An improved RF-based multi-source rainfall fusion method, device and storage medium

Abstract

The present disclosure relates to the technical field of data processing, and discloses a multi-source rainfall fusion method based on improved RF, equipment and a storage medium. The method comprises: performing scale reduction processing on first rainfall data corresponding to rainfall products of a target region based on a target scale reduction model and first influence factor data of rainfall influence factors of the target region to obtain third rainfall data; performing rainfall state and / or rainfall level classification on the third rainfall data based on the first influence factor data and a target classification model to obtain first classification data; and performing fusion on the third rainfall data and second rainfall data corresponding to a rain gauge station of the target region based on the first influence factor data, the first classification data and a target fusion model to obtain fused rainfall data; wherein at least one of the target scale reduction model, the target classification model and the target fusion model is obtained based on a random forest model. The present disclosure can solve the fusion problem of rain gauge rainfall data and satellite rainfall data.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of data processing, in particular to a multi-source rainfall fusion method based on improved RF, a device and a storage medium. BACKGROUND

[0002] At present, rainfall data is mainly observed by rainfall stations and satellite rainfall products. The rainfall data measured by the rainfall stations is affected by uneven distribution of rainfall at the stations, and the rainfall data of the rainfall stations can only reflect the local spatial distribution of rainfall, and it is difficult to accurately depict the precipitation in the entire region. The satellite rainfall data measured by the satellite rainfall products has insufficient quantitative accuracy. Therefore, in order to make up for the deficiency of a single data source, a fusion method for rainfall data of rainfall stations and satellite rainfall data is urgently needed. SUMMARY

[0003] Therefore, the present disclosure provides a multi-source rainfall fusion method based on improved RF, an electronic device and a storage medium to solve the fusion problem of rainfall data of rainfall stations and satellite rainfall data.

[0004] In a first aspect, the present disclosure provides a multi-source rainfall fusion method based on improved RF, the method comprising:

[0005] obtaining first rainfall information of a target region; wherein the first rainfall information comprises first rainfall data, second rainfall data and first influence factor data of a rainfall influence factor; the first rainfall data is obtained by a rainfall product of the target region; the second rainfall data is obtained by a rainfall station of the target region;

[0006] performing downscaling processing on the first rainfall data based on the first influence factor data and a target downscaling model to obtain third rainfall data;

[0007] performing rainfall classification on the third rainfall data based on the first influence factor data and a target classification model to obtain first classification data; wherein the rainfall classification comprises rainfall state classification and / or rainfall intensity classification;

[0008] performing fusion on the second rainfall data and the third rainfall data based on the first influence factor data, the first classification data and a target fusion model to obtain fused rainfall data; wherein at least one of the target downscaling model, the target classification model and the target fusion model is obtained based on a random forest model.

[0009] In a second aspect, the present disclosure provides an electronic device, comprising a memory and a processor, which are communicatively connected with each other, and the memory stores computer instructions, and the processor executes the computer instructions to perform the improved RF-based multi-source rainfall fusion method of the first aspect or any of the corresponding embodiments.

[0010] In a third aspect, the present disclosure provides a computer-readable storage medium, which stores computer instructions for causing a computer to perform the improved RF-based multi-source rainfall fusion method of the first aspect or any of the corresponding embodiments.

[0011] The improved RF-based multi-source rainfall fusion method provided by the embodiments of the present disclosure first performs scale reduction on the first rainfall data based on the first influence factor data and a target scale reduction model to obtain third rainfall data with spatial correlation. Then, the third rainfall data is classified in terms of rainfall state and / or rainfall level based on the first influence factor data and a target classification model to obtain first classification data. Subsequently, the second rainfall data and the third rainfall data are fused based on the first influence factor data, the first classification data, and a target fusion model to obtain fused rainfall data. Therefore, the classification data of the rainfall state and the rainfall level can be introduced when fusing the rainfall data, and the nonlinear relationship between the influence factor data of different rainfall states, rainfall levels, and rainfall influence factors can be utilized to improve the problem that the fusion accuracy of the rainfall data decreases with the increase of the rainfall level, thereby effectively fusing the rainfall data of the rain gauge and the rainfall data of the rainfall product and improving the fusion accuracy of the rainfall data.

[0012] The beneficial effects of the electronic device and the storage medium correspond to those of the improved RF-based multi-source rainfall fusion method, which will not be described herein again. BRIEF DESCRIPTION OF DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the specific embodiments of the present disclosure or the prior art, the drawings needed in the specific embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present disclosure, and other drawings can be obtained by those skilled in the art without creative labor.

[0014] Figure 1 is a flowchart of an improved RF-based multi-source rainfall fusion method according to an embodiment of the present disclosure;

[0015] Figure 2 is a flowchart of a construction method of a target scale reduction model according to an embodiment of the present disclosure;

[0016] Figure 3 is a flowchart of a construction manner of a target classification model according to an embodiment of the present disclosure;

[0017] Figure 4 is a flowchart of another multi-source rainfall fusion method based on improved RF according to an embodiment of the present disclosure;

[0018] Figure 5 is a structural block diagram of a multi-source rainfall fusion device based on improved RF according to an embodiment of the present disclosure;

[0019] Figure 6 is a structural block diagram of an electronic device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0020] To make the objectives, technical solutions, and superiorities of the embodiments of the present disclosure clearer, the technical solutions in the embodiments of the present disclosure will be described below in a clear and complete manner with reference to the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are some but not all of the embodiments of the present disclosure. Based on the embodiments in the present disclosure, all other embodiments obtained by a person of ordinary skill in the art without creative effort are within the protection scope of the present disclosure.

[0021] Continuous and accurate observation of rainfall can obtain a comprehensive surface environment to provide key information for the spatio-temporal distribution of global water cycle, and help to optimize hydrological model parameters and improve the prediction accuracy of hydrological models. At present, rainfall data is mainly observed through rain gauges (i.e., ground stations) and satellite rainfall products. Among them, the rainfall data measured by rain gauges is affected by uneven distribution of rainfall at the station, and the rainfall data can only reflect the local spatial distribution of rainfall, and it is difficult to accurately depict the rainfall in the entire region. Compared with the rainfall data of rain gauges, satellite rainfall products can greatly improve the ability to capture the spatial distribution characteristics of rainfall, but the quantitative accuracy is still insufficient. For example, the uncorrected IMERG (Integrated Multi-satellitE Retrievals for GPM) data has the problem of high quantitative accuracy for rainfall data with small rainfall amount and low quantitative accuracy for rainfall data with large rainfall amount.

[0022] Based on this, it is considered to fuse rain gauge rainfall data and satellite rainfall data to make up for the shortcomings of a single data source in a way of taking the advantages and making up the disadvantages, so as to improve the accuracy of spatio-temporal distribution of rainfall, provide more reliable input data for hydrological models, and improve the accuracy and application effect of hydrological prediction.

[0023] In related technologies, rainfall fusion methods mainly include two categories of statistical methods and data-driven methods.

[0024] Among them, statistical methods such as geographically weighted regression, quantile mapping and Bayesian model averaging have good application effects in some cases, but these methods are limited by strong data assumptions. For example, quantile mapping can eliminate bias within a statistical period, but cannot reflect the duration of dry and wet days and interannual variability of rainfall. Most importantly, statistical methods are difficult to depict the relationship between rainfall process and complex environmental variables.

[0025] In contrast, data-driven methods can more accurately capture complex rainfall patterns by automatically learning high-dimensional nonlinear relationships between different rainfall data, and are therefore widely used. Satellite rainfall products used by data-driven methods (i.e., products that use satellite remote sensing technology to obtain rainfall data) have different capture capabilities for rainfall spatial distribution characteristics, which are affected by inversion algorithms, satellite sensors and sampling frequencies. Different satellite rainfall products have significant differences in capturing rainfall spatial distribution characteristics, making each satellite rainfall product have certain limitations in capturing rainfall spatial distribution characteristics. Therefore, the fusion effect of rainfall station rainfall data and single satellite rainfall data varies at different times, regions and product types. Since no satellite rainfall product can perform well at all times and spatial scales, this may also lead to less than ideal hydrological simulation results in different regions.

[0026] Therefore, in order to better exploit the advantages of various satellite rainfall products, fusing multiple satellite rainfall data with rainfall station rainfall data may further improve the accuracy of rainfall spatial estimation.

[0027] Most current data-driven methods use machine learning methods to fuse rainfall data. Since the time series of rainfall data fused by this method is usually 10 to 20 years, it is often difficult to meet the demand for long sequence rainfall data in hydrological applications. Moreover, the fusion accuracy of rainfall data fused by machine learning methods often decreases significantly with increasing rainfall intensity, especially for extreme rainfall events. The reason for this decrease in accuracy may be that when using data-driven methods such as Random Forest (RF) to fuse rainfall data, only the regression relationship between rainfall state classification and environmental variables is considered, while different rainfall intensities and environmental variables may have different regression relationships, resulting in a decrease in fusion accuracy of rainfall data with increasing rainfall intensity.

[0028] It can be seen that when using machine learning methods to fuse rainfall data, rainfall state and rainfall intensity are the key to fundamentally improving the fusion accuracy of rainfall data.

[0029] Therefore, in the embodiments of the present disclosure, it is intended to improve the fusion accuracy of rainfall data by classifying the rainfall state and the rainfall intensity level at the same time, and constructing the regression relationship between different rainfall intensity levels and environmental variables respectively by using a machine learning method.

[0030] Therefore, according to the embodiments of the present disclosure, an improved RF-based multi-source rainfall fusion method is provided. It should be noted that the steps shown in the flowchart can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in a different order.

[0031] In the embodiments, an improved RF-based multi-source rainfall fusion method is provided, which can be used in electronic devices such as personal computers, server computers, mobile devices, etc., and the electronic device is used to fuse rainfall data from multiple sources. Figure 1 is a flowchart of an improved RF-based multi-source rainfall fusion method according to the embodiments of the present disclosure, as shown in Figure 1 The flowchart includes the following steps:

[0032] In step S101, first rainfall information of a target area is obtained; wherein the first rainfall information includes first rainfall data, second rainfall data, and first impact factor data of a rainfall impact factor; the first rainfall data is obtained by a rainfall product of the target area; the second rainfall data is obtained by a rain gauge of the target area. At least one of the target downscaling model, the target classification model, and the target fusion model is obtained based on a random forest model.

[0033] The rainfall impact factor refers to an environmental factor that affects the rainfall data. Optionally, the rainfall impact factor includes at least one of a climate factor and a geographical spatial factor, the first impact factor data includes at least one of first climate data and first geographical spatial data, the first geographical spatial data includes at least one of first longitude, first latitude, first slope, first angle, and first elevation data, which can be adjusted according to actual conditions.

[0034] It should be noted that the rainfall product is a product for measuring rainfall data, and the rainfall product includes at least one of a satellite rainfall product, a remote sensing rainfall product, and a reanalysis rainfall product. The satellite rainfall product includes a product obtained by using satellite technology to obtain rainfall data, the remote sensing rainfall product includes a product obtained by using remote sensing technology to obtain rainfall data, and the reanalysis rainfall product includes a product obtained by fusing multi-source observation data and numerical model output meteorological data. In addition, the rainfall product can also include other types of products for measuring rainfall data, which are not limited herein. At present, most of the rainfall products are low-resolution rainfall products.

[0035] Specifically, the step S101 comprises:

[0036] The step a1 acquires original rainfall information of the target area; wherein, the original rainfall information comprises original rainfall product data, original rain gauge data and original impact factor data of rainfall impact factors; the original rainfall product data comprises rainfall data measured by rainfall products of the target area; the original rain gauge data comprises rainfall data measured by rain gauges of the target area.

[0037] The step a2 pre-processes the original rainfall information to obtain first rainfall information; wherein, the spatial resolution of the first rainfall information is higher than that of the original rainfall information.

[0038] Optionally, the pre-processing comprises at least one of data interpolation extension, data spatial interpolation and resampling. In addition, the pre-processing also comprises data conversion and the like, which can be adjusted according to actual conditions.

[0039] Specifically, the resampling comprises using a bilinear interpolation method to resample longitude, latitude, slope, angle and elevation data of a Digital Elevation Model (DEM) with an original spatial resolution to a first spatial resolution (such as 1km spatial resolution) and / or a second spatial resolution consistent with the rainfall product, wherein the first spatial resolution is higher than the second spatial resolution.

[0040] Specifically, the data conversion comprises performing daily scale accumulation on data in the original rainfall information to obtain daily scale original rainfall information. Wherein, before performing the daily scale accumulation, the time of data (such as original rainfall product data, climate data in original impact factor data) in the original rainfall information can be converted into a preset standard time. Then, the daily scale original rainfall information is accumulated in a daily scale to obtain daily scale original rainfall information.

[0041] The step S102 performs downscaling processing on the first rainfall data based on the first impact factor data and a target downscaling model to obtain third rainfall data.

[0042] Specifically, the first impact factor data and the first rainfall data are input into the target downscaling model for downscaling processing to obtain the third rainfall data.

[0043] It should be noted that the data-driven method such as the random forest in the related art does not consider the geographical position or the spatial structure of the rainfall data when performing spatial prediction on the rainfall data, but regards the rainfall data as independent data, so that the consideration of spatial autocorrelation is lacked in spatial data analysis. Therefore, the embodiment constructs a target downscaling model considering spatial correlation to perform downscaling processing on the first rainfall data, so as to estimate the third rainfall data with spatial correlation.

[0044] The target downscaling model is constructed based on a regression relationship among the original rainfall data of the rainfall product, the rainfall data of the rainfall product of the second spatial resolution, and the influence factor data of the rainfall influence factor of the second spatial resolution.

[0045] In step S103, the third rainfall data is classified based on the first influence factor data and the target classification model to obtain first classification data; wherein the rainfall classification includes rainfall state classification and / or rainfall intensity classification.

[0046] Specifically, the first influence factor data and the third rainfall data are input into the target classification model for rainfall classification to obtain the first classification data.

[0047] The target classification model is constructed based on a regression relationship among the classification data of the original rainfall data of the rain gauge at the same position, the output data of the target downscaling model, and the influence factor data of the rainfall influence factor of the first spatial resolution.

[0048] It should be noted that the rainfall state classification refers to dividing the rainfall data input into the target classification model into no rain data and rain data according to the rainfall state. The rainfall intensity classification refers to classifying the rain data according to the rainfall intensity, such as moderate rain event, heavy rain event, and extreme rainfall event.

[0049] In step S104, the second rainfall data and the third rainfall data are fused based on the first influence factor data, the first classification data, and the target fusion model to obtain fused rainfall data; wherein at least one of the target downscaling model, the target classification model, and the target fusion model is obtained based on a random forest model.

[0050] Specifically, the first influence factor data, the second rainfall data, the third rainfall data, and the first classification data are input into the target fusion model for rainfall data fusion to obtain the fused rainfall data.

[0051] The target fusion model is constructed based on a regression relationship between the original rainfall data of the rainfall station at the same location, the influence factor data of the rainfall influence factor at the first spatial resolution, the output data of the target downscaling model, the rainfall data of the rainfall station at the first spatial resolution, and the classification data of the original rainfall data of the rainfall station. In this embodiment, the target fusion model is used to improve the classification accuracy of the rain event.

[0052] For example, the first influence factor data of all grid points at the 1km spatial resolution, the second rainfall data at the 1km spatial resolution, the third rainfall data of all grid points at the 1km spatial resolution, and the first classification data are input into the target fusion model to obtain the fused rainfall data at the 1km spatial resolution.

[0053] In actual applications, the target downscaling model can be constructed based on a random forest model, the target classification model and the target fusion model can be constructed based on machine learning models such as a Gradient Boosting Decision Tree (GBDT) and an eXtreme Gradient Boosting (XGBoost) model. Alternatively, the target downscaling model and the target classification model can be constructed based on a random forest model, and the target fusion model can be constructed based on machine learning models such as a Gradient Boosting Decision Tree and an eXtreme Gradient Boosting model. Alternatively, the target downscaling model and the target fusion model can be constructed based on a random forest model, and the target classification model can be constructed based on machine learning models such as a Gradient Boosting Decision Tree and an eXtreme Gradient Boosting model. Alternatively, the target classification model and the target fusion model can be constructed based on a random forest model, and the target downscaling model can be constructed based on machine learning models such as a Gradient Boosting Decision Tree and an eXtreme Gradient Boosting model. Alternatively, the target downscaling model, the target classification model, and the target fusion model can be constructed based on a random forest model, and no limitation is made herein.

[0054] The multi-source rainfall fusion method provided in this embodiment first performs downscaling processing on the first rainfall data based on the first influence factor data and the target downscaling model to obtain the third rainfall data with spatial correlation. The third rainfall data is classified based on the first influence factor data and the target classification model to obtain the first classification data. Then, the second rainfall data and the third rainfall data are fused based on the first influence factor data, the first classification data, and the target fusion model to obtain the fused rainfall data. Therefore, the classification data of the rainfall state and the rainfall intensity can be introduced when the rainfall data is fused, the nonlinear relationship between the influence factor data of different rainfall states, rainfall intensities, and rainfall influence factors is utilized, the problem that the fusion accuracy of the rainfall data decreases with the increase of the rainfall intensity is improved, and thus the rainfall data of the rainfall station and the rainfall data of the rainfall product can be effectively fused, and the fusion accuracy of the rainfall data is improved.

[0055] In some optional embodiments, referring to Figure 2 The target downscaling model is obtained by the following way:

[0056] In step S201, second rainfall information of one or more sample areas is obtained; the second rainfall information includes first rainfall product data and second impact factor data of a rainfall impact factor; the first rainfall product data includes rainfall data measured by a rainfall product of the sample area.

[0057] It should be noted that the sample area can be consistent with the target area, or can not be consistent, which is not limited here.

[0058] Optionally, the second impact factor data includes at least one of second climate data and second geospatial data, and the second geospatial data includes at least one of second longitude, second latitude, second slope, second angle and second elevation data, which can be adjusted according to actual conditions.

[0059] In step S202, the second rainfall information is preprocessed to obtain third rainfall information; the spatial resolution of the third rainfall information is consistent with that of the second rainfall information; the third rainfall information includes second rainfall product data and third impact factor data.

[0060] The third rainfall information is consistent with the spatial resolution of the rainfall product. The second rainfall product data corresponds to the first rainfall product data, and the third impact factor data corresponds to the second impact factor data.

[0061] For example, the preprocessing includes data space interpolation, the second rainfall product data is obtained by data space interpolation on the first rainfall product data, and the third impact factor data is obtained by data space interpolation on the second impact factor data.

[0062] It should be noted that the related steps of preprocessing the second rainfall information can refer to the description of preprocessing above, which will not be repeated here.

[0063] In step S203, a target downscaling model is constructed based on the regression relationship between the first rainfall product data, the second rainfall product data and the third impact factor data.

[0064] Optionally, a regression model is constructed based on the regression relationship between the first rainfall product data, the second rainfall product data and the third impact factor data to obtain the target downscaling model.

[0065] Optionally, a first random forest model is used to construct the regression relationship between the first rainfall product data, the second rainfall product data and the third impact factor data to obtain the target downscaling model.

[0066] Specifically, the downscaling formula of the target downscaling model is represented by the following formula:

[0067]

[0068] wherein f downscale represents a regression relationship between the first rainfall product data, the second rainfall product data and the third influence factor data, MSP is the first rainfall product data, is the second rainfall product data, is the third influence factor data, e L is the first preset residual, is the third rainfall data, is the first rainfall data, is the first influence factor data, e 1km is the second preset residual.

[0069] It should be noted that the residual correction can be ignored in the downscaling process of the target downscaling model.

[0070] The multi-source rainfall fusion method provided in the embodiment is based on the regression relationship between the first rainfall product data originally collected by the rainfall product, the second rainfall product data obtained after preprocessing and the third influence factor data of the rainfall influence factor, and a target downscaling model is constructed. Therefore, the target downscaling model can learn the correlation between the rainfall data of the rainfall product and the rainfall influence factor, so as to accurately estimate the spatial correlation of the input rainfall data.

[0071] In some optional embodiments, the second rainfall information further includes first rain gauge data, and the first rain gauge data includes rainfall data measured by a rain gauge in the sample area. Referring to Figure 3 , the target classification model is obtained by the following way:

[0072] In step S301, the first rain gauge data is classified according to the rainfall state and the rainfall level to obtain second classification data.

[0073] In some optional embodiments, the above step S301 includes:

[0074] In step b1, the first rain gauge data is classified according to the rainfall state to obtain first classification data; wherein the first classification data includes rain data and no rain data.

[0075] Specifically, the rainfall state includes rain state and no rain state, and the first rain gauge data is divided into rain data and no rain data according to the rainfall state.

[0076] Specifically, the preset rainfall can be used as a classification standard of rain and no rain to classify the first rainfall station data.

[0077] Optionally, the preset rainfall is 0.1 mm / d, that is, the rainfall data of [0, 0.1) mm / d in the first rainfall station data is classified as no rain data, and the rainfall data greater than or equal to 0.1 mm / d in the first rainfall station data is classified as rain data. In actual application, the preset rainfall can be adjusted according to actual conditions, for example, 0.09 mm / d, 0.11 mm / d, etc., which is not limited herein.

[0078] In step b2, the rain data is classified according to the rain level to obtain secondary classification data, so as to obtain second classification data; wherein the second classification data includes the secondary classification data and the no rain data.

[0079] Specifically, the rain data is classified according to the rain level to obtain rainfall data of different rain levels, so as to obtain second classification data.

[0080] Further, the classification of the rain data according to the rain level in step b2 to obtain the secondary classification data includes: obtaining a mapping relationship between at least one preset rain level and a rainfall data range; and classifying the rain data in the rain data according to the rain level based on the mapping relationship to obtain the secondary classification data.

[0081] Specifically, the at least one preset rain level includes a first rain level, a second rain level and a third rain level, and the mapping relationship indicates that the first rain level corresponds to a first rain data range, the second rain level corresponds to a second rain data range, and the third rain level corresponds to a third rain data range. If the rain data in the rain data is located in the first rain data range, the rain data is classified as the first rain level. If the rain data in the rain data is located in the second rain data range, the rain data is classified as the second rain level. If the rain data in the rain data is located in the third rain data range, the rain data is classified as the third rain level.

[0082] Optionally, the first rain level represents a moderate rain event, the second rain level represents a heavy rain event, and the third rain level represents an extreme rain event. The first rain data range is [0.1, 20) mm / d, the second rain data range is [20, u) mm / d, and the third rain data range is greater than u mm / d. Wherein, u is the 98th percentile of the daily rainfall of each rainfall station.

[0083] Further, the rain data of different rainfall levels in the rain data without rain and the rain data with rain can be marked by using classification identifiers, for example, the classification identifier of the rain data without rain is assigned as 0, the classification identifier of the rain data of the first rainfall level is assigned as 1, the classification identifier of the rain data of the second rainfall level is assigned as 2, and the classification identifier of the rain data of the third rainfall level is assigned as 3.

[0084] In step S302, the second rainfall information is preprocessed to obtain fourth rainfall information; wherein the spatial resolution of the fourth rainfall information is higher than the spatial resolution of the second rainfall information; and the fourth rainfall information includes third rainfall product data, fourth influence factor data, and second rain gauge data.

[0085] The fourth rainfall information is high-resolution data. The third rainfall product data corresponds to the first rainfall product data, the fourth influence factor data corresponds to the second influence factor data, and the second rain gauge data corresponds to the first rain gauge data.

[0086] For example, the preprocessing includes data spatial interpolation, the third rainfall product data is obtained by performing data spatial interpolation on the first rainfall product data, the fourth influence factor data is obtained by performing data spatial interpolation on the second influence factor data, and the second rain gauge data is obtained by performing data spatial interpolation on the first rain gauge data.

[0087] It should be noted that the related steps of preprocessing the second rainfall information in step S302 can refer to the description of the preprocessing described above, and will not be described here.

[0088] In some optional embodiments, the multi-source rainfall fusion method of the present disclosure further includes: taking the grid of the digital elevation model of the sample area as a reference, performing position matching on the third rainfall information and the fourth rainfall information to obtain a position matching result.

[0089] The multi-source rainfall fusion method provided in the embodiment takes the grid of the digital elevation model of the sample area as a reference to perform position matching on the third rainfall information and the fourth rainfall information, so that the different data source grids can be ensured to be one-to-one corresponding in space, so as to facilitate corresponding processing of the data in the third rainfall information and the fourth rainfall information.

[0090] In step S303, the third rainfall product data is processed by a target downscaling model based on the fourth influence factor data to obtain downscaling data.

[0091] Specifically, the third rainfall product data and the fourth influence factor data are input into the target downscaling model to estimate the downscaling data with spatial correlation.

[0092] It should be noted that the spatial resolution of the target downscaling model is consistent with the spatial resolution of the third rainfall product data. For example, if the spatial resolution of the third rainfall product data is 1 km, then the downscaling data output by the target downscaling model is 1 km rainfall data.

[0093] In step S304, a target classification model is constructed based on the regression relationship between the second classification data, the downscaling data, and the fourth influence factor data at the same location.

[0094] Specifically, the regression relationship between the second classification data, the downscaling data, and the fourth influence factor data at the same location is constructed based on a preset classification model to obtain the target classification model.

[0095] Optionally, the preset classification model is a second random forest model, that is, the regression relationship between the second classification data, the downscaling data, and the fourth influence factor data at the same location is constructed based on the second random forest model to obtain the target classification model. In addition, other classification models can also be selected according to actual conditions, for example, gradient boosting decision tree, extreme gradient boosting model, etc., which are not limited herein.

[0096] It should be noted that the same location in step S304 refers to the location of the grid point where the first rain gauge station is located.

[0097] Specifically, the target classification model is represented by the following formula:

[0098]

[0099] wherein f class represents the regression relationship between the second classification data, the downscaling data, and the fourth influence factor data at the same location, Gauge class is the second classification data, is the downscaling data of the grid point where the first rain gauge station is located, is the fourth influence factor data of the grid point where the first rain gauge station is located, is the first classification data, is the third rainfall data, is the first influence factor data.

[0100] The multi-source rainfall fusion method provided in the embodiment is based on the regression relationship between the second classification data, the downscaling data, and the fourth influence factor data at the same location to construct a target classification model. Therefore, the nonlinear relationship between the influence factor data of different rainfall states, rainfall levels, and rainfall influence factors can be constructed to improve the classification accuracy of the rainfall data.

[0101] In some optional embodiments, the target fusion model is obtained by constructing a regression relationship among the first rain gauge data, the downscaling data, the second rain gauge data, the fourth influence factor data and the second classification data at the same location.

[0102] It should be noted that the same location refers to the location of the grid point where the same rain gauge is located.

[0103] Optionally, a regression relationship among the second classification data at the same location, the downscaling data and the fourth influence factor data is constructed based on the third random forest model to obtain a target classification model. In addition, other fusion models can also be selected according to actual conditions, for example, gradient boosting decision tree, extreme gradient boosting model, etc., which are not limited herein.

[0104] Specifically, the target fusion model is represented by the following formula:

[0105]

[0106] wherein f merge represents a regression relationship among the first rain gauge data, the downscaling data, the second rain gauge data, the fourth influence factor data and the second classification data at the same location, Gauge is the first rain gauge data, is the downscaling data of the grid point where the rain gauge is located, is the second rain gauge data of the grid point where the rain gauge is located, is the fourth influence factor data of the grid point where the rain gauge is located, Gauge class is the second classification data, MSMP 1km is the fused rainfall data, is the third rainfall data, is the second rainfall data, is the first influence factor data, is the first classification data.

[0107] The multi-source rainfall fusion method provided in the embodiment is based on a regression relationship among the first rain gauge data, the downscaling data, the second rain gauge data, the fourth influence factor data and the second classification data at the same location to construct a target fusion model. Therefore, the target fusion model can learn the regression relationship among the rainfall influence factor, the rainfall state and the rainfall level, thereby improving the fusion accuracy of the rainfall data.

[0108] In some optional embodiments, the regression relationship between the first rainfall product data, the second rainfall product data, and the third influence factor data in step S203 is used to build the target downscaling model, including: using the first random forest model to build the regression relationship between the first rainfall product data, the second rainfall product data, and the third influence factor data, to obtain the target downscaling model. The regression relationship between the second classification data, the downscaling data, and the fourth influence factor data at the same location in step S304 is used to build the target classification model, including: using the second random forest model to build the regression relationship between the second classification data, the downscaling data, and the fourth influence factor data at the same location, to obtain the target classification model. The regression relationship between the first rain gauge data, the downscaling data, the second rain gauge data, the fourth influence factor data, and the second classification data at the same location is used to build the target fusion model, including: using the third random forest model to build the regression relationship between the first rain gauge data, the downscaling data, the second rain gauge data, the fourth influence factor data, and the second classification data at the same location, to obtain the target fusion model.

[0109] It can be understood that at least one of the target downscaling model, the target classification model, and the target fusion model is obtained based on the regression relationship between the input and the output of the random forest model. In actual application, the regression relationship between the first rainfall product data, the second rainfall product data, and the third influence factor data can be obtained by using only the first random forest model to build the regression relationship, to obtain the target downscaling model. The regression relationship between the second classification data, the downscaling data, and the fourth influence factor data at the same location can be obtained by using other machine learning models to build the regression relationship, to obtain the target classification model, and the regression relationship between the first rain gauge data, the downscaling data, the second rain gauge data, the fourth influence factor data, and the second classification data at the same location can be obtained by using other machine learning models to build the regression relationship, to obtain the target fusion model.

[0110] Alternatively, the regression relationship between the second classification data, the downscaling data, and the fourth influence factor data at the same location can be obtained by using only the second random forest model to build the regression relationship, to obtain the target classification model. The regression relationship between the first rainfall product data, the second rainfall product data, and the third influence factor data can be obtained by using other machine learning models to build the regression relationship, to obtain the target downscaling model, and the regression relationship between the first rain gauge data, the downscaling data, the second rain gauge data, the fourth influence factor data, and the second classification data at the same location can be obtained by using other machine learning models to build the regression relationship, to obtain the target fusion model, without limitation.

[0111] In order to facilitate a deep understanding of the multi-source rainfall data method of the present disclosure, the multi-source rainfall data method of the present disclosure is described below with a specific application example. In this example, the overall process of the multi-source rainfall data method of the present disclosure includes the following steps:

[0112] Step S1, according to the rainfall information shown in Table 1, collecting the original rainfall data MSP of the rainfall product of the target area 测 , the original rainfall data of the rain gauge Gauge 测 , the original influence factor data C of the rainfall influence factor 测 , to obtain the original rainfall information. Among them, the original rainfall data MSP 测 includes the rainfall of reanalysis data and the rainfall of remote sensing data. The original influence factor data C 测 includes climate data and geographic spatial data. The climate data includes total column water vapour (TCWV). The geographic spatial data includes longitude, latitude, slope, angle, and elevation data.

[0113] Table 1 Rainfall data and influence factor information to be collected in the target area

[0114]

[0115] Step S2, pre-processing the collected original rainfall information by data interpolation extension, data spatial interpolation, resampling, and data conversion.

[0116] Data interpolation extension: for the original rainfall data of the rain gauge Gauge 测 , considering the data sequence length and data integrity, the Gauge 测 is interpolated and extended according to the following requirements: (1) the data record starts before 1980. (2) the total missing data percentage <0.5%. Among them, for the rainfall data missing for 1-2 days, the average value of adjacent days is used for filling. For the continuously missing rainfall data, the average value of the long data sequence of the same day in different years is used for interpolation.

[0117] Data space interpolation: For the original rainfall data of rainfall products APHRODITE, CHIRPS, GSMaP, IMERG, ERA5Land, the time resolution is set to 1d, 1d, 1h, 0.5h, 1h respectively, and the spatial resolution is set to 0.25°x0.25°, 0.05°x0.05°, 0.1°x0.1°, 0.1°x0.1°, 0.25°x0.25° respectively. The climate data includes total column water vapour (TCWV) with a time resolution of 1h and a spatial resolution of 0.25°x0.25°. The geospatial data includes longitude, latitude, slope, angle and elevation data with a spatial resolution of 90m x 90m. On this basis, for the original rainfall data of rainfall products, the ordinary Kriging interpolation method is used to re-estimate the data with its 5-8 nearest neighboring data, so as to obtain the interpolated rainfall data. The interpolated rainfall data includes two kinds, the first kind is low-resolution rainfall data consistent with the original spatial resolution of the rainfall product The second kind is high-resolution rainfall data interpolated to 1km spatial resolution The original influence factor data C 测 The climate data is interpolated to 0.05°x0.05°, 0.1°x0.1°, 1km x 1km respectively in the same way, to obtain low-resolution influence factor data consistent with the original spatial resolution of the rainfall product and high-resolution influence factor data with 1km spatial resolution

[0118] Resampling: the longitude, latitude, slope, angle and elevation data in the geospatial data in C 测 are resampled to 1km x 1km, 0.25°x0.25°, 0.1°x0.1°, 0.05°x0.05° respectively by using the bilinear interpolation method.

[0119] Data conversion: the data in the original rainfall information is converted to standard time, and the converted data is accumulated on a daily scale to generate daily rainfall information.

[0120] Step S3: constructing a target downscaling model considering spatial correlation.

[0121] Specifically, a random forest model is used to construct the regression relationship between the original rainfall product MSP 测 , the low-resolution rainfall data and the low-resolution influence factor data , to obtain the target downscaling model. Then, the high-resolution rainfall data with 1km spatial resolution and high resolution impact factor data Input into the target downscaling model to estimate rainfall data with spatial correlation at 1 km spatial resolution

[0122] Step S4: Construct a classification model considering rainfall status and rainfall intensity.

[0123] Specifically, the original rainfall data Gauge 测 is divided into no-rain data and rain data according to the rainfall status. The rain data is further classified according to the rainfall intensity to obtain the classification data of the original rainfall data Gauge 测 of the rain gauge Then, a random forest model is used to construct a regression relationship between the classification data rainfall data of the grid point where the rain gauge is located and the high resolution impact factor data of the grid point where the rain gauge is located to obtain the target classification model. Then, the high resolution impact factor data and the rainfall data at 1 km spatial resolution are input into the target classification model to obtain the classification data at 1 km spatial resolution

[0124] Step S5: Construct a target fusion model.

[0125] Specifically, for the rain data, a random forest model is used to construct a regression relationship between the original rainfall data Gauge 测 of the rain gauge, the impact factor data of the grid point where the rain gauge is located, the rainfall data of the grid point where the rain gauge is located, the rainfall data of the rain gauge, the rainfall data of the grid point where the rain gauge is located, and the classification data to obtain the target fusion model. Then, the high resolution impact factor data rainfall data rainfall data and the classification data of all grid points are input into the target fusion model to obtain the fused rainfall data MSMP 测，1km at 1 km resolution. The rainfall data is the rainfall data at 1 km spatial resolution obtained by preprocessing the original rainfall data Gauge 测 of the rain gauge.

[0126] Step S6: evaluate the fusion effect of the above rainfall data.

[0127] Specifically, 70% of the rainfall stations are randomly selected as the training set, and the remaining 30% of the rainfall stations are selected as the validation set. Then, statistical indicators and classification indicators are used to evaluate the fusion effect.

[0128] The statistical indicators include correlation coefficient (CC), Kling-Gupta efficiency (KGE), and root mean square error (RMSE). The higher the value of CC, the stronger the linear correlation between the fused rainfall data and the observed rainfall information. KGE and CC are used together to reflect the overall fitting degree of the fused rainfall data and the original rainfall data of the rainfall station. RMSE is used to reflect the error between the fused rainfall data and the original rainfall data of the rainfall station. Among them, the closer CC and KGE are to 1, the better the fusion effect of the rainfall data, and the closer RMSE is to 0, the better the fusion effect of the rainfall data.

[0129] Specifically, the calculation formulas of CC, KGE and RMSE are as follows:

[0130]

[0131]

[0132] wherein S i is the rainfall amount in the fused rainfall data or the original rainfall data of the grid point where the ith rainfall station is located; O i is the original rainfall data of the ith rainfall station; is the mean value of the fused rainfall data or the original rainfall data; is the mean value of the original rainfall data of the rainfall station; n is the sample size; β is the mean ratio, μ s is the mean value of the fused rainfall data, μ o is the mean value of the rainfall data of the rainfall station; γ is the coefficient of variation ratio, σ s is the standard deviation of the rainfall amount in the fused rainfall data or the original rainfall data, σ o is the standard deviation of the original rainfall data of the rainfall station.

[0133] The classification indicators are used to quantitatively evaluate the detection ability of the original rainfall data and the fused rainfall data on the rain event. The classification indicators include the detection rate (POD), the false alarm rate (FAR), and the critical success index (CSI). Among them, the closer POD and CSI are to 1, the better the fusion effect of the rainfall data, and the closer FAR is to 0, the better the fusion effect of the rainfall data.

[0134] Specifically, the calculation formulas of POD, FAR and CSI are as follows:

[0135]

[0136] wherein H is the number of times that the original rainfall data of the rain gauge and the fused rainfall data both detect rain events, or the number of times that the original rainfall data of the rain gauge and the rainfall product both detect rain events; M represents a false negative, that is, the number of times that the rainfall data of the rain gauge detects rain events, but the fused rainfall data or the original rainfall data of the rainfall product does not detect rain events; and F represents a false positive, that is, the number of times that the rainfall data of the rain gauge does not detect rain events, but the fused rainfall data or the original rainfall data of the rainfall product detects rain events.

[0137] Further, the fusion accuracy of the rainfall data is analyzed according to the following two dimensions:

[0138] First, the accuracy of the improved RF-based multi-source rainfall fusion method of the present disclosure is compared with that of the original rainfall data. The original rainfall data MSP 测 (such as APHRODITE, ERA5Land, CHIRPS, GSMaP, and IMERG) and the fused rainfall data MSMP 测，1km are used to evaluate the accuracy, and the accuracy of the fused rainfall data MSMP 测，1km is quantified relative to the accuracy of the original rainfall data MSP 测 .

[0139] Referring to the evaluation effect of the statistical indicators shown in Table 2, the evaluation effect of the classification indicators shown in Table 3, and the evaluation effect of the classification indicators under different rainfall levels shown in Table 4, it can be seen that the accuracy of the rainfall data fused by the improved RF-based multi-source rainfall fusion method of the present disclosure is greatly improved compared with the original rainfall data, and the correction effect on extreme rainfall is significant.

[0140] Table 2 Statistical indicator data of five kinds of original rainfall data and fused rainfall data

[0141]

[0142] Table 3 Classification indicator data of five kinds of original rainfall data and fused rainfall data

[0143]

[0144] Table 4 Classification indicator data of five kinds of original rainfall data and fused rainfall data under different rainfall levels

[0145]

[0146]

[0147] Secondly, the fusion accuracy of the data-driven method in the improved RF-based multi-source rainfall fusion method and related technologies of the present disclosure is compared and analyzed. Specifically, the fusion effect of rainfall data under three machine learning methods of RF, XGBoost and GBDT and two classification methods of binary classification (BC) and multiple classification (MC) is compared and analyzed.

[0148] Referring to the different rainfall fusion schemes shown in Table 5, the evaluation effect of the statistical indicators shown in Table 6, and the evaluation effect of the classification indicators under different rainfall levels shown in Table 7, it can be seen that the rainfall data fused by the improved RF-based multi-source rainfall fusion method of the present disclosure has obvious improvement in statistical indicators and detection capability indicators compared with the binary classification model constructed by only considering the rainfall state classification in related technologies, especially for extreme rainfall.

[0149] Table 5 Different rainfall fusion schemes

[0150]

[0151] Table 6 Statistical indicator data of different rainfall fusion schemes

[0152] Indicator RF_BC GBDT_BC XGB_BC RF_MC GBDT_MC XGB_MC CC 0.85 0.79 0.78 0.88 0.87 0.85 KGE 0.83 0.75 0.77 0.83 0.84 0.83 RMSE 5.74 6.69 7.02 5.21 5.44 5.80

[0153] Table 7 Classification indicator data of different rainfall fusion schemes

[0154]

[0155]

[0156] It is worth noting that the multi-source rainfall product in related technologies has the following shortcomings: first, the time series of the multi-source rainfall data used in related technologies is short, which is difficult to meet the hydrological demand. Second, related technologies mostly focus on whether a rain event occurs, ignoring the rainfall level. In view of the above shortcomings, referring to Figure 4The improved direction of the improved RF-based multi-source rainfall fusion method of the present disclosure is as follows: first, downscaling. The original rainfall data of different data sources is down-scaled (e.g., to 1 km spatial resolution) by constructing a target downscaling model considering spatial correlation. Second, classification. A target classification model is constructed, and the down-scaled rainfall data is classified according to the rainfall state and rainfall level to obtain classified data. Third, fusion correction. A target fusion model is constructed, and the rainfall data is fused by combining the rainfall data of the high-resolution rain gauge (e.g., 1 km spatial resolution), the influence factor data, the down-scaled data of different data sources, and the classified data to obtain the fused rainfall data. Since the improved RF-based multi-source rainfall fusion method of the present embodiment introduces the classification of rainfall state and rainfall level in rainfall fusion, and constructs the nonlinear relationship between different rainfall states, rainfall levels, and rainfall influence factors, the problem of the decrease in rainfall precision with the increase in rainfall level in the fusion rainfall data can be effectively improved.

[0157] In the present embodiment, an improved RF-based multi-source rainfall fusion device is also provided, which is used to implement the above-mentioned embodiments and preferred embodiments, and will not be described again. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware, or a combination of software and hardware is also possible and contemplated.

[0158] The present embodiment provides an improved RF-based multi-source rainfall fusion device, as shown in Figure 5 , comprising:

[0159] The data acquisition module 501 is configured to acquire first rainfall information of a target area; wherein the first rainfall information comprises first rainfall data, second rainfall data, and first influence factor data of rainfall influence factors; the first rainfall data is obtained by a rainfall product of the target area; and the second rainfall data is obtained by a rain gauge of the target area;

[0160] The first processing module 502 is configured to perform downscaling processing on the first rainfall data based on the first influence factor data and a target downscaling model to obtain third rainfall data;

[0161] The second processing module 503 is configured to perform rainfall classification on the third rainfall data based on the first influence factor data and a target classification model to obtain first classification data; wherein the rainfall classification comprises rainfall state classification and / or rainfall level classification;

[0162] The third processing module 504 is configured to fuse the second rainfall data and the third rainfall data based on the first influence factor data, the first classification data, and the target fusion model to obtain fused rainfall data. At least one of the target downscaling model, the target classification model, and the target fusion model is obtained based on a random forest model.

[0163] In some optional embodiments, the improved RF-based multi-source rainfall fusion device of the present disclosure further comprises a first construction module, where the first construction module is configured to construct the target downscaling model. Specifically, the first construction module comprises:

[0164] The first obtaining unit is configured to obtain second rainfall information of one or more sample regions. The second rainfall information comprises first rainfall product data and second influence factor data of rainfall influence factors. The first rainfall product data comprises rainfall data measured by a rainfall product of the sample region.

[0165] The first processing unit is configured to pre-process the second rainfall information to obtain third rainfall information. The spatial resolution of the third rainfall information is consistent with that of the second rainfall information. The third rainfall information comprises second rainfall product data and third influence factor data.

[0166] The first construction unit is configured to construct the target downscaling model based on a regression relationship among the first rainfall product data, the second rainfall product data, and the third influence factor data.

[0167] In some optional embodiments, the second rainfall information further comprises first rain gauge station data, and the first rain gauge station data comprises rainfall data measured by a rain gauge station of the sample region. The improved RF-based multi-source rainfall fusion device of the present disclosure further comprises a second construction module, where the second construction module is configured to construct the target classification model. Specifically, the second construction module comprises:

[0168] The classification unit is configured to classify the first rain gauge station data according to rainfall states and rainfall levels to obtain second classification data.

[0169] The second processing unit is configured to pre-process the second rainfall information to obtain fourth rainfall information. The spatial resolution of the fourth rainfall information is higher than that of the second rainfall information. The fourth rainfall information comprises third rainfall product data, fourth influence factor data, and second rain gauge station data.

[0170] The third processing unit is configured to perform downscaling processing on the third rainfall product data based on the fourth influence factor data and the target downscaling model to obtain downscaling data.

[0171] The second constructing unit is configured to construct the target classification model based on a regression relationship among the second classification data, the downscaling data, and the fourth influence factor data of the same location.

[0172] In some optional embodiments, the classification unit comprises:

[0173] The first classification sub-unit is configured to perform rain state classification on the first rain gauge data to obtain primary classification data, wherein the primary classification data comprises rain data and no-rain data.

[0174] The second classification sub-unit is configured to perform rain intensity classification on the rain data to obtain secondary classification data, so as to obtain second classification data, wherein the second classification data comprises the secondary classification data and the no-rain data.

[0175] In some optional embodiments, the second classification sub-unit is specifically configured to: obtain a mapping relationship between at least one preset rain intensity and a rain data range; and perform rain intensity classification on the rain data in the rain data based on the mapping relationship to obtain the secondary classification data.

[0176] In some optional embodiments, the multi-source rain fusion device based on the improved RF also comprises a data matching module configured to perform location matching on the third rain information and the fourth rain information based on a grid of a digital elevation model of a sample area to obtain a location matching result.

[0177] In some optional embodiments, the multi-source rain fusion device based on the improved RF also comprises a third constructing module configured to construct a target fusion model. Specifically, the third constructing module comprises:

[0178] The third constructing unit is configured to construct the target fusion model based on a regression relationship among the first rain gauge data, the downscaling data, the second rain gauge data, the fourth influence factor data, and the second classification data of the same location.

[0179] Further function descriptions of the above-mentioned various modules and units are the same as those of the above-mentioned corresponding embodiments, which will not be described here again.

[0180] In some optional embodiments, the first construction unit is specifically configured to construct a regression relationship between the first rainfall product data, the second rainfall product data, and the third influence factor data based on the first random forest model to obtain a target downscaling model. The second construction unit is specifically configured to construct a regression relationship between the second classification data, the downscaling data, and the fourth influence factor data at the same location based on the second random forest model to obtain a target classification model. The third construction unit is specifically configured to construct a regression relationship between the first rain gauge station data, the downscaling data, the second rain gauge station data, the fourth influence factor data, and the second classification data at the same location based on the third random forest model to obtain a target fusion model.

[0181] The multi-source rainfall fusion device based on the improved RF in the embodiment is presented in the form of functional units. The units herein refer to ASIC (Application Specific Integrated Circuit) circuits, processors and memories that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0182] Figure 6 A structural block diagram of an electronic device is provided for the embodiments of the present disclosure.

[0183] Reference will now be made in detail to Figure 6 which shows a structural block diagram of an electronic device suitable for implementing the embodiments of the present disclosure. The electronic device can include a processor (such as a central processor, a graphics processor, etc.) 601, which can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 602 or loaded from a memory 608 into a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the electronic device are also stored. The processor 601, the ROM 602, and the RAM 603 are connected to each other through a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0184] Generally, the following devices can be connected to the I / O interface 605: input devices 606 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a memory 608 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 609. The communication device 609 can allow the electronic device to communicate with other devices wirelessly or by wire to exchange data. Although Figure 6 The electronic device is shown with various devices, but it should be understood that all the shown devices are not required to be implemented or possessed, and more or fewer devices can be alternatively implemented or possessed.

[0185] In particular, according to embodiments of the present disclosure, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, embodiments of the present disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods illustrated by the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network by the communication apparatus 609, or installed from the memory 608, or installed from the ROM 602. When the computer program is executed by the processor 601, the above-mentioned functions defined in the AAA method of the embodiments of the present disclosure are performed.

[0186] Figure 6 The electronic device shown is merely an example and should not impose any limitation on the functions and use range of the embodiments of the present disclosure.

[0187] The embodiments of the present disclosure also provide a computer-readable storage medium, and the above-mentioned method according to the embodiments of the present disclosure can be implemented in hardware, firmware, or as computer code recorded on a storage medium, or as computer code originally stored in a remote storage medium or a non-transitory machine-readable storage medium and downloaded through a network and to be stored in a local storage medium, so that the method described herein can be processed by such software on a storage medium using a general-purpose computer, a special-purpose processor, or programmable or special-purpose hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk or a solid state disk, etc.; further, the storage medium can also include a combination of the above-mentioned types of memories. It can be understood that the computer, processor, microprocessor controller or programmable hardware includes a storage component that can store or receive software or computer code, which, when accessed and executed by the computer, processor or hardware, implements the improved RF-based multi-source rainfall fusion method shown in the above-mentioned embodiments.

[0188] Part of the present disclosure can be applied as a computer program product, for example, computer program instructions, when executed by a computer, through the operation of the computer, the method and / or technical solutions according to the present disclosure can be invoked or provided. Those skilled in the art should understand that the form of computer program instructions in computer-readable medium includes but is not limited to source files, executable files, installation package files, etc., and accordingly, the way of computer program instructions executed by computer includes but is not limited to: the computer directly executes the instructions, or the computer executes the corresponding compiled program after compiling the instructions, or the computer reads and executes the instructions, or the computer reads and installs the instructions and then executes the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to the computer.

[0189] While embodiments of the present disclosure have been described in conjunction with the accompanying drawings, various modifications and changes can be suggested by those skilled in the art, and it is intended that the present disclosure encompass such modifications and changes as fall within the scope of the appended claims.

Claims

1. An improved RF-based multi-source rainfall fusion method, characterized in that, The method comprises: obtaining first rainfall information of a target area; wherein the first rainfall information comprises first rainfall data, second rainfall data and first impact factor data of a rainfall impact factor; the first rainfall data is obtained through rainfall products of the target area; the second rainfall data is obtained through rain gauges of the target area; based on the first impact factor data and a target downscaling model, performing downscaling processing on the first rainfall data to obtain third rainfall data; based on the first impact factor data and a target classification model, performing rainfall classification on the third rainfall data to obtain first classification data; wherein the rainfall classification comprises rainfall state classification and rainfall intensity classification; based on the first impact factor data, the first classification data and a target fusion model, fusing the second rainfall data and the third rainfall data to obtain fused rainfall data; wherein at least one of the target downscaling model, the target classification model and the target fusion model is obtained based on a random forest model.

2. The multi-source rainfall fusion method of claim 1, wherein, The target downscaling model is obtained by: obtaining second rainfall information of one or more sample areas; wherein the second rainfall information comprises first rainfall product data and second impact factor data of the rainfall impact factor; the first rainfall product data comprises rainfall data measured by rainfall products of the sample area; preprocessing the second rainfall information to obtain third rainfall information; wherein the spatial resolution of the third rainfall information is consistent with the spatial resolution of the second rainfall information; the third rainfall information comprises second rainfall product data and third impact factor data; based on the regression relationship between the first rainfall product data, the second rainfall product data and the third impact factor data, the target downscaling model is constructed.

3. The multi-source rainfall fusion method of claim 2, wherein, The second rainfall information further comprises first rain gauge data, the first rain gauge data comprising rainfall data measured by rain gauges of the sample area; the target classification model is obtained by: classifying the first rain gauge data according to rainfall state and rainfall intensity to obtain second classification data; preprocessing the second rainfall information to obtain fourth rainfall information; wherein the spatial resolution of the fourth rainfall information is higher than the spatial resolution of the second rainfall information; the fourth rainfall information comprises third rainfall product data, fourth impact factor data and second rain gauge data; based on the fourth impact factor data and the target downscaling model, performing downscaling processing on the third rainfall product data to obtain downscaling data; based on the regression relationship between the second classification data, the downscaling data and the fourth impact factor data at the same location, the target classification model is constructed.

4. The multi-source rainfall fusion method of claim 3, wherein, The classification of the first rain gauge data according to rainfall state and rainfall intensity to obtain second classification data comprises: The first rain station data is classified according to a rainfall state, and first classification data is obtained; the first classification data includes rain data and no-rain data; The rain data is classified according to a rainfall intensity level, and second classification data is obtained, so as to obtain the second classification data; the second classification data includes the second classification data and the no-rain data.

5. The multi-source rainfall fusion method of claim 4, wherein, The rain data is classified according to a rainfall intensity level, and second classification data is obtained, so as to obtain the second classification data; the second classification data includes the second classification data and the no-rain data. A mapping relationship between at least one preset rainfall intensity level and a rainfall data range is obtained; The rain data in the rain data is classified according to a rainfall intensity level based on the mapping relationship, and the second classification data is obtained.

6. The multi-source rainfall fusion method of claim 3, wherein, The method further includes: The third rainfall information and the fourth rainfall information are positionally matched based on a grid of a digital elevation model of the sample area, and a position matching result is obtained.

7. The multi-source rainfall fusion method of any one of claims 3 to 6, wherein, The target fusion model is obtained by: A regression relationship between the first rain station data, the downscaling data, the second rain station data, the fourth influence factor data and the second classification data at the same position is used to construct the target fusion model.

8. The multi-source rainfall fusion method of claim 7, wherein, The regression relationship between the first rainfall product data, the second rainfall product data and the third influence factor data is used to construct the target downscaling model, including: A first random forest model is used to construct the regression relationship between the first rainfall product data, the second rainfall product data and the third influence factor data, and the target downscaling model is obtained; A second random forest model is used to construct the regression relationship between the second classification data, the downscaling data and the fourth influence factor data at the same position, and the target classification model is obtained; A third random forest model is used to construct the regression relationship between the first rain station data, the downscaling data, the second rain station data, the fourth influence factor data and the second classification data at the same position, and the target fusion model is obtained. including: a memory and a processor, which are communicatively connected, the memory stores computer instructions, and the processor executes the computer instructions to perform the improved RF-based multi-source rainfall fusion method according to any one of claims 1 to 8.

9. An electronic device, comprising: The computer readable storage medium stores computer instructions for causing a computer to execute the improved RF-based multi-source rainfall fusion method according to any one of claims 1 to 8. ​ 10. A computer-readable storage medium, characterized in that, ​

Images