Rainfall early warning method and system based on weather radar

Abstract

The invention relates to the technical field of meteorological monitoring, and discloses a rainfall early warning method and system based on a weather radar, and the method comprises the steps: obtaining weather radar body sweeping arrival base data and rainfall station observation data, carrying out the alignment, carrying out the meshing of the radar base data, and obtaining the grid echo data; constructing a quasi-vertical profile to determine the height range and uncertainty of the melting layer, and carrying out weighted bright band correction on echoes in the melting layer to obtain corrected echo data; determining rainfall type partitions and performing inversion to obtain a radar rainfall intensity field; according to the rainfall station data and the radar rainfall intensity field, station residual data is obtained, a space deviation field is obtained, and finally a calibration rainfall intensity field is obtained; determining an accumulated precipitation field of each member according to the calibrated rainfall field; and obtaining an over-threshold probability published by early warning according to the accumulated precipitation field of each member, and outputting precipitation early warning information when a probability threshold is continuously met. According to the invention, the rainfall intensity overestimation and false alarm jitter are reduced, and the accuracy and stability of short temporary rainfall early warning are improved.

Description

Technical Field

[0001] This invention relates to the field of meteorological monitoring technology, and more specifically, to a precipitation early warning method and system based on weather radar. Background Technology

[0002] Short-term nowcasting precipitation warnings typically rely on weather radar as the core observation method. They utilize radar reflectivity and other echo information to obtain the spatial distribution and evolution characteristics of precipitation echoes, and achieve short-term precipitation intensity and location forecasts through echo tracking and extrapolation. Common operational approaches include cross-correlation or optical flow-based motion field estimation, echo translation extrapolation, and converting echoes into rainfall intensity and accumulating them for short periods to support warning issuance. To improve usability, engineering implementations also incorporate echo quality control, target identification and tracking, and product output. For example, publicly available documents include schemes for constructing nowcasting forecasting systems based on optical flow constraints, such as CN108646319B – A Method and System for Forecasting Short-Term Heavy Rainfall and CN104977584B – A Method and System for Nowcasting Convective Weather.

[0003] However, existing short-term warning systems, which primarily rely on echo tracking and extrapolation, are prone to reproducible biases and unstable triggering in certain scenarios. For example, when precipitation echoes are locally generated and their intensity and shape change rapidly, the quality of the motion vector field deteriorates, leading to tracking failures or the accumulation of extrapolation errors. Bright bands often appear in the melting layer during stratiform or mixed precipitation processes. These bright bands enhance radar reflectivity at narrow altitudes, causing overestimation of rainfall intensity based on direct echo inversion or empirical strategies such as maximizing rainfall at low elevation angles. This can induce false alarms and frequent jitter in warning chains triggered by rainfall intensity or cumulative rainfall thresholds. Bright bands are a typical radar feature of the melting layer, and radar precipitation estimation has explicitly indicated that they may lead to overestimation of precipitation, especially when the estimation strategy is superimposed on the bright band.

[0004] Therefore, it is necessary to design a precipitation early warning method and system based on weather radar to solve the problems existing in the current technology. Summary of the Invention

[0005] In view of this, the present invention proposes a precipitation early warning method and system based on weather radar, aiming to solve the problems in the current technology.

[0006] This invention proposes a precipitation early warning method based on weather radar, comprising: Acquire volume scan radar base data from weather radar and rain gauge observation data and perform time alignment; then grid the volume scan radar base data from weather radar to obtain grid echo data. Based on the grid echo data, a quasi-vertical profile is constructed to determine the range of the melt layer height and the uncertainty of the melt layer height. Then, the echoes within the melt layer are corrected by weighted brightness band based on the uncertainty of the melt layer height to obtain corrected echo data. Precipitation type zones are determined based on the grid echo data, and radar rainfall intensity fields are obtained by inversion based on the corrected echo data. Within a sliding time window, station residual data is obtained based on the rain gauge observation data and the radar rainfall field. Based on the station residual data, reflectivity and rainfall intensity conversion parameters are updated online in zones to obtain a spatial deviation field. The calibrated rainfall field is obtained by calibrating the spatial deviation field. The translation field is determined based on the calibration rainfall intensity field of adjacent time periods, and the residual field is obtained. Based on the translation field and the historical statistics of the residual field, a multi-member set is generated for extrapolation, and the cumulative precipitation field of each member is obtained. Based on the cumulative precipitation field of each member, the probability of exceeding the threshold for issuing an early warning is obtained. When the probability of exceeding the threshold continuously meets the probability threshold, precipitation early warning information is output, and when the uncertainty of the melting layer height exceeds the uncertainty threshold, a downgraded early warning is output.

[0007] Furthermore, when acquiring weather radar volume scan radar base data and rain gauge observation data and performing time alignment, and then gridding the weather radar volume scan radar base data to obtain gridded echo data, the process includes: Using the alignment time corresponding to the early warning cycle as a unified time reference, the rain gauge observation data is aggregated according to the preset statistical time period corresponding to the alignment time to obtain aligned rain gauge observation data; the weather radar volume scan radar base data with the time closest to the alignment time is selected as aligned radar base data; based on the radar site parameters and terrain obstruction data, the beam height and obstruction rate corresponding to each radar gate are calculated; the aligned radar base data is mapped to a preset horizontal grid, and the radar gate echoes falling into the grid are weighted and synthesized according to distance, beam height and obstruction rate in each grid to obtain the grid echo value of the grid, and the beam height and obstruction rate of the grid are used as the correlation attributes of the grid echo data.

[0008] Furthermore, when constructing a quasi-vertical profile and determining the range and uncertainty of the melt layer height, the following steps are taken: Within a preset distance ring, grid echo data with an occlusion rate not exceeding an occlusion rate threshold are selected. Azimuth convergence is performed at different height levels to obtain robust echo statistics for each height level and form a quasi-vertical profile. Candidate height intervals for echo enhancement zones are determined within these quasi-vertical profiles, and the upper and lower boundaries of these candidate height intervals are used as the range of the melting layer height. Based on the height dispersion between different azimuth convergence results and the height fluctuation between adjacent warning cycles, the uncertainty of the melting layer height is determined.

[0009] Furthermore, when performing weighted brightness band correction on the echo within the melt layer based on the uncertainty of the melt layer height and obtaining corrected echo data, the process includes: A preset reference height range is selected above the melt layer height range, and a reference echo profile is determined based on the quasi-vertical profile of the reference height range; for each height layer within the melt layer height range, a bright band correction amount is determined based on the difference between the quasi-vertical profile and the reference echo profile; a correction weight is determined based on the melt layer height uncertainty, and the bright band correction amount is weighted to correct the grid echo data within the melt layer height range, thereby obtaining the corrected echo data.

[0010] Furthermore, when determining precipitation type zones based on the grid echo data and retrieving the radar rainfall intensity field based on the corrected echo data, the process includes: Based on the grid echo data, the echo top height index, echo texture index and melt layer overlap index of each grid are calculated; When the echo top height index is greater than the preset top height threshold and the echo texture index is greater than the preset texture threshold, the corresponding grid will be assigned to the convective precipitation zone. When the melting layer overlap index is greater than the preset overlap threshold and the echo texture index is not greater than the preset texture threshold, the corresponding grid is assigned to the layered precipitation zone. The remaining grid cells are assigned to the mixed precipitation zone; The corrected echo data is extracted from the preset inversion height layer below the melting layer height range to obtain the inversion echo value. The inversion echo value is then converted into a rainfall intensity value using the reflectivity and rainfall intensity conversion parameters corresponding to the convective precipitation zone, the stratiform precipitation zone, and the mixed precipitation zone, respectively, to form the radar rainfall intensity field.

[0011] Furthermore, when obtaining station residual data based on the rain gauge observation data and radar rainfall intensity field, and updating the reflectivity and rainfall intensity conversion parameters online by region, the process includes: Collect the observed rainfall intensity value of each rain gauge station, and collect the grid rainfall intensity value of the radar rainfall intensity field corresponding to the location of the rain gauge station as the station radar rainfall intensity value. The difference between the observed rainfall intensity value at the rain gauge station and the radar rainfall intensity value at the station is used as the station residual data; Based on the precipitation type partition to which the corresponding grid of the rain gauge belongs, the residual data of the station are collected into the precipitation type partition; Within each precipitation type zone, the station weights are determined based on the shading rate and the uncertainty of the melting layer height. The reflectance and rainfall intensity conversion parameters are then updated recursively. When the number of valid stations in the corresponding zone does not reach the preset threshold, the reflectance and rainfall intensity conversion parameters of the previous warning cycle remain unchanged.

[0012] Furthermore, when obtaining the spatial bias field based on the station residual data and calibrating it to obtain the calibrated rainfall intensity field, the process includes: The station residual data is mapped to the preset horizontal grid, and the spatial deviation field is solved by weighted interpolation with adjacent grid variation constraints. The interpolation weight is determined according to the occlusion rate and the uncertainty of the melting layer height. When the occlusion rate exceeds the occlusion rate threshold, the corresponding station residual data is removed. The spatial deviation field is used as a grid-by-grid correction amount for the radar rainfall intensity field to compensate the radar rainfall intensity field and obtain the calibrated rainfall intensity field. The compensated grid rainfall intensity value is limited to not less than zero.

[0013] Furthermore, when determining the translation field and obtaining the residual field based on the calibrated rainfall intensity fields of adjacent time periods, and generating a multi-member set extrapolated based on the historical statistics of the translation field and the residual field to obtain the cumulative precipitation field of each member, the process includes: Using the calibrated rainfall field of two adjacent time intervals as input, the displacement is searched grid by grid within a preset matching window and a preset search range to minimize the sum of the absolute values ​​of the differences between the calibrated rainfall values ​​of two time intervals within the matching window, thus obtaining the translation field. The calibration rainfall field of the previous time period is extrapolated by the translation field to obtain the extrapolated rainfall field, and the residual field is determined by the difference between the calibration rainfall field of the current time period and the extrapolated rainfall field. Within a preset historical window, the residual field is statistically analyzed to obtain the residual mean and residual standard deviation of each grid. Multiple member sets are extrapolated based on a preset perturbation sequence. For each member, a residual perturbation is generated according to the residual mean and residual standard deviation, and a displacement perturbation is applied to the translation field. The perturbed translation field is used to extrapolate the current time-time calibration rainfall intensity field, and the corresponding residual perturbation is superimposed to obtain the member rainfall intensity forecast field. The member rainfall intensity forecast fields within a preset cumulative time period are accumulated over time to obtain the cumulative precipitation field of each member.

[0014] Furthermore, when obtaining the threshold probability of issuing a warning based on the cumulative precipitation field of each member and outputting precipitation warning information, it includes: For each early warning issuing unit, the cumulative precipitation values ​​of multiple grids covered by the early warning issuing unit are aggregated in the cumulative precipitation field of each member by area weighting to obtain the cumulative precipitation value of the issuing unit; The proportion of members whose cumulative precipitation value is not less than a preset precipitation threshold in the early warning unit is determined as the over-threshold probability of the early warning unit. When the over-threshold probability is not less than the probability threshold within a preset number of consecutive early warning cycles, the precipitation early warning information is output. When the uncertainty of the melting layer height exceeds the uncertainty threshold, the probability threshold is replaced with an increased probability threshold, and the output precipitation early warning information is marked as a downgraded early warning.

[0015] Compared with existing technologies, the advantages of this invention are as follows: Time alignment and gridding between volume scan radar and rain gauges ensure accurate registration between the radar field and ground observations. Quasi-vertical profiles are constructed using gridded echo data, and the range and uncertainty of the melting layer height are explicitly output. This makes bright band correction no longer a fixed empirical correction but a reliable correction adaptively weighted by uncertainty, reducing the tendency for overestimation and false alarms in rainfall intensity retrieval under layered or mixed precipitation from the source. Simultaneously, precipitation type zoning is introduced, and reflectivity and rainfall intensity conversion parameters are updated online within each zone. Then, a spatial bias field is generated using station residuals to spatially calibrate the radar rainfall intensity field, enabling the calibrated rainfall intensity field to adapt to regional biases, occlusion effects, and process evolution. The system converges to a level closer to the actual ground rainfall, reducing systemic bias between different regions. It uses adjacent time intervals to calibrate the rainfall intensity field to determine the translational field and combines it with historical statistics of the residual field to generate a multi-member set for extrapolation. It calculates the threshold probability through the cumulative precipitation field of the multi-member set and adopts a triggering rule that continuously satisfies the probability threshold. This effectively smooths the threshold jitter caused by target splitting and merging, local nascent development, or short-term noise, improving the temporal stability and spatial consistency of the warning. When the uncertainty of the melting layer height exceeds the uncertainty threshold, it automatically outputs a downgraded warning, avoiding the forced issuance of a high-confidence warning under conditions where the brightness band height is difficult to identify stably or the correction confidence is reduced. This reduces the cost of false alarms and improves the traceability of the warning product.

[0016] On the other hand, this application also provides a precipitation warning system based on weather radar, used to apply the above-mentioned precipitation warning method based on weather radar, including: The acquisition unit is configured to acquire volume scan radar base data of weather radar and observation data of rain gauge station and perform time alignment, and to grid the volume scan radar base data of weather radar to obtain grid echo data. The correction unit is configured to construct a quasi-vertical profile based on the grid echo data, determine the range of the melt layer height and the uncertainty of the melt layer height, and perform weighted brightness band correction on the echoes within the melt layer according to the uncertainty of the melt layer height to obtain corrected echo data. The inversion unit is configured to determine precipitation type partitions based on the grid echo data and to invert the radar rainfall intensity field based on the corrected echo data. The calibration unit is configured to obtain station residual data based on the rain gauge observation data and radar rainfall field within a sliding time window, perform partitioned online updates of reflectivity and rainfall intensity conversion parameters based on the station residual data and obtain a spatial deviation field, and calibrate the calibrated rainfall field using the spatial deviation field. The processing unit is configured to determine the translation field and obtain the residual field based on the calibration rainfall intensity field of adjacent time periods, and generate a multi-member set extrapolation based on the translation field and the historical statistics of the residual field to obtain the cumulative precipitation field of each member. The early warning unit is configured to obtain the over-threshold probability of early warning issuance based on the cumulative precipitation field of each member, output precipitation early warning information when the over-threshold probability continuously meets the probability threshold, and output a downgraded early warning when the uncertainty of the melting layer height exceeds the uncertainty threshold.

[0017] It is understandable that the above-mentioned precipitation warning methods and systems based on weather radar have the same beneficial effects, and will not be elaborated further here. Attached Figure Description

[0018] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A flowchart illustrating a precipitation warning method based on weather radar provided in an embodiment of the present invention; Figure 2 This is a functional block diagram of a precipitation early warning system based on weather radar provided in an embodiment of the present invention. Detailed Implementation

[0019] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey its scope to those skilled in the art. It should be noted that, unless otherwise specified, embodiments and features described herein can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] In some embodiments of this application, see Figure 1 As shown, this application proposes a precipitation early warning method based on weather radar, including: S100: Acquires volume scan radar base data from weather radar and rain gauge observation data and performs time alignment; then, grids the volume scan radar base data from weather radar to obtain gridded echo data.

[0021] S200: Based on grid echo data, a quasi-vertical profile is constructed to determine the range and uncertainty of the melt layer height. Then, the echoes within the melt layer are corrected by weighted brightness band correction based on the melt layer height uncertainty to obtain corrected echo data.

[0022] S300: Precipitation type zones are determined based on grid echo data, and radar rainfall intensity fields are obtained by inversion based on corrected echo data.

[0023] S400: Within a sliding time window, station residual data is obtained based on rain gauge observation data and radar rainfall intensity field. Based on the station residual data, reflectivity and rainfall intensity conversion parameters are updated online in zones and spatial deviation field is obtained. The calibrated rainfall intensity field is obtained by calibrating with the spatial deviation field.

[0024] S500: Determine the translation field and obtain the residual field based on the calibration rainfall intensity field of adjacent time periods. Generate a multi-member set extrapolation based on the historical statistics of the translation field and the residual field and obtain the cumulative precipitation field of each member.

[0025] S600: Based on the cumulative precipitation field of each member, the probability of exceeding the threshold for issuing early warnings is obtained. When the probability of exceeding the threshold continuously meets the probability threshold, precipitation early warning information is output. When the uncertainty of the melting layer height exceeds the uncertainty threshold, a downgraded early warning is output.

[0026] Specifically, in this embodiment, grid echo data refers to the echo intensity data and its associated attributes obtained by projecting volume scan radar base data onto a preset horizontal grid, with each grid as a unit; the quasi-vertical profile refers to the one-dimensional echo statistical sequence that varies with altitude after azimuth convergence of grid echo data at different altitude layers within a preset range ring; the melting layer height range refers to the upper and lower boundaries of the candidate altitude interval of the echo enhancement zone in the quasi-vertical profile; and the melting layer height uncertainty refers to the altitude instability amplitude jointly characterized by the maximum and minimum difference between the center altitude of the echo enhancement zone in different azimuth convergence results and the absolute difference in center altitude between adjacent warning cycles. The spatial bias field refers to the grid-by-grid bias correction obtained by weighted interpolation with adjacent grid variation constraints on the horizontal grid, which is obtained by the station residuals of rain gauge observations and radar rainfall intensity fields; the translation field refers to the grid-by-grid displacement result obtained by matching the rainfall intensity fields of two adjacent time periods within a preset matching window and a preset search range; the residual field refers to the difference between the current time period calibration rainfall intensity field and the previous time period calibration rainfall intensity field after translation extrapolation of the translation field; "multi-member ensemble extrapolation" refers to the process of generating multiple member rainfall intensity forecast fields under the constraints of historical statistics of translation field and residual field and further accumulating time to obtain the cumulative precipitation field of each member.

[0027] Specifically, in step S100, the alignment time corresponding to the early warning cycle is used as a unified time reference. The rain gauge observation data is aggregated according to the preset statistical period corresponding to the alignment time to obtain the aligned rain gauge observation data. The weather radar volume scan radar base data with the time closest to the alignment time is selected as the aligned radar base data. The beam height and obstruction rate corresponding to each radar gate are calculated based on the radar site parameters and terrain obstruction data. The aligned radar base data is mapped to a preset horizontal grid. In each grid, the radar gate echoes falling into the grid are weighted and synthesized according to the distance, beam height and obstruction rate to obtain the grid echo value. The beam height and obstruction rate are output as the associated attributes of the grid echo data. In step S200, grid echo data with occlusion rates not exceeding an occlusion rate threshold are selected within a preset distance ring. Azimuth convergence is performed by height layer, preferably using the median of the grid echo values ​​within the height layer as the robust echo statistics value to form a quasi-vertical profile. Candidate height intervals for echo enhancement zones are determined within the quasi-vertical profile, and the upper and lower boundaries of these candidate height intervals are used as the melting layer height range. Simultaneously, the height corresponding to the peak value of the robust echo statistics value within the candidate height interval is used as the center height of the echo enhancement zone. Further, the uncertainty of the melting layer height is determined based on the maximum and minimum differences between the center height of the echo enhancement zone in different azimuth convergence results and the absolute differences between adjacent warning cycles. Subsequently, a preset reference height range is selected above the melting layer height range, and a reference echo profile is determined based on the quasi-vertical profile of the reference height range. For each height layer within the melting layer height range, the brightness band correction amount is determined according to the difference between the quasi-vertical profile and the reference echo profile, and the correction weight is determined according to the melting layer height uncertainty. The brightness band correction amount is weighted and then used to correct the grid echo data in the melting layer to obtain corrected echo data. The greater the melting layer height uncertainty, the smaller the correction weight, so as to avoid over-correction when the melting layer identification is unstable and introduce new system bias. In step S300, echo top height index, echo texture index, and melting layer overlap index are calculated based on grid echo data. The echo top height index is the highest height above a preset echo threshold, the echo texture index is the standard deviation of the grid echo values ​​within a preset neighborhood, and the melting layer overlap index is the overlap height where the echo top height index falls within the melting layer height range. Accordingly, the grid is divided into convective precipitation zones, stratiform precipitation zones, or mixed precipitation zones. Corrected echo data is extracted from the preset inversion height layer below the melting layer height range as inversion echo values. The inversion echo values ​​are converted into rainfall intensity values ​​using reflectivity and rainfall intensity conversion parameters corresponding to each precipitation type zone to form a radar rainfall intensity field.

[0028] Specifically, taking a monitoring network consisting of a single weather radar and several rain gauges within its coverage area as an example, the warning cycle is taken as the time interval consistent with the radar volume scan update, and the preset horizontal grid resolution and preset inversion height layer are set according to operational needs. Within each warning cycle, step S100 completes the time alignment and gridding of rain gauge observation data and volume scan radar base data to obtain grid echo data containing grid echo values, beam height, and obstruction rate; step S200 constructs a quasi-vertical profile within a preset range ring and determines the range and uncertainty of the melting layer height, and performs weighted brightness band correction on the echoes within the melting layer in conjunction with the reference echo profile to obtain corrected echo data; step S300 completes the precipitation type zoning and forms the radar rainfall intensity field. Then, proceed to step S400. Within the sliding time window, select rain gauge observation data with valid quality identifiers. For each rain gauge, read the radar rainfall intensity value corresponding to its location grid as the station's radar rainfall intensity value, and use the difference between the observed rainfall intensity value and the station's radar rainfall intensity value as the station's residual data. Collect the station residual data according to the precipitation type partition corresponding to the rain gauge grid, and determine the station weight within each precipitation type partition based on the uncertainty of shading rate and melt layer height. Reflectivity and rainfall intensity conversion parameters are recursively updated, and when the number of valid stations is less than [a certain value], [further details are needed]. When the preset threshold is reached, the reflectivity and rainfall intensity conversion parameters of the previous warning cycle are kept unchanged to avoid parameter drift caused by insufficient samples. Then, the station residual data are mapped to a preset horizontal grid, and the spatial deviation field is solved by weighted interpolation with adjacent grid variation constraints. The interpolation weight is determined by the uncertainty of the occlusion rate and the melting layer height. When the occlusion rate exceeds the occlusion rate threshold, the residual data of the corresponding station is removed. Finally, the spatial deviation field is used as a grid-by-grid correction to compensate the radar rainfall intensity field, and the compensation result is limited to not less than zero to obtain the calibrated rainfall intensity field. In step S500, taking the calibration rainfall fields of two adjacent time periods as input, the displacement is searched grid by grid within the preset matching window and preset search range to minimize the sum of the absolute values ​​of the differences between the calibration rainfall values ​​of the two time periods within the matching window, thus obtaining the translation field. The calibration rainfall field of the previous time period is extrapolated by the translation field to obtain the extrapolated translation rainfall field, and the residual field is determined by the difference between the current calibration rainfall field and the extrapolated translation rainfall field. The residual field is statistically analyzed within the preset historical window to obtain the residual mean and residual standard deviation, and multiple member sets are generated for extrapolation based on this. For each member, a displacement perturbation is superimposed on the translation field, and a residual perturbation is generated based on the residual mean and residual standard deviation. The perturbed translation field is used to extrapolate the current calibration rainfall field and superimpose the residual perturbation to obtain the member rainfall forecast field. Then, the member rainfall forecast fields within the preset cumulative time period are accumulated over time to obtain the cumulative precipitation field of each member.In step S600, for each early warning issuing unit, the cumulative precipitation field of each member is aggregated by area weighting within the coverage area of ​​the early warning issuing unit to obtain the cumulative precipitation value of the issuing unit. The probability of exceeding the threshold is determined by the proportion of the number of members whose cumulative precipitation value of the issuing unit is not less than the preset precipitation threshold to the total number of members. When the probability of exceeding the threshold is not less than the probability threshold within a preset number of consecutive early warning cycles, precipitation early warning information is output. When the uncertainty of the melting layer height exceeds the uncertainty threshold, the probability threshold is replaced with an increased probability threshold and the output early warning information is marked as a downgraded early warning, thereby suppressing false alarms when the melting layer identification is unstable and the credibility of bright band correction is reduced.

[0029] Understandably, the bright band of the melting layer is considered a key error source causing overestimation of rainfall intensity and threshold fluctuations. The approach of "determining the range of melting layer height using quasi-vertical profiles and outputting height uncertainty—weighted bright band correction with uncertainty constraints" ensures that the correction strength is consistent with the identification stability, reducing systematic overestimation under layered or mixed precipitation conditions from the source. Simultaneously, using rain gauge observations as a closed-loop benchmark, reflectivity and rainfall intensity conversion parameters are updated online according to precipitation type, and the spatial deviation field is solved to calibrate the radar rainfall intensity field, enabling the calibrated rainfall intensity field to dynamically converge with process evolution and regional differences. Furthermore, the calibrated rainfall intensity field is used to construct a translational field and combined with historical statistics of the residual field to generate a multi-member set extrapolation. Warnings are triggered by continuously satisfying thresholds with an over-threshold probability, improving the temporal stability and spatial consistency of warning issuance. A degradation strategy triggered by melting layer height uncertainty explicitly transmits observational uncertainty to the issuance strategy, thus achieving an interpretable trade-off between accuracy and robustness and improving practical operational usability.

[0030] In some embodiments of this application, when acquiring weather radar volume scan radar base data and rain gauge observation data and performing time alignment, and then gridding the weather radar volume scan radar base data to obtain grid echo data, the process includes: using the alignment time corresponding to the warning cycle as a unified time reference, aggregating the rain gauge observation data according to the preset statistical time period corresponding to the alignment time to obtain aligned rain gauge observation data; selecting the weather radar volume scan radar base data whose time is closest to the alignment time as the aligned radar base data; calculating the beam height and obstruction rate corresponding to each radar gate according to the radar site parameters and terrain obstruction data; mapping the aligned radar base data to a preset horizontal grid, and weighting and synthesizing the radar gate echoes falling into the grid within each grid according to distance, beam height, and obstruction rate to obtain the grid echo value of the grid, and using the beam height and obstruction rate of the grid as the associated attributes of the grid echo data.

[0031] Specifically, step S100 is used to convert the volume-scanned radar base data of the weather radar and the observation data of the rain gauge into a unified data base that can directly participate in subsequent profile construction, brightness band correction, and rainfall intensity inversion. Specifically, "alignment time" refers to the target time selected in each warning cycle, which is used as a unified time reference for radar data and rain gauge data; "preset statistical period" refers to the rain gauge time aggregation interval set around the alignment time, which is used to convert the discrete observations of the rain gauge into precipitation or rainfall intensity characteristics that match the alignment time; "beam height" refers to the spatial height of the radar beam relative to the ground at a certain radar gate, which is used to characterize the vertical position detected by the radar gate; "obstruction rate" refers to the characterization of the degree of radar energy attenuation caused by terrain or building obstruction, which is used to reflect the availability of the radar gate observation; "preset horizontal grid" refers to a two-dimensional spatial grid covering the effective detection range of the radar, which is used to uniformly map the polar coordinate radar gate data to a planar grid coordinate system so that subsequent spatial calculations can be performed based on the grid. To avoid ambiguity, in this embodiment, "weighted synthesis" is defined as follows: within the same grid, multiple radar gate echoes falling into that grid are normalized and weighted by the product of range weight, beam height weight, and obstruction rate weight, thereby outputting a single grid echo value. At the same time, the weighted beam height and weighted obstruction rate participating in the synthesis within that grid are saved as associated attributes of the grid echo data for subsequent quality control and correction strategy invocation.

[0032] Specifically, taking a weather radar coverage area as an example, the warning cycle is consistent with the radar volume scan update. Within each warning cycle, the alignment time is determined as the end time of that cycle. A preset statistical period is set as a fixed time interval before the alignment time, used to perform time aggregation on rain gauge observation data to obtain aligned rain gauge observation data. The time aggregation involves summing the precipitation within the statistical period and converting it into rainfall intensity, and only rain gauge observation data with valid quality labels are selected for aggregation. Subsequently, for the weather radar volume scan radar base data, the volume scan with the time closest to the alignment time is selected from the data stream as the aligned radar base data. Based on radar site parameters and terrain obstruction data, the beam height and obstruction rate are calculated for each radar gate in the aligned radar base data: the beam height is determined based on the radar gate distance, corresponding elevation angle, and site altitude, and the obstruction rate is determined based on the radar beam pointing and terrain obstruction data. After completing the above calculations, the aligned radar base data is mapped to a preset horizontal grid: First, the spatial position of each radar gate is projected onto the horizontal grid coordinates, and the grid cell it falls into is determined; then, all radar gate echoes falling into each grid are collected, and the range weight, beam height weight, and obstruction rate weight are calculated respectively. A normalized weighted average is then performed on the radar gate echoes to obtain the grid echo value for that grid. Simultaneously, the beam heights of the participating radar gates are weighted equally to obtain the beam height of that grid, and the obstruction rates of the participating radar gates are weighted equally to obtain the obstruction rate of that grid. The beam height and obstruction rate are then output as associated attributes of the grid echo data. To ensure the usability of the results, this embodiment can also perform validity screening on the grids: when the number of participating radar gates in a grid is lower than a preset threshold or the obstruction rate of that grid exceeds an obstruction rate threshold, the grid is marked as a low-confidence grid, and its weight is reduced or it is removed from the calculation in subsequent steps.

[0033] Understandably, by using "rain gauge time aggregation with alignment time as a unified time reference" and "near-time selection of volume scan radar base data," radar observations and ground observations can be directly paired under the same temporal semantics, reducing systemic bias caused by time mismatch. By explicitly calculating and retaining beam height and occlusion rate as correlation attributes of grid echo data during the gridding process, subsequent profile construction and bright band correction can be constrained based on interpretable observation geometry and occlusion conditions, thereby improving the stability of melting layer identification, bright band correction, and rainfall intensity inversion. At the same time, by using weighted synthesis constrained by distance, beam height, and occlusion rate, the sensitivity of grid echo values ​​to distant high-altitude and highly occluded low-confidence observations is reduced, enhancing the usability and consistency of grid echo data and providing a more reliable data foundation for subsequent zonal inversion, station residual calibration, and ensemble extrapolation.

[0034] In some embodiments of this application, the construction of quasi-vertical profiles and the determination of the melting layer height range and melting layer height uncertainty include: selecting grid echo data with occlusion rates not exceeding an occlusion rate threshold within a preset distance ring, performing azimuth convergence by height layer to obtain robust echo statistics for each height layer and forming a quasi-vertical profile; determining candidate height intervals for echo enhancement zones within the quasi-vertical profiles, using the upper and lower boundaries of the candidate height intervals as the melting layer height range; determining the melting layer height uncertainty based on the height dispersion between different azimuth convergence results and the height fluctuation between adjacent warning cycles. The melting layer height uncertainty is also determined based on the difference between the maximum and minimum values ​​of the center height of the echo enhancement zone in different azimuth convergence results, and the absolute difference in the center height of the echo enhancement zone between adjacent warning cycles. The center height of the echo enhancement zone refers to the height corresponding to the peak value of the robust echo statistics within the candidate height interval.

[0035] In some embodiments of this application, when performing weighted brightness band correction on the echo within the melt layer based on the uncertainty of the melt layer height and obtaining corrected echo data, the following steps are included: A preset reference height range is selected above the melting layer height range, and a reference echo profile is determined based on the quasi-vertical profile of the reference height range. For each height layer within the melting layer height range, a brightness band correction is determined based on the difference between the quasi-vertical profile and the reference echo profile. A correction weight is determined based on the melting layer height uncertainty, and the brightness band correction is weighted before correcting the grid echo data within the melting layer height range to obtain corrected echo data. The larger the melting layer height uncertainty, the smaller the correction weight. Furthermore, when the melting layer height uncertainty exceeds a preset uncertainty threshold, the correction weight is limited to a preset minimum correction weight.

[0036] Specifically, this embodiment is used to identify the melting layer based on grid echo data and to controllably correct bright bands, thereby reducing the risk of overestimation of rainfall intensity under layered or mixed precipitation conditions. Specifically, the "preset distance ring zone" is used to define the sampling area of ​​the quasi-vertical profile, preferably avoiding near-field clutter areas and sparsely sampled areas at distant upper altitudes; "azimuth convergence" refers to the statistical convergence of grid echo data from different azimuths within the same height layer to eliminate local anomalies and obtain representative height-layer echo levels; the "robust echo statistics" preferably takes the median of the grid echo values ​​participating in the convergence at that height layer, thereby reducing the impact of strong convective cores or isolated noise on the profile; the "candidate height interval of the echo enhancement zone" refers to a continuous height interval in the quasi-vertical profile where the echo enhances with height and forms a peak, corresponding to the height zone where the melting layer may exist; the "center height of the echo enhancement zone" is limited to the height corresponding to the peak value of the robust echo statistics within the candidate height interval; the "melting layer height uncertainty" is used to quantify the stability of melting layer height identification. The determination rules are as follows: spatial inconsistency is characterized by the difference between the maximum and minimum values ​​of the center height in the convergence results of different orientations; temporal instability is characterized by the absolute difference of the center height between adjacent warning cycles; and the uncertainty of the melting layer height is determined based on these rules. The "preset reference height range" refers to the height band above the melting layer height range where the echo pattern is relatively stable, used to construct a reference echo profile unaffected by the bright band. The "bright band correction amount" refers to the portion of the quasi-vertical profile within the melting layer height range that exceeds the reference echo profile, used to characterize the additional enhancement caused by the bright band. The "correction weight" is used to control the correction intensity, and it is explicitly stated that the greater the uncertainty of the melting layer height, the smaller the correction weight. When the uncertainty of the melting layer height exceeds the preset uncertainty threshold, the correction weight is limited to the preset minimum correction weight to avoid over-correction under unstable identification conditions.

[0037] Specifically, using the grid echo data output in step S100 as input, the following processing is performed in each warning cycle: First, grid echo data with occlusion rates not exceeding an occlusion rate threshold are filtered within a preset distance ring, and a height layer set is established according to height, where the height originates from the beam height correlation attribute of the grid echo data; then, azimuth convergence is performed on each height layer, that is, the grid echo values ​​covered by the height layer in different azimuths are collected and the median is calculated as the robust echo statistics value of the height layer, and concatenated in height order to form a quasi-vertical profile. Subsequently, candidate height intervals for echo enhancement zones are searched in the quasi-vertical profile: preferably, the peak height of the profile is determined first, and then the height is expanded upward and downward until the robust echo statistics value drops to a certain proportion of the peak value or drops to a preset echo threshold, thereby obtaining continuous candidate height intervals, and the upper and lower boundaries of the candidate height intervals are used as the melting layer height range; at the same time, the height corresponding to the peak value of the robust echo statistics value within the candidate height interval is determined as the center height of the echo enhancement zone. To obtain the uncertainty of the melting layer height, within the same warning cycle, the preset distance ring is divided into multiple azimuth sectors according to azimuth. The above-mentioned azimuth convergence and center height extraction process is repeated for each azimuth sector to obtain multiple center heights. The difference between the maximum and minimum values ​​is used as the spatial inconsistency index. Then, the center height of the previous warning cycle is read and the absolute difference between it and the center height of the current warning cycle is calculated as the temporal instability index. The spatial inconsistency index and the temporal instability index are used together to determine the uncertainty of the melting layer height. For example, the larger of the two can be used as the uncertainty of the melting layer height, so that the uncertainty is sensitive to the worst-case scenario. After the melting layer is identified, brightness band correction is performed: a preset reference height range is selected above the melting layer height range, and a reference echo profile is determined based on the quasi-vertical profile of this range. The reference echo profile is preferably a representative value of the robust echo statistics within the reference height range and is kept consistent along the height. For each height layer within the melting layer height range, the difference between the robust echo statistics of this height layer and the reference echo profile is calculated to obtain the brightness band correction amount. The correction weight is determined according to the melting layer height uncertainty, and the brightness band correction amount is multiplied by the correction weight and used as the correction amount for this height layer. This correction is applied to the grid echo data within the melting layer height range to obtain the corrected echo data. When the melting layer height uncertainty is greater than a preset uncertainty threshold, the correction weight is limited to a preset minimum correction weight to ensure that the correction remains conservative and controllable.

[0038] Understandably, by selecting effective grids within a preset distance ring using an occlusion rate threshold and forming a quasi-vertical profile using the median, the profile becomes insensitive to local strong echo kernels and random noise, thereby improving the repeatability of melting layer identification. By decomposing the uncertainty of melting layer height into spatial inconsistency and temporal instability and quantifying it using unambiguous quantities such as "maximum-minimum difference" and "absolute difference between adjacent time intervals," the uncertainty becomes calculable, verifiable, and directly usable for correction control. Furthermore, through a chain mechanism of "reference echo profile—bright band correction amount—uncertainty constraint correction weight," the bright band correction is transformed from a fixed empirical rule into a correction strategy that adaptively adjusts according to the stability of identification. This effectively suppresses the reflectivity enhancement and rainfall intensity overestimation caused by the bright band when the melting layer is stable, and automatically converges to conservative correction when the melting layer height identification is unstable, reducing the risk of introducing new systematic errors through overcorrection. This provides more reliable corrected echo input for subsequent zonal inversion, site calibration, and probabilistic early warning triggering.

[0039] In some embodiments of this application, when determining precipitation type zoning based on grid echo data and obtaining radar rainfall intensity field based on corrected echo data, the process includes: calculating the echo top height index, echo texture index, and melting layer overlap index for each grid based on the grid echo data; wherein, the echo top height index is the highest height above a preset echo threshold, the echo texture index is the standard deviation of the grid echo values ​​within a preset neighborhood, and the melting layer overlap index is the overlap height where the echo top height index falls within the melting layer height range. When the echo top height index is greater than the preset top height threshold and the echo texture index is greater than the preset texture threshold, the corresponding grid is assigned to the convective precipitation zone; when the melt layer overlap index is greater than the preset overlap threshold and the echo texture index is not greater than the preset texture threshold, the corresponding grid is assigned to the stratiform precipitation zone; the remaining grids are assigned to the mixed precipitation zone; the corrected echo data is extracted from the preset inversion height layer below the melt layer height range to obtain the inversion echo value, and the reflectivity and rainfall intensity conversion parameters corresponding to the convective precipitation zone, stratiform precipitation zone, and mixed precipitation zone are used respectively to convert the inversion echo value into rainfall intensity value to form a radar rainfall intensity field.

[0040] Specifically, this embodiment is used to explicitly introduce echo morphology differences into the rainfall intensity inversion process after bright band correction, thereby avoiding systematic bias caused by "the same set of inversion parameters covering all precipitation patterns". Specifically, the "echo top height index" is used to characterize the vertical development intensity of the echo, and its value is limited to the highest beam height above the preset echo threshold in the vertical column at the same horizontal grid position; the "echo texture index" is used to characterize the spatial roughness and small-scale undulation of the echo, and its value is limited to the standard deviation of the grid echo values ​​within a preset neighborhood centered on the grid. In this embodiment, the range of the preset neighborhood is limited to a grid window of a fixed size; the "melting layer overlap index" is used to characterize the relative positional relationship between the echo top height and the melting layer, and its value is limited to the overlap height corresponding to when the echo top height index falls within the melting layer height range, otherwise it is zero. To avoid ambiguity, in this embodiment, "preset top height threshold, preset texture threshold, and preset overlap threshold" are all stored as configurable thresholds and remain consistent within the same warning cycle; "preset inversion height layer" is limited to a height layer located below the melting layer height range and strongly correlated with surface precipitation, used to extract inversion echo values ​​from corrected echo data; "reflectivity and rainfall intensity conversion parameters" are limited to a set of parameters configured or updated online for different precipitation type zones, used to convert inversion echo values ​​into rainfall intensity values, and in this embodiment, each zone corresponds to a fixed set of conversion parameters to ensure the determinism and reproducibility of zone inversion.

[0041] Specifically, using the grid echo data obtained in step S100, the melting layer height range obtained in step S200, and the corrected echo data as input, the following processing is performed on each horizontal grid in each early warning cycle: First, the grid echo values ​​corresponding to different beam heights are traversed in the vertical column of the grid, the set of heights higher than the preset echo threshold is filtered, and the highest beam height among them is taken as the echo top height index; then, a preset neighborhood window is determined with the grid as the center, the grid echo values ​​of each grid in the window are collected, and the standard deviation is calculated as the echo texture index; then, the echo top height index is compared with the melting layer height range. When the echo top height index is within the melting layer height range, the height difference between the echo top height index and the lower boundary of the melting layer height range is taken as the melting layer overlap index; otherwise, the melting layer overlap index is set to zero. After the index calculation is completed, the grid is partitioned according to threshold rules: when the echo top height index is greater than the preset top height threshold and the echo texture index is greater than the preset texture threshold, the grid is assigned to the convective precipitation partition; when the melt layer overlap index is greater than the preset overlap threshold and the echo texture index is not greater than the preset texture threshold, the grid is assigned to the stratiform precipitation partition; the remaining grids are assigned to the mixed precipitation partition. After partitioning, a preset inversion height layer is selected below the melt layer height range, and inversion echo values ​​are extracted from the corrected echo data at this inversion height layer; then, according to the precipitation type partition to which the grid belongs, the reflectivity and rainfall intensity conversion parameters corresponding to the partition are called to convert the inversion echo values ​​into rainfall intensity values, and the rainfall intensity values ​​of all grids are set to form the radar rainfall intensity field. To ensure the physical rationality of the output, in this embodiment, non-negativity constraint processing can be performed on the converted rainfall intensity values, that is, when the rainfall intensity value is less than zero, it is set to zero, and the input format is kept consistent with the subsequent station residual calculation.

[0042] Understandably, by combining the criteria of "echo top height index, echo texture index, and melting layer overlap index," the vertical development, spatial roughness, and relative positional relationship of the echo with the melting layer are incorporated into the precipitation type zoning, enabling differentiated processing of convective, stratiform, and mixed precipitation during the rainfall intensity inversion stage. Corrected echo data are extracted from a preset inversion height layer below the melting layer height range as the inversion echo value, avoiding the residual bright band effect caused by directly using echoes within the melting layer. Furthermore, by matching different reflectivity and rainfall intensity conversion parameters to different precipitation type zoning, the rainfall intensity inversion can simultaneously consider the strong vertical development of convective precipitation and the stable structural characteristics of stratiform precipitation, thereby reducing the systematic overestimation or underestimation of "single parameter inversion" in mixed precipitation and melting layer scenarios, and providing a more stable initial radar rainfall intensity field input for subsequent online zoning updates and spatial bias calibration based on rain gauges.

[0043] In some embodiments of this application, when obtaining station residual data based on rain gauge observation data and radar rainfall intensity field, and updating reflectivity and rainfall intensity conversion parameters online by region, the process includes: collecting the rain gauge observation rainfall intensity value of each rain gauge, and collecting the grid rainfall intensity value of the radar rainfall intensity field corresponding to the location of the rain gauge as the station radar rainfall intensity value; using the difference between the rain gauge observation rainfall intensity value and the station radar rainfall intensity value as the station residual data; aggregating the station residual data into the precipitation type region according to the precipitation type region to which the corresponding grid of the rain gauge belongs; within each precipitation type region, determining the station weight based on the shading rate and the uncertainty of the melting layer height, and recursively updating the reflectivity and rainfall intensity conversion parameters; and keeping the reflectivity and rainfall intensity conversion parameters of the previous warning cycle unchanged when the number of effective stations in the corresponding region does not reach the preset number threshold.

[0044] In some embodiments of this application, when obtaining a spatial deviation field based on station residual data and calibrating the spatial deviation field to obtain a calibrated rainfall intensity field, the process includes: mapping the station residual data to a preset horizontal grid; solving the spatial deviation field based on weighted interpolation with adjacent grid variation constraints, wherein the interpolation weights are determined according to the occlusion rate and the uncertainty of the melting layer height, and the corresponding station residual data is discarded when the occlusion rate exceeds the occlusion rate threshold; using the spatial deviation field as a grid-by-grid correction amount for the radar rainfall intensity field to compensate the radar rainfall intensity field to obtain the calibrated rainfall intensity field, and limiting the compensated grid rainfall intensity value to not less than zero.

[0045] Specifically, this embodiment uses rain gauge observations as an external "real benchmark" to introduce into the radar rainfall intensity inversion link, achieving online adaptive and spatial consistency calibration of zoning parameters. Specifically, "rain gauge observation rainfall intensity value" refers to the rainfall intensity representation obtained by converting precipitation from the rain gauge within a preset statistical period, with the statistical period aligned with the warning cycle; "station radar rainfall intensity value" refers to the grid rainfall intensity value read from the grid corresponding to the rain gauge location in the radar rainfall intensity field. If the rain gauge location falls near the grid boundary, it can be optionally implemented in the specification using distance-weighted synthesis of rainfall intensity values ​​from multiple nearby grids, but this embodiment uses single-grid reading as the definitive implementation; "station residual data" is limited to the difference between the rain gauge observation rainfall intensity value and the station radar rainfall intensity value, used to characterize the systematic deviation of radar rainfall intensity at the station; "zoning online update" refers to collecting the station residual data separately according to the precipitation type zoning of the corresponding grid of the rain gauge and updating it in each zoning. The reflectivity and rainfall intensity conversion parameters are recursively updated. "Site weight" is used to characterize the contribution of site residuals to parameter updates and interpolation solutions. In this embodiment, the site weight is determined by the shading rate and the uncertainty of the melting layer height. The larger the shading rate or the greater the uncertainty of the melting layer height, the smaller the site weight. "Number of effective sites" is limited to the number of rain gauges with effective quality identification within the current sliding time window that have not been eliminated by the shading rate threshold. "Spatial bias field" refers to the grid-by-grid correction field obtained by interpolating discrete site residuals on a preset horizontal grid. "Weighted interpolation with adjacent grid variation constraints" is limited to applying smoothing constraints to the variation amplitude of adjacent grid corrections while weighting the fitting of site residuals, thereby suppressing local spikes caused by sparse sites or outliers.

[0046] Specifically, taking a network of rain gauge stations within a radar coverage area as an example, in each warning cycle, rain gauge station observation data with valid quality identifiers are first selected within a sliding time window and aggregated to obtain the observed rainfall intensity value of each rain gauge station; then, the grid rainfall intensity value of the grid corresponding to the location of each rain gauge station is read from the radar rainfall intensity field as the station radar rainfall intensity value, and the difference between the observed rainfall intensity value of the rain gauge station and the station radar rainfall intensity value is calculated as the station residual data. Next, based on the precipitation type partition to which the corresponding grid of the rain gauge belongs, the station residual data are collected into the corresponding partition, and the reflectance and rainfall intensity conversion parameters are recursively updated within each partition. Specifically, a station weight is calculated for each station, which is determined based on the occlusion rate of the grid corresponding to that station and the uncertainty of the melting layer height in the current warning cycle. Within the partition, the station residuals are weighted and aggregated using the station weights and used to update the reflectance and rainfall intensity conversion parameters of that partition. The updated parameters are limited to a preset parameter range to avoid abnormal jumps. When the number of valid stations in a certain partition does not reach a preset threshold, the reflectance and rainfall intensity conversion parameters of the previous warning cycle are kept unchanged to avoid parameter drift under insufficient sample conditions. The spatial calibration phase then begins: the station residual data is mapped to a preset horizontal grid, station residual data with occlusion rates exceeding the occlusion rate threshold are removed, and the spatial bias field is solved based on weighted interpolation with adjacent grid variation constraints, where the interpolation weights are consistent with the station weights; after obtaining the spatial bias field, it is used as a grid-by-grid correction for the radar rainfall intensity field to compensate for the radar rainfall intensity field and obtain the calibrated rainfall intensity field, and non-negativity constraints are applied to the compensated grid rainfall intensity values ​​to ensure the physical interpretability of the calibrated rainfall intensity field and the numerical stability of subsequent processing.

[0047] Understandably, by constructing station residual data using the difference between rain intensity observed at rain gauge stations and radar rain intensity values, radar rain intensity errors are explicitly quantified and can be used for online learning. Through a mechanism of "regional residual aggregation by precipitation type—determining station weights based on uncertainty of shading rate and melting layer height—regional recursive updating of reflectivity and rain intensity conversion parameters," parameter updates can adaptively adjust to the error characteristics of different precipitation patterns and automatically reduce the impact of unreliable stations on updates when shading is strong or melting layer identification is unstable, thereby improving the robustness of parameter updates. Thirdly, by solving the spatial bias field through "weighted interpolation with adjacent grid variation constraints" and compensating the radar rain intensity field grid by grid, the station-scale bias correction is extended into a spatially continuous correction field, reducing local spikes and discontinuities caused by station sparsity. Under the premise of ensuring non-negative constraints, the spatial consistency and availability of the calibrated rain intensity field are improved, providing more reliable input for subsequent translational extrapolation, residual statistical set extrapolation, and probability threshold-triggered early warning.

[0048] In some embodiments of this application, when determining the translation field and obtaining the residual field based on the calibration rainfall intensity fields of adjacent time periods, and generating a multi-member set extrapolation based on the historical statistics of the translation field and the residual field to obtain the cumulative precipitation field of each member, the process includes: taking the calibration rainfall intensity fields of two adjacent time periods as input, searching for displacement grid by grid within a preset matching window and a preset search range, so as to minimize the sum of the absolute values ​​of the differences between the calibration rainfall intensity values ​​of the two time periods within the matching window, thereby obtaining the translation field; extrapolating the calibration rainfall intensity field of the previous time period according to the translation field to obtain the extrapolated translation rainfall intensity field, and using the calibration rainfall intensity of the current time period as the basis for the extrapolation. The difference between the field and the extrapolated rainfall intensity field determines the residual field. Within a preset historical window, the residual field is statistically analyzed to obtain the mean and standard deviation of the residuals for each grid. Multiple member sets are extrapolated based on a preset perturbation sequence. For each member, a residual perturbation is generated based on the mean and standard deviation of the residuals, and a displacement perturbation is applied to the extrapolated field. The perturbed extrapolated field is used to extrapolate the current time-based calibrated rainfall intensity field, and the corresponding residual perturbation is superimposed to obtain the member rainfall intensity forecast field. The member rainfall intensity forecast fields within a preset cumulative time period are accumulated over time to obtain the cumulative precipitation field for each member. The amplitudes of the displacement perturbation and residual perturbation decrease as the shading rate increases or the uncertainty of the melt layer height increases.

[0049] Specifically, this embodiment is used to simultaneously characterize "translational uncertainty" and "intensity evolution uncertainty" during the short-term near-term extrapolation stage, in order to reduce the error accumulation of single-path extrapolation under conditions of local echo generation, abrupt intensity changes, or rapid morphological changes. Specifically, the "preset matching window" refers to a local neighborhood window centered on the current grid, used to compare the local morphological consistency of the calibrated rainfall intensity fields of two adjacent time periods; the "preset search range" refers to the set of displacement candidates allowed for searching, used to find the best matching displacement within a certain radius or grid distance; "grid-by-grid search for displacement" is limited to traversing displacement candidates within the search range for each target grid, and selecting the displacement with the minimum cost using the sum of the absolute values ​​of the differences between the calibrated rainfall intensity values ​​of two time periods within the matching window as the cost function, thus obtaining the displacement result of that grid, thereby forming the translational field; the "residual field" is limited to the difference between the calibrated rainfall intensity field of the current time period and the extrapolated rainfall intensity field obtained by extrapolating the calibrated rainfall intensity field of the previous time period according to the translational field, used to characterize the intensity increases and decreases and morphological changes that the translational model cannot explain; the "preset history window" is limited to a set of recent... The time window consisting of the early warning cycle is used to perform grid-by-grid statistics on the residual field to obtain the residual mean and residual standard deviation; the "preset perturbation sequence" is limited to a pre-stored set of numerical sequences used to generate perturbations of set members, and each member corresponds to a fixed sequence index, thereby ensuring that the set generation process can be repeated under the same input conditions; "displacement perturbation" refers to the additional perturbation applied to the displacement result of the translation field to characterize the motion field estimation error, and "residual perturbation" refers to the additional perturbation generated in each grid based on the residual mean and residual standard deviation to characterize the intensity evolution error; and in order to reduce the adverse effects of low-confidence observations on extrapolation, in this embodiment the amplitudes of displacement perturbation and residual perturbation decrease as the occlusion rate increases or the uncertainty of the melting layer height increases, thereby automatically converging to a more conservative extrapolation and smaller random perturbation when the occlusion is strong or the melting layer identification is unstable.

[0050] Specifically, using the calibrated rainfall intensity field of the current warning cycle and the calibrated rainfall intensity field of the previous warning cycle as input, a translation field is first constructed: For each grid, a preset matching window between the previous and current times is extracted with the grid as the center, and displacement candidates are tried one by one within a preset search range; for each displacement candidate, the matching window of the previous time is aligned to the matching window position of the current time according to the displacement candidate, the sum of the absolute values ​​of the differences between the calibrated rainfall intensity values ​​at the corresponding positions is calculated as the cost, and the displacement candidate with the smallest cost is selected as the displacement result of the grid. All grid displacement results are summarized to form the translation field. Subsequently, a translation extrapolated rainfall intensity field and a residual field are generated: the calibrated rainfall intensity field of the previous time is extrapolated according to the translation field to obtain the translation extrapolated rainfall intensity field, and the residual field is determined by the difference between the calibrated rainfall intensity field of the current time and the translation extrapolated rainfall intensity field. Next, the residual field is statistically analyzed grid by grid within a preset historical window to obtain the residual mean and residual standard deviation for each grid. The residual field at each time point within the historical window is calculated and included in the statistics using the same caliber. Then, a multi-member set is generated for extrapolation: a preset perturbation sequence is called to generate multiple members. For each member, residual perturbations are generated in each grid based on the residual mean and residual standard deviation, and displacement perturbations are applied to the translation field. The amplitudes of the displacement perturbation and residual perturbation are adjusted according to the occlusion rate of the grid and the uncertainty of the melting layer height in the current warning cycle. The larger the occlusion rate or the larger the uncertainty of the melting layer height, the smaller the perturbation amplitude. The lower limit of the perturbation amplitude can be preset to the minimum perturbation ratio to prevent the extrapolation from not diverging due to set collapse. For each member, the perturbed translation field is used to extrapolate the current time-time calibration rainfall intensity field, and the corresponding residual perturbation is superimposed to obtain the member rainfall intensity forecast field. Finally, the member rainfall intensity forecast fields within the preset cumulative time period are accumulated over time to obtain the cumulative precipitation field of each member. The time accumulation is based on the warning cycle period as the step size. The rainfall intensity forecast fields of each step are summed on the time axis and converted into cumulative precipitation values. Non-negative constraints are applied to the accumulation results to ensure physical rationality.

[0051] Understandably, by solving the translation field grid-by-grid with a local matching window and search range, motion estimation can adapt to the local differences in echoes from different regions, reducing structural errors caused by single global displacements. By constructing a residual field and obtaining the residual mean and residual standard deviation within a historical window, the "intensity evolution error" is explicitly separated from the translation error and quantified in statistical form, thus providing an interpretable source for ensemble perturbations. Furthermore, by generating multi-member ensemble extrapolation through dual channels of "displacement perturbation and residual perturbation," the extrapolation results are expanded from a single deterministic field to a set of possible results in a probabilistic sense, which can more robustly cover situations such as local echo generation, splitting and merging, and rapid enhancement and attenuation. At the same time, the uncertainty of the occlusion rate and melting layer height is used to adaptively converge the perturbation amplitude, suppressing ensemble diffusion under low-confidence observation conditions, avoiding excessive discrepancies and false triggers caused by unreliable inputs, and providing a stable and interpretable uncertainty characterization for subsequent early warning issuance based on over-threshold probability.

[0052] In some embodiments of this application, when obtaining the over-threshold probability of early warning issuance and outputting precipitation early warning information based on the cumulative precipitation field of each member, the process includes: for each early warning issuance unit, performing area-weighted aggregation of the cumulative precipitation values ​​of multiple grids covered by the early warning issuance unit in the cumulative precipitation field of each member to obtain the cumulative precipitation value of the issuance unit; determining the proportion of the number of members whose cumulative precipitation value of the issuance unit is not less than a preset precipitation threshold to the total number of members as the over-threshold probability of the early warning issuance unit; when the over-threshold probability is not less than the probability threshold within a consecutive preset number of early warning cycles, outputting precipitation early warning information; when the uncertainty of the melting layer height exceeds the uncertainty threshold, replacing the probability threshold with an increased probability threshold, and marking the output precipitation early warning information as a downgraded early warning.

[0053] Specifically, this embodiment is used to convert the extrapolation results of a multi-member set into a publishable early warning conclusion, and suppresses threshold jitter through "probability triggering and continuity constraints," while explicitly transmitting the uncertainty of melting layer identification to the release strategy. Specifically, the "early warning release unit" refers to the release spatial unit of the early warning product, preferably an administrative division grid, business partition, or regularized raster area, whose coverage area corresponds to a preset horizontal grid; "area-weighted aggregation" refers to weighting and summing the cumulative precipitation values ​​of multiple grids covered by the early warning release unit according to the proportion of the grid's coverage area within the early warning release unit, thereby obtaining the cumulative precipitation value of the release unit. When a grid completely falls within the early warning release unit, its coverage area proportion is one; when a grid intersects with the boundary of the early warning release unit, its coverage area proportion is the ratio of the overlapping area to the grid area; "preset precipitation threshold" refers to the cumulative precipitation threshold used to determine whether the release unit meets the early warning conditions; "exceeding threshold probability" is limited to being issued among all members. The proportion of members whose cumulative precipitation value is not less than the preset precipitation threshold to the total number of members; "Probability threshold" is the probability threshold used to trigger the warning; "Number of consecutive preset times" is used to constrain the triggering to meet the condition in multiple consecutive warning cycles, thereby reducing false triggering caused by single noise; "Increased probability threshold" refers to a stricter triggering threshold used when the uncertainty of the melting layer height exceeds the uncertainty threshold. Its value can be preset and stored as a configuration item in the specification, forming a one-to-one correspondence with the original probability threshold to ensure the determinism and reproducibility of the release strategy; "Downgraded warning" means that the warning information output under the above uncertainty conditions is marked as a lower confidence level or a lower release level, used to indicate that the warning conclusion has higher uncertainty.

[0054] Specifically, taking the cumulative precipitation field of each member obtained in step S500 as input, a set of early warning release units is first established, and the coverage relationship between each early warning release unit and the preset horizontal grid and the coverage area ratio of each grid are pre-stored. For each early warning cycle, for each early warning release unit and each member, the cumulative precipitation values ​​of multiple grids falling within the coverage area of ​​the early warning release unit in the member's cumulative precipitation field are read, and area-weighted aggregation is performed according to the corresponding coverage area ratio to obtain the cumulative precipitation value of the release unit corresponding to the member; then it is determined whether the cumulative precipitation value of the release unit is not less than the preset precipitation threshold, and the determination results of all members are counted to obtain the number of members not less than the preset precipitation threshold; the ratio of the number of members to the total number of members is determined as the over-threshold probability of the early warning release unit in the current early warning cycle. Then, the trigger determination is entered: an over-threshold probability sequence updated according to the early warning cycle is maintained for each early warning release unit. When the sequence is not less than the probability threshold in a consecutive preset number of early warning cycles, the precipitation early warning information of the early warning release unit is output. Simultaneously, the uncertainty of the melting layer height in the current warning cycle is read. When the uncertainty exceeds the uncertainty threshold, the probability threshold used for triggering the judgment is replaced with an increased probability threshold. After the continuity triggering condition is met, precipitation warning information is output with an added downgrade warning flag. When the uncertainty of the melting layer height does not exceed the uncertainty threshold, regular precipitation warning information is output according to the original probability threshold. To ensure output consistency, in this embodiment, the same warning issuing unit outputs only one warning status within the same warning cycle. Furthermore, the output of a non-issuance or cancellation status when the triggering condition is not met is defined separately by business rules and uniformly constrained in the specification.

[0055] Understandably, by using area-weighted aggregation of cumulative precipitation for members at the early warning release unit scale, the early warning judgment is aligned with the operational release space, avoiding excessive influence of single-grid extreme values ​​on regional early warnings. By defining the over-threshold probability based on the proportion of members, the uncertainty of ensemble extrapolation is naturally mapped into an interpretable trigger quantity, thereby improving the robustness of early warnings without relying on a single extrapolation result. By constraining the probability threshold trigger with "continuous preset number of times," the early warning jitter and false triggers caused by short-term noise, sudden changes in local echoes, or instantaneous model biases are reduced. At the same time, the uncertainty of the melting layer height is used as a gating quantity for the release strategy. When the uncertainty increases, the probability threshold is automatically raised and a downgraded early warning flag is output, enabling the early warning system to converge to a more conservative and interpretable release strategy when observation conditions are unstable, thereby reducing the cost of false alarms and improving the credibility and operational availability of early warning products.

[0056] In summary, this application ensures accurate registration between the radar field and ground observations through time alignment and gridding of volume scan radar and rain gauges. It constructs quasi-vertical profiles using gridded echo data and explicitly outputs the range and uncertainty of the melting layer height. This transforms bright band correction from a fixed empirical correction into a reliable correction adaptively weighted by uncertainty, reducing the tendency for overestimation and false alarms in rainfall intensity retrieval under layered or mixed precipitation from the source. Simultaneously, it introduces precipitation type zoning and updates reflectivity and rainfall intensity conversion parameters online within each zone. Furthermore, it utilizes station residuals to generate a spatial bias field to spatially calibrate the radar rainfall intensity field, enabling the calibrated rainfall intensity field to dynamically converge to a closer approximation based on regional biases, occlusion effects, and process evolution. The system reduces systematic bias between different regions by accurately measuring the actual ground rainfall level; it determines the translational field by calibrating the rainfall intensity field at adjacent time intervals and generates a multi-member set extrapolation by combining the historical statistics of the residual field; it calculates the threshold exceedance probability through the cumulative precipitation field of the multi-member set and adopts a triggering rule that continuously meets the probability threshold, effectively smoothing the threshold jitter caused by target splitting and merging, local nascent development, or short-term noise, and improving the temporal stability and spatial consistency of the warning; when the uncertainty of the melting layer height exceeds the uncertainty threshold, it automatically outputs a downgraded warning, avoiding the forced issuance of a high-confidence warning under the condition that the brightness band height is difficult to identify stably or the correction confidence is reduced, thus reducing the cost of false alarms and improving the traceability of the warning product.

[0057] Based on another preferred embodiment described above, see [link to preferred embodiment]. Figure 2 As shown, this embodiment provides a precipitation warning system based on weather radar, used to apply the above-mentioned precipitation warning method based on weather radar, including: The acquisition unit is configured to acquire volume scan radar base data of weather radar and observation data of rain gauge station and perform time alignment, and to grid the volume scan radar base data of weather radar to obtain grid echo data. The correction unit is configured to construct a quasi-vertical profile based on grid echo data, determine the range of the melt layer height and the uncertainty of the melt layer height, and perform weighted brightness band correction on the echoes within the melt layer according to the uncertainty of the melt layer height to obtain corrected echo data. The inversion unit is configured to determine precipitation type zoning based on grid echo data and to invert radar rainfall intensity field based on corrected echo data; The calibration unit is configured to obtain station residual data based on rain gauge observation data and radar rainfall intensity field within a sliding time window, perform online updates of reflectivity and rainfall intensity conversion parameters in partitions based on station residual data, and obtain spatial deviation field, and calibrate the calibrated rainfall intensity field using spatial deviation field calibration. The processing unit is configured to determine the translation field and obtain the residual field based on the calibration rainfall intensity field of adjacent time periods, and generate a multi-member set extrapolation based on the historical statistics of the translation field and the residual field to obtain the cumulative precipitation field of each member. The early warning unit is configured to obtain the over-threshold probability of early warning issuance based on the cumulative precipitation field of each member. When the over-threshold probability continuously meets the probability threshold, it outputs precipitation early warning information, and when the uncertainty of the melting layer height exceeds the uncertainty threshold, it outputs a downgraded early warning.

[0058] Understandably, by aligning the volume-scanning radar with the rain gauges and using gridding, the radar field and ground observations can be accurately registered. A quasi-vertical profile is constructed using gridded echo data, and the range and uncertainty of the melting layer height are explicitly output. This makes the bright band correction no longer a fixed empirical correction but a reliable correction adaptively weighted by uncertainty, reducing the tendency for overestimation and false alarms in rainfall intensity retrieval under layered or mixed precipitation from the source. Simultaneously, precipitation type zoning is introduced, and reflectivity and rainfall intensity conversion parameters are updated online within each zone. Then, a spatial bias field is generated using station residuals to spatially calibrate the radar rainfall intensity field, enabling the calibrated rainfall intensity field to dynamically converge to a more accurate level as regional biases, occlusion effects, and process evolution occur. The level of near-surface actual rainfall is used to reduce system bias between different regions; the translation field is determined by calibrating the rainfall intensity field of adjacent time intervals and combined with the historical statistics of the residual field to generate a multi-member set extrapolation; the probability of exceeding the threshold is calculated by the cumulative precipitation field of the multi-member set and the triggering rule of continuously satisfying the probability threshold is adopted to effectively smooth the threshold jitter caused by target splitting and merging, local new formation and development or short-term noise, and improve the temporal stability and spatial consistency of the warning; when the uncertainty of the melting layer height exceeds the uncertainty threshold, a downgraded warning is automatically output, avoiding the forced issuance of a high-confidence warning under the condition that the brightness band height is difficult to identify stably or the correction confidence is reduced, reducing the cost of false alarms and improving the traceability of the warning product.

[0059] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A precipitation early warning method based on weather radar, characterized in that, include: Acquire volume scan radar base data from weather radar and rain gauge observation data and perform time alignment; then grid the volume scan radar base data from weather radar to obtain grid echo data. Based on the grid echo data, a quasi-vertical profile is constructed to determine the range of the melt layer height and the uncertainty of the melt layer height. Then, the echoes within the melt layer are corrected by weighted brightness band based on the uncertainty of the melt layer height to obtain corrected echo data. Precipitation type zones are determined based on the grid echo data, and radar rainfall intensity fields are obtained by inversion based on the corrected echo data. Within a sliding time window, station residual data is obtained based on the rain gauge observation data and the radar rainfall field. Based on the station residual data, reflectivity and rainfall intensity conversion parameters are updated online in zones to obtain a spatial deviation field. The calibrated rainfall field is obtained by calibrating the spatial deviation field. The translation field is determined based on the calibration rainfall intensity field of adjacent time periods, and the residual field is obtained. Based on the translation field and the historical statistics of the residual field, a multi-member set is generated for extrapolation, and the cumulative precipitation field of each member is obtained. Based on the cumulative precipitation field of each member, the probability of exceeding the threshold for issuing an early warning is obtained. When the probability of exceeding the threshold continuously meets the probability threshold, precipitation early warning information is output, and when the uncertainty of the melting layer height exceeds the uncertainty threshold, a downgraded early warning is output.

2. The precipitation early warning method based on weather radar according to claim 1, characterized in that, When acquiring volume-scanned radar base data from weather radar and rain gauge observation data and performing time alignment, and then gridding the weather radar base data to obtain gridded echo data, the process includes: Using the alignment time corresponding to the early warning cycle as a unified time reference, the rain gauge observation data is aggregated according to the preset statistical time period corresponding to the alignment time to obtain aligned rain gauge observation data; the weather radar volume scan radar base data with the time closest to the alignment time is selected as aligned radar base data; based on the radar site parameters and terrain obstruction data, the beam height and obstruction rate corresponding to each radar gate are calculated; the aligned radar base data is mapped to a preset horizontal grid, and the radar gate echoes falling into the grid are weighted and synthesized according to distance, beam height and obstruction rate in each grid to obtain the grid echo value of the grid, and the beam height and obstruction rate of the grid are used as the correlation attributes of the grid echo data.

3. The precipitation early warning method based on weather radar according to claim 2, characterized in that, When constructing a quasi-vertical profile and determining the range and uncertainty of the melt layer height, the following steps are included: Within a preset distance ring, grid echo data with an occlusion rate not exceeding an occlusion rate threshold are selected. Azimuth convergence is performed at different height levels to obtain robust echo statistics for each height level and form a quasi-vertical profile. Candidate height intervals for echo enhancement zones are determined within these quasi-vertical profiles, and the upper and lower boundaries of these candidate height intervals are used as the range of the melting layer height. Based on the height dispersion between different azimuth convergence results and the height fluctuation between adjacent warning cycles, the uncertainty of the melting layer height is determined.

4. The precipitation early warning method based on weather radar according to claim 3, characterized in that, When performing weighted brightness band correction on the echo within the melt layer based on the uncertainty of the melt layer height and obtaining the corrected echo data, the following steps are included: A preset reference height range is selected above the melt layer height range, and a reference echo profile is determined based on the quasi-vertical profile of the reference height range; for each height layer within the melt layer height range, a bright band correction amount is determined based on the difference between the quasi-vertical profile and the reference echo profile; a correction weight is determined based on the melt layer height uncertainty, and the bright band correction amount is weighted to correct the grid echo data within the melt layer height range, thereby obtaining the corrected echo data.

5. The precipitation early warning method based on weather radar according to claim 1, characterized in that, When determining precipitation type zones based on the grid echo data and retrieving the radar rainfall intensity field based on the corrected echo data, the process includes: Based on the grid echo data, the echo top height index, echo texture index and melt layer overlap index of each grid are calculated; When the echo top height index is greater than the preset top height threshold and the echo texture index is greater than the preset texture threshold, the corresponding grid will be assigned to the convective precipitation zone. When the melting layer overlap index is greater than the preset overlap threshold and the echo texture index is not greater than the preset texture threshold, the corresponding grid is assigned to the layered precipitation zone. The remaining grid cells are assigned to the mixed precipitation zone; The corrected echo data is extracted from the preset inversion height layer below the melting layer height range to obtain the inversion echo value. The inversion echo value is then converted into a rainfall intensity value using the reflectivity and rainfall intensity conversion parameters corresponding to the convective precipitation zone, the stratiform precipitation zone, and the mixed precipitation zone, respectively, to form the radar rainfall intensity field.

6. The precipitation early warning method based on weather radar according to claim 2, characterized in that, When obtaining station residual data based on the rain gauge observation data and radar rainfall intensity field, and updating the reflectivity and rainfall intensity conversion parameters online by region, the process includes: Collect the observed rainfall intensity value of each rain gauge station, and collect the grid rainfall intensity value of the radar rainfall intensity field corresponding to the location of the rain gauge station as the station radar rainfall intensity value. The difference between the observed rainfall intensity value at the rain gauge station and the radar rainfall intensity value at the station is used as the station residual data; Based on the precipitation type partition to which the corresponding grid of the rain gauge belongs, the residual data of the station are collected into the precipitation type partition; Within each precipitation type zone, the station weights are determined based on the shading rate and the uncertainty of the melting layer height. The reflectance and rainfall intensity conversion parameters are then updated recursively. When the number of valid stations in the corresponding zone does not reach the preset threshold, the reflectance and rainfall intensity conversion parameters of the previous warning cycle remain unchanged.

7. The precipitation early warning method based on weather radar according to claim 6, characterized in that, When obtaining a spatial bias field based on the station residual data and calibrating the spatial bias field to obtain a calibrated rainfall intensity field, the following steps are included: The station residual data is mapped to the preset horizontal grid, and the spatial deviation field is solved by weighted interpolation with adjacent grid variation constraints. The interpolation weight is determined according to the occlusion rate and the uncertainty of the melting layer height. When the occlusion rate exceeds the occlusion rate threshold, the corresponding station residual data is removed. The spatial deviation field is used as a grid-by-grid correction amount for the radar rainfall intensity field to compensate the radar rainfall intensity field and obtain the calibrated rainfall intensity field. The compensated grid rainfall intensity value is limited to not less than zero.

8. The precipitation early warning method based on weather radar according to claim 1, characterized in that, When determining the translational field and obtaining the residual field based on the calibrated rainfall intensity fields of adjacent time periods, and generating a multi-member set extrapolated based on the historical statistics of the translational field and the residual field to obtain the cumulative precipitation field of each member, the process includes: Using the calibrated rainfall field of two adjacent time intervals as input, the displacement is searched grid by grid within a preset matching window and a preset search range to minimize the sum of the absolute values ​​of the differences between the calibrated rainfall values ​​of two time intervals within the matching window, thus obtaining the translation field. The calibration rainfall field of the previous time period is extrapolated by the translation field to obtain the extrapolated rainfall field, and the residual field is determined by the difference between the calibration rainfall field of the current time period and the extrapolated rainfall field. Within a preset historical window, the residual field is statistically analyzed to obtain the residual mean and residual standard deviation of each grid. Multiple member sets are extrapolated based on a preset perturbation sequence. For each member, a residual perturbation is generated according to the residual mean and residual standard deviation, and a displacement perturbation is applied to the translation field. The perturbed translation field is used to extrapolate the current time-time calibration rainfall intensity field, and the corresponding residual perturbation is superimposed to obtain the member rainfall intensity forecast field. The member rainfall intensity forecast fields within a preset cumulative time period are accumulated over time to obtain the cumulative precipitation field of each member.

9. The precipitation early warning method based on weather radar according to claim 1, characterized in that, When obtaining the probability of exceeding the threshold for issuing a precipitation warning based on the cumulative precipitation fields of each member and outputting precipitation warning information, the following are included: For each early warning issuing unit, the cumulative precipitation values ​​of multiple grids covered by the early warning issuing unit are aggregated in the cumulative precipitation field of each member by area weighting to obtain the cumulative precipitation value of the issuing unit; The proportion of members whose cumulative precipitation value is not less than a preset precipitation threshold in the early warning unit is determined as the over-threshold probability of the early warning unit. When the over-threshold probability is not less than the probability threshold within a preset number of consecutive early warning cycles, the precipitation early warning information is output. When the uncertainty of the melting layer height exceeds the uncertainty threshold, the probability threshold is replaced with an increased probability threshold, and the output precipitation early warning information is marked as a downgraded early warning.

10. A precipitation warning system based on weather radar, used to apply the precipitation warning method based on weather radar as described in any one of claims 1-9, characterized in that, include: The acquisition unit is configured to acquire volume scan radar base data of weather radar and observation data of rain gauge station and perform time alignment, and to grid the volume scan radar base data of weather radar to obtain grid echo data. The correction unit is configured to construct a quasi-vertical profile based on the grid echo data, determine the range of the melt layer height and the uncertainty of the melt layer height, and perform weighted brightness band correction on the echoes within the melt layer according to the uncertainty of the melt layer height to obtain corrected echo data. The inversion unit is configured to determine precipitation type partitions based on the grid echo data and to invert the radar rainfall intensity field based on the corrected echo data. The calibration unit is configured to obtain station residual data based on the rain gauge observation data and radar rainfall field within a sliding time window, perform partitioned online updates of reflectivity and rainfall intensity conversion parameters based on the station residual data and obtain a spatial deviation field, and calibrate the calibrated rainfall field using the spatial deviation field. The processing unit is configured to determine the translation field and obtain the residual field based on the calibration rainfall intensity field of adjacent time periods, and generate a multi-member set extrapolation based on the translation field and the historical statistics of the residual field to obtain the cumulative precipitation field of each member. The early warning unit is configured to obtain the over-threshold probability of early warning issuance based on the cumulative precipitation field of each member, output precipitation early warning information when the over-threshold probability continuously meets the probability threshold, and output a downgraded early warning when the uncertainty of the melting layer height exceeds the uncertainty threshold.

Images