Satellite-ground cross-scale multi-source precipitation three-stage fusion method

Abstract

The invention provides a satellite-ground cross-scale multi-source precipitation three-stage fusion method. The method comprises the following steps: extracting DEM data and a drainage basin surface file of a drainage basin; local and global feature factors such as topography and meteorology of the drainage basin are obtained; extracting original satellite rainfall data of the watershed; a Kriging interpolation method is used for unifying the spatial resolution of the rainfall data of the multiple original satellites; constructing a hybrid geographically weighted regression model based on a kernel function, and establishing a mapping relationship between the multiple pieces of satellite rainfall data and global and local feature factors; calculating to obtain a plurality of primary fused precipitation products; and integrating a plurality of precipitation products fused once by using a Bayesian triangular cap method to finally obtain a high-precision and high-resolution precipitation fusion product obtained through three-stage and two-time fusion. Multi-source precipitation fusion is optimized through a refined data processing technology, the precipitation data precision and applicability are improved, high-quality data support is provided, and the method is suitable for different hydrological and meteorological research scenes.

Description

Technical Field

[0001] The present invention belongs to the field of hydrology and meteorology, and specifically relates to a three-stage fusion method for satellite-ground cross-scale multi-source precipitation. Background Art

[0002] Floods and waterlogging are frequent in China. Global warming is exacerbating uneven precipitation and its spatial and temporal distribution, leading to regional flooding and resulting in loss of life and property. Precipitation, as the core driving force of the global water cycle and the most important input to hydrological models, directly impacts the accuracy of simulation results. However, the spatial and temporal distribution of precipitation is highly heterogeneous and significantly influenced by topography, making its measurement challenging.

[0003] Precipitation data is primarily acquired through ground-based observations, radar-based quantitative precipitation estimation, and satellite remote sensing. Traditional ground-based observations rely on ground-based precipitation stations, often using limited measurements to represent precipitation conditions over a wide area. However, precipitation exhibits significant spatiotemporal variability. Particularly in areas with fewer ground stations, spatial variations in precipitation are difficult to fully capture, resulting in poor spatial representativeness of the observed data and a major challenge in hydrological research. Radar-based quantitative precipitation estimation offers high spatial resolution and strong real-time performance, but is subject to obstruction by ground objects and has limited coverage. Advances in satellite remote sensing technology, combining satellite and radar data, have gradually improved precipitation monitoring, effectively overcoming the spatial limitations of ground stations and providing a more comprehensive means of observing precipitation. Satellite remote sensing offers unique advantages in monitoring the spatiotemporal variations of global precipitation, resulting in the development of a variety of precipitation products, such as CMORPH, PERSIANN, and IMERG. Although satellite precipitation data has significant advantages in coverage and observation time continuity, due to the limitations of remote sensing instruments and inversion algorithms, its spatial resolution and accuracy are relatively low, and its inversion capability for solid precipitation is weak, and it still faces certain challenges. Summary of the Invention

[0004] In response to the problems existing in the prior art, the present invention provides a three-stage fusion method for satellite-ground cross-scale multi-source precipitation, which can give full play to the advantages of different precipitation sources and improve the accuracy of precipitation products.

[0005] To solve the above technical problems, the present invention provides the following technical solution: a three-stage fusion method for multi-source precipitation from satellites and ground across scales, comprising the following steps:

[0006] S1, extracting the DEM raster data and watershed surface file in the form of grid of the target watershed;

[0007] S2. Calculate the terrain steepness, terrain direction, surface roughness, and distance between the target watershed and the coastline based on the DEM raster data of the target watershed; extract the vegetation coverage index and wind speed of the watershed based on the watershed surface file;

[0008] S3. Extract the original satellite precipitation data P of the basin based on the basin surface file sat ;

[0009] S4, using Kriging interpolation method to the original satellite precipitation data P obtained in step S3 sat Perform interpolation with a spatial resolution of Lkm to obtain precipitation data P satdown Lkm ;

[0010] S5. Using the terrain steepness, terrain direction, surface roughness, and distance between the target watershed and the coastline obtained in step S2 as local characteristic factors, and the vegetation cover index and wind speed as global characteristic factors, a hybrid geographically weighted regression model is constructed, a mapping relationship between satellite precipitation data and the global characteristic factors and the local characteristic factors is established, and a kernel function is used to estimate the weight parameter of Lkm;

[0011] S6. Combine the weight parameter of Lkm, the global characteristic factor and the local characteristic factor to obtain the precipitation fusion product with a spatial resolution of Lkm;

[0012] S7, combining the precipitation fusion product P with a spatial resolution of Lkm obtained in step S6 1M_MGWR Lkm , using the Bayesian triangular hat method, the precipitation fusion products are integrated, and finally the precipitation fusion product with a spatial resolution of Lkm is obtained after three stages and two fusions:

[0013] P 2M_MGWR-BTCH Lkm (P satdown1 Lkm ,P satdown2 Lkm ,P satdown3 Lkm ,...,P satdownn Lkm ,P 实测 ).

[0014] Furthermore, the aforementioned step S1 includes the following sub-steps:

[0015] S1.1. Use ArcGIS software to import DEM data and perform depression filling, flow direction analysis, and water catchment calculation in the watershed;

[0016] S1.2. Determine the watershed exit point and extract the DEM data and watershed surface file of the watershed to be studied.

[0017] Furthermore, the aforementioned step S2 includes the following sub-steps:

[0018] S2.1. Calculate the watershed terrain steepness, terrain direction, and surface roughness in Lkm using the DEM data of the target watershed.

[0019] S2.2, using the DEM data of the target study basin and the China coastline file, calculate the distance between the basin and the coastline at a spatial resolution of 1 km;

[0020] S2.3. Use the surface file of the target study basin to crop the basin vegetation cover index data and wind speed data with a spatial resolution of Lkm from the global vegetation cover index data and the global wind speed data downloaded from the European Meteorological Center.

[0021] Furthermore, the aforementioned step S3 includes specifically: downloading the original satellite precipitation data P sat , based on the basin surface file in step S1 as the basin range, clip the satellite precipitation data of the basin to be studied.

[0022] Furthermore, the aforementioned step S4 includes the following sub-steps:

[0023] S4.1. Using the Kriging interpolation method, the original precipitation data P of different stations are sat The optimal weights are interpolated to the high-resolution grid to obtain the satellite precipitation weight corresponding to each grid cell at time T+1. Subsequently, this step is continued to calculate the weight of each satellite in the fine grid at time T+2;

[0024] S4.2. Based on the optimal weight, calculate the satellite precipitation data P in the basin to be studied that is downscaled to Lkm. satdown Lkm .

[0025] Furthermore, the aforementioned step S5 includes the following sub-steps:

[0026] S5.1. Construct a hybrid geographically weighted regression model to establish the mapping relationship between satellite precipitation data and global and local characteristic factors, as shown in the following formula:

[0027]

[0028] Where, P i The satellite precipitation data P is Lkm satdown Lkm , V im is the global characteristic factor of the mth Lkm in the ith grid, namely the watershed vegetation coverage index data and wind speed data, V inis the local characteristic factor of the nth global variable Lkm in the ith grid, namely, the steepness of the basin terrain, the direction of the terrain, the roughness of the surface, and the distance between the basin and the coastline. im is the weight parameter corresponding to the global feature factor, L in is the weight parameter corresponding to the local feature factor, δ i is a random deviation.

[0029] S5.2. Use the kernel function to calculate the weight parameter of Lkm, as follows:

[0030] p=V in (V in T X i V in ) -1 V in T X i

[0031] Where, X i is the weight coefficient matrix, which is calculated as follows;

[0032]

[0033] Where, l ab Represents the distance from site a to grid b; o is a non-negative attenuation parameter used to describe the mapping relationship between weight coefficient and distance;

[0034] The weight parameter corresponding to the global characteristic factor of grid i is calculated as follows:

[0035] G i =[V im T (Ip) T (Ip)V im ] -1 V im T (Ip) T (V im -p)P i

[0036] Where I is the identity matrix, P i The satellite precipitation data P is Lkm satdown Lkm ;

[0037] The weight parameter corresponding to the local characteristic factor of grid i is calculated as follows:

[0038] V i =(V in T X i Vin ) -1 V in T X i (P i -V im G i )

[0039] Where, P i The satellite precipitation data P is Lkm satdown Lkm .

[0040] Furthermore, in the aforementioned step S6, the weight parameter of Lkm, the global characteristic factor, and the local characteristic factor are combined to obtain a precipitation fusion product with a spatial resolution of Lkm, specifically:

[0041] P 1M_MGWR Lkm (P satdown Lkm ,P 实测 )=G i V im Lkm sat +L i V in Lkm sat +δ i

[0042] Where, P 1M_MGWR Lkm (P satdown Lkm ,P 实测 ) is P satdown Lkm The precipitation fusion product with a spatial resolution of Lkm is obtained by fusing the precipitation data at the measured stations. i V im Lkm is the global characteristic factor of Lkm, L i V in Lkm is the local characteristic factor of Lkm.

[0043] Furthermore, the aforementioned step S7 includes the following sub-steps:

[0044] S7.1. Determine and process the input format of precipitation data and convert it into a data format that complies with Bayesian tricorne processing. S7.2. Establish a probability density function for precipitation data and construct a Bayesian tricorne data fusion model as follows:

[0045]

[0046] Where, P iThe three precipitation fusion products P obtained in step S6 1M_MGWR Lkm , where i represents the input of different precipitation fusion products, P m P i The truth value of i and σ i P i | is random white noise with zero mean and error variance; F(A|B) is the likelihood function; P m for:

[0047] P 2M_MGWR-BTCH Lkm (P satdown1 Lkm ,P satdown2 Lkm ,P satdown3 Lkm ,...,P satdownn Lkm ,P 实测 ).

[0048] Furthermore, the aforementioned Bayesian tricorne method includes: by constructing a probability density function of precipitation data, establishing a balanced relationship between the prior distribution and the data source, thereby optimizing the accuracy of the final fusion result and reducing errors in the data fusion process.

[0049] Furthermore, the aforementioned three-stage fusion method of multi-source precipitation across scales between satellite and ground, based on the adaptive capabilities of Kriging interpolation and Bayesian triangular hat method, adjusts model parameters according to basin characteristics, data source changes, and local and global influencing factors to optimize precipitation estimation accuracy.

[0050] Compared with the existing technology, the beneficial technical effects of the above technical solutions adopted in the present invention are as follows: the present invention provides a three-stage fusion method for multi-source precipitation across satellite and ground scales, which combines Kriging interpolation, hybrid geographically weighted regression and Bayesian triangular hat method to achieve efficient fusion of multi-source data and improve the spatial consistency and temporal dynamics of precipitation data; the MGWR model fully considers multi-dimensional local and global characteristic factors such as terrain and meteorology, and improves the accuracy of precipitation data.

[0051] The Bayesian triangular hat method optimizes the fusion results under the constraints of prior distribution, reduces errors and improves data reliability;

[0052] This method can adjust model parameters according to different watershed characteristics and data sources, and is suitable for different hydrological and meteorological research scenarios.

[0053] This invention introduces refined data processing technology to optimize the multi-source precipitation fusion process, improve the accuracy and applicability of precipitation data, provide high-quality data support for hydrological models and meteorological models, and has important practical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is a flow chart of a three-stage fusion method for satellite-ground cross-scale multi-source precipitation according to the present invention.

[0055] Figure 2 This is the precipitation fusion product at different stages in the specific implementation and and P 2M_MGWR-HTCH 1km (P CMORPH 1km ,P PERSIANN 1km ,P IMERG 1km ,P 实测 ))'s evaluation index box plot, Figure 3 P CMORPH 1km , measured rainfall station data and P 2M_MGWR-BTCH 1km Spatial distribution of basin precipitation in spring, summer, autumn and winter ((a) shows the spring P CMORPH 1km (b) shows the watershed diagram of P in summer. CMORPH 1km (c) shows the watershed map of autumn P CMORPH 1km The watershed diagram of (d) shows the winter P CMORPH 1km (e) shows the watershed map of the measured rainfall station data in spring, (f) shows the watershed map of the measured rainfall station data in summer, (g) shows the watershed map of the measured rainfall station data in autumn, and (h) shows the watershed map of the measured rainfall station data in winter; (i) shows the watershed map of the P 2M_MGWR-BTCH 1km The watershed diagram (j) shows the summer P 2M_MGWR-BTCH 1km The watershed diagram (k) shows the autumn P 2M_MGWR-BTCH 1km The basin map (l) shows the winter P 2M_MGWR-BTCH 1km watershed map). DETAILED DESCRIPTION

[0056] In order to better understand the technical content of the present invention, specific embodiments are given and described below with reference to the accompanying drawings.

[0057] Various aspects of the present invention are described herein with reference to the accompanying drawings, which show a number of illustrative embodiments. The embodiments of the present invention are not limited to those described in the accompanying drawings. It should be understood that the present invention can be implemented by any of the various concepts and embodiments described above, as well as the concepts and implementations described in detail below, because the concepts and embodiments disclosed herein are not limited to any particular implementation. In addition, some aspects disclosed herein may be used alone or in any appropriate combination with other aspects disclosed herein.

[0058] Take the Fuping River Basin in my country as an example. Located on the Shahe River, a tributary of the southern branch of the Daqing River, the Fuping River lies between longitudes 113°39′ and 114°34′ east and latitudes 38°10′ and 39°30′ north. The basin receives an average annual precipitation of approximately 650 mm, with most of the rainfall occurring in the summer, particularly in July and August. The Fuping River Basin is mountainous, with elevations ranging from 247 to 2,765 meters above sea level. Its V-shaped riverbed is almost entirely free of terraces, and the riverbed is primarily rocky, with large boulders and gravel covering the riverbed. Vegetation within the basin is poor, with some small patches of forest. Due to the influence of the mountainous and hilly terrain, precipitation in the Fuping River Basin is unevenly distributed. Furthermore, capturing spatial rainfall at observation stations is difficult, resulting in low-precipitation satellite telemetry data accuracy.

[0059] refer to Figure 1 Flowchart of the present invention. The present invention provides a three-stage fusion method for satellite-ground cross-scale multi-source precipitation, comprising the following steps:

[0060] S1. Extract the grid-formatted DEM raster data and watershed surface file of the target watershed. This includes the following sub-steps:

[0061] S1.1, using ArcGIS software, import 90m resolution DEM data to perform depression filling, flow direction analysis, and water catchment calculation in the watershed;

[0062] S1.2. Determine the watershed exit point and extract the DEM data and watershed surface file of the watershed to be studied.

[0063] S2. Calculate the terrain steepness, terrain direction, surface roughness, and distance from the coastline of the target watershed based on the DEM raster data of the target watershed; and extract the vegetation coverage index and wind speed of the watershed based on the watershed surface file. The specific sub-steps are as follows:

[0064] S2.1. Use the DEM data of the watershed to be studied to calculate the steepness, direction, and surface roughness of the watershed terrain at a depth of 1 km.

[0065] S2.2. Using the DEM data of the watershed to be studied and the China coastline file, calculate the distance between the watershed and the coastline at a spatial resolution of 1 km;

[0066] S2.3. Using the surface file of the watershed to be studied, crop the watershed vegetation cover index data and wind speed data with a spatial resolution of 1 km from the global vegetation cover index data and the global wind speed data downloaded from the European Centre for Meteorological and Geological Sciences (ECWMF).

[0067] S3. Extract the original satellite precipitation data P of the basin based on the basin surface file sat Specifically, the original CMORPH satellite precipitation data, the original PERSIANN satellite precipitation data and the original IMERG satellite precipitation data are downloaded; then, based on the basin surface file in step S1 as the basin range, the satellite precipitation data P of the basin to be studied is cut out. CMORPH 、P PERSIANN 、P IMERG .

[0068] S4, use Kriging interpolation method to interpolate the original satellite precipitation data obtained in step S3 with a spatial resolution of 1km to obtain precipitation data P satdown 1km , including three types of satellite precipitation data P CMORPH 1km 、P PERSIANN 1km 、P IMERG 1km The specific sub-steps are as follows:

[0069] S4.1. Use Kriging interpolation method to convert the original P values ​​of different stations into CMORPH 、P PERSIANN 、P IMERG Optimal weight (σ CMORPH , σ PERSIANN , σ IMERG ) to the high-resolution grid to obtain the satellite precipitation weight corresponding to each grid cell at time T+1. Then, continue this step to calculate the weight of each satellite in the fine grid at time T+2;

[0070] S4.2, based on the optimal weight σ sat , that is, σ CMORPH , σ PERSIANN , σ IMERG , calculate the satellite precipitation data P in the basin to be studied that is downscaled to 1 km satdown 1km :P CMORPH 1km 、P PERSIANN 1km 、P IMERG 1kmS5. Using the terrain steepness, terrain direction, surface roughness, and distance between the target watershed and the coastline obtained in step S2 as local characteristic factors, and the vegetation coverage index and wind speed as global characteristic factors, a hybrid geographically weighted regression model is constructed. A mapping relationship is established between satellite precipitation data and the global and local characteristic factors, and a kernel function is used to estimate the 1km weight parameter. The specific sub-steps are as follows:

[0071] S5.1. Construct a hybrid geographically weighted regression model to establish the mapping relationship between satellite precipitation data and global and local characteristic factors, as shown in the following formula:

[0072]

[0073] Where, P i P is the 1km satellite precipitation data satdown 1km , that is, P CMORPH 1km 、P PERSIANN 1km 、P IMERG 1km , V im is the global characteristic factor of the mth 1km in the i-th grid, namely the watershed vegetation coverage index data and wind speed data, V in is the local characteristic factor of the nth global variable 1km in the ith grid, namely, the steepness of the basin terrain, the direction of the terrain, the roughness of the surface, and the distance between the basin and the coastline. im is the weight parameter corresponding to the global feature factor, L in is the weight parameter corresponding to the local feature factor, δ i is a random deviation.

[0074] S5.2. Use the kernel function to calculate the weight parameter of 1km, as follows:

[0075] p=V in (V in T X i V in ) -1 V in T X i

[0076] Where, X i is the weight coefficient matrix, which is calculated as follows;

[0077]

[0078] Where, l abrepresents the distance from site a to grid b; o is a non-negative attenuation parameter used to describe the mapping relationship between weight coefficient and distance; Describes how the weights between sites and grids are calculated, and this design is done through the index internal In the form of , the weight of distant points decreases more slowly, thereby improving the modeling ability of spatial remote correlation;

[0079] The weight parameter corresponding to the global characteristic factor of grid i is calculated as follows:

[0080] G i =[V im T (Ip) T (Ip)V im ] -1 V im T (Ip) T (V im -p)P i

[0081] Where I is the identity matrix, P i P is the 1km satellite precipitation data satdown 1km , that is, P CMORPH 1km 、P PERSIANN 1km 、P IMERG 1km ;

[0082] The weight parameter corresponding to the local characteristic factor of grid i is calculated as follows:

[0083] V i =(V in T X i V in ) -1 V in T X i (P i -V im G i )

[0084] Where, P i The satellite precipitation data P is Lkm satdown 1km , that is, P CMORPH 1km 、P PERSIANN 1km 、P IMERG 1kmS6. Combining the 1km weight parameter, global characteristic factor and local characteristic factor, a precipitation fusion product with a spatial resolution of 1km is obtained, specifically:

[0085]

[0086] Where: P CMORPH 1km The precipitation fusion product with a spatial resolution of 1 km is obtained by fusing the precipitation data at the measured stations. 1M_MGWR 1km (P PERSIANN 1km ,P 实测 ) is P PERSIANN 1km The precipitation fusion product with a spatial resolution of 1 km is obtained by fusing the precipitation data at the measured stations. 1M_MGWR 1km (P IMERG 1km ,P 实测 ) is P IMERG 1km The precipitation fusion product with a spatial resolution of 1 km is obtained by fusing the precipitation data of the measured stations. i V im 1km is the global characteristic factor of 1 km, L i V in 1km is the local characteristic factor of 1 km.

[0087] S7: Combine the precipitation fusion product with a spatial resolution of 1 km obtained in step S6 and The Bayesian triangular hat method is used to integrate the precipitation fusion products, and finally the precipitation fusion product P with a spatial resolution of 1 km is obtained after three stages and two fusions. 2M_MGWR-BTCH 1km (P CMORPH 1km ,P PERSIANN 1km ,P IMERG 1km ,P 实测 ), the specific sub-steps are as follows:

[0088] S7.1. Determine and process the input format of precipitation data and convert it into a data format that complies with Bayesian tricorne processing;

[0089] S7.2. Establish the probability density function of precipitation data and construct the Bayesian tricorne data fusion model to obtain P 2M_MGWR-BTCH 1km (P CMORPH1km ,P PERSIANN 1km ,P IMERG 1km ,P 实测 )

[0090]

[0091] Where: P i The three precipitation fusion products obtained in step S6 and Where i represents the input of different precipitation fusion products, P m P i The truth value of i and σ i P i | is random white noise with zero mean and error variance; F(A|B) is the likelihood function; P m For: P 2M_MGWR-BTCH 1km (P CMORPH 1km ,P PERSIANN 1km ,P IMERG 1km ,P 实测 By constructing the probability density function of precipitation data, the Bayesian tricorn hat method establishes a balance between multiple data sources and improves the accuracy of the final fusion result.

[0092] like Figure 2 The precipitation fusion products at different stages are shown and ), (a) shows the BIAS box plot of these precipitation fusion products, (b) shows the RMSE box plot of these precipitation fusion products, and (c) shows the R box plot of these precipitation fusion products.

[0093] like Figure 3 As shown, (a) shows the spring P CMORPH 1km (b) shows the watershed diagram of P in summer. CMORPH 1km (c) shows the watershed map of autumn P CMORPH 1km The watershed diagram of (d) shows the winter P CMORPH 1km(e) shows the watershed map of the measured rainfall station data in spring, (f) shows the watershed map of the measured rainfall station data in summer, (g) shows the watershed map of the measured rainfall station data in autumn, and (h) shows the watershed map of the measured rainfall station data in winter; (i) shows the watershed map of the P 2M_MGWR-BTCH 1km The watershed diagram (j) shows the summer P 2M_MGWR-BTCH 1km The watershed diagram (k) shows the autumn P 2M_MGWR-BTCH 1km The basin map (l) shows the winter P 2M_MGWR-BTCH 1km Basin map.

[0094] The present invention can better describe the spatial variation of precipitation, give full play to the advantages of different precipitation sources, improve the quality of precipitation products, construct a high-precision, fine-spatiotemporal resolution hydrological model driving field, and use a hybrid weighted stepwise regression method to downscale the original satellite precipitation and an adaptive dynamic Bayesian fusion method to fuse the downscaled precipitation of multiple satellites with the station precipitation. This improves the accuracy of precipitation products and fully combines the characteristics of different precipitation data. The Kriging interpolation method is used to unify the spatial resolution of several original satellite precipitation data, the Hybrid Geographically Weighted Regression (MGWR) method is used to fuse multiple satellite precipitation data with measured station data, and the Bayesian Triangular Hat (BTCH) method is used to aggregate the fused satellite precipitation data, thus constructing a technical framework for a high-precision and high-resolution precipitation data fusion method.

[0095] While the present invention has been described above with reference to preferred embodiments, this is not intended to limit the present invention. Persons skilled in the art will readily appreciate that various modifications and variations can be made without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A three-stage fusion method for multi-source precipitation across satellite and ground scales, characterized by: The following steps are involved: S1, extracting the DEM raster data and watershed surface file in the form of grid of the target watershed; S2. Calculate the terrain steepness, terrain direction, surface roughness, and distance between the target watershed and the coastline based on the DEM raster data of the target watershed; Extract the vegetation coverage index and wind speed of the watershed based on the watershed surface file; S3. Extract the original satellite precipitation data P of the basin based on the basin surface file sat ; S4, use Kriging interpolation method to interpolate the original satellite precipitation data obtained in step S3 with a spatial resolution of Lkm to obtain precipitation data P satdown Lkm ; S5. Using the terrain steepness, terrain direction, surface roughness, and distance between the target watershed and the coastline obtained in step S2 as local characteristic factors, and the vegetation cover index and wind speed as global characteristic factors, a hybrid geographically weighted regression model is constructed, a mapping relationship between satellite precipitation data and the global characteristic factors and the local characteristic factors is established, and a kernel function is used to estimate the weight parameter of Lkm; S6. Combine the weight parameter of Lkm, the global characteristic factor and the local characteristic factor to obtain the precipitation fusion product with a spatial resolution of Lkm; S7, combining the precipitation fusion product P with a spatial resolution of Lkm obtained in step S6 1M_MGWR Lkm , using the Bayesian triangular hat method, the precipitation fusion products are integrated, and finally the precipitation fusion product P with a spatial resolution of Lkm is obtained after three stages and two fusions. 2M_MGWR-BTCH Lkm .

2. The three-stage fusion method for multi-source precipitation from satellites and ground across scales according to claim 1 is characterized in that: Step S1 includes the following sub-steps: S1.

1. Use ArcGIS software to import DEM data and perform depression filling, flow direction analysis, and water catchment calculation in the watershed; S1.

2. Determine the watershed exit point and extract the DEM data and watershed surface file of the watershed to be studied.

3. The three-stage fusion method for multi-source precipitation from satellites and ground across scales according to claim 1 is characterized in that: Step S2 includes the following sub-steps: S2.

1. Calculate the watershed terrain steepness, terrain direction, and surface roughness in Lkm using the DEM data of the target watershed. S2.2, using the DEM data of the target study basin and the China coastline file, calculate the distance between the basin and the coastline at a spatial resolution of 1 km; S2.

3. Use the surface file of the target study basin to crop the basin vegetation cover index data and wind speed data with a spatial resolution of Lkm from the global vegetation cover index data and the global wind speed data downloaded from the European Meteorological Center.

4. The three-stage fusion method for multi-source precipitation from satellites and ground across scales according to claim 1 is characterized in that: Step S3 includes: downloading the original satellite precipitation data P sat , based on the basin surface file in step S1 as the basin range, clip the satellite precipitation data of the basin to be studied.

5. The three-stage fusion method for multi-source precipitation from satellites and ground across scales according to claim 1 is characterized in that: Step S4 includes the following sub-steps: S4.

1. Using the Kriging interpolation method, the original precipitation data P of different stations are sat The optimal weights are interpolated to the high-resolution grid to obtain the satellite precipitation weight corresponding to each grid cell at time T+1. Subsequently, this step is continued to calculate the weight of each satellite in the fine grid at time T+2; S4.

2. Based on the optimal weight, calculate the satellite precipitation data P in the basin to be studied that is downscaled to Lkm. satdown Lkm .

6. The three-stage fusion method for satellite-ground cross-scale multi-source precipitation according to claim 1 is characterized in that: Step S5 includes the following sub-steps: S5.

1. Construct a hybrid geographically weighted regression model to establish the mapping relationship between satellite precipitation data and global and local characteristic factors, as shown in the following formula: Where, P i The satellite precipitation data P is Lkm satdown Lkm , V im is the global characteristic factor of the mth Lkm in the ith grid, namely the watershed vegetation coverage index data and wind speed data, V in is the local characteristic factor of the nth global variable Lkm in the ith grid, namely, the steepness of the basin terrain, the direction of the terrain, the roughness of the surface, and the distance between the basin and the coastline. im is the weight parameter corresponding to the global feature factor, L in is the weight parameter corresponding to the local feature factor, δ i is a random deviation. S5.

2. Use the kernel function to calculate the weight parameter of Lkm, as follows: p=V in (V in T X i V in ) -1 V in T X i Where, X i is the weight coefficient matrix, which is calculated as follows; Where, l ab Represents the distance from site a to grid b; o is a non-negative attenuation parameter used to describe the mapping relationship between weight coefficient and distance; The weight parameter corresponding to the global characteristic factor of grid i is calculated as follows: G i =[V im T (I-p) T (I-p)V im ] -1 V im T (I-p) T (V im -p)P i Where I is the identity matrix, P i The satellite precipitation data P is Lkm satdown Lkm ; The weight parameter corresponding to the local characteristic factor of grid i is calculated as follows: V i =(V in T X i V in ) -1 V in T X i (P i -V im G i ) Where, P i The satellite precipitation data P is Lkm satdown Lkm .

7. The three-stage multi-source precipitation fusion method based on refined data according to claim 1 is characterized in that: In step S6, the weight parameter of Lkm, the global characteristic factor and the local characteristic factor are combined to obtain a precipitation fusion product with a spatial resolution of Lkm, specifically: P 1M_MGWR Lkm (P satdown Lkm ,P 实测 )=G i V im Lkm sat +L i V in Lkm sat +δ i Where, P 1M_MGWR Lkm (P satdown Lkm ,P 实测 ) is P satdown Lkm The precipitation fusion product with a spatial resolution of Lkm is obtained by fusing the precipitation data at the measured stations. i V im Lkm is the global characteristic factor of Lkm, L i V in Lkm is the local characteristic factor of Lkm.

8. The three-stage fusion method for multi-source precipitation from satellites and ground across scales according to claim 1 is characterized in that: Step S7 includes the following sub-steps: S7.

1. Determine and process the input format of precipitation data and convert it into a data format that complies with Bayesian tricorne processing; S7.

2. Establish the probability density function of precipitation data and construct the Bayesian tricorne data fusion model as follows: Where, P i The three precipitation fusion products P obtained in step S6 1M_MGWR Lkm , where i represents the input of different precipitation fusion products, P m P i The truth value of ε i and σ i P i ∣ is random white noise with zero mean and error variance; F(A|B) is the likelihood function; P m P 2M_MGWR-BTCH Lkm (P satdown1 Lkm ,P satdown2 Lkm ,P satdown3 Lkm ,...,P satdownn Lkm ,P 实测 ).

9. The three-stage fusion method for multi-source precipitation from satellites and ground across scales according to claim 1 is characterized in that: The Bayesian tricorne hat method includes: constructing the probability density function of precipitation data and establishing a balanced relationship between the prior distribution and the data source, thereby optimizing the accuracy of the final fusion result and reducing the error in the data fusion process.

10. The three-stage fusion method for multi-source precipitation from satellites and ground across scales according to claim 1 is characterized in that: Based on the adaptive capabilities of Kriging interpolation and Bayesian tricorn hat method, model parameters are adjusted according to watershed characteristics, data source changes, and local and global influencing factors to optimize precipitation estimation accuracy.

Images