Multi-source precipitation depth spatio-temporal fusion method and device, electronic equipment and storage medium

Abstract

The application belongs to the technical field of remote sensing hydrology, and specifically discloses a multi-source precipitation depth spatio-temporal fusion method and device, an electronic device and a storage medium. The method comprises the following steps: performing data preprocessing on multi-source precipitation data of a to-be-measured region to determine multi-source precipitation data of a target spatio-temporal resolution; inputting the multi-source precipitation data of the target spatio-temporal resolution into a precipitation classification regression model to obtain a precipitation prediction result of the to-be-measured region output by the precipitation classification regression model; the precipitation classification regression model is constructed based on a deep learning model and a machine learning model, is trained according to multi-source precipitation data samples and corresponding precipitation data labels, is used to identify each precipitation position in the to-be-measured region based on fusion features obtained by performing feature extraction and fusion on the multi-source precipitation data, and predict the precipitation amount of each precipitation position. Through the application, the effect of multi-source precipitation data depth spatio-temporal fusion can be effectively improved, and high-precision and high-spatio-temporal-resolution precipitation prediction can be realized.

Description

Technical Field

[0001] This application belongs to the field of remote sensing hydrology technology, and more specifically, relates to a multi-source precipitation depth spatiotemporal fusion method, device, electronic device and storage medium. Background Technology

[0002] Precipitation is an important component of the water cycle and one of the most significant factors affecting the accuracy of hydrological model simulations. Obtaining accurate and reliable precipitation data is of great importance for studying the spatiotemporal variability of precipitation and improving the accuracy of hydrological model simulations.

[0003] Currently, the main sources of precipitation data include observations from ground meteorological stations, estimations from meteorological radar, and satellite inversion. While ground meteorological station observations provide long-term, high-precision precipitation information, the spatial heterogeneity of these stations leads to high spatiotemporal variability in the obtained precipitation data. Meteorological radar and satellite remote sensing, as large-scale, periodic, real-time or near-real-time precipitation observation methods, offer advantages such as wide spatiotemporal coverage and easy data acquisition. However, they typically suffer from coarse spatial resolution, low accuracy, and inadequate identification of precipitation events, limiting their application. Therefore, to overcome the limitations of single data sources, researchers have proposed data fusion and other processing methods to generate higher-quality precipitation data for subsequent hydrological simulations and other related studies.

[0004] Currently, existing precipitation fusion methods are mainly divided into two categories. One category is based on statistical and physical models, such as weighted average and optimal interpolation. This method relies on strong assumptions and has limited ability to handle nonlinear relationships, resulting in low model accuracy. The other category is based on machine learning methods, but most of them use a single machine learning method for precipitation fusion prediction, which has problems such as overlearning, resulting in low prediction accuracy of the model.

[0005] Therefore, how to better achieve the fusion and processing of precipitation data has become a technical problem that the industry urgently needs to solve. Summary of the Invention

[0006] In view of the shortcomings of the existing technology, the purpose of this application is to better realize the fusion processing of precipitation data, and to solve the problem of low prediction accuracy in the existing precipitation fusion methods.

[0007] To achieve the above objectives, in a first aspect, this application provides a multi-source precipitation depth spatiotemporal fusion method, comprising:

[0008] Data preprocessing is performed on the multi-source precipitation data of the area to be measured to determine the multi-source precipitation data with the target spatiotemporal resolution.

[0009] The multi-source precipitation data with the target spatiotemporal resolution is input into the precipitation classification and regression model to obtain the precipitation prediction results of the area to be measured output by the precipitation classification and regression model.

[0010] The precipitation classification regression model is constructed based on deep learning and machine learning models. It is trained on multi-source precipitation data samples and their corresponding precipitation data labels. It is used to identify each precipitation location in the area to be measured and predict the precipitation amount at each precipitation location based on the fusion features obtained by feature extraction and fusion of multi-source precipitation data at the target spatiotemporal resolution. The multi-source precipitation data includes remote sensing precipitation data, meteorological database data and digital elevation data.

[0011] Optionally, the meteorological database data includes precipitation data and meteorological data; the multi-source precipitation data of the area to be measured undergoes data preprocessing to determine the multi-source precipitation data with the target spatiotemporal resolution, including:

[0012] The precipitation data in the meteorological database is accumulated and aggregated hourly to obtain the precipitation data of the meteorological database at the target time scale, and the average value of the meteorological data is calculated to obtain the meteorological data at the target time scale.

[0013] Using interpolation, the precipitation data and meteorological data of the meteorological database at the target time scale are respectively transformed into precipitation data and meteorological data of the meteorological database at the target spatiotemporal resolution;

[0014] The digital elevation data is resampled according to the target spatiotemporal resolution to obtain digital elevation data with the target spatiotemporal resolution; the target spatiotemporal resolution is the spatiotemporal resolution of the remote sensing precipitation data.

[0015] Optionally, the precipitation classification and regression model includes a classification module and a regression module; the step of inputting the multi-source precipitation data with the target spatiotemporal resolution into the precipitation classification and regression model to obtain the precipitation prediction results of the area to be measured output by the precipitation classification and regression model includes:

[0016] The multi-source precipitation data with the target spatiotemporal resolution is input into the classification module to obtain each precipitation location in the area to be measured, as output by the classification module.

[0017] The multi-source precipitation data with target spatiotemporal resolution corresponding to each precipitation location is input into the regression module to obtain the precipitation prediction result for each precipitation location in the area to be measured, output by the regression module.

[0018] Optionally, the classification module is a CNN-LSTM model, and the regression module is a random forest model.

[0019] Optionally, both the classification module and the regression module are joint models composed of a random forest model, an XGBoost model, a CNN-LSTM model, and a ConvLSTM model in parallel; the step of inputting the multi-source precipitation data of the target spatiotemporal resolution into the classification module to obtain each precipitation location in the area to be measured output by the classification module includes:

[0020] The multi-source precipitation data with the target spatiotemporal resolution are respectively input into the random forest model, the XGBoost model, the CNN-LSTM model and the ConvLSTM model to obtain the first recognition result output by the random forest model, the second recognition result output by the XGBoost model, the third recognition result output by the CNN-LSTM model and the fourth recognition result output by the ConvLSTM model;

[0021] Using a voting method, each precipitation location in the area to be measured is determined based on the first identification result, the second identification result, the third identification result, and the fourth identification result.

[0022] Optionally, the step of inputting multi-source precipitation data with target spatiotemporal resolution corresponding to each precipitation location into the regression module to obtain the precipitation prediction result for each precipitation location in the area to be measured output by the regression module includes:

[0023] The multi-source precipitation data with target spatiotemporal resolution corresponding to each precipitation location are respectively input into the random forest model, the XGBoost model, the CNN-LSTM model and the ConvLSTM model to obtain the first prediction result output by the random forest model, the second prediction result output by the XGBoost model, the third prediction result output by the CNN-LSTM model and the fourth prediction result output by the ConvLSTM model;

[0024] The precipitation prediction result for each precipitation location in the area to be measured is determined by averaging the first prediction result, the second prediction result, the third prediction result, and the fourth prediction result.

[0025] Optionally, before inputting the multi-source precipitation data with the target spatiotemporal resolution into the precipitation classification and regression model to obtain the precipitation prediction results for the area to be measured output by the precipitation classification and regression model, the method further includes:

[0026] Data preprocessing is performed on the multi-source precipitation data samples of the area to be measured to determine the multi-source precipitation data samples with the target spatiotemporal resolution;

[0027] Obtain precipitation monitoring data from ground stations corresponding to the multi-source precipitation data samples, and determine the precipitation data labels corresponding to the multi-source precipitation data samples with the target spatiotemporal resolution based on the precipitation monitoring data from the ground stations according to the target precipitation threshold.

[0028] Multiple sets of training samples are obtained by using the multi-source precipitation data samples with the target spatiotemporal resolution and their corresponding precipitation data labels as a set of training samples.

[0029] The precipitation classification regression model was trained using multiple sets of training samples to obtain a well-trained precipitation classification regression model.

[0030] Secondly, this application provides a multi-source precipitation depth spatiotemporal fusion device, comprising:

[0031] The preprocessing module is used to preprocess multi-source precipitation data of the area to be measured and determine the target spatiotemporal resolution of the multi-source precipitation data.

[0032] The prediction module is used to input the multi-source precipitation data with the target spatiotemporal resolution into the precipitation classification and regression model to obtain the precipitation prediction results of the area to be measured output by the precipitation classification and regression model.

[0033] The precipitation classification regression model is constructed based on deep learning and machine learning models. It is trained on multi-source precipitation data samples and their corresponding precipitation data labels. It is used to identify each precipitation location in the area to be measured and predict the precipitation amount at each precipitation location based on the fusion features obtained by feature extraction and fusion of multi-source precipitation data at the target spatiotemporal resolution. The multi-source precipitation data includes remote sensing precipitation data, meteorological database data and digital elevation data.

[0034] Thirdly, this application provides an electronic device, comprising: at least one memory for storing a program; and at least one processor for executing the program stored in the memory, wherein when the program stored in the memory is executed, the processor is configured to execute the method described in the first aspect or any possible implementation thereof.

[0035] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to perform the method described in the first aspect or any possible implementation thereof.

[0036] Fifthly, this application provides a computer program product that, when run on a processor, causes the processor to perform the method described in the first aspect or any possible implementation thereof.

[0037] It is understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here.

[0038] Overall, the technical solutions conceived in this application have the following beneficial effects compared with the prior art:

[0039] This application provides a method, apparatus, electronic device, and storage medium for multi-source precipitation depth spatiotemporal fusion. By integrating the excellent spatiotemporal feature extraction capabilities of deep learning models with the advantages of strong generalization and stability of machine learning models, the multi-source precipitation data of the area to be measured is processed into high spatiotemporal resolution multi-source precipitation data. The precipitation classification and regression model constructed and trained by the deep learning model and the machine learning model is used for multi-source precipitation depth spatiotemporal fusion prediction, which can effectively improve the effect of multi-source precipitation data depth spatiotemporal fusion and achieve high-precision and high spatiotemporal resolution precipitation prediction. Attached Figure Description

[0040] Figure 1 This is one of the flowcharts of the multi-source precipitation depth spatiotemporal fusion method provided in the embodiments of this application;

[0041] Figure 2 This is the second flowchart illustrating the spatiotemporal fusion method for multi-source precipitation depth provided in this application embodiment;

[0042] Figure 3 This is a general view of the construction of the model input dataset provided in the embodiments of this application;

[0043] Figure 4 This is a schematic diagram showing the precipitation fusion accuracy results of different combined models provided in the embodiments of this application;

[0044] Figure 5 This is a schematic diagram of the box-shaped results of the site indicators provided in the embodiments of this application;

[0045] Figure 6 This is a line graph diagram of the indexes of different combined models / precipitation sources under different rainfall intensities provided in the embodiments of this application;

[0046] Figure 7 This is a schematic diagram of the multi-source precipitation depth spatiotemporal fusion device provided in the embodiments of this application;

[0047] Figure 8 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0049] The terms "first" and "second," etc., used in the specification and claims of this application are used to distinguish different objects, not to describe a specific order of objects. For example, "first identification result" and "second identification result," etc., are used to distinguish different identification results, not to describe a specific order of identification results.

[0050] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0051] In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more, for example, multiple processing units means two or more processing units, multiple elements means two or more elements, etc.

[0052] The embodiments of this application are described below with reference to the accompanying drawings.

[0053] Figure 1 This is a flowchart illustrating the spatiotemporal fusion method for multi-source precipitation depth provided in this application embodiment, as shown below. Figure 1 As shown, the method includes:

[0054] Step S1: Perform data preprocessing on the multi-source precipitation data of the area to be measured to determine the multi-source precipitation data with the target spatiotemporal resolution;

[0055] Step S2: Input the multi-source precipitation data with the target spatiotemporal resolution into the precipitation classification and regression model to obtain the precipitation prediction results of the area to be measured output by the precipitation classification and regression model.

[0056] Among them, the precipitation classification regression model is built based on deep learning and machine learning models. It is trained based on multi-source precipitation data samples and their corresponding precipitation data labels. It is used to identify each precipitation location in the area to be measured and predict the precipitation amount at each precipitation location based on the fusion features obtained by feature extraction and fusion of multi-source precipitation data at the target spatiotemporal resolution. The multi-source precipitation data includes remote sensing precipitation data, meteorological database data and digital elevation data.

[0057] Specifically, the multi-source precipitation data described in this application refers to precipitation data acquired through various precipitation data products, which may include remote sensing precipitation data, meteorological database data, and Digital Elevation Model (DEM) data. Specifically, the remote sensing precipitation data may be GPM remote sensing precipitation data, and the meteorological database data may be ERA5 reanalysis precipitation data.

[0058] The target spatiotemporal resolution described in the embodiments of this application is used to characterize high spatiotemporal resolution. Specifically, it can be determined based on the spatiotemporal resolution of the collected remote sensing precipitation data. For example, the target spatiotemporal resolution can be a time resolution of day and a spatial resolution of 0.1°.

[0059] The precipitation classification regression model described in this application is built based on deep learning models and machine learning models. Specifically, the deep learning models may include CNN-LSTM models, ConvLSTM models, etc.; the machine learning models may include Random Forest (RF) models, XGBoost boosting tree models, etc.

[0060] Here, the precipitation classification regression model can be trained based on multi-source precipitation data samples and their corresponding precipitation data labels to learn the spatiotemporal correlations between precipitation data from different regional grids. It is understood that multi-source precipitation data samples can specifically include remote sensing precipitation data samples, meteorological database data samples, and DEM data samples; precipitation data labels can be determined based on measured data from ground stations.

[0061] In the embodiments of this application, in step S1, multi-source precipitation data is formed by retrieving GPM remote sensing precipitation data, precipitation data from the ERA5 reanalysis dataset, and DEM data of the area to be measured. Then, these multi-source precipitation data can be preprocessed using data preprocessing methods, including spatiotemporal scale unification processing and normalization processing, to obtain multi-source precipitation data with the target spatiotemporal resolution.

[0062] In the embodiments of this application, in step S2, the precipitation classification and regression model is pre-trained using multi-source precipitation data samples and their corresponding precipitation data labels to obtain a trained precipitation classification and regression model. Therefore, by inputting the multi-source precipitation data with the target spatiotemporal resolution obtained in step S1 into the precipitation classification and regression model, the model can extract and fuse features from the multi-source precipitation data with the target spatiotemporal resolution, obtain corresponding fused features, and identify each precipitation location in the area to be measured by recognizing these fused features. This allows for the prediction of precipitation at each location, thereby obtaining the precipitation prediction result for the area to be measured.

[0063] The multi-source precipitation depth spatiotemporal fusion method of this application integrates the excellent spatiotemporal feature extraction capability of deep learning models with the advantages of strong generalization and stability of machine learning models. It processes the multi-source precipitation data of the area to be measured into high spatiotemporal resolution multi-source precipitation data, and performs multi-source precipitation depth spatiotemporal fusion prediction by combining the precipitation classification and regression models constructed and trained by deep learning models and machine learning models. This can effectively improve the effect of multi-source precipitation data depth spatiotemporal fusion and achieve high-precision and high spatiotemporal resolution precipitation prediction.

[0064] Based on the above embodiments, as an optional embodiment, the meteorological database data includes meteorological database precipitation data and meteorological data; step S1, performing data preprocessing on the multi-source precipitation data of the area to be measured to determine the multi-source precipitation data with target spatiotemporal resolution, including:

[0065] The precipitation data in the meteorological database is accumulated and aggregated hourly to obtain the precipitation data of the meteorological database at the target time scale, and the average value of the meteorological data is calculated to obtain the meteorological data at the target time scale.

[0066] Using interpolation, precipitation data and meteorological data in the meteorological database at the target time scale are transformed into precipitation data and meteorological data in the meteorological database at the target spatiotemporal resolution, respectively.

[0067] The digital elevation data is resampled according to the target spatiotemporal resolution to obtain digital elevation data with the target spatiotemporal resolution; the target spatiotemporal resolution is the spatiotemporal resolution of the remote sensing precipitation data.

[0068] Specifically, in the embodiments of this application, the meteorological database data, namely ERA5 reanalysis data and DEM data, need to undergo spatiotemporal scale unification processing to ensure that the spatiotemporal resolution of the two types of data is consistent with that of the remote sensing precipitation data. The ERA5 reanalysis data includes precipitation data and meteorological data. The meteorological data specifically includes five dynamic meteorological auxiliary factors: the u-component of 10m wind speed, the v-component of 10m wind speed, the 2m dew point temperature, the 2m temperature, and surface pressure. The DEM data specifically includes five static meteorological auxiliary factors: altitude, slope, aspect, longitude, and latitude.

[0069] More specifically, in terms of temporal resolution, for the ERA5 reanalysis dataset, precipitation data is aggregated into a target time scale, such as a daily scale, by accumulating hourly. At the same time, the intraday average of meteorological data such as temperature, air pressure and wind speed is taken to preserve the climate characteristics at the daily scale.

[0070] Regarding spatial resolution, interpolation methods, such as nearest neighbor interpolation, can be used to reduce the spatial resolution of precipitation and meteorological data at the target time scale to the target spatial resolution, such as 0.1°, to ensure spatial consistency. For DEM data with a conventional spatial resolution of 90m, a geographic coordinate system can be added using ArcGIS software. Through operations such as projection and clipping, geographic factors, including altitude, slope, aspect, and latitude and longitude, can be extracted. Then, cubic convolution interpolation can be used for resampling, converting the DEM data into a terrain feature dataset with a spatial resolution of 0.1°. Through these processes, a multi-source precipitation dataset with a target spatiotemporal resolution of 1 day and a spatial resolution of 0.1° can be constructed.

[0071] The method in this application embodiment unifies the spatiotemporal scale of precipitation data, meteorological data, and DEM data from the meteorological database to construct high spatiotemporal resolution model input data, which is beneficial to improving the spatiotemporal resolution of precipitation prediction results of precipitation classification regression models.

[0072] Based on the above embodiments, as an optional embodiment, the precipitation classification and regression model includes a classification module and a regression module; step S2, inputting multi-source precipitation data with a target spatiotemporal resolution into the precipitation classification and regression model to obtain the precipitation prediction results for the area to be measured output by the precipitation classification and regression model, including:

[0073] The multi-source precipitation data with the target spatiotemporal resolution is input into the classification module to obtain the precipitation location of each precipitation location in the area to be measured, output by the classification module.

[0074] The multi-source precipitation data with target spatiotemporal resolution corresponding to each precipitation location is input into the regression module to obtain the precipitation prediction results for each precipitation location in the area to be measured, output by the regression module.

[0075] Specifically, in the embodiments of this application, a classification module and a regression module are introduced to construct a precipitation classification and regression model. Precipitation events are identified based on the classification module, and rainfall is predicted based on the regression module. The classification module can be constructed based on a combination of deep learning and machine learning models, or it can be constructed based solely on a deep learning model. Similarly, the regression module can be constructed based on a combination of deep learning and machine learning models, or it can be constructed solely on a machine learning model.

[0076] In the embodiments of this application, the precipitation classification regression model includes a classification module and a regression module. By inputting multi-source precipitation data with a target spatiotemporal resolution into the classification module for precipitation event identification, each precipitation location in the area to be measured can be identified.

[0077] Furthermore, in the embodiments of this application, multi-source precipitation data with target spatiotemporal resolution corresponding to each precipitation location are input into the regression module, and precipitation is predicted through the regression module to obtain the precipitation prediction result for each precipitation location in the area to be measured.

[0078] The method in this application embodiment introduces independent classification and regression modules to perform multi-source precipitation depth spatiotemporal fusion, separating the binary classification (precipitation presence or absence) and regression / classification (precipitation magnitude) tasks. This can effectively avoid the accuracy loss caused by a single-stage model processing two types of tasks simultaneously. At the same time, by classifying and screening precipitation areas through the classification module, precipitation prediction can be carried out in a targeted manner, saving computational resources.

[0079] Based on the above embodiments, as an optional embodiment, the classification module is a CNN-LSTM model and the regression module is an RF model.

[0080] Specifically, in the embodiments of this application, the classification module can be selected as a CNN-LSTM model, and the regression module can be selected as an RF model.

[0081] Specifically, after obtaining the trained CNN-LSTM model and RF model, the aforementioned high spatiotemporal resolution multi-source precipitation data can be input into the CNN-LSTM model. Leveraging the excellent spatiotemporal feature extraction capabilities of the CNN-LSTM model, rain / no rain identification can be performed at each monitoring grid point in the area under test, effectively identifying each precipitation location in the area. Furthermore, the high spatiotemporal resolution multi-source precipitation data corresponding to each precipitation location is input into the RF model for regression prediction, obtaining the predicted precipitation amount for each precipitation location in the area under test.

[0082] In this embodiment of the application, the strategy of setting the classification module as a CNN-LSTM model and the regression module as an RF model can be achieved by introducing deep learning models (such as CNN-LSTM models, ConvLSTM models, etc.) and machine learning models (such as RF models, XGBoost models, etc.) in pairs, with one model serving as the classification module and the other as the regression module. The combined model fusion results are evaluated from multiple dimensions using accuracy indicators (such as correlation coefficient (CC), mean absolute error (MAE), root mean square error (RMSE)) and precipitation event detection capability indicators (such as probability of detection (POD), false alarm rate (FAR), critical success index (CSI)).

[0083] The method in this application introduces different deep learning models and machine learning models to combine them, explores the precipitation fusion prediction performance of different model combinations, and selects the optimal model combination, namely, the CNN-LSTM model as the classification module and the RF model as the regression module, which can further improve the accuracy and effect of deep spatiotemporal fusion prediction of multi-source precipitation data.

[0084] Based on the above embodiments, as an optional embodiment, both the classification module and the regression module are joint models composed of parallel combinations of the RF model, XGBoost model, CNN-LSTM model, and ConvLSTM model; multi-source precipitation data with a target spatiotemporal resolution are input into the classification module to obtain each precipitation location in the area to be measured, as output by the classification module, including:

[0085] Multi-source precipitation data with target spatiotemporal resolution were input into the RF model, XGBoost model, CNN-LSTM model and ConvLSTM model respectively to obtain the first recognition result output by the RF model, the second recognition result output by the XGBoost model, the third recognition result output by the CNN-LSTM model and the fourth recognition result output by the ConvLSTM model.

[0086] Using a voting method, each precipitation location in the area to be measured is determined based on the first, second, third, and fourth identification results.

[0087] Understandably, in the embodiments of this application, the first identification result is the result of precipitation event identification by the RF model based on multi-source precipitation data with target spatiotemporal resolution; the second identification result is the result of precipitation event identification by the XGBoost model based on multi-source precipitation data with target spatiotemporal resolution; the third identification result is the result of precipitation event identification by the CNN-LSTM model based on multi-source precipitation data with target spatiotemporal resolution; and the fourth identification result is the result of precipitation event identification by the ConvLSTM model based on multi-source precipitation data with target spatiotemporal resolution.

[0088] Specifically, in the embodiments of this application, both the classification module and the regression module can be joint models composed of parallel combinations of the RF model, XGBoost model, CNN-LSTM model, and ConvLSTM model. Inputting multi-source precipitation data with a target spatiotemporal resolution into the classification module is equivalent to inputting high spatiotemporal resolution multi-source precipitation data into the RF model, XGBoost model, CNN-LSTM model, and ConvLSTM model respectively. Using each model to identify precipitation events yields the identification results output by each model, namely, a first identification result, a second identification result, a third identification result, and a fourth identification result.

[0089] Furthermore, in the embodiments of this application, a voting method can be used to determine whether there is precipitation at each monitoring grid point in the area to be measured, based on the first identification result, the second identification result, the third identification result, and the fourth identification result. If there is no precipitation, the output is 0 directly; if there is precipitation, the output is each precipitation location in the area to be measured, and the result is transmitted to the regression module for rainfall prediction.

[0090] The method in this application embodiment introduces RF model, XGBoost model, CNN-LSTM model and ConvLSTM model to jointly process precipitation classification tasks. This can complement the advantages of different types of models, couple machine learning models with stable prediction performance and deep learning models with strong spatiotemporal feature extraction capabilities, and effectively improve the precipitation recognition capability of the classification module.

[0091] Based on the above embodiments, as an optional embodiment, multi-source precipitation data with target spatiotemporal resolution corresponding to each precipitation location is input into the regression module to obtain the precipitation prediction results for each precipitation location in the area to be measured, output by the regression module, including:

[0092] The multi-source precipitation data with target spatiotemporal resolution corresponding to each precipitation location are input into the RF model, XGBoost model, CNN-LSTM model and ConvLSTM model respectively, to obtain the first prediction result output by the RF model, the second prediction result output by the XGBoost model, the third prediction result output by the CNN-LSTM model and the fourth prediction result output by the ConvLSTM model.

[0093] The precipitation forecast for each precipitation location in the area to be measured is determined by averaging the first, second, third, and fourth prediction results.

[0094] Understandably, in the embodiments of this application, the first prediction result is the result of the RF model predicting the rainfall of the precipitation event based on the multi-source precipitation data with the target spatiotemporal resolution; the second prediction result is the result of the XGBoost model predicting the rainfall of the precipitation event based on the multi-source precipitation data with the target spatiotemporal resolution; the third prediction result is the result of the CNN-LSTM model predicting the rainfall of the precipitation event based on the multi-source precipitation data with the target spatiotemporal resolution; and the fourth prediction result is the result of the ConvLSTM model predicting the rainfall of the precipitation event based on the multi-source precipitation data with the target spatiotemporal resolution.

[0095] Specifically, in the embodiments of this application, the multi-source precipitation data with target spatiotemporal resolution corresponding to each precipitation location is input into the regression module. This is equivalent to inputting the multi-source precipitation data with target spatiotemporal resolution corresponding to each precipitation location into the RF model, XGBoost model, CNN-LSTM model, and ConvLSTM model respectively. By using each model to predict precipitation, the prediction results output by each model can be obtained, namely the first prediction result, the second prediction result, the third prediction result, and the fourth prediction result.

[0096] Furthermore, in the embodiments of this application, the current precipitation prediction result of each precipitation point in the area to be measured can be calculated by averaging the first prediction result, the second prediction result, the third prediction result and the fourth prediction result.

[0097] The method in this application embodiment, by introducing RF model, XGBoost model, CNN-LSTM model and ConvLSTM model to jointly process precipitation regression prediction task, can also complement the advantages of different types of models, couple the learning capabilities of machine learning model and deep learning model, and effectively improve the precipitation prediction performance of regression module.

[0098] Based on the above embodiments, as an optional embodiment, before inputting multi-source precipitation data with target spatiotemporal resolution into the precipitation classification and regression model to obtain the precipitation prediction results for the area to be measured output by the precipitation classification and regression model, the method further includes:

[0099] Data preprocessing is performed on multi-source precipitation data samples from the area to be measured to determine the target spatiotemporal resolution of the multi-source precipitation data samples.

[0100] Acquire precipitation monitoring data from ground stations corresponding to multi-source precipitation data samples, and determine the precipitation data labels corresponding to the multi-source precipitation data samples with the target spatiotemporal resolution based on the precipitation monitoring data from ground stations according to the target precipitation threshold.

[0101] Multiple sets of training samples are obtained by using multi-source precipitation data samples with target spatiotemporal resolution and their corresponding precipitation data labels as a set of training samples.

[0102] The precipitation classification regression model was trained using multiple sets of training samples to obtain a well-trained precipitation classification regression model.

[0103] Specifically, in the embodiments of this application, before using the precipitation classification regression model to predict the precipitation data of the area to be measured, it is also necessary to train the precipitation classification regression model.

[0104] In the embodiments of this application, it is first necessary to obtain multi-source precipitation data samples of the area to be measured, and then perform data preprocessing based on the multi-source precipitation data samples of the area to be measured to determine the multi-source precipitation data samples with the target spatiotemporal resolution. For example... Figure 2 As shown in the embodiments of this application, according to the aforementioned data preprocessing method, the ERA5 reanalysis data (including precipitation data and dynamic auxiliary factor data) and DEM data (including static auxiliary factor data) in the multi-source precipitation data samples can be processed in a unified spatiotemporal scale, and finally a multi-source precipitation data sample dataset with a time resolution of 1 day, a spatial resolution of 0.1°, and a time range of historical years can be constructed.

[0105] Furthermore, by acquiring precipitation monitoring data from ground stations corresponding to each multi-source precipitation data sample, and classifying the multi-source precipitation data sample dataset at the target spatiotemporal resolution according to a target precipitation threshold, such as 0.1 mm / day, a rainy / no-rain dataset is obtained. Each multi-source precipitation data sample in this dataset carries its corresponding precipitation data label. Understandably, the precipitation data label includes a classification label and a regression label. The classification label includes a no-rain label and a rainy label, while the regression label includes a precipitation amount label under rainy conditions.

[0106] Furthermore, the multi-source precipitation data samples and their corresponding precipitation data labels at the target spatiotemporal resolution are used as a set of training samples. Multiple sets of training samples are obtained from the rainy / no-rain datasets, and the precipitation classification regression model is trained using the training sample dataset formed by these multiple sets of training samples. Specifically, in the regression prediction stage, multiple sets of training samples from the rainy / no-rain datasets are selected and normalized to accelerate model convergence and improve prediction accuracy.

[0107] In the embodiments of this application, the precipitation classification regression model can be constructed and trained by introducing two machine learning models, such as the RF model and the XGBoost model, and two deep learning models, such as the CNN-LSTM model and the ConvLSTM model.

[0108] Furthermore, based on the model's input features, the model input dataset is constructed using multiple sets of training samples. The inputs to the RF model and XGBoost model are two-dimensional matrices, which can be represented as [ N,C ],in, N The total sample size is represented by the product of the spatial dimension of meteorological stations (e.g., 60 stations) and the time series dimension (e.g., 4383 days). N =262980; C This indicates the dimension of the feature variables, covering multi-source precipitation data and related environmental covariates.

[0109] It should be noted that the embodiments of this application select C =12 characteristic variables, namely 5 static auxiliary factors, 5 dynamic auxiliary factors, GPM remote sensing precipitation and ERA5 reanalysis precipitation, for a total of 12 characteristic variables.

[0110] The input to both CNN-LSTM and ConvLSTM models is a five-dimensional tensor, which can be defined as [ N,t,l,w,C ],in, t The time step is defined as the number of historical time periods required to predict precipitation in the next period, i.e., the prediction time. T n Rainfall needs to be input at all times T n-t to T n The specific values ​​of the independent variable for a time period can be determined according to the candidate set, for example, the candidate set is {1,3,5,7,9}; l and w These represent the length and width of the spatial neighborhood window centered on the ground station location, respectively. In this embodiment, a square neighborhood window (i.e.) can be used. l = wIts specific value can be determined according to the candidate set, for example, the candidate set is {1,3,5,7,9}; N and C These represent the total number of samples and the dimension of the feature variables, respectively, consistent with traditional machine learning models.

[0111] In the data preprocessing stage, such as Figure 3 As shown, for each ground station, including the station i、 Site i+ 1. ...site i + n, the original observation data of each feature variable corresponding to each station can be first processed into a three-dimensional tensor structure. N,l, w Then, by superimposing the various feature variables, it is processed into a four-dimensional tensor structure. N,l,w,C Subsequently, the time dimension was dynamically constructed using the time-series sliding window technique. t Finally, a five-dimensional tensor conforming to the input specification of a deep learning model is generated. N,t, l,w,C This processing workflow effectively preserves the spatiotemporal evolution characteristics of the precipitation field, laying a data foundation for subsequent spatiotemporal feature extraction. The time step... t With spatial neighborhood window l , w The hyperparameters can be optimized to determine the optimal combination of parameters.

[0112] In this embodiment of the application, the model performance is significantly sensitive to the spatial neighborhood window parameter. The optimal spatial window size of each combined model differs greatly, while the sensitivity to the time step is low. The optimal time step tends to be consistent. Considering both accuracy and computational cost, a time step of 5 and a spatial neighborhood window of 3×3 are ultimately selected.

[0113] Furthermore, continue to refer to Figure 2 In the embodiments of this application, the precipitation classification regression model can be divided into a classification module and a regression module. During the training phase of the classification module, a four-fold cross-validation method can be used. The training sample dataset is randomly divided into four parts based on the ground station number, with three parts used for training (e.g., 210,384 samples) and one part used for testing (e.g., 52,596 samples). This process is repeated four times to ensure that precipitation from each station participates in the testing. Next, the precipitation event classification task is trained using four models: RF, XGBoost, CL, and ConvLSTM. After each model is trained, a corresponding trained model is obtained.

[0114] here, Figure 2The diagram illustrates the model structures of four models: RF, XGBoost, CL, and ConvLSTM. Detailed descriptions of the model structures can be found in relevant existing technologies, and this application will not describe them in detail.

[0115] During the regression module training phase, the input data partitioning and sample size are consistent with those of the classification module training phase. A four-fold cross-validation method can also be used. Four models—RF, XGBoost, CL, and ConvLSTM—are trained to perform regression tasks on precipitation events. After repeated iterations, each of the four models generates its corresponding trained model. Finally, the results from the classification and regression modules are combined to generate a high-precision, high-resolution precipitation product spanning historical years.

[0116] Furthermore, by combining the models in pairs, the trained classification and regression models can be combined to obtain 16 combined models, namely RF-RF, RF-XGBoost, RF-ConvLSTM, RF-CL, XGBoost-RF, XGBoost-XGBoost, XGBoost-ConvLSTM, XGBoost-CL, ConvLSTM-RF, ConvLSTM-XGBoost, ConvLSTM-ConvLSTM, ConvLSTM-CL, CL-RF, CL-XGBoost, CL-ConvLSTM, and CL-CL. Here, CL stands for CNN-LSTM model.

[0117] Finally, the fusion results of the aforementioned 16 combined models were evaluated. First, based on ground monitoring data, indicators were selected to evaluate the precipitation fusion results of the 16 combined models (named as classification model-regression model) over a historical time span (e.g., 2006-2017) in terms of rainfall accuracy and precipitation event detection capability. Ultimately, the optimal combined model was selected to generate high-precision, high spatiotemporal resolution precipitation products for the entire study area, providing a data foundation for subsequent research.

[0118] Figure 4 This is a schematic diagram showing the precipitation fusion accuracy results of different combined models provided in the embodiments of this application. Figure 5 This is a schematic diagram of the box layout of the site indicator results provided in the embodiments of this application, such as... Figure 4 , Figure 5As shown, under different evaluation methods, the model combination method is highly sensitive to the fusion effect. Among them, the deep learning model-machine learning model combination performs the best, that is, the correlation coefficient CC is larger, the mean absolute error MAE and the root mean square error RMSE are smaller, and the optimal combination model is CL-RF. The next best are machine learning model-machine learning model and machine learning model-deep learning model, while deep learning model-deep learning model performs the worst. The experiment shows that introducing spatiotemporal features can improve the model accuracy by more than 5%, highlighting the importance of spatiotemporal correlation in precipitation fusion.

[0119] Figure 6 This is a line graph illustrating the indices of different combined models / precipitation sources under different rainfall intensities, as provided in the embodiments of this application. Figure 6 As shown, it can be seen that the optimal combination model CL-RF performs best in precipitation prediction regardless of the rainfall scenario.

[0120] from Figures 4 to 6 As can be seen from this, the method provided in the embodiments of this application has three major advantages:

[0121] (1) It can complement the advantages of different types of models, couple machine learning models with stable prediction performance and deep learning models with strong spatiotemporal feature extraction capabilities, and fully consider the spatiotemporal correlation of features to solve problems such as weak precipitation identification ability and low accuracy of rainfall intensity.

[0122] (2) Explore the effect of combining different types of models, give full play to the advantages of different models, construct a fusion model that can characterize the spatiotemporal dynamics between multi-source precipitation and characteristic factors, and generate high-precision, high spatiotemporal resolution precipitation products.

[0123] (3) The proposed method is significantly better than the original precipitation data (GPM remote sensing precipitation data and ERA5 reanalysis precipitation data) in terms of overall fusion accuracy, spatial accuracy distribution characteristics and different rainfall intensities. Among them, the optimal model CL-RF achieved CC, RMSE, MAE and CSI of 0.871, 4.337 mm / day, 1.717 mm / day and 0.591, respectively, which are more than 85%, 45%, 45% and 19% higher than the original data. This shows that the fusion model has significant advantages in precipitation identification ability and fusion accuracy and has broad application value.

[0124] The method in this application embodiment improves the accuracy of the precipitation classification and regression model by iteratively training each component model within the precipitation classification and regression model using multiple sets of training samples.

[0125] The multi-source precipitation depth spatiotemporal fusion device provided in this application is described below. The multi-source precipitation depth spatiotemporal fusion device described below can be referred to in correspondence with the multi-source precipitation depth spatiotemporal fusion method described above.

[0126] Figure 7 This is a schematic diagram of the multi-source precipitation depth spatiotemporal fusion device provided in the embodiments of this application, as shown below. Figure 7 As shown, the device includes:

[0127] Preprocessing module 10 is used to preprocess multi-source precipitation data of the area to be measured and determine the multi-source precipitation data with target spatiotemporal resolution.

[0128] Prediction module 20 is used to input multi-source precipitation data with target spatiotemporal resolution into precipitation classification and regression model to obtain precipitation prediction results for the area to be measured output by precipitation classification and regression model;

[0129] Among them, the precipitation classification regression model is built based on deep learning and machine learning models. It is trained based on multi-source precipitation data samples and their corresponding precipitation data labels. It is used to identify each precipitation location in the area to be measured and predict the precipitation amount at each precipitation location based on the fusion features obtained by feature extraction and fusion of multi-source precipitation data. Multi-source precipitation data includes remote sensing precipitation data, meteorological database data and digital elevation data.

[0130] It is understood that the detailed functional implementation of each of the above units / modules can be found in the description in the aforementioned method embodiments, and will not be repeated here.

[0131] It should be understood that the above-described device is used to execute the methods in the above embodiments. The implementation principle and technical effect of the corresponding program modules in the device are similar to those described in the above methods. The working process of the device can be referred to the corresponding process in the above methods, and will not be repeated here.

[0132] The multi-source precipitation depth spatiotemporal fusion device of this application integrates the excellent spatiotemporal feature extraction capability of deep learning models with the advantages of strong generalization and stability of machine learning models. It processes the multi-source precipitation data of the area to be measured into high spatiotemporal resolution multi-source precipitation data, and performs multi-source precipitation depth spatiotemporal fusion prediction by combining the precipitation classification regression model constructed and trained by deep learning models and machine learning models. This can effectively improve the effect of multi-source precipitation data depth spatiotemporal fusion and achieve high-precision and high spatiotemporal resolution precipitation prediction.

[0133] Based on the methods in the above embodiments, this application provides an electronic device, such as... Figure 8As shown, the electronic device may include a processor 810, a communications interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communications interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 can call logical instructions in the memory 830 to execute the methods in the above embodiments.

[0134] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0135] Based on the methods in the above embodiments, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0136] Based on the methods in the above embodiments, this application provides a computer program product that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0137] It is understood that the processor in the embodiments of this application can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. A general-purpose processor can be a microprocessor or any conventional processor.

[0138] The method steps in this application embodiment can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, portable hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can reside in an ASIC.

[0139] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0140] It is understood that the various numerical designations used in the embodiments of this application are merely for the convenience of description and are not intended to limit the scope of the embodiments of this application.

[0141] It should be understood that expressions such as “comprising” and “may include” used in this application indicate the existence of the disclosed functions, operations, or constituent elements, and do not limit one or more additional functions, operations, and constituent elements. In this application, terms such as “comprising” and / or “having” are to be interpreted as indicating a particular characteristic, number, operation, constituent element, component, or combination thereof, but not to exclude the existence or possibility of adding one or more other characteristics, numbers, operations, constituent elements, components, or combinations thereof.

[0142] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A multi-source precipitation depth spatiotemporal fusion method, characterized in that, include: Data preprocessing is performed on the multi-source precipitation data of the area to be measured to determine the multi-source precipitation data with the target spatiotemporal resolution. The multi-source precipitation data with the target spatiotemporal resolution is input into the precipitation classification and regression model to obtain the precipitation prediction results of the area to be measured output by the precipitation classification and regression model. The precipitation classification regression model is constructed based on deep learning and machine learning models. It is trained using multi-source precipitation data samples and their corresponding precipitation data labels. It is used to identify each precipitation location in the area to be measured based on the fused features obtained by feature extraction and fusion of multi-source precipitation data at the target spatiotemporal resolution, and to predict the precipitation amount at each precipitation location. The multi-source precipitation data includes remote sensing precipitation data, meteorological database data, and digital elevation data. The precipitation classification and regression model includes a classification module and a regression module. The classification module is used to identify precipitation events, and the regression module is used to predict rainfall. The classification module is constructed based on a combination of a deep learning model and a machine learning model, or based solely on a deep learning model. The regression module is constructed based on a combination of a deep learning model and a machine learning model, or based solely on a machine learning model. The step of inputting the multi-source precipitation data with the target spatiotemporal resolution into the precipitation classification and regression model to obtain the precipitation prediction results for the area to be measured output by the precipitation classification and regression model includes: The multi-source precipitation data with the target spatiotemporal resolution is input into the classification module to obtain each precipitation location in the area to be measured, as output by the classification module. The multi-source precipitation data with target spatiotemporal resolution corresponding to each precipitation location is input into the regression module to obtain the precipitation prediction result for each precipitation location in the area to be measured, output by the regression module.

2. The multi-source precipitation depth spatiotemporal fusion method according to claim 1, characterized in that, The meteorological database data includes precipitation data and meteorological data; the multi-source precipitation data of the area to be measured undergoes data preprocessing to determine the multi-source precipitation data with the target spatiotemporal resolution, including: The precipitation data in the meteorological database is accumulated and aggregated hourly to obtain the precipitation data of the meteorological database at the target time scale, and the average value of the meteorological data is calculated to obtain the meteorological data at the target time scale. Using interpolation, the precipitation data and meteorological data of the meteorological database at the target time scale are respectively transformed into precipitation data and meteorological data of the meteorological database at the target spatiotemporal resolution; The digital elevation data is resampled according to the target spatiotemporal resolution to obtain digital elevation data with the target spatiotemporal resolution; the target spatiotemporal resolution is the spatiotemporal resolution of the remote sensing precipitation data.

3. The multi-source precipitation depth spatiotemporal fusion method according to claim 1, characterized in that, The classification module is a CNN-LSTM model, and the regression module is a random forest model.

4. The multi-source precipitation depth spatiotemporal fusion method according to claim 1, characterized in that, Both the classification module and the regression module are joint models composed of a random forest model, an XGBoost model, a CNN-LSTM model, and a ConvLSTM model in parallel. The step of inputting the multi-source precipitation data with the target spatiotemporal resolution into the classification module to obtain each precipitation location in the area to be measured, as output by the classification module, includes: The multi-source precipitation data with the target spatiotemporal resolution are respectively input into the random forest model, the XGBoost model, the CNN-LSTM model and the ConvLSTM model to obtain the first recognition result output by the random forest model, the second recognition result output by the XGBoost model, the third recognition result output by the CNN-LSTM model and the fourth recognition result output by the ConvLSTM model; Using a voting method, each precipitation location in the area to be measured is determined based on the first identification result, the second identification result, the third identification result, and the fourth identification result.

5. The multi-source precipitation depth spatiotemporal fusion method according to claim 4, characterized in that, The step of inputting multi-source precipitation data with target spatiotemporal resolution corresponding to each precipitation location into the regression module to obtain the precipitation prediction result for each precipitation location in the area to be measured, output by the regression module, includes: The multi-source precipitation data with target spatiotemporal resolution corresponding to each precipitation location are respectively input into the random forest model, the XGBoost model, the CNN-LSTM model and the ConvLSTM model to obtain the first prediction result output by the random forest model, the second prediction result output by the XGBoost model, the third prediction result output by the CNN-LSTM model and the fourth prediction result output by the ConvLSTM model; The precipitation prediction result for each precipitation location in the area to be measured is determined by averaging the first prediction result, the second prediction result, the third prediction result, and the fourth prediction result.

6. The multi-source precipitation depth spatiotemporal fusion method according to any one of claims 1-5, characterized in that, Before inputting the multi-source precipitation data with the target spatiotemporal resolution into the precipitation classification and regression model to obtain the precipitation prediction results for the area to be measured output by the precipitation classification and regression model, the method further includes: Data preprocessing is performed on the multi-source precipitation data samples of the area to be measured to determine the multi-source precipitation data samples with the target spatiotemporal resolution; Obtain precipitation monitoring data from ground stations corresponding to the multi-source precipitation data samples, and determine the precipitation data labels corresponding to the multi-source precipitation data samples with the target spatiotemporal resolution based on the precipitation monitoring data from the ground stations according to the target precipitation threshold. Multiple sets of training samples are obtained by using the multi-source precipitation data samples with the target spatiotemporal resolution and their corresponding precipitation data labels as a set of training samples. The precipitation classification regression model was trained using multiple sets of training samples to obtain a well-trained precipitation classification regression model.

7. A multi-source precipitation depth spatiotemporal fusion device, characterized in that, include: The preprocessing module is used to preprocess multi-source precipitation data of the area to be measured and determine the target spatiotemporal resolution of the multi-source precipitation data. The prediction module is used to input the multi-source precipitation data with the target spatiotemporal resolution into the precipitation classification and regression model to obtain the precipitation prediction results of the area to be measured output by the precipitation classification and regression model. The precipitation classification regression model is constructed based on deep learning and machine learning models. It is trained using multi-source precipitation data samples and their corresponding precipitation data labels. It is used to identify each precipitation location in the area to be measured based on the fused features obtained by feature extraction and fusion of multi-source precipitation data at the target spatiotemporal resolution, and to predict the precipitation amount at each precipitation location. The multi-source precipitation data includes remote sensing precipitation data, meteorological database data, and digital elevation data. The precipitation classification and regression model includes a classification module and a regression module. The classification module is used to identify precipitation events, and the regression module is used to predict rainfall. The classification module is constructed based on a combination of a deep learning model and a machine learning model, or based solely on a deep learning model. The regression module is constructed based on a combination of a deep learning model and a machine learning model, or based solely on a machine learning model. The step of inputting the multi-source precipitation data with the target spatiotemporal resolution into the precipitation classification and regression model to obtain the precipitation prediction results for the area to be measured output by the precipitation classification and regression model includes: The multi-source precipitation data with the target spatiotemporal resolution is input into the classification module to obtain each precipitation location in the area to be measured, as output by the classification module. The multi-source precipitation data with target spatiotemporal resolution corresponding to each precipitation location is input into the regression module to obtain the precipitation prediction result for each precipitation location in the area to be measured, output by the regression module.

8. An electronic device, characterized in that, include: At least one memory for storing computer programs; At least one processor is configured to execute a program stored in the memory, wherein when the program stored in the memory is executed, the processor is configured to perform the method as described in any one of claims 1-6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is run on the processor, it causes the processor to perform the method as described in any one of claims 1-6.

Images