Sichuan-Tibet Railway. Step 13: Based on the rectangular boundary determined by the four vertices, in order to fully utilize the precipitation data of the observation stations and better reflect the influence of terrain on the precipitation distribution in mountainous areas, a buffer zone is established by extending 0.25° outward along the boundary of the study area. The APHRODITE and IMERG precipitation information within the buffer zone is read respectively, and the APHRODITE and IMERG precipitation data of the study area with a time interval of one day and a spatial resolution of 0.25° and 0.1° are obtained. Step 14: Count the extracted satellite precipitation product information pixel by pixel to obtain the daily APHRODITE and IMERG precipitation data for each pixel; Step 15: Organize the daily precipitation monitored by observation stations in the buffer study area.

3. The method for back-calculating satellite precipitation data in mountainous areas according to claim 1, wherein: The step 3 specifically includes the following steps: Step 31, calculate the daily average precipitation value of APHRODITE and IMERG data for 15 years; Step 32: Select a 3×3 moving window size, and calculate the ratio of the daily rainfall of the central grid, i.e., the average APHRODITE precipitation data of a certain day, to the daily average rainfall of the corresponding 3×3 window, i.e., the average IMERG precipitation data of a certain day, to obtain a daily spatial decomposition weight of 0.1° based on IMERG.

4. The method for back-calculating satellite precipitation data in mountainous areas according to claim 1, wherein: The step 4 specifically includes the following steps: Step 41: Multiply the daily spatial decomposition weight value of IMERG at the 0.1° scale obtained in step 32 by the NIMERG precipitation data at the 0.25° scale obtained in step 210 to obtain the NIMERG daily precipitation data at 0.1° on the corresponding date. In this way, the NIMERG daily precipitation data between the corresponding years are obtained. Step 42, based on the precipitation data of the observation station, the accuracy verification analysis of the NIMERG daily precipitation data is performed using the correlation coefficient R and the root mean square error RMSE related indicators; R represents the correlation between the two. The closer the value is to 1, the better the correlation is, and the higher the accuracy of the downscaling model is. RMSE is used to measure the deviation between the precipitation obtained by the downscaling model and the measured precipitation data, reflecting the overall level of assessment error. The smaller the value, the closer the two are. The calculation formulas for each indicator are as follows: (4) (5) Where: P c 、P o They represent the NIMERG daily precipitation value and the precipitation value at the observation station, respectively, and n is the number of observation stations.

Images