A seamless splicing correction method and system for multi-radar networking QPE of gridded terrain partitioning

Abstract

This invention belongs to the field of meteorological radar data processing technology, specifically relating to a method and system for seamless stitching and correction of multi-radar network QPE (Quality, Equipment, and Partition) in a gridded terrain zoning system. The method includes: obtaining gridded terrain multipliers based on digital elevation measurements; obtaining single-area precipitation estimates based on gridded terrain multipliers, mean precipitation, standard deviation of precipitation, radar standard deviation, real-time linear measurements, and radar linear mean; obtaining fused precipitation estimates based on the first statistical variance, second statistical variance, and error covariance of the radar equipment ensemble; obtaining blind-area precipitation estimates by combining ensemble measurements such as blind-area precipitation mean, blind-area standard deviation, ensemble correlation coefficient, and ensemble standard deviation; obtaining a grid-stitched total field based on single-area precipitation estimates, fused precipitation estimates, and blind-area precipitation estimates; and obtaining initial and station-level mean values ​​based on the grid-stitched total field and actual ground measurements to output a stitched correction total field. Based on the stitched correction total field, this invention achieves seamless grid-based stitching.

Description

Technical Field

[0001] This invention relates to the field of meteorological radar data processing technology. More specifically, this invention relates to a method and system for seamless stitching and correction of multi-radar network QPE in gridded terrain partitioning. Background Technology

[0002] In the daily operation of meteorological disaster early warning and refined weather forecasting, gridded precipitation estimation is a core component to ensure disaster prevention and mitigation and the accurate evolution of hydrological models. Its main task is to conduct all-weather monitoring and quantitative calculation of sounding echoes and precipitation distribution across the entire spatial area, so as to promptly capture abnormal water vapor evolution when localized rainstorms or sudden heavy precipitation occur, thereby preventing extreme weather evolution from causing floods or secondary geological disasters.

[0003] Existing gridded precipitation estimation typically uses a fixed radar reflectivity-precipitation intensity conversion relationship and applies traditional simple distance weighting or arithmetic mean methods to calculate precipitation distribution across the entire region using multiple radar observation data and ground measurements.

[0004] However, in actual meteorological observation and complex terrain environments, precipitation evolution and meteorological signals fluctuate frequently due to terrain undulations. This leads to conventional extrapolation methods lacking computational boundary protection and anomaly interception mechanisms. When faced with stable, clear skies or extreme drought conditions, key calculation indicators are prone to division by zero due to numerical values, resulting in computational logic deadlock or crash, and ultimately failing to generate a globally aligned and stitched correction total field. Summary of the Invention

[0005] To address the technical problem that conventional precipitation estimation methods lack computational boundary protection under extreme weather conditions, which can easily lead to division-by-zero deadlock in the underlying calculations and prevent the generation of a globally aligned and stitched correction field, this invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a method for seamless stitching and correction of multi-radar network QPE in a gridded terrain partition, comprising: acquiring digital elevation measurements of the entire spatial domain to generate a terrain grid set, and thereby acquiring a target terrain grid and its grid elevation measurements; acquiring grid terrain multipliers based on the grid elevation measurements; acquiring historical linear sequences, historical precipitation sequences, and real-time linear measurements through the target terrain grid, and acquiring radar standard deviation and radar linear mean through the historical linear sequences, and simultaneously acquiring precipitation standard deviation and precipitation mean through the historical precipitation sequences; combining precipitation mean, grid terrain multipliers, precipitation standard deviation, radar standard deviation, real-time linear measurements, and radar linear mean to obtain a single-area precipitation estimate; and obtaining a recalculated terrain grid based on the terrain grid set. A radar array of topographic grids is used to obtain first precipitation estimates, second precipitation estimates, first statistical variance, second statistical variance, and error covariance, ultimately yielding a fused precipitation estimate. Based on the topographic grid array, blind zone topographic grids and grid boundary arrays are obtained, and the blind zone precipitation mean, ensemble correlation coefficient, blind zone standard deviation, ensemble standard deviation, real-time ensemble measurements, and ensemble precipitation mean are obtained, ultimately yielding a blind zone precipitation estimate. The single-zone precipitation estimate, fused precipitation estimate, and blind zone precipitation estimate are combined to obtain a grid-stitched total field. Real ground measurements from all surface rain gauges are extracted to obtain the station measurement mean, and combined with the grid-stitched total field to obtain an initial mean. Finally, a stitched correction total field is obtained based on the grid-stitched total field, the station measurement mean, and the initial mean.

[0007] Preferably, the step of obtaining the grid terrain multiplier based on the grid elevation measurement includes: obtaining a preset baseline elevation determined by the lowest geographical elevation; adding the baseline elevation to a preset zero-divide-off protection constant to obtain a safety elevation base; and dividing the grid elevation measurement by the safety elevation base to obtain the grid terrain multiplier.

[0008] Preferably, the step of combining the mean precipitation, grid topography multiplier, precipitation standard deviation, radar standard deviation, real-time linear measurement, and radar linear mean to obtain a single-area precipitation estimate includes: measuring the degree of deviation of the real-time linear measurement from the radar linear mean to obtain the real-time fluctuation deviation; dividing the precipitation standard deviation by the radar standard deviation to obtain the discrete conversion coefficient; multiplying the grid topography multiplier, the discrete conversion coefficient, and the real-time fluctuation deviation to obtain the anomaly offset term; adding the mean precipitation and the anomaly offset term, and if the sum is greater than or equal to zero, the sum is used as the single-area precipitation estimate; if the sum is less than zero, the single-area precipitation estimate is forcibly reset to zero.

[0009] Preferably, the step of obtaining a radar device set based on an overlapping terrain grid, and using this set to acquire a first precipitation estimate, a second precipitation estimate, a first statistical variance, a second statistical variance, and an error covariance, ultimately obtaining a fused precipitation estimate, includes: extracting two independent radar devices from the radar device set as a first overlapping radar and a second overlapping radar, respectively, and outputting the first precipitation estimate and the second precipitation estimate; extracting the historical error sequences of the first overlapping radar and the second overlapping radar for the overlapping terrain grid, obtaining the first statistical variance, the second statistical variance, and the error covariance; and then using the second statistical variance as the basis for obtaining the fused precipitation estimate. Subtracting the error covariance from the variance yields the first effective cross weight corresponding to the first precipitation estimate; multiplying the first effective cross weight by the first precipitation estimate yields the first effective physical contribution component; subtracting the error covariance from the first statistical variance yields the second effective cross weight corresponding to the second precipitation estimate; multiplying the second effective cross weight by the second precipitation estimate yields the second effective physical contribution component; adding the first and second effective physical contribution components forms a weighted sum; dividing the weighted sum by the sum of the effective cross weights of the two radars, the fused precipitation estimate is normalized and output.

[0010] Preferably, the step of obtaining the blind area terrain grid and grid boundary set based on the terrain grid set, and thereby obtaining the blind area precipitation mean, ensemble correlation coefficient, blind area standard deviation, ensemble standard deviation, real-time ensemble measurement, and ensemble precipitation mean, and finally obtaining the blind area precipitation estimate, includes: extracting the grid boundary set containing a preset number of edge terrain grids from the terrain grid set using a coordinate traversal algorithm based on a preset search radius; converting the single-area precipitation estimate and fused precipitation estimate within the grid boundary set into real-time ensemble measurement using an inverse distance weighted interpolation algorithm; and statistically analyzing the historical precipitation sequence of the blind area terrain grid and the corresponding ensemble historical sequence of the grid boundary set. The blind zone precipitation mean, ensemble precipitation mean, blind zone standard deviation, and ensemble standard deviation are obtained, and the ensemble correlation coefficient is obtained using the Pearson correlation coefficient algorithm. If the ensemble standard deviation is lower than the preset spatial steady-state threshold, the blind zone precipitation mean is used as the blind zone precipitation estimate. Otherwise, the numerical difference is obtained by subtracting the ensemble precipitation mean from the real-time ensemble measurement, and the spatial fluctuation mapping ratio is obtained by dividing the blind zone standard deviation by the ensemble standard deviation. The ensemble correlation coefficient, the spatial fluctuation mapping ratio, and the numerical difference are multiplied to obtain the dynamic driving term. If the sum of the blind zone precipitation mean and the dynamic driving term is greater than or equal to zero, it is used as the blind zone precipitation estimate; if it is less than zero, the blind zone precipitation estimate is forcibly reset to zero.

[0011] Preferably, the step of combining single-area precipitation estimation, fused precipitation estimation, and blind-area precipitation estimation to obtain the grid-stitched total field includes: using a grid indexing algorithm to extract the latitude and longitude coordinates of each grid in the terrain grid set as the index of the bottom array; and writing the single-area precipitation estimation, fused precipitation estimation, and blind-area precipitation estimation into the global blank matrix according to the corresponding bottom array index to construct a continuous and complete grid-stitched total field.

[0012] Preferably, obtaining the splicing correction total field based on the grid splicing total field, the station measurement mean, and the initial mean includes: when the initial mean is greater than or equal to a preset global starting baseline, dividing the station measurement mean by the initial mean to calculate the original correction ratio; when the initial mean is less than the preset global starting baseline, forcibly setting the original correction ratio to the value 1; constructing amplitude limiting logic by introducing an upper limit truncation factor and a lower limit truncation factor, selecting the larger value between the lower limit truncation factor and the original correction ratio, and then selecting the smaller value between the upper limit truncation factor and the larger value as the global correction multiplier; performing a multiplicative superposition operation on the grid splicing total field and the global correction multiplier to output the splicing correction total field.

[0013] Secondly, the present invention provides a multi-radar networked QPE seamless stitching correction system for gridded terrain partitioning, characterized in that it is used to implement the method described in any one of claims 1 to 7, comprising: an environment construction and feature analysis module, used to acquire digital elevation measurements of the entire spatial domain to generate a terrain grid set, and thereby acquire a target terrain grid and its grid elevation measurements; acquire grid terrain multipliers based on grid elevation measurements; acquire historical linear sequences, historical precipitation sequences, and real-time linear measurements through the target terrain grid, and acquire radar standard deviation and radar linear mean through the historical linear sequences, and simultaneously acquire precipitation standard deviation and precipitation mean through the historical precipitation sequences; a partitioned precipitation collaborative estimation module, used to combine precipitation mean, grid terrain multipliers, precipitation standard deviation, radar standard deviation, real-time linear measurements, and radar linear mean to obtain single-area precipitation estimation; based on the terrain grid... The system first obtains a radar equipment set with overlapping terrain grids, and uses this set to obtain the first precipitation estimate, the second precipitation estimate, the first statistical variance, the second statistical variance, and the error covariance, ultimately obtaining a fused precipitation estimate. Based on the terrain grid set, it obtains the blind area terrain grid and grid boundary set, and uses this set to obtain the blind area precipitation mean, ensemble correlation coefficient, blind area standard deviation, ensemble standard deviation, real-time ensemble measurement, and ensemble precipitation mean, ultimately obtaining a blind area precipitation estimate. The global stitching and benchmark correction module combines the single-area precipitation estimate, the fused precipitation estimate, and the blind area precipitation estimate to obtain the grid stitching total field. It extracts the actual ground measurements from all surface rain gauges in the entire region to obtain the station measurement mean, and combines this with the grid stitching total field to obtain the initial mean. Finally, it obtains the stitching correction total field based on the grid stitching total field, the station measurement mean, and the initial mean.

[0014] Preferably, the regional precipitation collaborative estimation module includes: a single-region extrapolation submodule, used to generate a single-region precipitation estimate by superimposing physical precipitation increments that conform to geographical undulation characteristics onto the precipitation mean by measuring real-time fluctuation deviation, discrete transformation coefficient and anomaly offset term; and a cross-fusion submodule, used to extract the positive correlation interference of multi-source radar errors by using a non-negative truncation operator after subtracting the error covariance from the first statistical variance and the second statistical variance, construct the first effective cross weight and the second effective cross weight, and output the fused precipitation estimate based on the effective physical contribution components.

[0015] The blind zone reconstruction submodule is used to perform state cutoff judgment on the ensemble standard deviation using the spatial steady-state threshold. When the large-scale uniform meteorological field switching state is not triggered, the external meteorological abrupt change characteristics are mapped into precipitation increments that conform to the spatial transmission law by using the spatial fluctuation mapping ratio and the ensemble correlation coefficient, so as to output the blind zone precipitation estimate.

[0016] Preferably, the global stitching and baseline correction module includes: a matrix mapping submodule, used to convert latitude and longitude coordinates into underlying array indices using a grid indexing algorithm, and to write the directional assignment values ​​of single-area precipitation estimation, fused precipitation estimation, and blind-area precipitation estimation into a global blank matrix to prevent cross-area faults and construct a continuous and complete grid stitching total field; a deviation measurement submodule, used to benchmark against real ground measurements and verify grid values, measure the multiplicative deviation of the calculation model at the global level to obtain the original correction ratio, and when the initial mean is lower than the global starting baseline, trigger a deadlock avoidance mechanism for the global seamless stitching division formula, forcibly setting the original correction ratio to the value 1; and a amplitude limiting correction submodule, used to generate a global correction multiplier by introducing upper and lower limit truncation factors to constrain the safety boundary of the original correction ratio, and to perform multiplicative alignment between the grid stitching total field and the global correction multiplier to output the stitching correction total field.

[0017] The beneficial effects of this invention are as follows: This invention extracts elevation measurements to obtain grid topographic multipliers and calculates single-area precipitation estimates, measuring the spatial lifting gain and nonlinear evolution properties of precipitation intensity within the target topographic grid, thus mitigating the risk of computational divergence under stable weather conditions; it introduces a non-negative truncation operator within overlapping observation grids to obtain fused precipitation estimates, measuring the joint fault tolerance properties and confidence evaluation of multi-radar networks in overlapping coverage areas, and preventing abnormal negative weight backlash caused by error coupling in the same direction; it utilizes coordinate traversal and Pearson correlation coefficient algorithms to output blind zone precipitation estimates, measuring the detection... The spatial steady-state transmission law and water vapor co-evolution attributes between missing blocks and the outer boundary network dynamically isolate the erroneous polarity transmission caused by mountain blockage; combined with the measurement values ​​of ground rain gauges across the entire region, a global baseline is introduced and amplitude-limited multiplicative correction is performed to output a spliced ​​correction total field. This measures the multiplicative deviation attribute between the global grid field and the actual ground measurement values ​​and the macroscopic precipitation facts, avoiding the risk of low-level computational deadlock during drought stagnation periods. It ensures the smooth alignment of the global baseline, the coherent distribution of data throughout the entire cycle, and the seamless splicing smoothness of multi-source networking under complex terrain conditions for the automatic precipitation estimation and control system. Attached Figure Description

[0018] Figure 1 This is a flowchart of a multi-radar network QPE seamless stitching correction method for gridded terrain partitioning;

[0019] Figure 2 This is a multi-radar network stitching and correction effect diagram of gridded terrain zoning.

[0020] Figure 3 This is a functional module diagram of a multi-radar networked QPE seamless stitching correction system for gridded terrain partitioning;

[0021] Figure 4 Comparison of grid profile changes between existing conventional techniques and the present invention during the seamless stitching and correction process of multi-radar network QPE;

[0022] Figure 5 A comparison of the grid distribution of traditional interpolation methods and the present invention for reconstructing blind areas in complex terrain. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0025] This invention discloses a method for seamless stitching and correction of multi-radar network QPE in gridded terrain zoning, referring to... Figure 1 This includes steps S1-S4:

[0026] S1. Obtain digital elevation measurements of the entire spatial area to generate a set of terrain grids, extract grid elevation measurements to determine grid terrain multipliers, and use a non-negative truncation operator to extrapolate and fuse the average precipitation with real-time radar measurements to output a single-area precipitation estimate.

[0027] It should be noted that the detected echoes and actual precipitation intensity within the target terrain grid exhibit exponentially nonlinear physical characteristics due to the elevation undulations of the terrain grid. Fixed transformation rules cannot adapt to the nonlinear changes in water vapor caused by the elevation differences in the gridded terrain. This computational limitation leads to logical omissions in conventional extrapolation methods when facing stable weather conditions, as the standard deviation may return to zero, causing division failures. Therefore, this invention measures the nonlinear evolution of precipitation intensity within the target terrain grid using the grid terrain multiplier and the effective fluctuation lower limit, effectively mitigating the computational divergence risk under stable conditions and providing a basis for single-area precipitation estimation.

[0028] Specifically, digital elevation measurements of the entire spatial area are extracted from a geographic information database; a spatial grid partitioning algorithm is used to divide the entire spatial area based on the digital elevation measurements to generate a set of terrain grids containing multiple terrain grids;

[0029] Traverse the terrain grid set and use the current terrain grid to be processed as the target terrain grid;

[0030] Extract the grid elevation measurements of the target terrain grid; obtain the preset baseline elevation determined by the lowest geographical elevation; add the baseline elevation to the preset zero-division protection constant to obtain the safety elevation base; divide the grid elevation measurements by the safety elevation base to obtain the grid terrain multiplier;

[0031] Extract the historical logarithmic sequence of the target terrain grid within a preset past time window; extract the historical precipitation sequence of the target terrain grid within a preset past time window;

[0032] The historical linear sequence is obtained by exponentiation of the historical logarithmic sequence with a preset fixed transformation base; real-time logarithmic measurements of the corresponding target grid positions in the meteorological radar detection equipment are obtained; real-time linear measurements are obtained by exponentiation of the real-time logarithmic measurements with a fixed transformation base; radar standard deviation is obtained by calculating the standard deviation of the historical linear sequence; radar linear mean is obtained by calculating the arithmetic mean of the historical linear sequence; precipitation standard deviation is obtained by calculating the standard deviation of the historical precipitation sequence; precipitation mean is obtained by calculating the arithmetic mean of the historical precipitation sequence; single-area precipitation estimation is obtained based on precipitation mean, grid topographic multiplier, precipitation standard deviation, radar standard deviation, real-time linear measurements, and radar linear mean.

[0033] It should be further explained that the real-time logarithmic measurement represents the intensity of the meteorological radar echo received within the target grid at the current moment. Its construction principle is that the meteorological radar equipment emits electromagnetic waves into space and receives the backscattered energy of precipitation particles. Given the extremely large physical range of the scattered energy from meteorological targets, the detection system converts the received raw feedback power into a logarithmic form for encoding and storage to adapt to the wide dynamic range of meteorological signal characteristics.

[0034] It should be further explained that the real-time linear measurement represents the true physical reflectivity factor of precipitation particles within the target grid. Its construction principle is to prevent distortion of physical quantities caused by directly performing arithmetic statistics in the logarithmic domain. It uses an exponential operation model to inversely transform radar detection data with logarithmic properties, restoring the true physical quantity with linear superposition properties based on a fixed transformation base. This provides a data foundation with a consistent physical reference for subsequent calculations of precipitation mean and variance.

[0035] Specifically, the precipitation estimation for a single area satisfies the following relationship:

[0036] ;

[0037] In the formula, For single-area precipitation estimation, This is the average precipitation. For grid terrain multipliers, For precipitation standard deviation, For radar standard deviation, For real-time linear measurement, This represents the linear mean of the radar.

[0038] It should be added that, The discrete transformation coefficient is characterized by the physical proportional mapping law between the standard deviation of precipitation and the standard deviation of radar, which transforms the discrete fluctuation characteristics of radar detection signals into physical fluctuation components of the real precipitation dimension.

[0039] It characterizes real-time fluctuation deviations and is constructed by measuring the degree to which real-time linear measurements deviate from the radar linear mean, thus measuring the transient change difference of a specific weather evolution process compared to a normal physical baseline.

[0040] The anomalous offset term is constructed by comprehensively considering the ratio constraint between the spatial uplift gain induced by the grid terrain multiplier and the discrete conversion coefficient, and mapping the real-time fluctuation deviation in the radar signal domain into a physical precipitation increment that conforms to the geographical undulation characteristics.

[0041] in, The larger the value, the more dramatic the elevation changes of the target terrain grid. The larger the value, the more dynamically the terrain gain on the anomalous offset is amplified, thus increasing the sensitivity of the windward slope terrain grid to heavy precipitation sensing. The smaller the value, the more likely the target terrain grid is located in a low-lying plain. The smaller the value, the smoother the terrain gain on the anomalous offset term is, thus achieving the goal of filtering out small clutter interference from the plain terrain grid.

[0042] The larger the value, the higher the concentration of meteorological feedback signals within the target terrain grid. The greater the increase, the more likely it is to achieve the goal of outputting high-intensity single-area precipitation estimates; The smaller the value, the thinner the meteorological feedback signal within the target terrain grid, making it more likely to be detected. The more it tends to decrease, the more it tends to achieve the goal of ensuring a non-negative baseline for precipitation physical quantities.

[0043] The larger the value, the higher the historical precipitation within the target terrain grid exhibits high-frequency fluctuations. The greater the increase, the more likely it is to improve the sensitivity of precipitation event detection; The smaller the value, the more stable the historical precipitation within the target terrain grid is, making... The more it decreases, the more it achieves the goal of filtering out ground clutter interference.

[0044] The larger the value, the more drastic the fluctuations in historical radar echoes within the target terrain grid. The more it decreases, the more it achieves the goal of suppressing high-frequency clutter from the radar equipment itself; The smaller the value, the more stable the historical radar echoes within the target terrain grid are, making... The larger the value, the more likely it is to enhance the ability to reconstruct weak meteorological signals.

[0045] The larger the value, the higher the historical radar observation baseline within the target terrain grid, indicating that the baseline is at a high level. The smaller the signal, the more likely it is to eliminate false echoes caused by normal ground features. The smaller the value, the clearer the historical radar observation baseline within the target terrain grid, making... The greater the intensity, the more sensitive the ability to detect sudden precipitation events.

[0046] For example, the baseline elevation determines the physical reference zero point when calculating the terrain multiplier of the target terrain grid. Empirically, the baseline elevation ranges from -500 to 0. In this embodiment, the baseline elevation is set to -100. Implementers can set the baseline elevation based on the lowest geographical elevation characteristics of the region. For instance, when the detection area includes extremely deep basins or terrain below sea level, the baseline elevation can be appropriately lowered to cover all concave terrains; when the detection area is entirely located on a high-altitude plateau, the baseline elevation can be appropriately raised to narrow the relative elevation span.

[0047] For example, the division-by-zero protection constant determines the basic safety margin for multiplier division operations on the terrain grid, with an empirical range of [0.01, 1]. In this embodiment, the division-by-zero protection constant is set to 0.1. Implementers can set the division-by-zero protection constant according to the required accuracy of the overall terrain elevation calculation. For instance, when the elevation measurement accuracy is extremely high and the risk of floating-point overflow is high, the division-by-zero protection constant can be appropriately increased to enhance the stability of the underlying calculation against division-by-zero. When the terrain is gently undulating and it is necessary to retain sensitivity to small elevation changes, the division-by-zero protection constant can be appropriately decreased to avoid smoothing out effective physical differences.

[0048] For example, the past time window determines the length of the historical meteorological samples referenced by the algorithm, with an empirical range of [1, 6]. In this embodiment, the past time window is set to 1. Implementers can set the past time window according to seasonal climate characteristics. For example, when it is a large-scale stable precipitation season and the meteorological evolution is slow, the past time window can be appropriately increased to obtain more stable statistical characteristics; when it is a flood season with frequent strong convection and sudden changes, the past time window can be appropriately decreased to improve the response speed to sudden rainstorms.

[0049] For example, the fixed conversion base determines the conversion benchmark ratio for restoring logarithmic decibels to physical linearity, with an empirical range of [1.1, 1.5]. In this embodiment, the fixed conversion base is set to 1.26. Implementers can set the fixed conversion base according to the radar hardware's underlying calibration protocol. For instance, when the radar base data uses a non-standard decibel mapping with compressed dynamic range, the fixed conversion base can be appropriately increased to restore the true physical fluctuation amplitude; when there is a systematic overestimation bias in radar calibration, the fixed conversion base can be appropriately decreased to suppress the dimensional amplification effect.

[0050] For example, the effective fluctuation lower limit determines the signal-to-noise ratio threshold for the algorithm to initiate dynamic mapping, with an empirical range of [10, 15]. In this embodiment, the effective fluctuation lower limit is set to 10. Implementers can set the effective fluctuation lower limit according to the radar hardware noise floor characteristics. For example, when the detection conditions face a large area of ​​clear sky and prominent thermal noise, the effective fluctuation lower limit can be appropriately increased to prevent clutter from triggering falsely; when facing abundant precipitation and needing to capture the edge of weak precipitation, the effective fluctuation lower limit can be appropriately decreased to improve the sensitivity of light rain detection.

[0051] S2. Extract the first and second historical errors of the overlapping terrain grids, calculate the statistical variance and error covariance, and perform cross-weighting on the first and second precipitation estimates in combination with the average weight mutual exclusion threshold to output the fused precipitation estimate.

[0052] It should be noted that overlapping terrain grids that jointly observe the same weather evolution process within a multi-radar network exhibit a positively correlated common attenuation characteristic. A simple measure of individual variance ignores the limitation of error aggregation generated by multi-source radars for cloud clusters within the same overlapping terrain grid. This limitation leads to uncontrolled misjudgments caused by weight collapse when conventional weighting methods face covariance exceeding a single variance. Therefore, this invention measures the joint fault tolerance attribute of a multi-radar network within an overlapping terrain grid by using a weighted mutual exclusion threshold. This effectively blocks the backlash of negative abnormal weights and provides a basis for judgment in fused precipitation estimation.

[0053] Specifically, the following steps are taken: First, extract the terrain grids covered by multiple radars from the terrain grid set as overlapping terrain grids. Second, obtain the radar equipment set covering the overlapping terrain grids. Third, extract two independent radar equipment from the radar equipment set as the first overlapping radar and the second overlapping radar, respectively. Fourth, extract the first historical estimation sequence of the overlapping terrain grids from the first overlapping radar within a preset past time window. Fifth, extract the second historical estimation sequence of the overlapping terrain grids from the second overlapping radar within a preset past time window. Sixth, extract the historical precipitation sequence of the overlapping terrain grids within the preset past time window. Finally, subtract the historical precipitation sequence from the first historical estimation sequence to obtain the final result. First error sequence; subtract historical precipitation sequence from second historical estimation sequence to obtain second error sequence; calculate variance of first error sequence to obtain first statistical variance; calculate variance of second error sequence to obtain second statistical variance; calculate covariance of first error sequence and second error sequence to obtain error covariance; obtain single-area precipitation estimation output by first overlapping radar for overlapping terrain grid as first precipitation estimation; obtain single-area precipitation estimation output by second overlapping radar for overlapping terrain grid as second precipitation estimation; obtain fused precipitation estimation based on first precipitation estimation, second precipitation estimation, first statistical variance, second statistical variance and error covariance.

[0054] It should be further clarified that the first and second overlapping radars are used only to distinguish any two independent detection devices covering the same overlapping terrain grid. The designation of "first" and "second" is merely for identification and does not imply any specific ordering based on detection performance, physical distance, or signal processing sequence. When performing cross-fusion calculations, personnel can arbitrarily designate two independent devices as the first and second overlapping radars respectively; this will not affect the final fused precipitation estimation results.

[0055] Specifically, the integrated precipitation estimation satisfies the following relationship:

[0056] ;

[0057] In the formula, To integrate precipitation estimation, The second statistical variance, For error covariance, For the first precipitation estimate, The first statistical variance, This is an estimate of the second precipitation event.

[0058] It should be added that, The effective cross weights of the first precipitation estimation within the overlapping terrain grid are characterized by extracting the second statistical variance and subtracting the error covariance. The non-negative truncation operator is used to remove the abnormal negative weights caused by the same-direction coupling of errors, thus transforming the discrete fluctuations of the second overlapping radar into a confidence evaluation of the first overlapping radar.

[0059] The effective physical contribution component that represents the first precipitation estimate output to the overlapping terrain grid is constructed by performing a multiplication operation on the first precipitation estimate and its corresponding effective cross weight, and scaling the observed precipitation intensity according to the relative reliability of the first overlapping radar.

[0060] The effective cross weight of the second precipitation estimation in the overlapping terrain grid is characterized by extracting the first statistical variance and subtracting the error covariance, using a non-negative truncation operator to isolate the positive correlation interference of multi-source radar errors, and transforming the discrete fluctuation of the first overlapping radar into a confidence evaluation of the second overlapping radar.

[0061] The effective physical contribution component of the second precipitation estimate output to the overlapping terrain grid is characterized by multiplying the second precipitation estimate with its corresponding effective cross weights and scaling the observed precipitation intensity according to the relative reliability of the second overlapping radar.

[0062] The fusion output value of precipitation from multiple radars within an overlapping terrain grid after statistical variance trade-off is characterized by adding the first physical contribution component and the second physical contribution component to form a weighted sum, and then dividing by the sum of the effective cross weights of the two radars to perform normalization calculation. By suppressing single-point anomalies of individual radars through cross-suppression, a fusion precipitation estimate with a stable physical chassis is output.

[0063] Furthermore, The larger the value, the more likely the multi-source radar error exhibits a positively coupled, co-current attenuation characteristic within the overlapping terrain grid, making... and The more it tends to decrease, the more it tends to achieve the goal of preventing abnormal reversal of overlapping terrain grid data; The smaller the value, the more likely the multi-source radar error exhibits an independent random distribution. and The greater the potential, the more likely it is to achieve the goal of outputting high-confidence fused precipitation estimates.

[0064] Furthermore, The larger the value, the higher the independent error of the first precipitation estimate exhibits high-frequency oscillation characteristics, making it... The greater the increase, the more likely it is to achieve the goal of outputting a bias towards the second precipitation estimate. The smaller the value, the more stable and reliable the independent error of the first precipitation estimate is. The more it tends to decrease, the more it tends to achieve the goal of outputting a bias towards the first precipitation estimate.

[0065] Furthermore, The larger the value, the more likely it is that the first precipitation estimation sensed a high-intensity local precipitation event within the overlapping terrain grid, making... The greater the increase, the more likely it is to achieve the goal of improving the intensity of integrated precipitation estimation; The smaller the value, the more likely the initial precipitation estimate detected a weak precipitation condition. The more it tends to decrease, the more it tends to achieve the goal of suppressing the estimated intensity of fused precipitation.

[0066] The larger the value, the more pronounced the volatile independent error in the second precipitation estimate becomes. The greater the increase, the more likely it is to achieve the goal of outputting a bias towards the first precipitation estimate. The smaller the value, the more stable and reliable the independent error of the second precipitation estimate is. The more it tends to decrease, the more it tends to achieve the goal of deeply integrating the second precipitation estimation facts.

[0067] The larger the value, the more likely it is that the second precipitation estimation sensed independently developing convective events within the overlapping topographic grid, making... The larger it tends to be, the more it tends to achieve the purpose of compensating for the detection attenuation blind zone of a single radar; The smaller the value, the more likely the second precipitation estimation sensed the absence of precipitation echo characteristics, making... The more it decreases, the more likely it is to eliminate ground feature artifacts in the first precipitation estimate.

[0068] For example, the weighted mutual exclusion threshold determines the lower limit of weight aggregation that triggers the weighted backoff logic, with an empirical range of [0.05, 0.2]. In this embodiment, the weighted mutual exclusion threshold is set to 0.1. Implementers can adjust the weighted mutual exclusion threshold according to the homogeneity of the multi-radar network devices. For instance, when the detection devices are prone to attenuation due to similar terrain and highly coupled errors, the weighted mutual exclusion threshold can be appropriately increased to accelerate the weighted backoff response speed; when the radar network devices are highly heterogeneous and have high independent detection capabilities, the weighted mutual exclusion threshold can be appropriately decreased to fully leverage the advantages of cross-weighting.

[0069] S3. Extract the historical precipitation sequence of the blind zone and the historical precipitation sequence of the grid boundary set, combine the spatial steady-state threshold and the standard deviation of the set to make inference and judgment, and use the non-negative truncation operator to fuse the mean precipitation of the blind zone with the set measurement to output the estimated precipitation of the blind zone.

[0070] It should be noted that the precipitation distribution within the blind area topographic grid is spatially non-uniform and heterogeneous due to the influence of meteorological wind field advection. Conventional extrapolation methods identify negatively correlated topographic partition blockages as unidirectional transmission, thus generating spatial artifact correlation limitations. This computational limitation leads to logical omissions in extrapolation calculations when the standard deviation is divided by zero under large-scale uniform clear skies. Therefore, this invention extracts high-confidence grid data sources and measures the water vapor co-evolution attributes of the blind area topographic grid and surrounding topographic grid partitions through spatial steady-state thresholds. It effectively isolates erroneous polarity transmission caused by topographic blockages, providing a high-fidelity dynamic reconstruction basis for blind area precipitation estimation.

[0071] Specifically, terrain grids without radar coverage are extracted from the terrain grid set as blind zone terrain grids;

[0072] Historical precipitation sequences within a preset past time window are extracted from the topographic grid of the blind area as the historical sequence of the blind area;

[0073] The coordinate traversal algorithm is used to extract a set of grid boundaries containing a preset number of edge terrain grids from the terrain grid set according to a preset search radius;

[0074] It should be further explained that the coordinate traversal algorithm is used to transform spatial neighborhood search into underlying matrix distance determination. Its principle is to extract the physical center coordinates of the blind zone terrain grid as the starting point, and then extract the physical center coordinates of the remaining grids by traversing the terrain grid set, calculating the physical spatial distance between the remaining grids and the starting point. When this physical spatial distance is less than a preset search radius, and the target grid has valid detection data, it is filtered as an edge terrain grid. For example, when performing data matching, valid grids with single-area precipitation estimation or fused precipitation estimation are extracted, sorted in ascending order from nearest to farthest in physical spatial distance, and the grid objects at the front of the list that meet a preset number are selected to construct a grid boundary set. This avoids sample distortion caused by interference from distant terrain, providing reliable underlying data support for subsequent reconstruction and inference.

[0075] Extract single-area precipitation estimation from the edge terrain grids within the grid boundary set; extract fused precipitation estimation from the edge terrain grids within the grid boundary set;

[0076] The inverse distance weighted interpolation algorithm is used to convert single-area precipitation estimation and fused precipitation estimation into real-time ensemble measurements.

[0077] It should be further explained that the inverse distance weighted interpolation algorithm is used to smoothly map discrete edge data to the center of the blind zone. Implementers calculate the physical straight-line distance from each edge terrain grid within the grid boundary set to the blind zone terrain grid. Using the reciprocal of this physical straight-line distance as the weight, they perform a weighted summation of single-area precipitation estimates and fused precipitation estimates. For example, edge terrain grids closer to the blind zone center are assigned higher computational weights, while those farther away are assigned lower weights, thus constructing a smoothly transitioning real-time ensemble measurement.

[0078] Extract historical precipitation sequences of each edge terrain grid within a preset past time window; calculate the arithmetic mean of the historical precipitation sequence values ​​of each edge terrain grid at the same time node to construct the ensemble historical sequence; calculate the arithmetic mean of the values ​​at each time node in the blind zone historical sequence to obtain the blind zone precipitation mean; calculate the standard deviation of the blind zone historical sequence to obtain the blind zone standard deviation; calculate the arithmetic mean of the values ​​at each time node in the ensemble historical sequence to obtain the ensemble precipitation mean; calculate the standard deviation of the ensemble historical sequence to obtain the ensemble standard deviation.

[0079] The correlation coefficient between the blind zone historical sequence and the ensemble historical sequence is obtained by using the Pearson correlation coefficient algorithm.

[0080] It should be further explained that the Pearson correlation coefficient algorithm is used to assess the degree of linear synchronization between the historical climate of the blind area and the historical climate of the surrounding area. Implementers extract the historical sequences of the blind area and the ensemble historical sequences, calculate their covariance, and divide by the product of their standard deviations. For example, when the blind area topographic grid and the grid boundary ensemble are in the same windward slope climate zone and their water vapor evolution is synchronized, the calculated output is a positive correlation coefficient approaching positive one; when the two are blocked by towering mountains, forming a leeward slope climate difference, the calculated output is a correlation coefficient approaching zero or negative, thus dynamically isolating erroneous spatial physical transmission caused by topographic obstruction.

[0081] Determine whether the ensemble standard deviation is lower than the preset spatial steady-state threshold; if it is lower, it is determined to be a large-scale uniform meteorological field, and the mean precipitation in the blind zone is used as the estimate of the blind zone precipitation; if it is not lower, the estimate of the blind zone precipitation is obtained based on the mean precipitation in the blind zone, the ensemble correlation coefficient, the blind zone standard deviation, the ensemble standard deviation, the real-time ensemble measurement, and the mean precipitation in the ensemble.

[0082] Specifically, when the state cutoff mechanism is not triggered, the estimated precipitation in the blind zone satisfies the following relationship:

[0083] ;

[0084] In the formula, For blind spot precipitation estimation, This represents the average precipitation in the blind area. For ensemble correlation coefficient, The standard deviation of the blind zone. For the set standard deviation, For real-time ensemble measurements, This represents the ensemble precipitation mean.

[0085] It should be added that, The spatial fluctuation mapping ratio between the topographic grid of the blind zone and the grid boundary set is characterized by extracting the standard deviation of the blind zone and the standard deviation of the set and performing a division calculation to transform the physical dispersion of the external meteorological field into the precipitation dispersion inside the blind zone.

[0086] It characterizes the degree of deviation of the current observation state within the grid boundary set from the historical average baseline. Its construction principle is to extract the real-time set measurement value and subtract the set precipitation mean value, calculate the numerical difference of the external weather system from the normal physical benchmark, and establish a measurement benchmark for meteorological change characteristics.

[0087] The dynamic driving term is characterized by using numerical difference as the basis for calculation, and performing a series of multiplication operations to superimpose the synergistic gain of fluctuation mapping proportional constraint and set correlation coefficient, thereby mapping the evolution characteristics of the grid boundary set into precipitation increments that conform to the spatial steady-state transmission law.

[0088] in, The larger the value, the more consistent and unidirectional the evolution characteristics of the blind area terrain grid and the grid boundary set are, thus tending to achieve the goal of high-fidelity dynamic inference of blind area terrain grid data; The smaller the value, the more likely it is to achieve the purpose of blocking false spatial artifacts that cross mountain terrain barriers, as the blind area terrain grid and grid boundary set exhibit the deviation characteristics caused by terrain partitioning and blocking.

[0089] The larger the value, the more pronounced the water vapor fluctuations in the historical meteorological field within the blind area's topographic grid. The larger the value, the more likely it is to improve the sensitivity of response to small weather increments within the terrain grid in the blind area; The smaller the value, the more stable and dry the historical meteorological field within the blind area's topographic grid appears, making it easier to observe changes in weather patterns. The more it tends to decrease, the more it tends to achieve the goal of suppressing false triggering by external large-scale precipitation interference.

[0090] The larger the value, the more likely it is that the grid boundary set exhibits precipitation increments exceeding historical averages. The greater the increase, the more likely it is to raise the estimated precipitation value in the blind area; The smaller the value, the lower the characteristic of the mesh boundary set, indicating that it exhibits features below the historical baseline. The more it tends to decrease, the more it tends to achieve the goal of suppressing the estimated precipitation value in the blind area.

[0091] The larger the value, the higher the frequency of historical precipitation evolution within the grid boundary set, indicating that it exhibits high-frequency oscillation characteristics. The more it tends to decrease, the more it tends to achieve the goal of smoothly extrapolating the blind zone of preventing external strong convective interference; The smaller the value, the more stable the meteorological background evolution within the grid boundary set becomes. The larger the value, the more likely it is to achieve the goal of smooth interpolation and reconstructing a stable precipitation field distribution.

[0092] The larger the value, the more likely the grid boundary set has been exposed to high-intensity precipitation over a long period. The smaller the precipitation level, the more effectively the influence of real-time small precipitation fluctuations is weakened, thus achieving the goal of maintaining a steady state of high precipitation background in the blind area. The smaller the value, the more likely the mesh boundary set remains in a dry state for a long time. The larger the cloud clusters, the more prominent the weather increments caused by sudden convective clouds become, thus enhancing the ability of blind spots to perceive new precipitation events.

[0093] For example, the search radius determines the spatial coverage span of the peripheral reference data during blind zone reconstruction, with an empirical range of [5, 20]. In this embodiment, the search radius is set to 10. Implementers can set the search radius based on the distribution density of surrounding observation stations. For instance, when the observation stations are extremely sparsely distributed, the search radius can be appropriately increased to cross the blank area and obtain sufficient reference targets; when the observation stations are densely distributed and local micro-scale meteorological characteristics are significant, the search radius can be appropriately reduced to prevent the inclusion of distant, non-homogeneous meteorological samples.

[0094] For example, the preset quantity determines the effective boundary reference sample size for blind zone reconstruction, with an empirical range of [4, 12]. In this embodiment, the preset quantity is set to 8. Implementers can set the preset quantity according to the complexity of the surrounding terrain. For example, when the terrain around the blind zone is relatively open and flat and the climate characteristics are uniformly distributed, the preset quantity can be appropriately increased to improve the smoothness of spatial interpolation; when the surrounding terrain is undulating and there are multiple natural climate blocking zones, the preset quantity can be appropriately reduced to limit the participation of far-end distorted samples in reconstruction.

[0095] For example, the spatial steady-state threshold determines the lower limit of the driving force of fluctuations in the peripheral meteorological field, with an empirical range of [0.1, 0.5]. In this embodiment, the spatial steady-state threshold is set to 0.2. Implementers can set the spatial steady-state threshold according to the regional topographic zoning characteristics. For example, when facing extremely arid and low-rainfall terrain conditions and persistent large-scale clear skies, the spatial steady-state threshold can be appropriately increased to cut off spatial artifacts caused by weak disturbances; when facing mountainous areas with complex terrain and frequent micro-scale convection, the spatial steady-state threshold can be appropriately decreased to enhance the ability to reconstruct and capture local water vapor abrupt changes.

[0096] S4. Use the grid indexing algorithm to merge the various precipitation estimates to obtain the grid splicing total field. Extract the actual ground measurement values ​​to calculate the station mean and initial mean. Combine the global starting baseline and cutoff factor to perform multiplicative calculations on the grid splicing total field to output the splicing corrected total field.

[0097] It should be noted that the meteorological dynamic evolution in the gridded terrain zoning and stitching scenario causes precipitation errors to exhibit a multiplicative diffusion physical characteristic; conventional compensation rules have a bottom line limitation of initial mean returning to zero when facing extreme drought conditions; this physical limitation leads to numerical lag in the output deadlock caused by the accumulation of gridded seamless stitching calculations at the global level; therefore, this invention measures the multiplicative deviation attribute between the gridded network and ground monitoring stations through a global starting baseline; effectively prevents the risk of division-to-zero deadlock during drought stagnation periods, and provides a basis for judging the stable alignment of the global baseline for global precipitation estimation.

[0098] Specifically, the grid indexing algorithm is used to merge single-area precipitation estimation, fused precipitation estimation, and blind-area precipitation estimation to obtain the grid splicing total field.

[0099] It should be further explained that the grid indexing algorithm is used to restore locally scattered inference data to their physical location within the global matrix. Its construction principle involves extracting the latitude and longitude coordinates of the terrain grid as the index of the bottom-level array. Precipitation estimates from single areas, merged precipitation estimates, and blind area precipitation estimates are then oriented and written into the global blank matrix according to their corresponding bottom-level array indices. For example, when the computer system constructs a global precipitation field, based on pre-calculated physical coordinates, estimated values ​​from different sources are directly assigned to the matching pixel grids in the overall field matrix. This prevents data gaps caused by cross-regional calculations, constructing a continuous and complete grid-stitched overall field, and providing a reliable underlying data container for global benchmark correction.

[0100] The actual ground measurements of all ground rainfall monitoring stations within the topographic grid set are extracted to form the actual measurement set; the arithmetic mean of all discrete values ​​in the actual measurement set is calculated to obtain the station mean; the verification grid values ​​corresponding to the physical locations of ground rainfall monitoring stations in the grid splicing total field are extracted to form the verification grid set; the arithmetic mean of all discrete values ​​in the verification grid set is calculated to obtain the initial mean.

[0101] It should be further explained that the verification grid numerical representation represents the inferred precipitation intensity at specific coordinate points of the grid spliced ​​total field. Its construction principle involves extracting the grid precipitation estimation results corresponding to the actual coordinates of ground rainfall monitoring stations, establishing a one-to-one correspondence between the inferred data and the actual ground measurements in spatial coordinates, and providing a benchmark for subsequently measuring the multiplicative bias of the calculation model.

[0102] It should be further explained that the discrete numerical values ​​in the actual measurement set and the verification grid set represent the sampling data of independent monitoring stations within the entire spatial area. The construction principle is to extract independent monitoring values ​​from different geographical coordinates as the underlying data, transform the data of physical monitoring points into set elements, and provide a reliable statistical sample for calculating the systematic mean.

[0103] The splicing correction total field is obtained based on the grid splicing total field, the preset upper limit truncation factor, the preset lower limit truncation factor, the station measurement mean, the initial mean, and the preset global starting baseline.

[0104] Specifically, the splicing and correction of the total field satisfies the following relation:

[0105] ;

[0106] when hour, ;when hour, ;

[0107] In the formula, To assemble and correct the main site, For the grid splicing total field, The upper limit cutoff factor, The lower cutoff factor is used. This is the original correction ratio. This is the average value measured at each station. The initial mean, This serves as the baseline for the entire region.

[0108] It should be noted that, The global correction multiplier is characterized by the principle of constructing a limiting logic by introducing an upper and lower truncation factor to constrain the original correction ratio within a preset safety boundary, thus preventing global multiplicative divergence caused by local radar signal distortion. It characterizes the actual physical precipitation in the target area after being constrained by the ground reference. Its construction principle is to perform multiplicative superposition operation on the grid splicing total field and the global correction multiplier, and use the water vapor base of the ground rain gauge to align the radar inversion error in the whole domain, and output a spatial field matrix that conforms to the precipitation facts.

[0109] in, The larger the value, the less likely the grid stitching total field has fallen into an echo-free stagnant state. The closer it is to the normal range, the closer it is to achieving a seamless splicing of the global water vapor benchmark; The smaller the value, the more likely the grid splicing field is to fall into an extremely arid stagnation state, thus achieving the goal of avoiding deadlock in the division formula of the seamless splicing of the whole-domain terrain partition.

[0110] The larger the value, the more it indicates that the grid splicing total field has a negative spatial error, which effectively and proportionally increases and compensates for all the terrain grid values ​​on the base surface of the grid splicing total field, thus becoming more and more likely to achieve the goal of fully compensating for the total amount of warm cloud precipitation that was not detected by the radar network. The smaller the value, the more likely it is that the total field of the grid splicing has a positive deviation in spatial error, which effectively compresses and reduces the topographic grid value on the base surface of the total field of the grid splicing, thereby achieving the goal of comprehensively suppressing the false excess of water volume caused by abnormal topographic bright band echoes.

[0111] The larger the value, the stronger the precipitation feedback captured by the grid splicing total field, making... The larger the value, the more likely it is to achieve the goal of outputting a high-throughput splicing correction total field; The smaller the value, the lower the feedback state of the total field of the mesh stitching. The smaller the value, the more likely it is to achieve the goal of limiting the overload of the splicing correction total field.

[0112] For example, the upper limit cutoff factor determines the extreme boundary of upward compensation in the multiplicative correction calculation, with an empirical range of [1.5, 5]. In this embodiment, the upper limit cutoff factor is set to 3. Implementers can set the upper limit cutoff factor according to the normal degree of missed detections and underreporting in the radar network. For example, when it is the season dominated by warm cloud and low centroid precipitation and the radar is prone to serious underestimation, the upper limit cutoff factor can be appropriately increased to release sufficient ground compensation space; when the radar detection hardware performance is excellent and occasional overestimation clutter is dominant, the upper limit cutoff factor can be appropriately decreased to prevent abnormally excessive upward adjustment.

[0113] For example, the lower cutoff factor determines the extreme boundary of downward compression and adjustment in the multiplicative correction calculation, with an empirical value range of [0.1, 0.8]. In this embodiment, the lower cutoff factor is set to 0.3. Implementers can set the lower cutoff factor according to the frequency of the radar network encountering bright bands at the zero-degree layer or the frequency of ground object super-refraction. For example, when large-area bright bands are very likely to occur in winter, causing the radar to seriously overestimate, the lower cutoff factor can be appropriately reduced to give the algorithm sufficient compression and adjustment authority; when the climate is dry and the radar usually underestimates, the lower cutoff factor can be appropriately increased to maintain the basic precipitation physical quantity baseline.

[0114] For example, the global starting baseline determines the baseline for global average rainfall activation in the multiplicative correction calculation, with an empirical range of [0.05, 0.2]. In this embodiment, the global starting baseline is set to 0.1. Implementers can set the global starting baseline according to the drought and flood season characteristics of the regional topography and climate. For example, when it is an extremely dry season with very little precipitation, the global starting baseline can be appropriately increased to prevent the division calculation from collapsing and deadlocking; when it is a season with abundant precipitation and sufficient global water vapor reference, the global starting baseline can be appropriately decreased to expand the activation range of the multiplicative correction calculation.

[0115] This invention also discloses a multi-radar network QPE seamless stitching correction system for gridded terrain partitioning, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a multi-radar network QPE seamless stitching correction method for gridded terrain partitioning according to the present invention.

[0116] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0117] Figure 2The image shows the correction effect of multi-radar network stitching for gridded terrain partitioning. The initial topographic precipitation coupling field in the image reflects the physical baseline precipitation distribution, which is greatly affected by complex terrain obstruction. The uncorrected single-area precipitation estimation not only exhibits numerical oscillations at the echo center but also is prone to data loss and precipitous drops in obstructed areas. In contrast, this invention extracts nonlinear evolution characteristics in the early stages of the simulation by using grid terrain multipliers and effective fluctuation lower limits, thus maintaining physical constraints on single-area precipitation estimation and reducing the risk of computational divergence under statically stable conditions. In the cross-fusion area facing overlapping multi-source echoes, weighted calculations are performed using a weighted mutual exclusion threshold to output the fused precipitation estimation, reducing the risk of abnormal weight divergence caused by error aggregation. In the ambiguous boundary areas without detection data, spatial steady-state thresholds are identified and dynamic driving terms are reconstructed, resulting in a smooth transition characteristic of isolating and blocking errors in the output blind zone precipitation estimation. Subsequent processing merges the data from each area into a grid stitched total field and uses amplitude limiting logic combined with station averages to perform multiplicative correction, generating a smoothly aligned stitched corrected total field. The results show that the present invention effectively balances the needs of terrain compensation and blind zone reconstruction, suppresses fault jumps across radar boundaries, and enhances the numerical stability and reliable control of the seamless splicing process of multi-source meteorological networks while ensuring the continuous distribution of precipitation fields across the entire region.

[0118] like Figure 3 The diagram shown is a functional block diagram of a multi-radar network QPE seamless stitching correction system for gridded terrain zoning provided by an embodiment of the present invention. The multi-radar network QPE seamless stitching correction system for gridded terrain zoning described in this invention can be installed in an electronic device. Depending on the functions implemented, the multi-radar network QPE seamless stitching correction system 100 for gridded terrain zoning may include: an environment construction and feature analysis module 101, a zoned precipitation collaborative estimation module 102, and a global stitching and benchmark correction module 103.

[0119] The module described in this invention, also referred to as a unit, refers to a series of computer program segments that can be executed by the processor of an electronic device and perform a fixed function, and are stored in the memory of the electronic device. In this embodiment, the functions of each module / unit are as follows:

[0120] The environment construction and feature analysis module 101 is used to generate a set of terrain grids based on digital elevation measurements, extract the target terrain grid and its grid elevation measurements to calculate the grid terrain multiplier; calculate the historical linear sequence and the real-time logarithmic measurement by exponentiation respectively to obtain the historical linear sequence and the real-time linear measurement; calculate the radar standard deviation and radar linear mean based on the historical linear sequence, and calculate the precipitation standard deviation and precipitation mean based on the historical precipitation sequence;

[0121] The partitioned precipitation collaborative estimation module 102 is used to obtain single-area precipitation estimates based on precipitation mean, grid topographic multiplier, precipitation standard deviation, radar standard deviation, real-time linear measurement, and radar linear mean; extract overlapping topographic grids and their radar equipment sets, and obtain fused precipitation estimates based on first precipitation estimates, second precipitation estimates, first statistical variance, second statistical variance, and error covariance; extract blind area topographic grids and grid boundary sets, and obtain blind area precipitation estimates based on blind area precipitation mean, ensemble correlation coefficient, blind area standard deviation, ensemble standard deviation, real-time ensemble measurement, and ensemble precipitation mean;

[0122] The global splicing and benchmark correction module 103 is used to obtain the grid splicing total field based on single-area precipitation estimation, fused precipitation estimation and blind area precipitation estimation; extract the real ground measurement values ​​of the global ground rainfall monitoring stations to calculate the station measurement mean, and combine it with the grid splicing total field to calculate the initial mean; and obtain the splicing correction total field based on the grid splicing total field, the station measurement mean and the initial mean.

[0123] Figure 4 The figure shows a comparison of the grid profile changes between existing conventional techniques and the present invention during the seamless stitching and correction process of multi-radar network QPE. Existing conventional techniques use fixed rules to transform the detected echoes, which are greatly affected by terrain undulations. Their estimated values ​​not only exhibit numerical oscillations in terrain-undulating areas but are also prone to signal interruption and precipitous drops when passing through elevation blind zones. In contrast, the present invention, in the single-area extrapolation stage, uses grid terrain multipliers to solve for abnormal offset terms, forcing a zero-based determination in single-area precipitation estimation, reducing the risk of division-by-zero failure under statically stable conditions. In the cross-fusion stage facing overlapping terrain grids, it utilizes error covariance to construct first and second effective cross weights, reducing the risk of weight distortion caused by multi-source radar error correlation. In the global stitching and baseline correction stages, it identifies station averages and introduces a global starting baseline, presenting a seamless stitching feature with stable baseline alignment through global correction multipliers.

[0124] Figure 5The figure shows a comparison of the grid distribution of traditional interpolation methods and the present invention under the reconstruction of blind areas in complex terrain. Traditional interpolation methods use distance-weighted algorithms for spatial mapping, which are significantly affected by spatial non-uniformity and heterogeneity. This results in abnormal diffusion of grid values ​​across boundaries and data distortion and structural faults within the blind area. In contrast, the present invention extracts real-time ensemble measurements of the grid boundary set during the blind area reconstruction stage using a coordinate traversal algorithm and an inverse distance-weighted interpolation algorithm. This ensures that the precipitation estimation in the blind area maintains geographical undulation characteristics and reduces the risk of cross-regional faults. During the transition stage across terrain zones, the Pearson correlation coefficient algorithm is used to calculate the ensemble correlation coefficient, reducing the risk of cross-regional data jumps caused by spatial data discontinuity. Within the blind area terrain grid, a spatial steady-state threshold is identified, and a state cutoff judgment is performed on the ensemble standard deviation. Through spatial fluctuation mapping ratios and dynamic driving term constraints, a smooth precipitation distribution characteristic is presented.

Claims

1. A method for seamless stitching and correction of multi-radar networked QPE systems with gridded terrain partitioning, characterized in that, include: Digital elevation measurements of the entire spatial domain are obtained to generate a set of terrain grids, and the target terrain grid and its grid elevation measurements are obtained accordingly. The grid topography multiplier is obtained based on the grid elevation measurements; historical linear sequences, historical precipitation sequences, and real-time linear measurements are obtained through the target topography grid, and the radar standard deviation and radar linear mean are obtained through the historical linear sequences, while the precipitation standard deviation and precipitation mean are obtained through the historical precipitation sequences. Precipitation estimates for a single area are obtained by combining the mean precipitation, grid topographic multiplier, standard deviation of precipitation, standard deviation of radar precipitation, real-time linear measurements, and the linear mean of radar. The radar equipment set with overlapping terrain grids is obtained based on the terrain grid set, and the first precipitation estimate, the second precipitation estimate, the first statistical variance, the second statistical variance and the error covariance are obtained from it, and finally the fused precipitation estimate is obtained. Based on the terrain grid set, the terrain grid and grid boundary set of the blind area are obtained, and the mean precipitation of the blind area, ensemble correlation coefficient, standard deviation of the blind area, standard deviation of the ensemble, real-time ensemble measurement and mean precipitation of the ensemble are obtained, and finally the precipitation of the blind area is estimated. The grid splicing total field is obtained by combining single-area precipitation estimation, fused precipitation estimation and blind area precipitation estimation. The real ground measurement values ​​of all surface rain gauges in the region are extracted to obtain the station measurement mean, and the initial mean is obtained by combining the grid splicing total field. The spliced ​​correction total field is obtained based on the grid splicing total field, the station measurement mean, and the initial mean.

2. The method for seamless stitching and correction of multi-radar network QPE in gridded terrain partitioning according to claim 1, characterized in that, The process of obtaining the grid terrain multiplier based on grid elevation measurements includes: Obtain the preset baseline elevation determined by the lowest geographical elevation; The baseline elevation is added to the preset zero-division protection constant to obtain the safety elevation base. Divide the grid elevation measurement by the safe elevation base to obtain the grid terrain multiplier.

3. The method for seamless stitching and correction of multi-radar network QPE in gridded terrain partitioning according to claim 1, characterized in that, The method of combining precipitation mean, grid topographic multiplier, precipitation standard deviation, radar standard deviation, real-time linear measurement and radar linear mean to obtain single-area precipitation estimation includes: The real-time fluctuation deviation is obtained by measuring the degree to which the real-time linear measurement deviates from the radar linear mean. Divide the standard deviation of precipitation by the standard deviation of radar to obtain the discrete conversion coefficient; The anomaly offset term is obtained by multiplying the grid terrain multiplier, the discrete transformation coefficient, and the real-time fluctuation deviation. The mean precipitation value is added to the anomalous offset term. If the sum is greater than or equal to zero, the sum is used as the single-area precipitation estimate. If the sum is less than zero, the single-area precipitation estimate is forcibly reset to zero.

4. The method for seamless stitching and correction of multi-radar network QPE in gridded terrain partitioning according to claim 1, characterized in that, The radar equipment set based on the terrain grid set is obtained by creating an overlapping terrain grid, and then using this set to obtain a first precipitation estimate, a second precipitation estimate, a first statistical variance, a second statistical variance, and an error covariance, ultimately obtaining a fused precipitation estimate, including: Two independent radar devices are extracted from the radar device set and used as the first overlapping radar and the second overlapping radar, respectively, to output the first precipitation estimate and the second precipitation estimate. Historical error sequences of the first and second overlapping radars for the overlapping terrain grid are extracted to obtain the first statistical variance, the second statistical variance, and the error covariance. Subtract the error covariance from the second statistical variance to obtain the first effective cross weight corresponding to the first precipitation estimate; multiply the first effective cross weight by the first precipitation estimate to obtain the first effective physical contribution component; Subtract the error covariance from the first statistical variance to obtain the second effective cross weight corresponding to the second precipitation estimate; multiply the second effective cross weight by the second precipitation estimate to obtain the second effective physical contribution component; The first effective physical contribution component and the second effective physical contribution component are added together to form a weighted sum. The weighted sum is then divided by the sum of the effective cross weights of the two radars, and the normalized output is the fused precipitation estimate.

5. The method for seamless stitching and correction of multi-radar network QPE in gridded terrain zoning according to claim 1, characterized in that, The process involves obtaining the blind area terrain grid and grid boundary set based on the terrain grid set, and then using this to obtain the blind area precipitation mean, ensemble correlation coefficient, blind area standard deviation, ensemble standard deviation, real-time ensemble measurement, and ensemble precipitation mean, ultimately obtaining the blind area precipitation estimate, including: Based on a preset search radius, a coordinate traversal algorithm is used to extract a set of grid boundaries containing a preset number of edge terrain grids from the terrain grid set; The inverse distance weighted interpolation algorithm is used to convert single-area precipitation estimation and fused precipitation estimation within the grid boundary set into real-time ensemble measurements. The historical precipitation sequences of the topographic grid in the blind area are statistically analyzed and compared with the ensemble historical sequences corresponding to the grid boundary set. The mean precipitation in the blind area, the mean precipitation in the ensemble, the standard deviation of the blind area, and the standard deviation of the ensemble are obtained. The ensemble correlation coefficient is obtained using the Pearson correlation coefficient algorithm. If the ensemble standard deviation is lower than the preset spatial steady-state threshold, the mean precipitation in the blind zone is used as the estimate of the blind zone precipitation; otherwise, the numerical difference is obtained by subtracting the mean precipitation from the real-time ensemble measurement, the spatial fluctuation mapping ratio is obtained by dividing the blind zone standard deviation by the ensemble standard deviation, and the dynamic driving term is obtained by multiplying the ensemble correlation coefficient, the spatial fluctuation mapping ratio, and the numerical difference. If the sum of the mean precipitation in the blind zone and the dynamic driving term is greater than or equal to zero, it is used as the estimated precipitation in the blind zone; if it is less than zero, the estimated precipitation in the blind zone is forcibly reset to zero.

6. The method for seamless stitching and correction of multi-radar network QPE in gridded terrain partitioning according to claim 1, characterized in that, The method of combining single-area precipitation estimation, fused precipitation estimation, and blind-area precipitation estimation to obtain a grid-stitched total field includes: The latitude and longitude coordinates of each grid in the terrain grid set are extracted using a grid indexing algorithm and used as the index of the bottom array. Single-area precipitation estimates, fused precipitation estimates, and blind-area precipitation estimates are written into a global blank matrix according to the corresponding underlying array subscripts to construct a continuous and complete grid splicing total field.

7. The method for seamless stitching and correction of multi-radar network QPE in gridded terrain partitioning according to claim 1, characterized in that, The process of obtaining the spliced ​​correction total field based on the grid splicing total field, the station measurement mean, and the initial mean includes: When the initial mean is greater than or equal to the preset global baseline, the station mean is divided by the initial mean to calculate the original correction ratio. When the initial mean is less than the preset global baseline, the original correction ratio will be forcibly set to the value 1. Amplitude limiting logic is constructed by introducing upper and lower cutoff factors. The larger value between the lower cutoff factor and the original correction ratio is selected, and the smaller value between the upper cutoff factor and the larger value is selected as the global correction multiplier. Perform a multiplicative superposition operation between the grid stitched total field and the global correction multiplier to output the stitched corrected total field.

8. A multi-radar networked QPE seamless stitching correction system for gridded terrain zoning, characterized in that, To implement the method according to any one of claims 1 to 7, comprising: The environment construction and feature parsing module is used to obtain digital elevation measurements of the entire space to generate a set of terrain grids, and to obtain the target terrain grid and its grid elevation measurements; to obtain the grid terrain multiplier based on the grid elevation measurements; to obtain historical linear sequences, historical precipitation sequences and real-time linear measurements through the target terrain grid, and to obtain the radar standard deviation and radar linear mean through the historical linear sequence, and to obtain the precipitation standard deviation and precipitation mean through the historical precipitation sequence; The regional precipitation collaborative estimation module is used to combine the precipitation mean, grid topographic multiplier, precipitation standard deviation, radar standard deviation, real-time linear measurement, and radar linear mean to obtain the precipitation estimate for a single area; it obtains a radar equipment set with overlapping topographic grids based on the topographic grid set, and uses this to obtain the first precipitation estimate, the second precipitation estimate, the first statistical variance, the second statistical variance, and the error covariance, and finally obtains the fused precipitation estimate; it obtains the blind area topographic grid and grid boundary set based on the topographic grid set, and uses this to obtain the blind area precipitation mean, ensemble correlation coefficient, blind area standard deviation, ensemble standard deviation, real-time ensemble measurement, and ensemble precipitation mean, and finally obtains the blind area precipitation estimate; The global splicing and baseline correction module combines single-area precipitation estimation, fused precipitation estimation, and blind-area precipitation estimation to obtain a grid splicing total field. It extracts the real ground measurements from all surface rain gauges in the region to obtain the station average and combines them with the grid splicing total field to obtain the initial average. Based on the grid splicing total field, the station average, and the initial average, it obtains the splicing correction total field.

9. A multi-radar networked QPE seamless stitching correction system for gridded terrain zoning according to claim 8, characterized in that, The regional precipitation collaborative estimation module includes: The single-area extrapolation submodule is used to generate a single-area precipitation estimate by superimposing physical precipitation increments that conform to geographical undulation characteristics onto the precipitation mean by measuring real-time fluctuation deviation, discrete transformation coefficient and anomaly offset. The cross-fusion submodule is used to extract the positive correlation interference of multi-source radar errors by using a non-negative truncation operator after subtracting the error covariance from the first statistical variance and the second statistical variance, respectively, to construct the first effective cross weight and the second effective cross weight, and to output the fused precipitation estimate based on the effective physical contribution components. The blind zone reconstruction submodule is used to perform state cutoff judgment on the ensemble standard deviation using the spatial steady-state threshold. When the large-scale uniform meteorological field switching state is not triggered, the external meteorological abrupt change characteristics are mapped into precipitation increments that conform to the spatial transmission law by using the spatial fluctuation mapping ratio and the ensemble correlation coefficient, so as to output the blind zone precipitation estimate.

10. A multi-radar networked QPE seamless stitching correction system for gridded terrain zoning according to claim 8, characterized in that, The global stitching and benchmark correction module includes: The matrix mapping submodule is used to convert latitude and longitude coordinates into underlying array indices using a grid indexing algorithm, and to write the directional assignments of single-area precipitation estimation, fused precipitation estimation, and blind area precipitation estimation into a global blank matrix to prevent cross-area faults and construct a continuous and complete grid splicing total field. The deviation measurement submodule is used to benchmark against real ground measurements and verify grid values. It measures the multiplicative deviation of the calculation model at the global level to obtain the original correction ratio. When the initial mean is lower than the global starting baseline, it triggers the global seamless splicing division formula deadlock avoidance mechanism to force the original correction ratio to be set to the value 1. The amplitude-limiting correction submodule is used to generate a global correction multiplier by constraining the safety boundary of the original correction ratio by introducing upper and lower truncation factors, and to perform multiplicative alignment between the grid stitched total field and the global correction multiplier to output the stitched correction total field.

Images