Multi-source rainfall measurement product data fusion method, system, equipment and medium

Abstract

The invention discloses a multi-source rainfall measurement product data fusion method, system and device and a medium, and belongs to the technical field of meteorological data processing, and the method comprises the steps: obtaining meteorological satellite rainfall measurement data, weather radar rainfall measurement data and ground rain gauge rainfall measurement data, and constructing a multi-source rainfall data set; performing preprocessing through the multi-source rainfall data set to obtain satellite rainfall data with a unified coordinate system and resolution, radar rainfall data after noise reduction and rainfall gauge rainfall data after interpolation completion; and in combination with the preprocessed multi-source rainfall data, constructing a time-space consistency data set. According to the method, meteorological satellite, radar and rain gauge data are integrated, a multi-source rainfall data set is constructed, a coordinate system and resolution are unified, noise reduction interpolation ensures time-space consistency, a Kriging interpolation method is adopted to fuse data, resolution and accuracy are improved, an evaluation model is subjected to cross validation, and accurate support is provided for weather forecast and disaster prevention and reduction.

Description

Technical Field

[0001] This invention relates to the field of meteorological data processing technology, specifically to a method, system, equipment, and medium for fusing multi-source precipitation measurement product data. Background Technology

[0002] Multi-source precipitation data fusion technology is widely used in meteorology for weather forecasting, flood warning, and water resource management. By integrating multiple data sources such as ground rain gauges, weather radar, and meteorological satellites, it provides more comprehensive precipitation information. Existing technologies typically employ weighted summation methods based on mathematical statistics, achieving data fusion by assigning weights to different data sources. These weights are often determined based on statistical patterns or historical data.

[0003] However, this method often ignores the characteristics of precipitation data in physical spatiotemporal distribution, leading to discrepancies between the fusion results and actual precipitation conditions, resulting in low data accuracy. Furthermore, ground-based rain gauges, weather radars, and weather satellites all have limitations when used independently: ground-based rain gauges are easily damaged and have limited coverage, making it difficult to provide large-scale precipitation information; weather radar measurements are affected by terrain and clutter, making reflectivity data prone to errors; while weather satellites have wide coverage and long forecast periods, their inversion accuracy is low, especially for solid precipitation, where they have almost no inversion capability. Summary of the Invention

[0004] To address the aforementioned technical issues, a multi-source precipitation measurement product data fusion method is proposed. This method includes: acquiring meteorological satellite precipitation measurement data, weather radar precipitation measurement data, and ground rain gauge precipitation measurement data to construct a multi-source precipitation dataset; preprocessing the multi-source precipitation dataset to obtain satellite precipitation data with a unified coordinate system and resolution, denoised radar precipitation data, and interpolated and completed rain gauge precipitation data; combining the preprocessed multi-source precipitation data to construct a spatiotemporally consistent data set; using the spatiotemporally consistent data set, employing surface-to-point co-kriging interpolation to obtain fused precipitation data; and using the fused precipitation data, constructing a fusion model evaluation system using cross-validation to evaluate the precipitation data fusion results.

[0005] As a preferred embodiment of the multi-source precipitation measurement product data fusion method described in this invention, the method includes: preprocessing the multi-source precipitation dataset to obtain satellite precipitation data with a unified coordinate system and resolution, including: acquiring meteorological satellite precipitation measurement data and auxiliary variable data to construct a multi-channel input dataset; using the multi-channel input dataset, extracting multi-scale main variable and auxiliary variable features using a convolutional neural network to obtain high-resolution satellite precipitation data; and based on the high-resolution satellite precipitation data, projecting it onto a ground plane coordinate system using satellite orbit parameters and a positioning algorithm to obtain satellite precipitation data consistent with the coordinate system of the ground rain gauge.

[0006] As a preferred embodiment of the multi-source precipitation measurement product data fusion method described in this invention, the denoised radar precipitation data includes: constructing a dual-polarization parametric dataset based on the horizontal and vertical channel echo power of weather radar precipitation measurement data; using the dual-polarization parametric dataset, employing a dual-polarization parametric filtering method to determine echo correlation and remove noise sources to obtain denoised radar precipitation data; and based on the denoised radar precipitation data, converting it to a ground plane coordinate system through radar coordinate position and beam scanning relationship to obtain radar precipitation data consistent with the coordinate system of the ground rain gauge.

[0007] As a preferred embodiment of the multi-source precipitation measurement product data fusion method described in this invention, the interpolated and completed rain gauge precipitation data includes: constructing an initial rain gauge dataset based on ground rain gauge precipitation measurement data; performing integrity, consistency, and representativeness checks and standardization on the initial rain gauge dataset to obtain a preprocessed rain gauge dataset; and using terrain-coordinated Kriging interpolation to complete missing data based on the preprocessed rain gauge dataset and regional elevation data to obtain the interpolated and completed ground rain gauge precipitation data.

[0008] The beneficial effects of this preferred technical solution are as follows: By constructing an initial rain gauge dataset and performing integrity, consistency, and representativeness checks and standardization, data noise and outliers can be effectively eliminated, ensuring a reliable foundation for the preprocessed rain gauge dataset. This allows for accurate capture of the heterogeneous influence of terrain on precipitation distribution when using terrain-coordinated kriging interpolation to complete missing data in conjunction with regional elevation data. It avoids spatial bias caused by traditional interpolation methods ignoring terrain factors, significantly improving the spatial continuity and accuracy of ground rain gauge precipitation data after interpolation completion in mountainous or complex terrain scenarios, and providing a high-quality input foundation for the fusion of multi-source precipitation measurement product data.

[0009] As a preferred embodiment of the multi-source precipitation measurement product data fusion method described in this invention, the method involves: constructing a spatiotemporally consistent data set by combining the preprocessed multi-source precipitation data, including: constructing a multi-source preprocessed dataset based on preprocessed meteorological satellite precipitation data, weather radar precipitation data, and ground rain gauge precipitation data; performing pairwise correlation analysis on the multi-source preprocessed dataset and selecting spatiotemporally synchronized data that meet the correlation conditions to obtain a spatiotemporally synchronized dataset; and performing logarithmic processing on the spatiotemporally synchronized dataset and dividing the target area into gridded grids of equal area to construct a spatiotemporally consistent data set.

[0010] As a preferred embodiment of the multi-source precipitation measurement product data fusion method described in this invention, the fused precipitation data is obtained by using a surface-to-point co-kriging interpolation method based on the spatiotemporally consistent data set. This includes: constructing a raster-format dataset based on meteorological satellite precipitation data and weather radar precipitation data in the spatiotemporally consistent data set; converting the raster-format dataset into discrete-point format data that is spatiotemporally synchronized with ground rain gauge precipitation data to obtain a synchronized discrete-point dataset; and using ground rain gauge precipitation data as the main variable, constructing a co-kriging weight function through a variogram and covariance to obtain the fused precipitation data.

[0011] As a preferred embodiment of the multi-source precipitation measurement product data fusion method described in this invention, the method includes: constructing a fusion model evaluation system using the fused precipitation data and employing cross-validation to evaluate the precipitation data fusion results; constructing a cross-validation dataset by fusing precipitation data and some ground rain gauge precipitation data; using the cross-validation dataset, setting some ground rain gauge precipitation data as unknown, and predicting precipitation values ​​using co-kriging interpolation to obtain multi-source fused prediction values; based on the multi-source fused prediction values, predicting precipitation values ​​using single-source ground rain gauge, weather radar, and meteorological satellite precipitation data respectively using single-source kriging interpolation to obtain single-source prediction values; and combining the multi-source fused prediction values ​​and single-source prediction values, calculating and comparing the root mean square error with the measured values ​​to achieve an accuracy evaluation of the fused precipitation data.

[0012] The beneficial effects of this preferred technical solution are as follows: By fusing precipitation data and using cross-validation to construct a fusion model evaluation system, in a scenario simulating the absence of some ground rain gauge data, multi-source fusion prediction values ​​are generated using co-kriging interpolation and compared with single-source kriging prediction values ​​to calculate the root mean square fusion error. This quantifies the superiority of the method of fusing multi-source precipitation data and verifies its robustness and accuracy in complex or sparsely populated regions. It also improves the effectiveness of constructing the evaluation system and can effectively integrate information from multiple data sources such as ground rain gauges, weather radar, and meteorological satellites, thereby improving the spatial coverage and accuracy of precipitation data.

[0013] As a preferred embodiment of the multi-source precipitation measurement product data fusion system described in this invention, it is characterized by comprising: a data acquisition module for acquiring meteorological satellite precipitation measurement data, weather radar precipitation measurement data, and ground rain gauge precipitation measurement data to construct a multi-source precipitation dataset; a data preprocessing module for preprocessing the multi-source precipitation dataset to obtain satellite precipitation data with a unified coordinate system and resolution, noise-reduced radar precipitation data, and interpolated and completed rain gauge precipitation data; a data analysis module for constructing a spatiotemporally consistent data set by combining the preprocessed multi-source precipitation data; a data fusion module for obtaining fused precipitation data based on the spatiotemporally consistent data set using surface-to-point co-kriging interpolation; and a model evaluation module for constructing a fusion model evaluation system using cross-validation based on the fused precipitation data to obtain the accuracy analysis results of the fused precipitation data.

[0014] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the multi-source precipitation measurement product data fusion method.

[0015] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the multi-source precipitation measurement product data fusion method.

[0016] The beneficial effects of this invention are as follows: By integrating precipitation measurement data from meteorological satellites, weather radars, and ground rain gauges, a multi-source precipitation dataset is constructed, effectively improving data coverage and diversity; by unifying the coordinate system, resolution, and performing noise reduction and interpolation completion processing, the consistency of multi-source data in terms of spatiotemporal scale is ensured, laying the foundation for high-precision fusion; the use of surface-to-point co-kriging interpolation for data fusion fully utilizes the complementary advantages of different data sources, improving the spatial resolution and accuracy of precipitation data; and the fusion model evaluation system constructed through cross-validation can objectively assess the reliability of the fusion results, thereby providing more accurate precipitation data support for applications such as weather forecasting and disaster prevention and mitigation. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a method for fusing multi-source precipitation measurement product data according to an embodiment of the present invention. Detailed Implementation

[0019] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0020] Example 1, referring to Figure 1 This is the first embodiment of the present invention, which provides a method for fusing multi-source precipitation measurement product data, including:

[0021] S1: Acquire precipitation measurement data from meteorological satellites, weather radars, and ground rain gauges to construct a multi-source precipitation dataset;

[0022] S2: Preprocess the multi-source precipitation dataset to obtain satellite precipitation data with unified coordinate system and resolution, noise-reduced radar precipitation data, and interpolated and completed rain gauge precipitation data;

[0023] S3: Combine the preprocessed multi-source precipitation data to construct a spatiotemporally consistent data set;

[0024] S4: Based on the aforementioned spatiotemporal consistent data set, the surface-to-point co-kriging interpolation method is used to obtain fused precipitation data;

[0025] S5: Using the fused precipitation data, a fusion model evaluation system is constructed using the cross-validation method to evaluate the precipitation data fusion results.

[0026] It should be noted that existing multi-source heterogeneous precipitation measurement data fusion methods generally obtain fused precipitation data by assigning weights to different precipitation data measurement sources. However, the methods for obtaining these weights are mostly based on mathematical statistics, often neglecting the impact of the physical and spatiotemporal distribution of precipitation measurement sources on the fusion results. This leads to fusion results that do not conform to actual physical conditions and have low data fusion accuracy. At the same time, existing precipitation measurement products are mainly concentrated in three types: ground rain gauges, weather radars, and meteorological satellites. Each of these products has limitations to varying degrees when used independently. For example, ground rain gauges are easily damaged and have a small measurement range; the reflectivity of the measurement band of weather radar is easily affected by terrain and other signals; and although meteorological satellites have a wide spatial range and a long forecast period, their inversion accuracy is low and they have virtually no inversion capability for solid precipitation. Existing multi-source precipitation data fusion methods are generally based on weighted summation methods using mathematical statistics. The results obtained by this method often ignore the physical and spatiotemporal distribution characteristics of the precipitation measurement sources, resulting in large deviations in the fusion results.

[0027] Therefore, to address the aforementioned problems, this invention, through steps S1-S5, first acquires meteorological satellite precipitation measurement data, weather radar precipitation measurement data, and ground rain gauge precipitation measurement data to construct a multi-source precipitation dataset, achieving preliminary integration of multi-source data. Then, the multi-source precipitation dataset is preprocessed to obtain satellite precipitation data with a unified coordinate system and resolution, noise-reduced radar precipitation data, and interpolated rain gauge precipitation data, eliminating data heterogeneity and noise interference to ensure consistent data quality. Next, a spatiotemporally consistent data set is constructed using the preprocessed multi-source precipitation data to enhance the spatiotemporal synchronization and correlation of the data. Subsequently, based on the spatiotemporally consistent data set, a surface-to-point co-kriging interpolation method is used to obtain fused precipitation data, fully considering the physical spatiotemporal distribution characteristics and improving fusion accuracy. Finally, a fusion model evaluation system is constructed using the fused precipitation data and a cross-validation method to evaluate the precipitation data fusion results, thereby verifying the model's effectiveness and leveraging the advantages of multi-source data to achieve high-precision, high-reliability precipitation data fusion, particularly suitable for precipitation measurement applications under complex terrain and variable weather conditions.

[0028] Example 2, refer to Figure 1 This is the second embodiment of the present invention, which differs from the first embodiment in that: a method for fusing multi-source precipitation measurement product data further includes:

[0029] In this embodiment, step S1 involves acquiring meteorological satellite precipitation measurement data, weather radar precipitation measurement data, and ground rain gauge precipitation measurement data. Constructing a multi-source precipitation dataset can be achieved by collecting satellite precipitation measurement product data, radar precipitation measurement product data, and ground rain gauge precipitation measurement product data. Specifically, this process aims to integrate precipitation measurement information from different sources, such as downloading satellite synchronous data from the National Meteorological Satellite Center website and radar and ground rain gauge synchronous data from the China National Meteorological Data Network. This ensures that the data covers the spatiotemporal range of the study area and avoids the limitations of single sources, such as the small measurement range of ground rain gauges or the low accuracy of satellite inversion, thus providing a comprehensive foundation for subsequent fusion.

[0030] In an alternative implementation, the construction of the multi-source precipitation dataset in step S1 can also be improved by removing outliers or inconsistent records through data integrity checks and preliminary screening, such as by performing a preliminary resolution assessment on satellite data.

[0031] In another alternative implementation, the construction of the multi-source precipitation dataset in step S1 can also be achieved by time series alignment to match the collection periods of data from different sources. For example, the wide-area coverage of satellites can be combined with the local high precision of radar to form a preliminary spatiotemporal synchronized set.

[0032] Furthermore, in step S2, the multi-source precipitation dataset is preprocessed to obtain satellite precipitation data with a unified coordinate system and resolution, denoised radar precipitation data, and interpolated and completed rain gauge precipitation data, including the following steps A1-A9:

[0033] A1: Obtain meteorological satellite precipitation measurement data and auxiliary variable data, and construct a multi-channel input dataset;

[0034] A2: Using the multi-channel input dataset, a convolutional neural network is used to extract multi-scale main and auxiliary variable features to obtain high-resolution satellite precipitation data;

[0035] A3: Based on the high-resolution satellite precipitation data, satellite precipitation data is projected onto the ground plane coordinate system through satellite orbit parameters and positioning algorithms to obtain satellite precipitation data consistent with the coordinate system of the ground rain gauge.

[0036] A4: Construct a dual polarization parameter dataset based on the horizontal and vertical channel echo power of weather radar precipitation measurement data;

[0037] A5: Using the aforementioned dual-polarization parametric dataset, the dual-polarization parametric filtering method is employed to determine echo correlation and eliminate noise sources, thereby obtaining denoised radar precipitation data.

[0038] A6: Based on the noise-reduced radar precipitation data, the radar coordinate position and beam scanning relationship are converted into a ground plane coordinate system to obtain radar precipitation data consistent with the coordinate system of the ground rain gauge.

[0039] A7: Construct an initial rain gauge dataset based on precipitation measurement data from ground rain gauges;

[0040] A8: The initial rain gauge dataset is checked for completeness, consistency, and representativeness, and then standardized to obtain a preprocessed rain gauge dataset.

[0041] A9: Based on the preprocessed rain gauge dataset, the missing data is supplemented by terrain-coordinated Kriging interpolation using regional elevation data to obtain the interpolated and supplemented ground rain gauge precipitation data.

[0042] In this embodiment of the application, step A2, to improve the resolution of satellite precipitation data, is achieved by extracting multi-scale main and auxiliary variable features through a convolutional neural network to obtain high-resolution satellite precipitation data, including the following steps A211-A213:

[0043] A211: Meteorological satellite precipitation measurement data and auxiliary variable data are used as multi-channel inputs to construct an input dataset. The auxiliary variable data includes digital elevation model (DEM, which refers to digital data reflecting terrain height), slope, aspect, infrared brightness temperature, and vegetation index data. These data are in high-resolution format and are used to enhance the spatial feature extraction of precipitation data.

[0044] A212: A convolutional neural network (CNN) is used to process multi-channel input datasets. The network structure consists of an encoder, a decoder, and skip connections. The encoder progressively extracts multi-scale main and auxiliary variable features from the input data through convolutional and pooling layers. The decoder decodes low-resolution features into high-resolution features through upsampling and convolutional layers. Skip connections pass low-level details from the encoder to the decoder, ensuring the output retains both high-level semantics and specific details. For example, when processing satellite precipitation data for a certain region, the data is decomposed into small 128×128 pixel image patches, divided into training, validation, and test sets. Simultaneous ground-based rain gauge and radar precipitation measurements are used as high-resolution references to verify the output accuracy of the convolutional neural network.

[0045] A213: The trained convolutional neural network outputs high-resolution satellite precipitation data and feeds the results back to the model for weight updates, thereby improving the accuracy and stability of the model when processing large-scale precipitation data fields.

[0046] In an optional embodiment, to improve the resolution of satellite precipitation data, another approach is to unify the resolution using an arithmetic mean interpolation method, including the following steps A221-A223:

[0047] A221: Divide meteorological satellite precipitation measurement data into multiple regions, and calculate the arithmetic mean of precipitation data in each region as the representative value of that region.

[0048] A222: The arithmetic mean interpolation method is used to unify the resolution of different satellite data to the target resolution range. For example, for satellite data with inconsistent resolution in a certain area, a precipitation data field with uniform resolution is generated by calculating the average value of adjacent pixels.

[0049] A223: Compare the processed data with ground rain gauge precipitation measurements to verify the accuracy of the interpolation results. This method has low computational complexity and is suitable for quickly processing small-scale precipitation data, but its accuracy may not be as good as convolutional neural network methods for complex terrain or areas with heavy precipitation.

[0050] In another alternative embodiment, to improve the resolution of satellite precipitation data, the method may also be to unify the resolution through distance-weighted interpolation, including the following steps A231-A234:

[0051] A231: Based on the spatial distribution of precipitation measurement data from meteorological satellites, determine the distance between each data point and the target point.

[0052] A232: The distance-weighted interpolation method (a spatial distance-based interpolation method where points closer to each other have higher weights) is used to assign weights to each data point, with the weights being inversely proportional to the distance.

[0053] A233: Calculate high-resolution precipitation data for a target point using a weighted average. For example, when processing satellite data for a region, select multiple data points close to the target point and calculate their weighted average as the precipitation value for the target point.

[0054] A234: Compare the interpolation results with ground rain gauge and radar data to verify the interpolation accuracy. This method is highly dependent on spatial distribution and is suitable for uniformly distributed data fields, but may have biases in areas with drastic precipitation variations.

[0055] It should be noted that convolutional neural network (CNN) methods extract multi-scale features through deep learning, making them suitable for handling complex terrain and heavy precipitation scenarios with high accuracy but high computational complexity. Arithmetic mean interpolation and distance-weighted interpolation methods are computationally simple and suitable for fast processing but have lower accuracy. The choice of a specific method requires a trade-off between data scale, computational resources, and accuracy requirements. Employing CNN methods can significantly improve the resolution of satellite precipitation data. Furthermore, validation using ground-based rain gauge and radar data ensures that the results conform to actual physical conditions, providing high-quality input data for subsequent multi-source data fusion.

[0056] Furthermore, in step S3, a spatiotemporally consistent data set is constructed by combining the preprocessed multi-source precipitation data, including the following steps B1-B3:

[0057] B1: Construct a multi-source preprocessed dataset based on preprocessed meteorological satellite precipitation data, weather radar precipitation data, and ground rain gauge precipitation data;

[0058] B2: Using the multi-source preprocessed dataset, perform pairwise correlation analysis and select spatiotemporal synchronized data that meet the correlation conditions to obtain a spatiotemporal synchronized dataset.

[0059] B3: Based on the aforementioned spatiotemporal synchronization dataset, logarithmic processing is performed and the target region is divided into gridded grids of equal area to construct a spatiotemporally consistent data set.

[0060] In this embodiment of the application, step B3 involves dividing the target area into a grid of equal areas. The division method is based on the uniform distribution of the area where the rain gauge is located, and includes the following steps B211-B213:

[0061] B211: First, determine the overall boundary of the target area, and calculate the total area of ​​the area and the coverage of the rain gauge points based on the spatial distribution of the rain gauges, so as to ensure that the division process covers all relevant measurement points.

[0062] B212: Divide the target area into multiple rectangular or square grid units of equal area. The size of each grid unit is obtained by dividing the total area of ​​the area by the preset number of grid units, and the boundaries are adjusted to avoid overlap or omission.

[0063] For example, in a research area covering hundreds of square kilometers, the system divides the area into 100 grid cells of equal area, each cell representing an equal spatial extent, for subsequent data allocation and fusion calculations.

[0064] B213: Each grid cell after division is associated with precipitation measurement points in the spatiotemporal synchronization dataset to form a structured grid data framework, which facilitates subsequent multi-source data fusion operations.

[0065] In an optional embodiment, for dividing the target area into gridded grids of equal area, the division method can also be adaptive adjustment based on terrain features, including the following steps B221-B223:

[0066] B221: First, analyze the topographic elevation and slope distribution of the target area to identify areas with complex terrain, such as mountainous or riverine areas. These areas may require a more detailed grid to capture precipitation variability.

[0067] B222: Dynamically adjust the grid size according to the complexity of the terrain. Use larger grid cells in flat areas, while reducing the grid area in areas with variable terrain to maintain the level of detail while adhering to the principle of equal area.

[0068] B223: The final generated adaptive grid set is matched with the spatiotemporally synchronized dataset to ensure that the fusion process can better reflect the impact of terrain on precipitation distribution. Compared with uniform distribution, this method improves the accuracy of complex terrain but increases the computational complexity.

[0069] In another optional embodiment, for dividing the target region into a grid of equal areas, the division method can also be dynamic division based on an event-triggered mechanism, including the following steps B231-B234:

[0070] B231: Pre-set trigger conditions, such as initiating partitioning adjustments when a heavy precipitation event is detected or when data density changes exceed a threshold.

[0071] B232: Real-time monitoring of precipitation intensity and distribution uniformity in spatiotemporal synchronized datasets; once trigger conditions are met, immediately recalculate the regional boundary and start the grid generation program.

[0072] B233: Based on the type of the triggered event, selectively refine the mesh division of a specific sub-region while maintaining the principle of equal overall area, thus avoiding the computational overhead caused by global re-division.

[0073] B234: Store the dynamically generated mesh sequence and associate it with the dataset to form a responsive mesh framework. Compared with adaptive mesh adjustment, this method focuses more on real-time event response but may lead to mesh instability and requires additional verification steps.

[0074] It should be noted that by constructing a spatiotemporally consistent dataset, we can ensure the spatiotemporal alignment and standardized representation of multi-source precipitation data, avoid fusion deviations caused by spatiotemporal inconsistencies, support the accurate application of subsequent Kriging interpolation, and improve the reliability and practicality of the overall data fusion model.

[0075] Furthermore, in step S4, based on the spatiotemporally consistent data set, a surface-to-point co-kriging interpolation method is used to obtain fused precipitation data, including the following steps C1-C3:

[0076] C1: Construct a raster-formatted dataset based on meteorological satellite precipitation data and weather radar precipitation data in the aforementioned spatiotemporal consistent data set;

[0077] C2: The raster format dataset is converted into discrete point format data that is spatiotemporally synchronized with the ground rain gauge precipitation data to obtain a synchronized discrete point dataset;

[0078] C3: Based on the aforementioned synchronous discrete point dataset, ground rain gauge precipitation data are used as the main variable, and a co-kriging weighting function is constructed using the variogram and covariance to obtain the fused precipitation data.

[0079] In this embodiment of the application, step C3, concerning the construction of the co-kriging weight function, is constructed using the standard co-kriging interpolation method with ground rain gauge data as the principal variable, including the following steps C311-C313:

[0080] C311: Using ground-based rain gauge precipitation data as the main variable, and selecting satellite precipitation data and radar precipitation data as auxiliary variables, a variogram and covariance model are constructed to determine the spatial correlation between the various data sources.

[0081] C312: Based on a synchronous discrete point dataset, several random points are generated within each grid area. The average precipitation data within the grid area is assigned to these random points to ensure that the random points are aligned with the ground rain gauge data in time and space, forming a unified analysis dataset.

[0082] C313: The spatial distance weights between data points are calculated using the variogram function. Combined with the covariance model, a kriging weight function is constructed to calculate the fused precipitation value of the target point and generate a fused precipitation data field.

[0083] For example, in a precipitation event, the system processes an area containing 100 ground rain gauge points, 1000 radar data points, and 1000 satellite data points. Through variogram analysis, it determines that the weight of the ground rain gauge points to the target point is higher than that of the radar and satellite data, and generates a fused precipitation data field for regional precipitation assessment.

[0084] In an optional embodiment, the co-kriging weight function can also be constructed using a dynamic weighted co-kriging interpolation method based on precipitation intensity and topographic features, including the following steps C321-C323:

[0085] C321: Using ground-based rain gauge precipitation data as the main variable, and adjusting the weights of satellite and radar data in conjunction with precipitation intensity and topographic features (such as elevation and slope), a dynamic variogram model is constructed to enhance the focus on areas of heavy precipitation.

[0086] C322: Based on a synchronous discrete point dataset, precipitation data within a raster area is classified according to precipitation intensity. Random point density is increased in areas of heavy precipitation, and higher weights are assigned to ensure that the fusion result more accurately reflects the distribution of heavy precipitation.

[0087] C323: By using dynamically adjusted variograms and covariance models, it calculates the fused precipitation values ​​of target points, generates a fused precipitation data field, and optimizes its adaptability to complex terrain and heavy precipitation scenarios.

[0088] For example, in mountainous areas with heavy rainfall, the system adjusts the radar data weights based on elevation data, increases the density of random points in areas of heavy rainfall, and generates a more accurate fused rainfall data field for flood early warning analysis.

[0089] In another alternative embodiment, the co-kriging weight function can be constructed using a spatiotemporally adaptive co-kriging interpolation method, including the following steps C331-C334:

[0090] C331: Using ground rain gauge precipitation data as the main variable, analyze the spatiotemporal characteristics of precipitation events, determine the type of variogram (such as spherical model or exponential model), and adaptively select model parameters according to the precipitation type (such as heavy precipitation or light precipitation).

[0091] C332: Based on a synchronized discrete point dataset, it dynamically adjusts the generation strategy of random points within a grid area, prioritizing random points aligned with the peak of precipitation events in the time dimension to improve time synchronization accuracy.

[0092] C333: By using an adaptive variogram and covariance model, the spatial and temporal weights of each data point to the target point are calculated, and a spatiotemporal co-kriging weight function is constructed.

[0093] C334: Based on the weighting function, calculate the fused precipitation value of the target point, generate a fused precipitation data field, and optimize the ability to capture rapidly changing precipitation events.

[0094] For example, in a rapidly moving rainstorm event, the system adjusts the variogram parameters based on the precipitation time characteristics, prioritizes random points aligned with the peak of the rainstorm, and generates a high temporal resolution fused precipitation data field for real-time precipitation monitoring.

[0095] It should be noted that by using the area-to-point co-kriging interpolation method for multi-source precipitation data fusion, the physical and spatiotemporal distribution characteristics of precipitation data sources can be fully considered. Compared with the traditional weighted summation method based on mathematical statistics, the fusion result is more consistent with the actual precipitation distribution, which improves the accuracy of data fusion and adaptability to complex precipitation scenarios, and helps to improve the accuracy of precipitation monitoring and forecasting.

[0096] Furthermore, in step S5, using the fused precipitation data, a fusion model evaluation system is constructed using cross-validation to evaluate the precipitation data fusion results, including the following steps E1-E4:

[0097] E1: Integrate precipitation data and some ground rain gauge precipitation data to construct a cross-validation dataset;

[0098] E2: Using the cross-validation dataset, some ground rain gauge precipitation data are set to unknown, and co-kriging interpolation is used to predict precipitation values ​​to obtain multi-source fusion prediction values;

[0099] E3: Based on the multi-source fusion prediction value, precipitation values ​​are predicted by single-source kriging interpolation using precipitation data from ground rain gauges, weather radars, and meteorological satellites, respectively, to obtain single-source prediction values;

[0100] E4: Combine the multi-source fusion prediction values ​​and single-source prediction values, calculate and compare the root mean square error with the measured values, and evaluate the accuracy of the precipitation data fusion results.

[0101] In this embodiment of the application, in step E2, for the predicted precipitation value, the method of co-kriging interpolation (co-kriging numerical fusion model) is adopted by setting some ground rain gauge precipitation data (ground rain gauge precipitation measurement data) as unknown and substituting it into the model, including the following steps E211-E213:

[0102] E211: Select a certain point as the verification point, temporarily set the measured value of the precipitation data from the ground rain gauge at that point as an unknown value, and retain all other precipitation data from ground rain gauges, precipitation measurement data from weather radar, and precipitation measurement data from meteorological satellites as known inputs.

[0103] E212: Substitute the known input data into the pre-constructed co-kriging interpolation method, and use the model's variogram and covariance calculation mechanism to perform spatial interpolation prediction of precipitation values ​​at unknown locations, forming multi-source fusion prediction values.

[0104] For example, in a typical study area, a ground rain gauge point located at the center of the grid is selected as a validation point. The measured precipitation value at this point is unknown. Other surrounding points and radar and satellite data are fully input into the model. The model generates the predicted precipitation value for this point based on this data, which is then used for subsequent error assessment.

[0105] E213: The generated predicted precipitation values ​​are initially compared with the actual recovered measured values ​​as the basis for cross-validation, ensuring that the prediction results conform to the physical characteristics in terms of spatiotemporal distribution.

[0106] In an optional embodiment, for predicted precipitation values, the co-kriging interpolation method can also be a time-series-based rolling verification path, including the following steps E221-E223:

[0107] E221: The cross-validation dataset is split into continuous sequences in chronological order. Ground rain gauge precipitation data for a specific time period in the sequence are set to unknown, while other data from the same period are retained as input to highlight the dynamic consistency over time.

[0108] E222: The model makes hourly predictions of precipitation values ​​for unknown time periods, and uses the covariance function to capture time correlations, forming a rolling multi-source fusion prediction value sequence.

[0109] E223: By comparing the deviations between the rolling forecast sequence and the measured sequence, this path is highlighted for its continuity advantage in handling the evolution of precipitation events, complementing the static site verification.

[0110] In another alternative embodiment, for predicted precipitation values, the co-kriging interpolation method can also be used as a batch verification path based on spatial grids, including the following steps E231-E234:

[0111] E231: Divide the entire study area into multiple sub-blocks, randomly select multiple ground rain gauge locations within each sub-block, set the precipitation data of these locations to unknown in batches, and retain the data outside the sub-block as input to cover spatial heterogeneity.

[0112] E232: Simultaneously run model predictions for unknown locations in each sub-block, and integrate radar and satellite grid information using a weighting function to form batch multi-source fusion prediction values.

[0113] E233: Calculate the spatial distribution error between predicted and measured values ​​within a sub-block and identify local deviation areas.

[0114] E234: Summarize the results of all sub-blocks to generate an overall spatial evaluation map, and compare it with the single-point verification path to highlight the difference in the applicability of the path in large-scale areas.

[0115] It should be noted that by using cross-validation to evaluate the multi-source fusion model, the accuracy advantage of the fusion results can be fully quantified, the limitations of single-source data influencing decision-making can be avoided, and the reliable application of precipitation forecasting in the field of hydropower automation can be supported.

[0116] In summary, S1: Acquire precipitation measurement data from meteorological satellites, weather radars, and ground rain gauges to construct a multi-source precipitation dataset. These data come from different sensors and have different spatial resolutions and temporal frequencies, providing diverse inputs for subsequent processing.

[0117] S2: Preprocessing the multi-source precipitation dataset yields satellite precipitation data with a unified coordinate system and resolution, denoised radar precipitation data, and interpolated rain gauge precipitation data. For satellite precipitation data, it is first projected onto a ground plane coordinate system, and the resolution is unified using arithmetic mean or distance mean interpolation methods. To further improve the resolution of satellite data, a convolutional neural network (CNN) is used for processing. The CNN takes satellite precipitation data and auxiliary variable data (such as high-resolution digital elevation models, slope, aspect, infrared brightness temperature, and vegetation index) as multi-channel inputs, decomposing the large-scale precipitation data field into small image patches of 128×128 pixels, which are then divided into training, validation, and test sets.

[0118] Simultaneous ground radar data and rain gauge data were used as high-resolution inputs to validate the accuracy of the convolutional neural network and update the model weights. The convolutional neural network consists of an encoder, a decoder, and skip connections. The encoder extracts multi-scale main and auxiliary variable features through convolutional layers, pooling functions, and pooling layers. The decoder decodes low-resolution features into high-resolution features through upsampling layers, convolutional layers, and activation functions. Skip connections transmit low-level details through channel concatenation.

[0119] For radar data, a dual-polarization parametric filtering method is used for noise reduction. Noise is eliminated by analyzing the correlation between the horizontal and vertical channel echo power (with a correlation threshold set to 0.85), and the echo power difference is used as an auxiliary variable. For ground rain gauge data, after checking the completeness, consistency, and representativeness of the data, a terrain-coordinated kriging interpolation method is used to complete the missing data. The formula for interpolating rainfall data for target points is as follows:

[0120]

[0121] in, The rainfall was measured using a rain gauge at the target point. The weighting of the actual measurements from the rain gauge; This refers to the actual precipitation measured by a rain gauge. Weights for regional elevation data; These are measured values ​​of regional elevation.

[0122] S3: Combining the preprocessed multi-source precipitation data, a spatiotemporally consistent dataset is constructed. Data with correlation coefficients greater than 0.5 are selected through pairwise correlation analysis, and their temporal characteristics are standardized and logarithmically converted. The area where the rain gauges are located is divided into a grid of equal areas to ensure spatiotemporal synchronization of the multi-source data.

[0123] S4: Based on the aforementioned spatiotemporally consistent dataset, a surface-to-point co-kriging interpolation method is used to obtain fused precipitation data. Specifically, random points are generated within each polygonal region, the average data value within the region is assigned to each random point, and points that are spatially synchronized with rainfall data are selected for analysis.

[0124] The variability, covariance, and kriging weighting function are constructed as follows:

[0125]

[0126]

[0127]

[0128] in, Represents the theoretical variability function; Indicates distance as Point-to-point number function; Indicates the value of a known point; This indicates the distance from the known point is... The point values; This represents the estimated value of the target point; Indicates distance as Covariance of points; The sill value is the variogram model. This indicates the weighting of the rain gauges; Assign weights to the radar; Assign weights to the satellites; n, m, and z represent the rain gauge precipitation measurement data points, radar precipitation measurement data blocks, and satellite precipitation measurement data blocks, respectively. , , These are precipitation measurement data from rain gauges, radar, and satellite, respectively.

[0129] Construct constraints:

[0130]

[0131]

[0132]

[0133] in The standard deviation of the error; The variance at the target point.

[0134] A system of co-Kriging equations is constructed by taking partial derivatives of the Lagrange operator and the weighting variables:

[0135]

[0136]

[0137]

[0138] in, For Lagrange operators; The objective function is... , , These are the components of the Lagrange operator; Represents a known point With the specified point Covariance between them; Indicates a specified point The covariance between the unknown point x and the unknown point x.

[0139] Obtain the fused data value:

[0140]

[0141] S5: Using the fused precipitation data, a cross-validation method is employed to construct an evaluation system for the fusion model and assess the fusion results. The measured data from a specific rain gauge point is set as unknown. Precipitation at that point is predicted using other rain gauge data, radar data, and satellite data. The root mean square error (RMSE) between the observed and estimated values ​​is calculated. The RMSE of the fused data is compared with that of a single data source (rain gauge, radar, satellite) to evaluate the performance of the fusion model.

[0142] Example 3 is the third embodiment of the present invention. It differs from the previous two embodiments in that: the multi-source precipitation measurement product data fusion method further includes, in order to verify and explain the technical effect adopted in this method, comparing the experimental results with scientific demonstration methods to verify the real effect of this method.

[0143] This method encompasses four steps: data acquisition, preprocessing, fusion, and effectiveness evaluation. In the data acquisition phase, satellite precipitation measurement data, radar precipitation measurement data, and ground rain gauge precipitation data are obtained by logging into the National Meteorological Satellite Center and the China Meteorological Data Network, ensuring the diversity and reliability of data sources. In the data preprocessing phase, optimization is performed based on the characteristics of different data sources: satellite data undergoes resolution enhancement through convolutional neural networks, and the coordinate system is unified by combining auxiliary variables such as elevation and slope; radar data undergoes noise removal through echo power correlation analysis, with a correlation threshold of 0.85 set to ensure data quality; ground rain gauge data uses multi-factor co-kriging interpolation to fill in missing values, and regional elevation data is combined to improve interpolation accuracy. In the data fusion phase, data is screened through correlation analysis, a standard of correlation coefficient greater than 0.5 is set, and various data formats are unified into discrete points. Co-kriging is used to achieve spatiotemporal synchronous fusion processing. In the effectiveness evaluation phase, the prediction accuracy of multi-source fused data is compared with that of single-source data using root mean square error to verify the effectiveness of the method.

[0144] To scientifically verify the effectiveness of this method, the following comparative experiment was designed. Assuming a typical watershed region, 10 ground rain gauge locations were selected, and synchronous satellite, radar, and ground rain gauge precipitation data were collected over a certain time period. In the experiment, precipitation data from one random ground rain gauge location was set as an unknown value. A multi-source fusion model was used to predict the precipitation at this location. Simultaneously, predictions were performed using single data sources (satellite only, radar only, and ground rain gauge only) through single-source Kriging models. The experiment ensured that data selection met the criterion of a correlation coefficient greater than 0.5, and data consistency was guaranteed through logarithmic processing and spatiotemporal synchronization. The complementarity of multi-source data and the refinement of preprocessing were comprehensively considered.

[0145] Example 4, the fourth embodiment of the present invention, differs from the previous three embodiments in that it provides a multi-source precipitation measurement product data fusion system, comprising: a data acquisition module for acquiring meteorological satellite precipitation measurement data, weather radar precipitation measurement data, and ground rain gauge precipitation measurement data to construct a multi-source precipitation dataset; a data preprocessing module for preprocessing the multi-source precipitation dataset to obtain satellite precipitation data with a unified coordinate system and resolution, denoised radar precipitation data, and interpolated and completed rain gauge precipitation data; a data analysis module for combining the preprocessed multi-source precipitation data to construct a spatiotemporally consistent data set; a data fusion module for obtaining fused precipitation data based on the spatiotemporally consistent data set using surface-to-point co-kriging interpolation; and a model evaluation module for constructing a fusion model evaluation system using cross-validation to evaluate the precipitation data fusion results.

[0146] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, 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 invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0147] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0148] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0149] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented in combination with any of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0150] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for fusing multi-source precipitation measurement product data, characterized in that: include, Acquire precipitation measurement data from meteorological satellites, weather radars, and ground rain gauges to construct a multi-source precipitation dataset; By preprocessing the multi-source precipitation dataset, satellite precipitation data with a unified coordinate system and resolution, noise-reduced radar precipitation data, and rain gauge precipitation data after interpolation and completion are obtained. By combining the preprocessed multi-source precipitation data, a spatiotemporally consistent data set is constructed. Based on the aforementioned spatiotemporal consistent data set, a surface-to-point co-kriging interpolation method is used to obtain fused precipitation data; Using the fused precipitation data, a fusion model evaluation system was constructed using cross-validation to evaluate the precipitation data fusion results.

2. The method for data fusion of multi-source precipitation measurement products as described in claim 1, characterized in that: Preprocessing the multi-source precipitation dataset yields satellite precipitation data with a unified coordinate system and resolution, including: Acquire meteorological satellite precipitation measurement data and auxiliary variable data, and construct a multi-channel input dataset; Using the multi-channel input dataset, a convolutional neural network is employed to extract multi-scale main and auxiliary variable features to obtain high-resolution satellite precipitation data. Based on the high-resolution satellite precipitation data, satellite precipitation data is projected onto the ground plane coordinate system through satellite orbit parameters and positioning algorithms to obtain satellite precipitation data consistent with the coordinate system of the ground rain gauge.

3. The method for data fusion of multi-source precipitation measurement products as described in claim 2, characterized in that: The noise-reduced radar precipitation data includes: A dual polarization parameter dataset is constructed based on the horizontal and vertical channel echo power of weather radar precipitation measurement data. Using the aforementioned dual-polarization parametric dataset, the dual-polarization parametric filtering method is employed to determine echo correlation and eliminate noise sources, thereby obtaining denoised radar precipitation data. Based on the noise-reduced radar precipitation data, the radar coordinates and beam scanning relationship are converted into a ground plane coordinate system to obtain radar precipitation data consistent with the coordinate system of the ground rain gauge.

4. The method for data fusion of multi-source precipitation measurement products as described in claim 3, characterized in that: The interpolated and completed rain gauge precipitation data includes, An initial rain gauge dataset was constructed based on precipitation measurement data from ground rain gauges. The initial rain gauge dataset is then checked for completeness, consistency, and representativeness, and standardized to obtain a preprocessed rain gauge dataset. Based on the preprocessed rain gauge dataset, the missing data is supplemented by terrain-coordinated Kriging interpolation using regional elevation data to obtain the interpolated and supplemented ground rain gauge precipitation data.

5. The method for data fusion of multi-source precipitation measurement products as described in claim 4, characterized in that: Combining the preprocessed multi-source precipitation data, a spatiotemporally consistent data set is constructed, including: A multi-source preprocessed dataset is constructed based on preprocessed meteorological satellite precipitation data, weather radar precipitation data, and ground rain gauge precipitation data. Using the multi-source preprocessed dataset, pairwise correlation analysis is performed and spatiotemporal synchronized data that meet the correlation conditions are selected to obtain the spatiotemporal synchronized dataset. Based on the spatiotemporal synchronization dataset, logarithmic processing is performed and the target region is divided into gridded grids of equal area to construct a spatiotemporally consistent data set.

6. The method for data fusion of multi-source precipitation measurement products as described in claim 5, characterized in that: Based on the aforementioned spatiotemporally consistent dataset, using the area-to-point co-kriging interpolation method, the fused precipitation data includes: Based on the meteorological satellite precipitation data and weather radar precipitation data in the aforementioned spatiotemporal consistent data set, a raster-formatted dataset is constructed. The raster-format dataset is converted into discrete-point format data that is spatiotemporally synchronized with ground rain gauge precipitation data to obtain a synchronized discrete-point dataset. Based on the aforementioned synchronous discrete point dataset, ground rain gauge precipitation data are used as the main variable, and a co-kriging weighting function is constructed using the variogram and covariance to obtain fused precipitation data.

7. The method for data fusion of multi-source precipitation measurement products as described in claim 6, characterized in that: Using the fused precipitation data, a cross-validation method was employed to construct an evaluation system for the fusion model. The evaluation of the precipitation data fusion results included... A cross-validation dataset was constructed by integrating precipitation data and some ground rain gauge precipitation data. Using the cross-validation dataset, some ground rain gauge precipitation data were set to unknown, and co-kriging interpolation was used to predict precipitation values ​​to obtain multi-source fusion prediction values. Based on the multi-source fusion prediction value, precipitation values ​​are predicted by single-source Kriging interpolation using precipitation data from ground rain gauges, weather radars, and meteorological satellites, respectively, to obtain single-source prediction values. By combining the multi-source fusion prediction values ​​and the single-source prediction values, the root mean square error between the predicted and measured values ​​is calculated and compared to evaluate the accuracy of the multi-source fusion precipitation data.

8. A multi-source precipitation measurement product data fusion system, employing the multi-source precipitation measurement product data fusion method as described in any one of claims 1 to 7, characterized in that... include, The data acquisition module is used to acquire precipitation measurement data from meteorological satellites, weather radars, and ground rain gauges to build a multi-source precipitation dataset. The data preprocessing module is used to preprocess the multi-source precipitation dataset to obtain satellite precipitation data with a unified coordinate system and resolution, noise-reduced radar precipitation data, and interpolated and completed rain gauge precipitation data. The data analysis module is used to combine the preprocessed multi-source precipitation data to construct a spatiotemporally consistent data set; The data fusion module is used to obtain fused precipitation data based on the spatiotemporally consistent data set using the surface-to-point co-kriging interpolation method. The model evaluation module is used to construct a fusion model evaluation system using the fused precipitation data and employing cross-validation to evaluate the precipitation data fusion results.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the multi-source precipitation measurement product data fusion method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the multi-source precipitation measurement product data fusion method according to any one of claims 1 to 7.

Images