Method for precision correction and downscaling of multi-source precipitation data based on causality constraints

Abstract

The application provides a multi-source precipitation data precision correction and downscaling method based on a causal relationship constraint, relates to the meteorological and hydrological fields, and the method comprises the following steps: establishing a characteristic causal relationship structure tree for multi-source precipitation data, and adding a constraint of the characteristic causal relationship on the basis of an LSTM deep learning correction model; building a CLSTM precipitation data precision correction model under the constraint of the causal relationship, fusing a GRASPS precipitation data downscaling method, generating a causal relationship constrained precipitation correction and downscaling model Casual-GRASPS, and generating a high-precision, high-resolution, high-spatiotemporal-continuity fused precipitation data set conforming to a statistical causal relationship. The application has the beneficial effects that a reliable method is provided for the downscaling of meteorological data such as precipitation, and a reliable data basis is provided for fine-scale precipitation research in a basin and a city region.

Description

Technical Field

[0001] This application relates to the field of meteorology and hydrology, and in particular to a method for accuracy correction and downscaling of multi-source precipitation data based on causal constraints. Background Technology

[0002] Precipitation is a crucial component of the Earth's climate system, directly impacting agriculture, water resources, and natural disasters. However, precipitation observation and forecasting face numerous challenges. Traditional observation methods suffer from low spatial and temporal resolution and local sampling biases, increasing the uncertainty of precipitation data quality. Therefore, accurately and reliably acquiring and analyzing precipitation data has become a hot topic and a significant challenge in current climate change and extreme hydrological research.

[0003] Traditional precipitation observation methods include satellite measurements, ground-based radar measurements, meteorological data reanalysis, and ground-based meteorological station monitoring, each with its own advantages and disadvantages. Satellite precipitation measurements have a wide range and high spatiotemporal continuity, but their accuracy is easily affected by factors such as climate region, seasonal variation, altitude, and precipitation intensity. Ground-based radar measurements are more accurate than satellite measurements, but their measurement range is limited, and their accuracy is greatly affected by environmental and climatic conditions. Meteorological reanalysis data has a broad spatial range and long-term series, but it differs significantly from the actual precipitation field. Ground-based meteorological station data has the highest observation accuracy, but it can only measure precipitation at single points, and due to economic, topographical, and other factors, its spatial distribution is uneven, making it difficult to obtain spatially continuous precipitation data.

[0004] To leverage the advantages of multi-source precipitation data and obtain more accurate and reliable high-quality precipitation data, precipitation data correction and downscaling based on multi-source precipitation data fusion methods have become important approaches to address precipitation data quality issues. Current fusion methods can be divided into two categories: traditional mathematical methods and machine learning methods. Traditional mathematical methods mainly rely on interpolation and regression to fuse precipitation data. In recent years, deep learning methods from machine learning have also been widely used for data fusion. Because precipitation data is currently widely available and readily accessible, it aligns with the requirement of deep learning to train on large amounts of data. Therefore, combining multi-source precipitation data and fusing it using deep learning methods can better compensate for the deficiencies of various data sources, improve the quality and accuracy of precipitation data, and thus provide reliable data support for research in related fields. However, the internal structure and parameters of deep learning models are similar to black boxes. Learning models are obtained by directly fitting a large amount of known data, making their internal mechanisms difficult to understand, the interactions between features unclear, and the model's generalization performance dependent on a large amount of training data.

[0005] Based on this, studying the statistical causal relationships between multi-source precipitation data and related factors, and exploring precipitation data correction and downscaling methods suitable for integrating different data sources, can provide a more accurate and reliable data foundation for precipitation research, and provide strong support and assistance for fields such as meteorological forecasting, natural disaster monitoring and water resource management. Summary of the Invention

[0006] The purpose of this invention is to address the technical problems of traditional machine learning methods, such as the difficulty in understanding the internal mechanisms, the unclear interactions between features, and the dependence of model generalization performance on a large amount of training data. This invention provides a method for accuracy correction and downscaling of multi-source precipitation data based on causal constraints.

[0007] The above-mentioned objective of this application is achieved through the following technical solution:

[0008] S1: Collect and preprocess multi-source precipitation data and precipitation auxiliary data; multi-source precipitation data includes: precipitation data from ground meteorological stations, satellite precipitation data, and meteorological reanalysis data;

[0009] S2: Construct a GCA-based feature causal relationship structure tree for each ground meteorological station in the study area;

[0010] S3: Construct an LSTM precipitation data correction model;

[0011] S4: By using the causal relationship information in the feature causal relationship tree structure, the LSTM precipitation data correction model is improved to obtain the CLSTM precipitation data accuracy correction model; the CLSTM precipitation data accuracy correction model includes: the linear CLSTM-Linar precipitation data correction model and the nonlinear CLSTM-Nolinar precipitation data correction model.

[0012] S5: Based on the Thiessen polygon theory, the entire study area is divided into a predetermined number of sub-regions; each sub-region contains at least one ground meteorological station; the CLSTM precipitation data accuracy correction model within the sub-region is used to perform precipitation correction on the multi-source precipitation data of each sub-region.

[0013] S6: Based on the GRASPS method, multi-source precipitation data are downscaled to obtain downscaled precipitation data;

[0014] S7: Construct the precipitation correction and downscaling model Casual-GRASPS using CLSTM-Nolinar precipitation data correction model and downscaled precipitation data;

[0015] S8: A high-precision, high-resolution, and highly spatiotemporally continuous fused precipitation dataset is obtained through precipitation correction and the Casual-GRASPS downscaling model.

[0016] An electronic device includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to enable the electronic device to perform a method for multi-source precipitation data accuracy correction and downscaling based on causal constraints.

[0017] A computer-readable storage medium storing instructions that, when executed, perform a method for accuracy correction and downscaling of multi-source precipitation data based on causal constraints.

[0018] The beneficial effects of the technical solution provided in this application are:

[0019] A causal relationship tree between precipitation and its auxiliary data features was established, and a CLSTM precipitation data accuracy correction model with corresponding causal relationship constraints was constructed. The GRASPS method was integrated to generate Casual-GRASPS, a causal relationship-constrained multi-source precipitation data accuracy correction and downscaling method, providing a reliable method for downscaling precipitation and other meteorological data. Attached Figure Description

[0020] The present application will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings:

[0021] Figure 1 This is a flowchart illustrating the implementation of the method for multi-source precipitation data accuracy correction and downscaling under causal constraints in the embodiments of this application.

[0022] Figure 2 This is a map showing the locations of meteorological stations in the embodiments of this application;

[0023] Figure 3 This is a graph showing the Pearson correlation coefficient analysis results in the embodiments of this application;

[0024] Figure 4 This is a comparison chart of the feature causal relationship analysis capabilities in the embodiments of this application;

[0025] Figure 5 This is a comparative analysis chart of the correction results of multi-source precipitation data in the embodiments of this application;

[0026] Figure 6 This is a performance result comparison and analysis chart from the embodiments of this application;

[0027] Figure 7 This is a comparative analysis chart showing the performance results of precipitation accuracy in the embodiments of this application with the original satellite precipitation data;

[0028] Figure 8This is a comparison diagram of the spatial distribution effects in the embodiments of this application;

[0029] Figure 9 This is a comparative analysis of the Casual-GRASPS, GRASPS, and LSTM precipitation fusion downscaling results, the original satellite precipitation, and the precipitation measured by rain gauges under four seasonal conditions in the embodiments of this application.

[0030] Figure 10 This is a schematic diagram of the electronic device structure in the embodiments of this application. Detailed Implementation

[0031] To provide a clearer understanding of the technical features, objectives, and effects of this application, the specific embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0032] The embodiments of this application provide a method for accuracy correction and downscaling of multi-source precipitation data based on causal constraints.

[0033] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating the steps of a method for accuracy correction and downscaling of multi-source precipitation data based on causal constraints in an embodiment of this application, including:

[0034] S1: Collect and preprocess multi-source precipitation data and precipitation auxiliary data; multi-source precipitation data includes: precipitation data from ground meteorological stations, satellite precipitation data, and meteorological reanalysis data;

[0035] Step S1 includes:

[0036] The satellite precipitation data includes: IMERG satellite precipitation data with a spatial resolution of 0.1°, CHIRPS satellite precipitation data with a spatial resolution of 0.05°, and PERSIANN-CDR satellite precipitation data with a spatial resolution of 0.25°; the temporal resolution of the satellite precipitation data is daily, and the spatial resolutions are 0.1°, 0.05°, and 0.25°, respectively.

[0037] Specifically, using IMERG satellite precipitation data, CHIRPS satellite precipitation data, and PERSIANN-CDR satellite precipitation data can retain the advantages of each to compensate for the shortcomings of existing precipitation products.

[0038] The meteorological reanalysis data includes: cloud base height, predicted albedo, surface pressure, two-meter temperature, total cloud cover, 100-meter horizontal wind speed, and 100-meter vertical wind speed; the time resolution of the meteorological reanalysis data is hourly, and the spatial resolution is 0.25°.

[0039] Specifically, the meteorological reanalysis data includes cloud base height (CBH), forecast albedo (FAL), surface pressure (SP), 2-meter temperature (T2M), total cloud cover (TCC), 100-meter horizontal wind speed (U100), and 100-meter vertical wind speed (V100) from ERA5, all with a temporal resolution of hourly and a spatial resolution of 0.25°.

[0040] The precipitation-related auxiliary data include: normalized difference vegetation index, digital elevation model, and land use data.

[0041] Specifically, the precipitation data from ground meteorological stations comes from the precipitation values ​​measured by rain gauges at 81 meteorological stations that have undergone quality control screening, with time intervals of 12 hours and 24 hours.

[0042] Step S1 also includes:

[0043] The preprocessing includes:

[0044] Perform format and unit conversion, anomaly handling, and missing value processing on precipitation data from ground meteorological stations;

[0045] The spatial resolution of satellite precipitation data is standardized to 0.1°.

[0046] Unify the temporal and spatial resolution of meteorological reanalysis data, and perform anomaly and missing value handling.

[0047] S2: Construct a GCA-based feature causal relationship structure tree for each ground meteorological station in the study area;

[0048] Step S2 includes:

[0049] S21: Perform linear and nonlinear Granger causality tests on multi-source precipitation data and precipitation auxiliary data to obtain the causal relationship between multi-source precipitation data and precipitation auxiliary data;

[0050] S22: By analyzing the Pearson correlation coefficient between multi-source precipitation data and precipitation auxiliary data, the differences in precipitation characteristics were obtained;

[0051] S23: Construct a characteristic causal relationship structure tree for each ground meteorological station by leveraging causal relationships and differences in precipitation characteristics.

[0052] Specifically, linear and nonlinear Granger causality tests were used to conduct an in-depth analysis of the causal relationships between multi-source precipitation data and precipitation auxiliary data from multiple perspectives. By analyzing the Pearson correlation coefficients between each input feature and precipitation, the differences in precipitation characteristics between different stations were revealed. A characteristic causal relationship structure tree for each ground meteorological station was constructed. The characteristic causal relationship structure trees constructed by linear and nonlinear GCA (Granger ausality analysis) were analyzed and compared, showing that the linear causal structure tree is more stable.

[0053] S3: Construct an LSTM precipitation data correction model;

[0054] Specifically, precipitation is a long-term changing process, and the LSTM model uses three different gates to pass long short-term memory to solve the problem of long-term dependencies, thus preserving long-term information and making it suitable for long-term series tasks such as precipitation.

[0055] S4: By using the causal relationship information in the feature causal relationship tree structure, the LSTM precipitation data correction model is improved to obtain the CLSTM precipitation data accuracy correction model; the CLSTM precipitation data accuracy correction model includes: the linear CLSTM-Linar precipitation data correction model and the nonlinear CLSTM-Nolinar precipitation data correction model.

[0056] Specifically, precipitation correction is performed on LSTM models based on causal constraints. The constructed LSTM precipitation data correction model is improved by incorporating linear and nonlinear GCA causal structures, resulting in linear and nonlinear CLSTM-Linar and CLSTM-Nolinar precipitation data correction models, respectively. Each node of the feature causal relationship structure tree is then analyzed. Introduce a causal relationship state Used to determine the nodes in the feature causal relationship tree structure Use the parent node to constrain the hidden state Furthermore, the causal relationship state of the parent node is applied to the fully connected layer for precipitation accuracy correction.

[0057] S5: Based on the Thiessen polygon theory, the entire study area is divided into a predetermined number of sub-regions; each sub-region contains at least one ground meteorological station; the CLSTM precipitation data accuracy correction model within the sub-region is used to perform precipitation correction on the multi-source precipitation data of each sub-region.

[0058] Specifically, by using a model of stations within the region, precipitation correction is performed on each sub-region, overcoming the problem that rain gauges can only measure precipitation at a single point and cannot seamlessly cover the entire study area.

[0059] S6: Based on the GRASPS method, multi-source precipitation data are downscaled to obtain downscaled precipitation data;

[0060] Specifically, the GRASPS method is a technique that combines point-area fusion and deep learning to generate high-precision, high-spatiotemporal resolution, and spatiotemporally continuous soil moisture data. The GRASPS downscaling method integrates multi-source precipitation datasets, including remote sensing products, model-based products, and diverse sensing capabilities from ground measurements. Its effectiveness has been well validated in downscaling variables such as soil moisture and temperature. Furthermore, this method adds three high-resolution data sources—DEM, land use, and NDVI—as auxiliary variables for downscaling, further enhancing the data downscaling effect.

[0061] S7: Construct the precipitation correction and downscaling model Casual-GRASPS using CLSTM-Nolinar precipitation data correction model and downscaled precipitation data;

[0062] Step S7 includes:

[0063] S71: Pixel segmentation is performed on the output image of the CLSTM-Nolinar precipitation data correction model using a nearest neighbor interpolation method;

[0064] Specifically, the first step is to divide the original data range into a grid based on the target resolution. Then, by mapping the grid to the original data pixels, the coordinates of each pixel in the new image within the original image are determined. Finally, based on the distance between the new pixel's coordinates and the original pixel's coordinates, the value of the original pixel is assigned to all new pixels within the corresponding region. This method improves the CLSTM resolution while preserving its accuracy.

[0065] S72: Based on the multiple linear regression model, secondary data fusion is performed on the output image after pixel segmentation and multi-source precipitation data;

[0066] S73: Construct a precipitation correction and downscaling model, Casual-GRASPS, by minimizing the sum of squared residuals.

[0067] Specifically, CLSTM cell segmentation based on nearest neighbor interpolation: Nearest neighbor interpolation cell segmentation is a commonly used method in image processing, especially during image scaling, rotation, or transformation. The basic idea of ​​this method is to find the cell in the original image that is closest to the cell in the target image and assign its value to the corresponding cell in the target image; Quadratic data fusion based on multiple linear regression: Multiple linear regression is a statistical method used to analyze the relationship between multiple independent variables and a dependent variable. It is a generalization of simple linear regression, which can consider the influence of multiple independent variables on the dependent variable.

[0068] Specifically, downscaling data accuracy verification: During dataset splitting, eight ground stations were reserved for non-training purposes. The datasets from these eight stations were used to perform multi-temporal and spatiotemporal scale analysis on the secondary fusion data. The resulting fused precipitation dataset is subsequently named the Casual-GRASPS dataset, meaning a high-precision, high-resolution, and highly spatiotemporally continuous fused precipitation dataset.

[0069] Step S72 includes:

[0070] The basic form of a multiple linear regression model is as follows:

[0071]

[0072] in, The dependent variable is the precipitation after secondary fusion. This represents downscaled precipitation data generated by the GRASPS method. This indicates high-precision precipitation correction data generated using the nonlinear CLSTM-Nolinar precipitation data correction model; , and It is the regression coefficient. This indicates the error term.

[0073] S8: A high-precision, high-resolution, and highly spatiotemporally continuous fused precipitation dataset is obtained through precipitation correction and the Casual-GRASPS downscaling model.

[0074] Specifically, Figure 1 This is a flowchart illustrating the implementation of a method for accuracy correction and downscaling of multi-source precipitation data based on causal constraints. Figure 2 It includes the location, elevation, land use classification, and location of meteorological stations used by the research institute in Hubei Province; Figure 3 This is the result of the Pearson correlation coefficient analysis between the input characteristics and the precipitation at each station; Figure 4 This is a comparison of the feature causal relationship analysis capabilities of three models: Random Forest (RF), Decision Tree (DT), and Support Vector Machine (SVM). Figure 5 This is a comparative analysis of the precipitation data correction results of three models: LSTM, CLSTM-Linear, and CLSTM-Nolinear. Figure 6 This is a comparative analysis of the performance results of three models, LSTM, CLSTM-Linear and CLSTM-Nolinear, at different spatial locations. Figure 7 This is a comparative analysis of the performance of three datasets—Casual-GRASPS, GRASPS, and CLSTM—in terms of precipitation accuracy with the original satellite precipitation data. Figure 8This is a comparison of the spatial distribution effects of the Casual-GRASPS, CLSTM, and GRASPS precipitation fusion downscaling results and the original satellite precipitation data during a typical precipitation period. Figure 9 This is a comparative analysis of the downscaling results of Casual-GRASPS, GRASPS, and LSTM precipitation fusion under four seasonal conditions, along with the original satellite precipitation and rain gauge measurements.

[0075] The implementation flowchart of the multi-source precipitation data accuracy correction and downscaling method based on causal constraints of the present invention is as follows: Figure 1 As shown, the specific steps are as follows:

[0076] Example (1): Collection of basic data;

[0077] In this embodiment, four types of precipitation-related data were collected: precipitation data from surface meteorological stations came from rain gauge measurements at the National Meteorological Information Center of the China Meteorological Administration. Considering the completeness of the dataset, this study selected daily-scale precipitation measurements from January 1, 2017 to December 31, 2019. Rain gauge data were acquired at 12-hour and 24-hour intervals. Precipitation data were acquired daily at 8:00 AM and 8:00 PM Beijing time (corresponding to 0:00 AM and 12:00 PM UTC), representing the sum of precipitation at the observation points within 12 hours. These two data points were then added together to obtain the total 24-hour precipitation. All three types of precipitation data underwent quality control. After screening, 81 meteorological stations within the study area were selected for this study. The specific distribution of surface meteorological stations is shown in [link to relevant documentation]. Figure 2 Table 1 provides detailed information:

[0078] Table 1 Information on Surface Meteorological Stations

[0079]

[0080] Satellite precipitation data includes data from three satellites: IMERG, CHIRPS, and PERSIANN-CDR. IMERG data has a spatial resolution of 0.1° and a daily temporal resolution; CHIRPS data has a spatial resolution of 0.05° and a daily temporal resolution; and PERSIANN-CDR data has a spatial resolution of 0.25° and a daily temporal resolution. The time range for all data is from January 1, 2017 to December 31, 2019. Meteorological reanalysis data includes cloud base height (CBH), predicted albedo (FAL), surface pressure (SP), two-meter temperature (T2M), total cloud cover (TCC), 100-meter horizontal wind speed (U100), and 100-meter vertical wind speed (V100) from ERA5. The data time range for all data is from January 1, 2017 to December 31, 2019, with an hourly temporal resolution and a spatial resolution of 0.25°. Precipitation ancillary data includes Normalized Difference Vegetation Index (NDVI), Digital Elevation Model (DEM), and land use data. NDVI data were obtained from the MODIS Terra product MOD13Q1 of the Moderate Resolution Imaging Spectroradiometer (MODIS), with a spatial resolution of 250 m and a temporal resolution of 16 days. In this embodiment, 2019 DEM data was selected, sourced from the United States Geological Survey (USGS), with a spatial resolution of 30 meters. Land use data from 2020, sourced from the National Geomatics Center of China, was also used, with a spatial resolution of 30 meters. The original land use data was divided into ten categories: cultivated land, forest land, grassland, shrubland, wetland, water bodies, tundra, artificial surfaces, bare land, and glaciers and permanent snow cover. To simplify the analysis process and highlight the main land use types, these categories were reclassified. After reclassification, land use types were integrated into five main categories: cultivated land, forest land, grassland, water bodies, and buildings. The area and proportion of different land types within the scope of this embodiment are shown in Table 2.

[0081] Table 2. Proportion of Land Use Types

[0082]

[0083] Example (2): Data preprocessing;

[0084] In this embodiment, the meteorological station data undergoes format and unit conversion. The TXT format is converted to CSV format, and then further converted to SHP format based on the latitude and longitude information of each station. The original station data's latitude and longitude fields are in "degrees and minutes," which are converted to floating-point "degrees" format. The original precipitation value unit is 0.1 mm, which is converted to mm. Individual stations with severe data gaps are removed. For stations missing data for a particular day, linear interpolation is used to supplement the missing data. The spatial resolution of the CHIRPS, IMERG, and PESIANN-CDR satellite precipitation data is resampled to 0.1° using bilinear interpolation. The ERA5 meteorological reanalysis data is divided into segments of 24 frames each, with the daily UTC 0:00 as the boundary, and the average value is calculated to obtain the daily resolution. The spatial resolution is resampled to 0.1° using bilinear interpolation, and null values ​​indicating "no clouds" are replaced with 0.

[0085] Example (3): Construction of causal structure of precipitation and its auxiliary data features based on causal relationships;

[0086] In this embodiment, leveraging the statistical causal relationships between multi-source precipitation and related data, and the advantages of machine learning, and combining the preprocessed data obtained in embodiment (2), the Pearson correlation coefficients between each input feature and precipitation at each station are analyzed, such as... Figure 3 As shown, IMERG is significantly more correlated with stations 57251, 57475, and 57583 than with other stations, while U100 is more strongly correlated with stations 57435, 57445, 57447, and 57543. This reveals differences in precipitation characteristics among different stations, necessitating the construction of a feature causal relationship structure tree for each station. Since precipitation is a highly nonlinear physical process, the interactions between various precipitation-related data do not follow simple linear laws; therefore, a nonlinear causal structure tree is more stable. Thus, three models—Random Forest (RF), Decision Tree (DT), and Support Vector Machine (SVM)—were selected to analyze and compare the causal structure trees constructed using nonlinear GCA. The causal relationship analysis capabilities of different models are compared. Figure 4 Comparing the three models, RF and DT are significantly better than SVM in the four feature analyses.

[0087] Example (4): Precipitation correction based on LSTM model;

[0088] In this embodiment, a precipitation data correction model is constructed based on the LSTM model. Since precipitation is a long-term changing process, and the LSTM model uses three different gates to pass long short-term memory to solve the long-term dependency problem, it can preserve long-term information and is suitable for long-term series tasks such as precipitation.

[0089] Example (5): Precipitation correction of LSTM model based on causal constraints;

[0090] In this embodiment, based on the LSTM precipitation data correction model of embodiment (4), it is improved by adding linear and nonlinear GCA causal structures respectively, forming linear and nonlinear CLSTM-Linar and CLSTM-Nolinar precipitation data correction models. In order to introduce causal relationship information into LSTM, a new state, namely the causal relationship state, is introduced for each node i. The hidden state is constrained based on the parent node of node i in the causal structure (denoted as...). To facilitate fusion, the causal state of the root node (i.e., precipitation in this study) is further applied to the fully connected layer for precipitation accuracy correction. Since the causal state contains both temporal interdependence and causal information, the CLSTM model can learn the temporal dependence and causal information of each node in the causal structure.

[0091] Example (6): Regional precipitation correction based on station location;

[0092] In this embodiment, based on the LSTM precipitation correction model obtained in embodiments (4) and (5) and the linear and nonlinear CLSTM-Linar and CLSTM-Nolinar precipitation correction models with causal constraints, the study area is divided into multiple regions according to the Thiessen polygon theory. Precipitation data in each region is corrected based on the three correction models trained on the stations within that region. The correction results of the three models are compared and analyzed based on four evaluation indicators: PCC, RMSE, MAE, and Bias. Line graphs are plotted for comparison. Figure 5 The specific results are shown in Table 3:

[0093] Table 3 Detailed information on model indicators

[0094]

[0095] Compared to LSTM and CLSTM-Linear, CLSTM-Nolinear exhibits superior performance. Firstly, from a PCC perspective, CLSTM-Nolinear achieves an average PCC of 0.55, significantly exceeding LSTM (0.43) and CLSTM-Linear (0.26). This indicates that CLSTM-Nolinear fits the actual precipitation more effectively and demonstrates a stronger linear relationship. Among the three error metrics, in terms of RMSE, CLSTM-Nolinear has an average of only 7.47 mm, significantly lower than LSTM (8.27 mm) and CLSTM-Linear (8.76 mm). Further examining MAE, CLSTM-Nolinear displays the smallest mean absolute error at only 3.47 mm. In comparison, LSTM and CLSTM-Linear have averages of 4.28 mm and 4.72 mm, respectively. Finally, regarding bias, the average bias of CLSTM-Nolinear is close to zero (-0.05 mm), which is lower than that of LSTM (0.28 mm) and CLSTM-Linear (0.95 mm). This indicates that CLSTM-Nolinear has a smaller correction error and is more accurate in correcting real precipitation data. It is worth noting that CLSTM-Linear performs poorly across all metrics, suggesting that linear causal relationship analysis of input features struggles to capture the causal relationships between features. The resulting causal structure is too simplistic, losing much useful information, to the point that its performance is even lower than LSTM. LSTM, by directly inputting all features, captures more precipitation information than CLSTM-Linear. Overall, the CLSTM-Nolinear model, obtained by adding nonlinear causal constraints to the LSTM structure, can more accurately capture various types of precipitation data and the potential connections between related data, and its correction results are closer to actual precipitation.

[0096] Given the wide coverage area of ​​this embodiment, the model's performance varies at different spatial locations. To further explore the performance changes of the model at different spatial locations, a more detailed analysis of the spatial distribution of the four evaluation indicators was conducted, and the results are as follows: Figure 6As shown, in the model performance comparison at different locations, CLSTM exhibited the best performance. It is worth noting that among all stations, the CLSTM-Nolinear bias was between -1.00 and -0.50 mm at 56 stations, indicating that this model slightly underestimates actual precipitation. However, compared to the other two models, CLSTM-Nolinear remains the best. Furthermore, the CLSTM-Nolinear model still demonstrates reliable precipitation correction performance under the influence of spatial factors such as different elevations and land types.

[0097] Example (7): Downscaling of precipitation data based on GRASPS;

[0098] In this embodiment, the GRASPS method is used to downscale precipitation data by leveraging the complementary advantages of multi-source precipitation data to obtain high-resolution meteorological data. Based on the CLSTM precipitation correction data obtained in Examples (4) and (5), the GRASPS method adds three high-resolution data types—DEM, land use, and NDVI—as auxiliary variables for downscaling. Spatially, to meet the high spatial resolution requirement of precipitation data, all auxiliary variables are resampled to 1 km. Then, the RF algorithm is used to perform pixel-by-pixel GRASPS fusion of the multi-source precipitation data products, and the average result algorithm is used to calculate the average value of all RF decision tree fusion results to obtain the final fusion result. Finally, under the GRASPS downscaling framework, a precipitation downscaled dataset with a spatial resolution of 1 km is generated.

[0099] Example (8): Downscaling of Casual-GRASPS precipitation data based on causal secondary fusion;

[0100] In this embodiment, the CLSTM-Nolinear precipitation data correction model of embodiment (5) and the GRASPS precipitation data downscaling method of embodiment (7) are combined. The optimal model, trained multiple times, is used to fuse the GRASPS and CLSTM data, forming the precipitation correction and downscaling model Casual-GRASPS. The performance of Casual-GRASPS, GRASPS, and CLSTM datasets in terms of precipitation accuracy is compared using four evaluation metrics: PCC, RMSE, MAE, and Bias. Figure 7As shown, GRASPS also demonstrates a significant improvement in accuracy compared to satellite precipitation data. Compared to the three types of satellite precipitation data, GRASPS's PCC improved by 65.71% to 93.33%. Secondly, the CLSTM model, as a calibration model, exhibits even higher data accuracy, with its PCC improving by 24.14% compared to GRASPS. In terms of the three error indices, RMSE and MAE, they decreased by 17.75% and 39.11%, respectively. Overall, both GRASPS and CLSTM overestimate and underestimate precipitation. GRASPS's bias is 0.32 mm, indicating a tendency to overestimate precipitation, while CLSTM's bias is -0.11, showing a significantly smaller underestimation than GRASPS's overestimation. These results fully demonstrate the superior performance of CLSTM in improving the accuracy of precipitation data. Finally, Casual-GRASPS combines the CLSTM and GRASPS models. Compared to the previous two models, Casual-GRASPS integrates the advantages of both CLSTM and GRASPS, achieving a significant improvement in data accuracy while maintaining a high spatial resolution of 1 km. Compared to CLSTM, Casual-GRASPS improves PCC by approximately 5.56%. Since Casual-GRASPS is derived from the fusion of CLSTM and GRASPS, it achieves a slight but further improvement in accuracy for specific cases where GRASPS outperforms CLSTM, while acquiring high-resolution GRASPS data. In summary, by comparing and analyzing the precipitation accuracy of the Casual-GRASPS, GRASPS, and CLSTM datasets, we have identified their respective advantages and characteristics. Casual-GRASPS, as a fusion of the two, not only possesses a fine spatial resolution of 1 km but also offers further improvements in accuracy.

[0101] To verify the downscaling effect of Casual-GRASPS, a typical precipitation period was selected, and the downscaling results for that day were evaluated. Figure 8 As shown. Figure 8 (a) for Casual-GRASPS and Figure 8 (b) The spatial resolution of GRASPS is 1 km, which is the highest among the six figures. Figure 8 (c) CLSTM is based on IMERG, and its spatial resolution and Figure 8 (d) IMERG is the same, both being 0.1°. Figure 8 (e) is CHIRPS, the highest resolution product among the three types of satellite precipitation data, reaching 0.05°. Figure 8 (f) is the lowest resolution product among the three satellite precipitation data, PERSIANN-CDR, with a resolution of 0.25°.

[0102] In terms of resolution, Casual-GRASPS and GRASPS reached a maximum of 1 km, demonstrating a significant advantage in spatial detail. Regarding the spatial distribution of precipitation, Casual-GRASPS, GRASPS, and CLSTM all exhibited similar trends. Their precipitation distribution across the entire study area showed a pattern of decreasing, increasing, and then decreasing again from west to east. Notably, Casual-GRASPS and CLSTM showed almost identical precipitation distributions, likely because Casual-GRASPS benefits from the accuracy of CLSTM, resulting in the most similar precipitation correction performance. Simultaneously, the three satellite precipitation data sets were evaluated. The results showed that IMERG underestimated actual precipitation in most areas, while CHIRPS and PERSIANN-CDR underestimated actual precipitation in the northwest and overestimated it in the east. In contrast, Casual-GRASPS effectively corrected these errors, more accurately reflecting the actual precipitation situation. It is noteworthy that GRASPS, while correcting satellite precipitation errors, also successfully integrated areas accurately monitored by satellite data. For example, in the southwest and central regions, GRASPS integrates the high precipitation trends from CHIRPS and PERSIANN-CDR; in the north and northeast, it integrates the high precipitation trends from IMERG. This integration strategy allows GRASPS to maintain high spatial resolution while also improving the accuracy of precipitation data. Overall, Casual-GRASPS successfully combines the correct precipitation components from three satellite precipitation data sets and corrects for erroneous components, thereby significantly improving both spatial resolution and the accuracy of precipitation data.

[0103] Precipitation exhibits distinct seasonal characteristics, and downscaling accuracy is closely related to precipitation intensity. Therefore, the performance of the fusion model may differ across seasons. To further analyze the performance of the fusion model in different seasons, the dataset was divided into four seasons—spring, summer, autumn, and winter—for comparative analysis. The season division was as follows: Spring: March, April, and May; Summer: June, July, and August; Autumn: September, October, and November; Winter: December, January, and February. Eight test sites were selected, and Taylor charts were used for comparative analysis. The results are as follows: Figure 9As shown. From a seasonal perspective, among the eight stations, all precipitation data showed a trend of being closest to the origin in winter. This is mainly because the study area experiences less precipitation and lower daily average precipitation in winter, resulting in a relatively smaller range of precipitation values. Conversely, in summer, except for station 57265, the precipitation data of the other seven stations were furthest from the origin. Summer precipitation is concentrated, with generally higher precipitation values, but there are also instances of no precipitation, leading to a wider range between the upper and lower limits of precipitation. Although the Casual-GRASPS fusion effect was relatively poor in autumn, its PCC reached a level close to 0.7 in all stations. This indicates that despite some errors, Casual-GRASPS still maintained relatively high accuracy in downscaling autumn precipitation data. From a station perspective, Casual-GRASPS had the best precipitation correction and downscaling results among all stations. In particular, station 57494 showed the most significant effect, with a PCC close to 0.9 in all four seasons, and even reaching 0.91 in winter. This demonstrates the stability of Casual-GRASPS across different seasons and highlights its superior performance at specific sites. Overall, Casual-GRASPS achieved excellent correction and downscaling results across different seasons and sites, showcasing its superior spatial and temporal performance.

[0104] The Casual-GRASPS precipitation accuracy correction and downscaling model of this invention is compared with existing precipitation accuracy correction and precipitation downscaling studies, and the results are shown in Table 4. In terms of spatial resolution, existing accuracy correction methods mostly have a spatial resolution between 0.05° and 0.5°, while existing downscaling methods and the Casual-GRASPS method proposed in this paper can downscale the data to 1 km. In terms of temporal resolution, existing downscaling methods mainly focus on monthly and yearly scales, while existing accuracy correction methods and the Casual-GRASPS method proposed in this paper can reach daily scales. In terms of the number of data sources selected, the Casual-GRASPS method proposed in this paper selects three types of satellite precipitation data for fusion, achieving complementary advantages, while most existing downscaling methods only correct one type of data. In terms of feature selection methods, most existing precipitation accuracy correction and downscaling studies adopt the method of directly inputting features, while this study analyzes the causal relationship between features and constructs a feature causal structure tree, realizing feature input that conforms to causal relationships. In terms of model generalization, the Casual-GRASPS method proposed in this paper can accurately capture heavy precipitation, with a PCC of up to 0.91, while most existing studies have not specifically studied heavy precipitation. In terms of factors considered, the Casual-GRASPS method takes into account the differences in feature importance caused by different station spatial locations, thus establishing a fusion model for each station. Overall, compared with previous studies, the Casual-GRASPS proposed in this paper has significantly improved precipitation accuracy correction and downscaling. Moreover, it innovatively introduces a feature causal relationship structure into the model, which enhances the model's generalization and interpretability.

[0105] Table 4 Comparison of Precipitation Data Accuracy Correction and Downscaling Methods

[0106]

[0107] This application also discloses an electronic device. (See reference...) Figure 10 , Figure 10 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of this application. The electronic device 500 may include: at least one processor 501, at least one network interface 504, a user interface 503, a memory 505, and at least one communication bus 502.

[0108] The communication bus 502 is used to enable communication between these components.

[0109] The user interface 503 may include a display screen, and optionally, the user interface 503 may also include a standard wired interface or a wireless interface.

[0110] The network interface 504 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0111] This application also discloses a computer-readable storage medium storing multiple instructions adapted for loading by a processor to execute the above-described method for accuracy correction and downscaling of multi-source precipitation data based on causal constraints.

[0112] The above are merely exemplary embodiments of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Other embodiments of this disclosure will readily conceive of those skilled in the art upon consideration of the specification and the disclosure of practical truths.

[0113] This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A method for accuracy correction and downscaling of multi-source precipitation data based on causal constraints, characterized in that, The method includes the following steps: S1: Collect and preprocess multi-source precipitation data and precipitation auxiliary data; multi-source precipitation data includes: precipitation data from ground meteorological stations, satellite precipitation data, and meteorological reanalysis data; S2: Construct a GCA-based feature causal relationship structure tree for each ground meteorological station in the study area; S3: Construct an LSTM precipitation data correction model; S4: By using the causal relationship information in the feature causal relationship tree structure, the LSTM precipitation data correction model is improved to obtain the CLSTM precipitation data accuracy correction model; the CLSTM precipitation data accuracy correction model includes: the linear CLSTM-Linar precipitation data correction model and the nonlinear CLSTM-Nolinar precipitation data correction model. S5: Based on the Thiessen polygon theory, the entire study area is divided into a predetermined number of sub-regions; each sub-region contains at least one ground meteorological station; the CLSTM precipitation data accuracy correction model within the sub-region is used to perform precipitation correction on the multi-source precipitation data of each sub-region. S6: Based on the GRASPS method, multi-source precipitation data are downscaled to obtain downscaled precipitation data; S7: Construct the precipitation correction and downscaling model Casual-GRASPS using CLSTM-Nolinar precipitation data correction model and downscaled precipitation data; S8: A high-precision, high-resolution, and highly spatiotemporally continuous fused precipitation dataset is obtained through precipitation correction and the Casual-GRASPS downscaling model.

2. The method for accuracy correction and downscaling of multi-source precipitation data based on causal constraints as described in claim 1, characterized in that, Step S1 includes: The satellite precipitation data includes: IMERG satellite precipitation data with a spatial resolution of 0.1°, CHIRPS satellite precipitation data with a spatial resolution of 0.05°, and PERSIANN-CDR satellite precipitation data with a spatial resolution of 0.25°; the temporal resolution of the satellite precipitation data is daily. The meteorological reanalysis data includes: cloud base height, predicted albedo, surface pressure, two-meter temperature, total cloud cover, 100-meter horizontal wind speed, and 100-meter vertical wind speed; the time resolution of the meteorological reanalysis data is hourly, and the spatial resolution is 0.25°. The precipitation-supporting data includes: Normalized Difference Vegetation Index (NDVI), Digital Elevation Model (DEM), and land use data.

3. The method for accuracy correction and downscaling of multi-source precipitation data based on causal constraints as described in claim 1, characterized in that, Step S1 also includes: The preprocessing includes: Perform format and unit conversion, anomaly handling, and missing value processing on precipitation data from ground meteorological stations; The spatial resolution of satellite precipitation data is standardized to 0.1°. Unify the temporal and spatial resolution of meteorological reanalysis data, and perform anomaly and missing value handling.

4. The method for accuracy correction and downscaling of multi-source precipitation data based on causal constraints as described in claim 1, characterized in that, Step S2 includes: S21: Perform linear and nonlinear Granger causality tests on multi-source precipitation data and precipitation auxiliary data to obtain the causal relationship between multi-source precipitation data and precipitation auxiliary data; S22: By analyzing the Pearson correlation coefficient between multi-source precipitation data and precipitation auxiliary data, the differences in precipitation characteristics were obtained; S23: Construct a characteristic causal relationship structure tree for each ground meteorological station by leveraging causal relationships and differences in precipitation characteristics.

5. The method for accuracy correction and downscaling of multi-source precipitation data based on causal constraints as described in claim 1, characterized in that, Step S7 includes: S71: Pixel segmentation is performed on the output image of the CLSTM-Nolinar precipitation data correction model using a nearest neighbor interpolation method; S72: Based on the multiple linear regression model, secondary data fusion is performed on the output image after pixel segmentation and multi-source precipitation data; S73: Construct a precipitation correction and downscaling model, Casual-GRASPS, by minimizing the sum of squared residuals.

6. The method for accuracy correction and downscaling of multi-source precipitation data based on causal constraints as described in claim 5, characterized in that, Step S72 includes: The basic form of a multiple linear regression model is as follows: in, The dependent variable is the precipitation after secondary fusion. This represents downscaled precipitation data generated by the GRASPS method. This indicates high-precision precipitation correction data generated using the nonlinear CLSTM-Nolinar precipitation data correction model; , and It is the regression coefficient. This indicates the error term.

7. An electronic device, characterized in that, The device includes a processor (501), a memory (505), a user interface (503), and a network interface (504). The memory (505) is used to store instructions. The user interface (503) and the network interface (504) are used to communicate with other devices. The processor (501) is used to execute the instructions stored in the memory (505) to cause the electronic device to perform the method as described in any one of claims 1-6.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed by a computer, perform the method as described in any one of claims 1-6.

Images