Rainfall erosivity inversion method integrating many-source satellite rainfall data

Abstract

The invention relates to a rainfall erosivity inversion method integrating crowd-source satellite rainfall data, which comprises the following steps: acquiring data such as a heterogenous remote sensing rainfall product and a DEM on a GEE remote sensing cloud platform, and acquiring characteristic variables such as characteristic multisource rainfall, sea and land positions and terrain after preprocessing such as numerical extraction, projection transformation and grid resampling; the method comprises the following steps: on the basis of meteorological station space distribution vector data, extracting characteristic variable information corresponding to a corresponding point as a covariable, and taking rainfall erosivity obtained by calculating rainfall data actually observed by a meteorological station as a dependent variable, so as to construct sample data; the sample set is randomly divided into a training set and a test set, and the training set is used for constructing a deep forest regression model for inversion of rainfall erosivity and performing spatial inversion mapping; and then verifying the rainfall erosivity inversion result by using the verification set. According to the method, high-precision inversion and mapping can be carried out on rainfall erosivity space distribution of a large-scale region and even a global scale.

Description

Technical Field

[0001] The present invention belongs to the technical fields of soil and water conservation and agricultural remote sensing, and particularly relates to a method for inversing rainfall erosivity based on integrated multi-source satellite precipitation data. Background Technique

[0002] Rainfall erosivity (R factor) is an important environmental parameter in disciplines such as hydrology and soil erosion science, and refers to the ability or intensity of rainfall (including rainfall amount, duration, and rainfall intensity) to erode and damage the surface soil. Since rainfall erosivity is the direct factor causing soil loss and sediment transport, it is also used as one of the indicators for quantitatively describing the scouring, loss, and erosion of soil by rainfall. High rainfall erosivity is a key indicator for quantitatively evaluating soil and water loss, and has important application values in fields such as soil and water conservation, ecological environment assessment, and sustainable land management. Accurately obtaining the spatio-temporal distribution of rainfall erosivity is of great significance for formulating effective soil and water loss prevention and control measures, evaluating ecosystem service functions, guiding sustainable land use, and coping with the impacts of climate change. However, due to the complexity of the formation mechanism of rainfall erosivity and its high spatio-temporal heterogeneity, accurately inversing rainfall erosivity has always been a challenging problem in the fields of soil and water conservation and earth science.

[0003] Rainfall erosivity is affected by the precipitation process and the underlying surface environment, and has longitude, latitude, and vertical zonality differentiation. The classical spatial interpolation method is based on the theory of spatial autocorrelation, which links the rainfall erosivity that is difficult to measure continuously in space with the erosivity measured at meteorological stations, and then infers the spatial variation information of rainfall erosivity. Among them, the point-based interpolation techniques represented by geostatistical methods (such as Kriging, Spline, Anuspline, inverse distance weighting, and geographically weighted) are the mainstream solutions for large-scale rainfall erosivity inversion due to their simple and convenient operation and intuitive spatial visualization effect. However, this method is severely restricted by the number, spatial layout, and distribution density of ground meteorological observation stations, resulting in low inversion accuracy in areas with sparse meteorological station distributions, inability to capture the details of the spatial variation of rainfall erosivity, obvious under-smoothing and numerical saturation phenomena. At the same time, this method fails to consider the complex non-linear relationship between rainfall erosivity and environmental factors, and can only reflect the macroscopic law of the spatial distribution of rainfall erosivity, and cannot reflect its distribution information at small scales. For example, it cannot depict the differentiation characteristics of rainfall erosivity with the change of the surface.

[0004] In the past 30 years, satellite remote sensing technology has provided reliable data for obtaining global precipitation information. Polar-orbiting and geostationary precipitation radar and infrared sensor satellites are the mainstream methods for detecting global precipitation due to their high timeliness, high spatio-temporal resolution, wide-area monitoring, and easy accessibility. The satellite precipitation remote sensing data contains rainfall amount and time information, which is used to invert the rainfall erosivity at the regional and even global scales. Among them, the representative satellite precipitation remote sensing data of Tropical Rainfall Measuring Mission (TRMM) and Global Satellite Mapping of Precipitation (GPM) are widely used to assist in the extraction of spatial distribution information of rainfall erosivity. This method for inverting rainfall erosivity based on single-source satellite precipitation data first replaces the site-based ground observation precipitation data with the single-source satellite precipitation grid domain data, and directly performs the preliminary inversion of rainfall erosivity in combination with the empirical calculation equation of rainfall erosivity. Then, the residual correction is performed using the true rainfall erosivity measured by the ground meteorological station data. Finally, the regionalized rainfall erosivity is obtained. Compared with the point-based spatial interpolation method, this method has higher inversion accuracy due to the characteristics of data spatial continuity and timeliness, and can capture more spatial detail information. However, this method has the following disadvantages: the inversion accuracy of rainfall erosivity depends on the performance of satellite sensors and the precipitation data itself, and it is impossible to avoid the inversion error caused by the quality of single-satellite precipitation data; based on the assumption of linear relationship, the non-linear relationship between precipitation distribution and environmental factors is ignored, and it is impossible to accurately describe the influence of macro (sea-land position, distance from the ocean) and micro (topography) factor changes on the spatial distribution of rainfall erosivity. Therefore, the interpretability of the above two methods is relatively low.

[0005] Fusing multi-source satellite precipitation data can overcome the inherent biases or limitations of single data sources, and then make full use of their complementarity to greatly extract their consistency information. However, there is currently no research on using multi-source satellite remote sensing precipitation data to improve the spatial inversion accuracy of rainfall erosivity. Based on this, there is an urgent need to develop an inversion method for rainfall erosivity that integrates multi-source satellite precipitation data to effectively solve the above problems. Summary of the Invention

[0006] The purpose of the present invention is to provide an inversion method for rainfall erosivity based on machine learning to integrate multi-source satellite precipitation data, so as to solve the problem of high-precision inversion of rainfall erosivity at large-scale regional and even global scales. The present invention incorporates multi-source satellite remote sensing precipitation data, sea-land position variables, and terrain variables as explanatory variables, follows the zonal distribution law of rainfall erosivity, and uses non-linear fitting of the random forest model to replace the traditional geostatistical fitting method, thereby improving the inversion accuracy of rainfall erosivity.

[0007] The purpose of the present invention is achieved through the following technical solutions:

[0008] A method for retrieving rainfall erosivity by integrating multi-source satellite precipitation data, comprising the following steps:

[0009] A. Acquisition and processing of precipitation data from ground meteorological stations: Obtain the observed precipitation data of ground meteorological stations, and then calculate the rainfall erosivity of each station;

[0010] B. Acquisition and processing of covariate data: Obtain 7 types of remote sensing precipitation data in the same period and the same region as the meteorological stations, and download the DEM data, and then extract the covariate dataset composed of multi-source satellite precipitation data, terrain and sea-land location;

[0011] B1. Process the remote sensing data of multi-source satellite precipitation;

[0012] B2. Process terrain variables;

[0013] B3. Process sea-land location variables;

[0014] B4. Preprocess covariate data;

[0015] C. Construct sample data: Take the annual rainfall erosivity obtained from meteorological station data as the dependent variable, combine the spatial location information of meteorological stations, extract the numerical information of 15 covariates at the corresponding positions, and then construct a sample set;

[0016] D. Construct a random forest regression model for spatial inversion of rainfall erosivity: Set the rainfall erosivity in the sample data as the dependent variable, and the above 15 covariates as the independent variables, train the random forest regression model, and use the grid search method to optimize the parameters of the random forest model;

[0017] D1. Randomly select 70% of the sample data as the training set, and the remaining 30% of the samples as the validation set;

[0018] D2. Set the dependent variable as the rainfall erosivity in the sample data, and the independent variables as the above 15 covariates, use the random forest regression algorithm for relationship fitting, and use the grid search method to optimize the hyperparameters mtry and ntree in the model;

[0019] E. Evaluate the accuracy of the rainfall erosivity inversion model:

[0020] Substitute the trained random forest model into the validation set, take the rainfall erosivity in the validation set as the true value, and the inversion of the random forest model as the predicted value, and calculate the determination coefficient R 2 , root mean square error RMSE, and mean error ME between the two as evaluation indicators to evaluate the inversion accuracy of rainfall erosivity, where the calculation process is as follows:

[0021]

[0022]

[0023] ME = y p -y o (9)

[0024]

[0025] where y p and y o are the predicted value and the actual rainfall erosivity value respectively; represents the average value of the predicted sequence and the actual sequence, and N is the number of test samples.

[0026] F, Spatial inversion of regional rainfall erosivity: Substitute the trained random forest model into the covariate set with a spatial resolution of 1 km to perform spatial prediction, and then obtain the spatial inversion result of rainfall erosivity.

[0027] Furthermore, in step A, the rainfall erosivity is calculated by the Xie equation:

[0028]

[0029] where P d represents the rainfall with a daily rainfall greater than 10 mm; α is a regulation parameter, with a value of 0.3937 in the warm season from May to September and a value of 0.3101 from October to April.

[0030] Furthermore, in step B1, the multi-source satellite precipitation remote sensing data includes 7 precipitation datasets, namely TerraClimate, TRMM3B43, CHIRPSv2.0, PERSIANN-CDR, GPMv6, ERA5, and CFSR. Remove the outliers in the data to obtain 7 satellite precipitation variables, and use these 7 multi-source satellite remote sensing precipitation data as independent variables to explain the spatial variation of rainfall erosivity.

[0031] Furthermore, in step B2, specifically: Perform pit filling on the DEM, and then extract the elevation, slope, and aspect variables.

[0032] Furthermore, in step B3, specifically: Based on the DEM data in B2, first convert the DEM raster data into vector point data, and then use the set calculation tool to calculate the geographic longitude and latitude information of each grid point in the attribute table; then, convert the longitude and latitude information in the attribute table into raster surface data through the method of feature to raster; finally, extract the longitude-latitude ratio and longitude-latitude product information through the raster calculator. Thus, 4 sea-land position variables, namely longitude, latitude, longitude-latitude ratio, and longitude-latitude product, of the region are generated.

[0033] Further, step B4 is specifically as follows: The raster data of the 7 satellite remote sensing precipitation variables, 3 topographic variables, and 5 sea-land position variables obtained are projected onto the China_Lambert_Conformal_Conic coordinate system, resampled to a spatial resolution of 1 km using the cubic method, and the spatial range and raster alignment of all raster data are unified, thereby obtaining 15 variable sets.

[0034] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0035] The present invention provides a rainfall erosivity inversion model that applies machine learning algorithms and integrates multi-source satellite remote sensing precipitation data and multi-source environmental variables, which can be used to accurately reconstruct the characteristics of rainfall erosivity in a specified area, and specifically has the following advantages:

[0036] 1. Based on the theory that rainfall erosivity has longitude, latitude, and vertical zonality differentiation characteristics, sea-land position variables, topographic variables, and water vapor variables are further incorporated as explanatory variables, following the complex non-linear law between rainfall erosivity and environmental variables, and using the non-linear fitting of the random forest model to replace the traditional geostatistical and linear relationship fitting methods, thereby improving the inversion accuracy of rainfall erosivity;

[0037] 2. Adding multi-source satellite precipitation data as an explanatory variable enables the inversion result of rainfall erosivity to inherit the advantages of the spatio-temporal distribution continuity of satellite precipitation data, making it possible to obtain high credibility even in areas with sparse ground meteorological stations or no observations; furthermore, it effectively avoids the influence of the distribution of ground meteorological stations on the inversion accuracy of rainfall erosivity;

[0038] 3. By using the random forest model to perform regression fitting on 7 multi-source satellite precipitation data, data noise is suppressed, abnormal or outlier values are filtered out, and the complementary and consistent information between various satellite precipitation data is fused; the influence caused by the deviation of a single satellite data source itself is effectively avoided, and the improvement of the inversion accuracy of rainfall erosivity is promoted;

[0039] 4. It can not only reflect the macro-distribution zonality law of rainfall erosivity in the meridional and zonal directions; but also depict its vertical zonality law, that is, the law of distribution with altitude and slope aspect; it can accurately depict the detailed characteristics of the spatial distribution of rainfall erosivity at the micro-scale;

[0040] Using this as a covariate to assist in rainfall erosivity modeling and spatial inversion greatly improves the inversion accuracy of rainfall erosivity factors and also provides a potential way to improve the spatial resolution of rainfall erosivity data;

[0041] 5. This method can freely introduce more relevant environmental parameters to invert rainfall erosivity, and explore the dependence relationship between other parameters and rainfall erosivity, such as the water vapor mass in clouds, surface temperature, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and thus should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other relevant drawings can also be obtained based on these drawings.

[0043] Figure 1 is the flowchart of the steps of the rainfall erosivity inversion method based on the random forest integrated multi-source satellite precipitation data;

[0044] Figure 2 is the spatial distribution of the collected meteorological station data;

[0045] Figures 3a - 3c is the topographic variable;

[0046] Figures 4a - 4d is the land-sea location variable;

[0047] Figures 5a - 5g is the satellite precipitation data;

[0048] Figure 6 is the parameter optimization process in the random forest model;

[0049] Figure 7 is the spatial inversion result of rainfall erosivity. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0050] The present invention will be further described below in conjunction with the embodiments:

[0051] The present invention will be further described in detail below in conjunction with the drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the present invention, rather than limiting the present invention. Additionally, it should be noted that for the convenience of description, only the parts related to the present invention are shown in the drawings, rather than all the structures.

[0052] It should be noted that: similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used for distinguishing descriptions, and cannot be understood as indicating or implying relative importance.

[0053] Rainfall erosivity is the direct factor causing soil loss and sediment transport, and is also a key physical quantity in soil erosion calculation and land degradation assessment. Most of the existing methods are based on spatial interpolation of meteorological station data or inversion using single-source satellite precipitation data, both of which have problems such as low inversion accuracy. Therefore, the present invention provides a remote sensing inversion method for rainfall erosivity that integrates multi-source remote sensing precipitation products and machine learning. On the Google earth engine remote sensing cloud platform, multi-source remote sensing precipitation products, DEM and other data are obtained. After preprocessing such as numerical extraction, projection transformation, and raster resampling, characteristic variables such as multi-source satellite precipitation, sea-land position, and terrain are obtained. Based on the spatial distribution vector data of meteorological stations, the information of the characteristic variables corresponding to the corresponding points is extracted as covariates, and the rainfall erosivity calculated from the actually observed precipitation data of meteorological stations is used as the dependent variable, and then sample data is constructed. The sample set is randomly divided into a training set (70%) and a validation set (30%). The training set is used to construct a random forest regression model for inverting rainfall erosivity and optimize the parameters. Then, the validation set is used to verify the rainfall erosivity inversion results and perform spatial inversion of rainfall erosivity. The present invention can perform high-precision inversion and mapping of the spatial distribution of rainfall erosivity at large-scale regions or even global scales.

[0054] Specifically, the rainfall erosivity inversion method integrating multi-source satellite precipitation data of the present invention includes the following steps:

[0055] 1. Acquisition and processing of precipitation data of meteorological stations.

[0056] The daily observed precipitation data of 618 ground meteorological stations in the study area in 2019 are obtained from the Resource and Environment Science and Data Center of the Chinese Academy of Sciences (https: / / www.resdc.cn / ), as Figure 2 shown; then, the rainfall erosivity is calculated based on the Xie equation in the prior art, as shown in formula (1).

[0057]

[0058] In the formula, P d represents rainfall with a daily rainfall greater than 10 mm; α is an adjustment parameter, and α takes a value of 0.3937 in the warm season from May to September and a value of 0.3101 from October to April. On this basis, the daily rainfall erosivity is accumulated to obtain the annual rainfall erosivity. The annual rainfall erosivity calculated at the station scale is used as the dependent variable.

[0059] 2. Obtain multi-source satellite precipitation data and DEM data, and extract satellite precipitation, terrain and sea-land position information to construct covariates.

[0060] 21. Obtain 7 kinds of satellite remote sensing precipitation data from the Google Earth Engine remote sensing cloud platform (the data characteristics are shown in Table 1), namely TerraClimate, TRMM3B43, CHIRPSv2.0, PERSIANN-CDR, GPMv6, ERA5, and CFSR (Figure 5). This satellite precipitation data is spatio-temporally consistent with the ground observation precipitation data. Use these 7 kinds of multi-source satellite remote sensing precipitation data as independent variables to explain the spatial variation of rainfall erosivity, giving play to their complementary advantages and avoiding the influence of the deviation of single-source satellite precipitation data on the inversion accuracy of rainfall erosivity.

[0061] 22. Obtain the digital elevation model (DEM) data SRTMDEM v3.0 of the study area from the Google Earth Engine remote sensing cloud platform, with a spatial resolution of 30m. Use SAGA GIS software to perform pit filling on the DEM data of the study area, and then extract elevation, slope, and aspect information (Figure 3). These 3 topographic variables are used as independent variables to explain the variation law of rainfall erosion with topography.

[0062] Table 1 Introduction to the data used for rainfall erosivity inversion

[0063]

[0064]

[0065] 23. Acquisition of the sea-land position variable. To describe the meridional and zonal laws of the spatial distribution of rainfall erosivity, based on the DEM data, convert the DEM grid pixels into point files (shp), and calculate the longitude and latitude corresponding to each spatial point in the WGS-1984 projection coordinate system. Then, extract longitude, latitude, the product of longitude and latitude, and the ratio of latitude to longitude to characterize the sea-land position as sea-land factors and use them as covariates, as shown in Figure 4.

[0066] 24. Covariate data processing. Preprocess the obtained 7 kinds of satellite precipitation grid data, 3 grid data of topographic factors (elevation, slope, aspect), and 4 grid data of sea-land position factors (longitude, latitude, product of longitude and latitude, and ratio of latitude to longitude). That is, use ArcGIS 10.8 software to perform grid reprojection, set it to the China_Lambert_Conformal_Conic projection system; then, use the cubic method for resampling, set its grid spatial resolution to 1km; finally, unify the spatial range of all grid data and perform grid alignment. Use these 14 grid variable sets as covariates.

[0067] 3. Construction of sample data.

[0068] Based on the spatial locations of ground meteorological stations, the covariate values ​​at the corresponding locations were extracted to obtain sample data with a dimension of 618*15 (14 covariates, 1 dependent variable, and 618 meteorological stations).

[0069] 4. Construction of spatial inversion model of rainfall erosivity and parameter optimization. The case area uses rainfall erosivity measured at 618 meteorological stations as the dependent variable, and takes 14 factors such as multi-source satellite remote sensing precipitation data, topography, and sea and land positions as input; first builds a spatial inversion model of rainfall erosivity based on random forest regression on the Rstudio platform; and quantitatively evaluates the accuracy of the model. The spatial inversion model of rainfall erosivity based on random forest makes full use of the advantages of nonlinear machine learning, and can automatically extract features from multi-source satellite remote sensing data, topography, and sea and land positions, establish complex nonlinear relationships, and achieve accurate and objective spatial inversion of rainfall erosivity.

[0070] The specific steps include:

[0071] 41. The sample data is divided into a training set (n=433) and a validation set (n=185) in a ratio of 7:3. The training set is used to build a random forest regression model, and the validation set is used to verify the model accuracy.

[0072] 42. The deepforest library in Python, an open source program platform, was used to construct a random forest model. The dependent variable in the model was rainfall erosivity, and the independent variables were the above 14 covariates. The training set was used to train the random forest regression model. The rainfall erosivity in the training samples of the case area was used as the dependent variable, and the 14 covariates were used as independent variables. A random forest regression algorithm was designed based on the 'caret' package of the Rstudio platform, and a spatial inversion model of rainfall erosivity was constructed.

[0073] During the model training process, the grid search method was used to optimize the hyperparameters ntree and mtry in the random forest algorithm. The search intervals were set to ntree[100, 500] and mtry[2, 8], respectively. When the cross-validation accuracy RMSE of the model training reached the minimum value, the corresponding ntree and mtry parameters were selected as 500 and 5, respectively.

[0074] 5. Accuracy evaluation of spatial inversion model of rainfall erosivity

[0075] In the validation set, the ground-observed precipitation combined with the rainfall erosivity calculated by the Xie equation was used as the true value, and the inversion result of the random forest model was used as the predicted value. The determination coefficient (R 2 ), mean absolute error (MAE), and root mean square error (RMSE) are used as evaluation indicators, and the calculation method is as follows:

[0076]

[0077]

[0078]

[0079] In the formula, y p and y o are the predicted values and the actual rainfall erosivity values calculated by the Xie model; y p and y o represent the average values of the predicted sequence and the actual sequence, and N is the number of test samples. In the present invention, N = 185. Generally, when R 2 is close to 1 and RMSE and MAE are close to 0, it indicates that the accuracy of the fusion model and the rainfall erosivity product is more reliable. Using this verification method, the present invention verifies the retrieved national rainfall erosivity. The results show that its R 2 reaches 0.8765, and MAE and RMSE are 864.56 and 1378.87 (MJ·mm / (hm 2 ·h·a)) respectively, which are better than the traditional methods.

[0080] 6. Spatial inversion results of rainfall erosivity. Substitute the trained random forest model into the raster environmental covariate set with a 1 km spatial resolution to perform spatial prediction, and then obtain the spatial inversion results of rainfall erosivity ( Figure 7 ). Obviously, the significant advantage of this result lies in providing its spatial detail expression ability, especially highlighting the distribution law of rainfall erosivity with changes in sea-land position and terrain undulation. For example, it can be seen that in the southern part of Tibet, it is better on the windward slope and closer to the Indian Ocean, and high-value areas of rainfall erosivity are formed under strong rainfall conditions. Obviously, the present invention reveals the detailed changes in the spatial distribution of rainfall erosivity, which is not available in the existing open-source data products of rainfall erosivity.

[0081] 7. Taking the rainfall erosivity inversion method based on the traditional interpolation method and applying single-source satellite remote sensing precipitation data as a control, select the model verification accuracy parameters R 2 , MAE and RMSE as evaluation indicators to evaluate the effect of the rainfall erosivity spatial inversion strategy based on machine learning and integrating multi-source satellite precipitation data proposed by the present invention.

[0082] 71. Control scheme for spatial inversion of rainfall erosion. Traditional interpolation method: Apply Kriging, inverse distance weighting, and Spline spatial interpolation methods to perform spatial interpolation on the rainfall erosivity in the training set in step 41; Rainfall erosivity inversion scheme using single-source satellite remote sensing precipitation data: Select one of the 7 satellite precipitation data, namely TerraClimate, TRMM3B43, CHIRPSv2.0, PERSIANN-CDR, GPMv6, ERA5, and CFSR, and combine 3 topographic factors (elevation, slope, and aspect) and 4 sea-land position variables (longitude, latitude, ratio of longitude to latitude, and product of longitude and latitude) as independent variables in the model; Use the random forest regression model to fit the relationship between the observed rainfall erosivity at the stations and the independent variables.

[0083] 72. Similarly, use step 5 to evaluate the accuracy of the control scheme for spatial inversion of rainfall erosion, and calculate the verification accuracy of each control strategy respectively, as shown in Table 2.

[0084] 73. Construct an evaluation index for the improvement effect of the method. The formula for calculating the effect improvement evaluation index P is as follows:

[0085] P = (x′ - x) / x × 100%

[0086] In the formula, P indicates the improvement effect of the present invention, x represents the verification accuracy parameters R 2 , MAE, and RMSE of the rainfall erosivity spatial inversion model based on the random forest integration of multi-source satellite remote sensing data proposed by the present invention; x′ is the verification accuracy parameter R 2 , MAE, and RMSE of the control model.

[0087] 74. Calculate the improvement effect of the present invention, as shown in Table 2.

[0088] Table 2 Evaluation of the improvement effect of this method

[0089]

[0090] According to Table 2, the R 2 of the method proposed by the present invention has increased by 26.53% - 31.90% compared with the spatial interpolation method, and the R 2 has increased by 11.43% - 31.90% compared with the single-source satellite precipitation data inversion method.

[0091] The MAE of the method proposed by the present invention has decreased by -44.75% - -41.58% compared with the spatial interpolation method, and the MAE has decreased by -44.75% - -12.37% compared with the single-source satellite precipitation data inversion method.

[0092] The RMSE of the method proposed by the present invention is reduced by -35.29% to -30.90% compared with the spatial interpolation method, and is reduced by 26.68% to -16.11% compared with the single-source satellite precipitation data inversion method.

[0093] The present invention intends to use a random forest model to integrate multi-source satellite precipitation data, terrain, and sea-land position variables to invert rainfall erosivity. Based on the theoretical basis that rainfall erosivity has characteristics of longitude, latitude, and vertical zonality differentiation, sea-land position variables and terrain variables are incorporated as explanatory variables; in addition, 7 new types of multi-source satellite precipitation data are added as explanatory variables; enabling the inversion result of rainfall erosivity to inherit the advantages of the spatio-temporal distribution continuity of satellite precipitation data. The present invention is based on the non-linear inversion of the random forest, which conforms to the complex relationship between the change of rainfall erosivity and sea-land position and terrain, avoiding the limitations of geostatistics and linear fitting, enabling the model to deeply explore the variation law of rainfall erosivity with environmental parameters, and inheriting the advantages of the spatio-temporal continuous distribution of the covariate set, thereby effectively improving the inversion accuracy. The present invention breaks the influence of the density of ground meteorological stations on the inversion accuracy of rainfall erosivity, providing a way for the accurate spatial inversion of rainfall erosivity in areas with sparse or missing station distributions. The present invention improves the spatial resolution of the inversion result of rainfall erosivity, breaks through the defects of fuzzy and generalized information in the inversion results of geostatistics and linear models, realizes the accurate characterization of the detailed information of the rainfall erosivity distribution at the micro-domain scale, and improves the spatial resolution of the rainfall erosivity raster product.

[0094] Note that the above is only the preferred embodiment of the present invention and the technical principles applied. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, re-adjustments, and substitutions can be made by those skilled in the art without departing from the protection scope of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments only. Without departing from the concept of the present invention, it can also include more other equivalent embodiments, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A rainfall erosivity inversion method integrating multi-source satellite precipitation data, characterized in that: The following steps are involved: A. Acquisition and processing of precipitation data at ground meteorological stations: Obtain the observed precipitation data at ground meteorological stations, and then calculate the rainfall erosivity of each station; B. Covariate data acquisition and processing: Obtain 7 types of remote sensing precipitation data of the same period and the same area as the meteorological station, download the DEM data, and then extract the covariate data set composed of multi-source satellite precipitation data, terrain, and sea and land positions; B1. Processing of multi-source satellite precipitation remote sensing data; B2, terrain variable processing; B3, sea route position variable processing; B4, covariate data preprocessing; C. Constructing sample data: Taking the annual rainfall erosivity obtained from the meteorological station data as the dependent variable, combined with the spatial location information of the meteorological station, extract the numerical information of 15 covariates at the corresponding location, and then construct the sample set; D. Constructing a random forest regression model for spatial inversion of rainfall erosivity: The rainfall erosivity in the sample data is set as the dependent variable, and the above 15 covariates are used as independent variables. The random forest regression model is trained, and the grid search method is used to optimize the parameters of the random forest model. D1, randomly select 70% of the sample data as the training set, and the remaining 30% of the samples as the validation set; D2, set the dependent variable to the rainfall erosivity in the sample data, and the independent variables to the above 15 covariates, use the random forest regression algorithm to fit the relationship, and use the grid search method to optimize the hyperparameters mtry and ntree in the model; E. Accuracy evaluation of rainfall erosivity inversion model: Substitute the trained random forest model into the validation set, taking the rainfall erosivity in the validation set as the true value and the random forest model inversion as the predicted value, and calculate the determination coefficient R between the two. 2 , root mean square error RMSE, and mean error ME are used as evaluation indicators to evaluate the inversion accuracy of rainfall erosivity, where the calculation process is as follows: ME=y p -and o (9) In the formula, y p ,y o is the predicted value and the actual rainfall erosivity value; It represents the average value of the predicted sequence and the actual sequence, and N is the number of test samples. F. Spatial inversion of regional rainfall erosivity: The trained random forest model is substituted into the covariate set with a spatial resolution of 1 km to perform spatial prediction and obtain the spatial inversion results of rainfall erosivity.

2. The method for inverting rainfall erosivity by integrating multi-source satellite precipitation data according to claim 1, characterized in that: Step A, rainfall erosivity is calculated by Xie equation: Where P d It indicates rainfall with daily rainfall greater than 10 mm. α is the adjustment parameter, which is 0.3937 in the warm season from May to September and 0.3101 in October to April.

3. The method for inverting rainfall erosivity by integrating multi-source satellite precipitation data according to claim 1, characterized in that: Step B1, the multi-source satellite precipitation remote sensing data includes 7 precipitation data sets, namely TerraClimate, TRMM3B43, CHIRPSv2.0, PERSIANN-CDR, GPMv6, ERA5 and CFSR. Outliers in the data are removed to obtain 7 satellite precipitation variables. These 7 multi-source satellite remote sensing precipitation data are used as independent variables to explain the spatial variation of rainfall erosivity.

4. The method for inverting rainfall erosivity by integrating multi-source satellite precipitation data according to claim 1, characterized in that: Step B2, specifically: fill the DEM and extract the altitude, slope and aspect variables.

5. The method for inverting rainfall erosivity by integrating multi-source satellite precipitation data according to claim 1, characterized in that: Step B3 is as follows: based on the DEM data in B2, first convert the DEM grid data into vector point data, and then use the set calculation tool to calculate the geographic longitude and latitude information of each grid point in the attribute table; then, convert the longitude and latitude information in the attribute table into grid surface data by using the method of converting elements to raster; finally, extract the longitude and latitude ratio and longitude and longitude product information by using the raster calculator. Thus, a total of four land and sea position variables are generated, namely the longitude, latitude, longitude and latitude ratio, and longitude and longitude product of the region.

6. The method for inverting rainfall erosivity by integrating multi-source satellite precipitation data according to claim 1, characterized in that: Step B4, specifically: project the obtained raster data of 7 satellite remote sensing precipitation variables, 3 terrain variables, and 5 land and sea position variables to the China_Lambert_Conformal_Conic coordinate system, and resample to 1km spatial resolution using the cubic method, and unify the spatial range and grid alignment of all raster data, thereby obtaining 15 variable sets.

Images