Satellite precipitation two-stage error correction method based on actual measurement site

Abstract

The invention discloses a satellite precipitation two-stage error correction method based on an actual measurement site, and the method comprises the steps: extracting drainage basin DEM data, obtaining a mask file of a drainage basin, and extracting the satellite precipitation of the drainage basin; performing inverse distance weighted interpolation on the rainfall data of the actual measurement site to a corresponding resolution to obtain grid data of the actual measurement site; error correction data are obtained through a deep learning model or dynamic quantile mapping; calculating a residual error between the rainfall data after error correction and grid rainfall of the actual measurement site, and learning a nonlinear relationship between the rainfall data and the residual error by using a gradient lifting regression tree model; and adding the precipitation residual error of the nonlinear residual error correction model to the error-corrected satellite precipitation error to obtain a residual error-corrected precipitation product. According to the method, the satellite precipitation product is corrected based on the actual measurement site data, the high-precision precipitation product is obtained, point-to-surface conversion of the precipitation data can be realized, and data support is provided for input of a refined hydrological model.

Description

Technical Field

[0001] This invention relates to the fields of hydrology and meteorology, and specifically to a method for two-stage error correction of satellite precipitation based on actual measurement stations. Background Technology

[0002] Precipitation is the core input variable of hydrological models, and the accuracy of its data directly determines the reliability of simulation results and the timeliness of forecasts and warnings.

[0003] Currently, mainstream precipitation observation methods (ground stations, weather radar, and satellite remote sensing) all suffer from inherent systematic errors and uncertainties. While ground-based observation data is considered a benchmark, its spatial representativeness is severely insufficient. Sparse station distribution makes it difficult to capture the spatial heterogeneity of precipitation, especially in complex terrain or areas with few stations, where observed values ​​may significantly deviate from actual precipitation conditions. Station maintenance, instrument bias, and missing records are also common problems. Radar quantitative precipitation estimation offers advantages in high spatiotemporal resolution and real-time performance, but its accuracy is easily affected by various factors, including coverage blind spots caused by ground cover obstruction, beam attenuation, ground clutter, vertical profile representativeness errors, and radar parameter calibration biases. These factors often lead to systematic overestimation or underestimation of precipitation in radar estimates. Satellite quantitative precipitation estimation provides valuable wide-area coverage and quasi-continuous observation capabilities, but its accuracy is generally lower than that of ground and radar observations. Errors mainly stem from limitations in remote sensing instrument sensitivity, insufficient sub-cloud / intra-cloud precipitation detection capabilities, difficulties in inverting complex surfaces, limitations of the inversion algorithm itself, and constraints in spatiotemporal resolution. Despite the continuous development of satellite products (such as the GPM, IMERG, and FY series), their raw data often contain significant regional or systematic biases. The inherent error characteristics of these observation methods make it difficult for raw precipitation data to directly meet the stringent accuracy requirements of hydrological models (especially high-precision distributed models). Uncorrected precipitation inputs can significantly amplify the uncertainties in hydrological simulations and forecasts, and may even lead to false alarms or missed alarms.

[0004] Therefore, developing effective precipitation correction techniques is crucial. The core objective is to identify and quantify systematic biases and random errors in precipitation data from different sources (especially remote sensing data such as satellite and radar). By utilizing more reliable data sources (such as high-density ground station networks) or advanced statistical / physical methods, bias corrections, accuracy improvements, and uncertainty quantifications can be applied to raw precipitation estimates. This provides hydrological models with more accurate and reliable input drivers, ultimately enhancing flood disaster simulation, forecasting, and early warning capabilities.

[0005] To better describe the spatial variation of precipitation, fully utilize the advantages of different precipitation sources, and improve the quality of precipitation products, we use high spatiotemporal resolution satellite quantitative precipitation estimates as the background field and ground station precipitation as a benchmark to perform error correction and residual correction on satellite precipitation. Summary of the Invention

[0006] The purpose of this invention is to provide a two-stage error correction method for satellite precipitation based on actual measurement stations. This method utilizes precipitation data from actual measurement stations, transforms the station data from point to area, and performs error correction and residual correction on the precipitation data of satellite precipitation products. The resulting precipitation products provide data support for the input of refined hydrological models.

[0007] To achieve the above functions, this invention designs a two-stage error correction method for satellite precipitation based on actual measurement stations, which executes the following steps S1-S9 to complete the error correction of satellite precipitation data:

[0008] Step S1: For the target area with actual measurement stations, extract the target watershed, the DEM raster data of the target watershed, and the mask file of the target watershed, and proceed to step S2;

[0009] Step S2: Using the mask file of the target watershed, extract the satellite precipitation data of the target watershed, and proceed to step S3;

[0010] Step S3: The precipitation data from the actual stations is interpolated to the spatial resolution of the satellite precipitation data through inverse distance weighted interpolation to obtain the actual station grid precipitation data with the same spatial resolution as the satellite precipitation data, and then proceed to step S4;

[0011] Step S4: Construct a deep learning model, using the original precipitation data from the actual measurement stations and the grid precipitation data from the actual measurement stations as inputs, and the predicted value of the precipitation data as the output, to train the deep learning model, and proceed to step S5;

[0012] Step S5: Determine whether the daily precipitation at the measured station is greater than the preset threshold. If it is greater than the preset threshold, use the precipitation data prediction value from step S4 as the precipitation data for the day and proceed to step S7; otherwise, proceed to step S6.

[0013] Step S6: For the predicted precipitation data, perform dynamic quantile mapping to satellite precipitation data and measured station data to obtain error-corrected precipitation data, and proceed to step S7;

[0014] Step S7: Calculate the precipitation residual between the error-corrected precipitation data and the measured station grid precipitation data, and proceed to step S8;

[0015] Step S8: Based on the gradient boosting regression tree, construct a nonlinear residual correction model, fit the nonlinear relationship between the error-corrected precipitation data and the precipitation residual, and proceed to step S9;

[0016] Step S9: Superimpose the precipitation residuals onto the error-corrected precipitation data to obtain the residual-corrected precipitation data, and output the final precipitation data.

[0017] Beneficial effects: Compared with the prior art, the advantages of the present invention include:

[0018] This invention designs a two-stage error correction method for satellite precipitation based on measured stations. The method obtains a watershed mask file from the watershed's DEM (Digital Elevation Model). Satellite precipitation data is extracted from the mask file, and the satellite precipitation data is combined with station-based precipitation data to perform error correction and residual correction. This method uses high spatiotemporal resolution precipitation product data as its base data, ensuring stable and reliable data sources. The training data process is smooth, and the method utilizes the deviation between station-based and satellite precipitation data to correct for satellite precipitation, guaranteeing the objectivity and rationality of the results. This yields high-precision precipitation products as input for a refined hydrological model. Attached Figure Description

[0019] Figure 1 This is a flowchart of a two-stage error correction method for satellite precipitation based on actual measurement stations, provided by an embodiment of the present invention;

[0020] Figure 2 This is a DEM map of the target watershed provided according to an embodiment of the present invention;

[0021] Figure 3 This is a mask file diagram of the target watershed provided according to an embodiment of the present invention;

[0022] Figure 4 This is a comparative diagram of evaluation indicators for raw satellite data, error correction data, and residual correction data provided in an embodiment of the present invention;

[0023] Figure 5 This is a schematic diagram comparing the areal rainfall of the original satellite data, error correction data, and residual correction data provided in the embodiments of the present invention. Detailed Implementation

[0024] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0025] This invention provides a method for two-stage error correction of satellite precipitation based on measured stations, referring to... Figure 1 Perform the following steps S1-S9 to complete the error correction of the satellite precipitation data:

[0026] Step S1: For the target area with actual measurement stations, extract the target watershed, the DEM (Digital Elevation Model) raster data of the target watershed, and the mask file of the target watershed, and proceed to step S2.

[0027] The specific steps of step S1 are as follows:

[0028] Step S1.1: Fill the depressions in the target area;

[0029] Step S1.2: Calculate the flow direction of the watershed in the target area;

[0030] Step S1.3: Calculate the runoff of the watershed;

[0031] Step S1.4: Determine the outlet station of the watershed;

[0032] Step S1.5: Extract the target watershed, its DEM raster data, and its mask file; taking the Huang-Huai-Hai River Basin as an example, the raw DEM data of the target watershed uses digital elevation data jointly provided by the U.S. Geological Survey (USUG) and the National Geomatics Center of China, and the DEM of the target watershed is referenced... Figure 2 The target watershed mask file is referenced. Figure 3 .

[0033] Step S2: Using the mask file of the target watershed, extract the satellite precipitation data of the target watershed, and proceed to step S3;

[0034] The specific steps of step S2 are as follows:

[0035] Step S2.1: Download satellite precipitation data and determine the coverage area, spatial resolution, and temporal resolution of the satellite precipitation data;

[0036] Step S2.2: Based on the mask file of the target watershed obtained in step S1, and using it as the boundary of the target watershed, extract the satellite precipitation data of the target watershed.

[0037] Step S3: Interpolate the precipitation data from the actual stations to the spatial resolution of the satellite precipitation data using inverse distance weighting (IDW) to obtain the actual station grid precipitation data with the same spatial resolution as the satellite precipitation data, and proceed to step S4;

[0038] Inverse distance weighting (IRW) is a distance-based spatial interpolation method. Its core idea is that the value of an unknown point is influenced by nearby known points, with the influence increasing the closer the points are. Raw, measured precipitation data from actual weather stations is interpolated using ILD to obtain gridded precipitation data with the same spatial resolution as the precipitation product.

[0039] The specific steps of step S3 are as follows:

[0040] Step S3.1: Assume (x i ,y i (x) represents the coordinates of the measured station i, and (x) represents the coordinates of the measured station i. j ,y j Let be the coordinates of grid point j. Calculate the distance matrix between the measured station coordinates and the grid point coordinates, as shown in the following formula:

[0041]

[0042] In the formula, d ij This represents the distance between the measured station i and grid point j.

[0043] Step S3.2: Calculate the weight of the measured station based on the distance between the measured station and the grid point, as shown in the following formula:

[0044]

[0045] In the formula, ω i The weight of the measured site i;

[0046] Step S3.3: Based on the weights of the measured sites, perform interpolation calculations, as shown in the following formula:

[0047]

[0048] In the formula, N is the number of sample points, z i Let z(x,y) be the observed value at station i, and z(x,y) be the interpolated data at grid point j. This yields the actual precipitation data for the grid at the measured station.

[0049] Step S4: Construct a deep learning model, using the original precipitation data from the actual measurement stations and the grid precipitation data from the actual measurement stations as inputs, and the predicted value of the precipitation data as the output, to train the deep learning model, and proceed to step S5;

[0050] The specific steps of step S4 are as follows:

[0051] Step S4.1: Construct a training dataset. The input feature X is spatiotemporal sequence data, which includes the filtered original measured precipitation data and the measured precipitation data of the grid. The training target Y is the predicted value of precipitation data, which is composed of the interpolated data z(x,y) of the grid points.

[0052] Step S4.2: Design a deep learning model using a ConvLSTM network structure, including an input layer, hidden layers, an LSTM layer, and an output layer. The input layer contains spatiotemporal sequence data in the form of T×H×W, where T is the time step, H is the height of the spatial grid, and W is the width of the spatial grid. The hidden layer consists of 3 convolutional layers, and the LSTM layer has 128 units to capture temporal dependencies. The output layer contains the predicted precipitation data.

[0053] Step S4.3: Train the deep learning model. The loss function used in the training is mean squared error. An early stopping strategy is designed. The training is terminated when the loss on the validation set does not decrease for 5 consecutive rounds. Finally, the predicted precipitation data is output.

[0054] Step S5: Determine whether the daily precipitation at the measured station is greater than the preset threshold. If it is greater than the preset threshold, use the precipitation data prediction value from step S4 as the precipitation data for the day and proceed to step S7; otherwise, proceed to step S6.

[0055] The specific steps of step S5 are as follows:

[0056] Step S5.1: Based on the precipitation data of the measured station, determine whether the daily precipitation of the measured station is greater than a preset threshold. In one embodiment, the preset threshold is 50 mm / day. If it is greater than the preset threshold, proceed to step S5.2; otherwise, proceed to step S6.

[0057] Step S5.2: Select a date when the daily precipitation at the measured station is greater than the preset threshold, use the precipitation data prediction value from step S4 as the precipitation data for that day, and use it as the precipitation data for error correction in step S7, and proceed to step S7.

[0058] Step S6: For the predicted precipitation data, perform dynamic quantile mapping to satellite precipitation data and measured station data to obtain error-corrected precipitation data, and proceed to step S7;

[0059] The specific steps of step S6 are as follows:

[0060] Step S6.1: Select dates where the daily precipitation at the measured stations is less than or equal to a preset threshold, and sort the satellite precipitation data;

[0061] Step S6.2: Sort the measured precipitation data from the grid at the stations;

[0062] Step S6.3: Calculate the quantile position of the precipitation data prediction value output by the deep learning model in the satellite precipitation data distribution, using the following formula:

[0063]

[0064] Where M is the number of satellite precipitation data samples, x is a single value of satellite precipitation data, source is the set of satellite precipitation data, value is the predicted value of satellite precipitation data, count(x∈source|x≤value) is the number of precipitation data that is less than the predicted value of precipitation data, and rank is the quantile ranking;

[0065] By mapping the quantile position of the predicted precipitation data in the satellite precipitation data distribution to the corresponding quantile in the actual station data distribution, the error-corrected precipitation data is obtained.

[0066]

[0067] in, y is the inverse cumulative distribution function of the target distribution, and y is the precipitation data after final error correction.

[0068] Step S7: Calculate the precipitation residual between the error-corrected precipitation data and the measured station grid precipitation data, and proceed to step S8;

[0069] The specific steps of step S7 are as follows:

[0070] Step S7.1: Based on the measured station grid, read the error-corrected precipitation data corresponding to the measured station grid;

[0071] Step S7.2: Subtract the corresponding error-corrected precipitation data from the measured station grid precipitation data to obtain the precipitation residual.

[0072] Step S8: Based on the Gradient Boosting Regression Tree (GBRT) model, construct a nonlinear residual correction model, fit the nonlinear relationship between the error-corrected precipitation data and the precipitation residual, and proceed to step S9;

[0073] The specific steps of step S8 are as follows:

[0074] Step S8.1: Construct a dataset with precipitation residuals as the dependent variable and error-corrected precipitation data as the independent variable, and automatically learn the nonlinear relationship;

[0075] Step S8.2: Construct a nonlinear residual correction model based on gradient boosting regression tree. Set the parameters of the nonlinear residual correction model as follows: tree depth is 5, number of trees is 100, learning rate is 0.1, and divide the dataset into training set and validation set. Train the nonlinear residual correction model to obtain the trained nonlinear residual correction model.

[0076] Step S8.3: Use the trained nonlinear residual correction model to predict the residuals of the error-corrected precipitation data for each grid point, and obtain the predicted precipitation residuals. The residual prediction formula is as follows;

[0077] R pred =f GBRT (P corr )

[0078] Among them, P corr R represents the predicted precipitation data from steps S5 and S6. pred For the predicted precipitation residual, f GBRTThis represents a gradient boosting regression tree.

[0079] Step S9: Superimpose the precipitation residuals onto the error-corrected precipitation data to obtain the residual-corrected precipitation data, and output the final precipitation data.

[0080] The method for calculating the residual-corrected precipitation data in step S9 is as follows:

[0081] Final precipitation = Error-corrected precipitation + R pred

[0082] In the formula, final precipitation is the precipitation data after residual correction, error-corrected precipitation is the precipitation data after error correction, and R... pred This represents the predicted precipitation residual.

[0083] A comparative diagram of evaluation metrics for raw satellite data, error-corrected data, and residual-corrected data is provided. Figure 4 The performance of each precipitation product was analyzed after correction using BIAS, CC, RMSE, KGE, and CSI. The results show that after error correction and further residual correction, all indicators of each precipitation product were improved, resulting in higher consistency with precipitation data from actual observation stations and significantly improved precipitation forecasting. A comparison diagram of areal precipitation from raw satellite data, error-corrected data, and residual-corrected data is provided. Figure 5 After correcting the data and adjusting the residuals, the areal rainfall data of the precipitation product is closer to the actual areal rainfall data at the measured stations, and the prediction is more accurate.

[0084] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A method for two-stage error correction of satellite precipitation based on measured stations, characterized in that, Perform the following steps S1-S9 to complete the error correction of the satellite precipitation data: Step S1: For the target area with actual measurement stations, extract the target watershed, the DEM raster data of the target watershed, and the mask file of the target watershed, and proceed to step S2; Step S2: Using the mask file of the target watershed, extract the satellite precipitation data of the target watershed, and proceed to step S3; Step S3: The precipitation data from the actual stations is interpolated to the spatial resolution of the satellite precipitation data through inverse distance weighted interpolation to obtain the actual station grid precipitation data with the same spatial resolution as the satellite precipitation data, and then proceed to step S4; Step S4: Construct a deep learning model, using the original precipitation data from the actual measurement stations and the grid precipitation data from the actual measurement stations as inputs, and the predicted value of the precipitation data as the output, to train the deep learning model, and proceed to step S5; Step S5: Determine whether the daily precipitation at the measured station is greater than the preset threshold. If it is greater than the preset threshold, use the precipitation data prediction value from step S4 as the precipitation data for the day and proceed to step S7; otherwise, proceed to step S6. Step S6: For the predicted precipitation data, perform dynamic quantile mapping to satellite precipitation data and measured station data to obtain error-corrected precipitation data, and proceed to step S7; Step S7: Calculate the precipitation residual between the error-corrected precipitation data and the measured station grid precipitation data, and proceed to step S8; Step S8: Based on the gradient boosting regression tree, construct a nonlinear residual correction model, fit the nonlinear relationship between the error-corrected precipitation data and the precipitation residual, and proceed to step S9; Step S9: Superimpose the precipitation residuals onto the error-corrected precipitation data to obtain the residual-corrected precipitation data, and output the final precipitation data.

2. The method for two-stage error correction of satellite precipitation based on measured stations according to claim 1, characterized in that, The specific steps of step S1 are as follows: Step S1.1: Fill the depressions in the target area; Step S1.2: Calculate the flow direction of the watershed in the target area; Step S1.3: Calculate the runoff of the watershed; Step S1.4: Determine the outlet station of the watershed; Step S1.5: Extract the target watershed, as well as the DEM raster data and mask file of the target watershed.

3. The method for two-stage error correction of satellite precipitation based on measured stations according to claim 1, characterized in that, The specific steps of step S2 are as follows: Step S2.1: Download satellite precipitation data and determine the coverage area, spatial resolution, and temporal resolution of the satellite precipitation data; Step S2.2: Based on the mask file of the target watershed obtained in step S1, and using it as the boundary of the target watershed, extract the satellite precipitation data of the target watershed.

4. The method for two-stage error correction of satellite precipitation based on measured stations according to claim 1, characterized in that, The specific steps of step S3 are as follows: Step S3.1: Assume (x i ,y i (x) represents the coordinates of the measured station i, and (x) represents the coordinates of the measured station i. j ,y j Let be the coordinates of grid point j. Calculate the distance matrix between the measured station coordinates and the grid point coordinates, as shown in the following formula: In the formula, d ij This represents the distance between the measured station i and grid point j. Step S3.2: Calculate the weight of the measured station based on the distance between the measured station and the grid point, as shown in the following formula: In the formula, ω i The weight of the measured site i; Step S3.3: Based on the weights of the measured sites, perform interpolation calculations, as shown in the following formula: In the formula, N is the number of sample points, z i Let z(x,y) be the observed value at station i, and z(x,y) be the interpolated data at grid point j. This yields the actual precipitation data for the grid at the measured station.

5. The method for two-stage error correction of satellite precipitation based on measured stations according to claim 1, characterized in that, The specific steps of step S4 are as follows: Step S4.1: Construct a training dataset. The input feature X is spatiotemporal sequence data, which includes the filtered original measured precipitation data and the measured precipitation data of the grid. The training target Y is the predicted value of precipitation data, which is composed of the interpolated data z(x,y) of the grid points. Step S4.2: Design a deep learning model using a ConvLSTM network structure, including an input layer, hidden layers, an LSTM layer, and an output layer. The input layer contains spatiotemporal sequence data in the form of T×H×W, where T is the time step, H is the height of the spatial grid, and W is the width of the spatial grid. The hidden layer consists of 3 convolutional layers, the LSTM layer has 128 units, and the output layer contains the predicted precipitation data. Step S4.3: Train the deep learning model. The loss function used in the training is mean squared error. An early stopping strategy is designed. The training is terminated when the loss on the validation set does not decrease for 5 consecutive rounds. Finally, the predicted precipitation data is output.

6. The method for two-stage error correction of satellite precipitation based on measured stations according to claim 1, characterized in that, The specific steps of step S5 are as follows: Step S5.1: Based on the precipitation data of the measured station, determine whether the daily precipitation of the measured station is greater than the preset threshold. If it is greater than the preset threshold, proceed to step S5.2; otherwise, proceed to step S6. Step S5.2: Select a date on which the daily precipitation at the measured station is greater than the preset threshold, use the precipitation data prediction value from step S4 as the precipitation data for that day, and proceed to step S7.

7. The method for two-stage error correction of satellite precipitation based on measured stations according to claim 1, characterized in that, The specific steps of step S6 are as follows: Step S6.1: Select dates where the daily precipitation at the measured stations is less than or equal to a preset threshold, and sort the satellite precipitation data; Step S6.2: Sort the measured precipitation data from the grid at the stations; Step S6.3: Calculate the quantile position of the precipitation data prediction value output by the deep learning model in the satellite precipitation data distribution, using the following formula: Where M is the number of satellite precipitation data samples, x is a single value of satellite precipitation data, source is the set of satellite precipitation data, value is the predicted value of satellite precipitation data, count(x∈source|x≤value) is the number of precipitation data that is less than the predicted value of precipitation data, and rank is the quantile ranking; By mapping the quantile position of the predicted precipitation data in the satellite precipitation data distribution to the corresponding quantile in the actual station data distribution, the error-corrected precipitation data is obtained. in, y is the inverse cumulative distribution function of the target distribution, and y is the precipitation data after final error correction.

8. The method for two-stage error correction of satellite precipitation based on measured stations according to claim 1, characterized in that, The specific steps of step S7 are as follows: Step S7.1: Based on the measured station grid, read the error-corrected precipitation data corresponding to the measured station grid; Step S7.2: Subtract the corresponding error-corrected precipitation data from the measured station grid precipitation data to obtain the precipitation residual.

9. The method for two-stage error correction of satellite precipitation based on measured stations according to claim 1, characterized in that, The specific steps of step S8 are as follows: Step S8.1: Construct a dataset with precipitation residuals as the dependent variable and error-corrected precipitation data as the independent variable; Step S8.2: Based on the gradient boosting regression tree model, construct a nonlinear residual correction model, set the parameters of the nonlinear residual correction model as follows: tree depth is 5, number of trees is 100, learning rate is 0.1, and divide the dataset into training set and validation set, train the nonlinear residual correction model, and obtain the trained nonlinear residual correction model. Step S8.3: Use the trained nonlinear residual correction model to predict the residuals of the error-corrected precipitation data for each grid point, and obtain the predicted precipitation residuals. The residual prediction formula is as follows; R pred =f GBRT (P corr ) Among them, P corr R represents the predicted precipitation data from steps S5 and S6. pred For the predicted precipitation residual, f GBRT This represents a gradient boosting regression tree.

10. The method for two-stage error correction of satellite precipitation based on measured stations according to claim 1, characterized in that, The method for calculating the residual-corrected precipitation data in step S9 is as follows: Final precipitation = Error-corrected precipitation + R pred In the formula, final precipitation is the precipitation data after residual correction, error-corrected precipitation is the precipitation data after error correction, and R... pred This represents the predicted precipitation residual.

Images