A quantitative precipitation estimation and error correction method for X-band phased array radar

Abstract

This invention relates to the field of hydrological and meteorological monitoring and information processing technology, and solves the technical problems of insufficient stability and accuracy of quantitative precipitation (QPE) in X-band phased array dual-polarization precipitation radar under strong attenuation, bright band crossing, complex terrain and rapid volume scanning scenarios. In particular, it relates to a method for quantitative precipitation estimation and error correction of X-band phased array radar. The method takes polarization-gated priority path attenuation correction as the core constraint, takes multi-parameter fusion based on weather type-rainfall intensity segmentation-adaptive weighting of cross-correlation coefficient as the main line, and converges through multiplicative (hourly scale) + additive (minute scale) hierarchical deviation correction closed loop, and finally outputs high spatiotemporal resolution precipitation products with pixel uncertainty and quality mask, which are suitable for real-time water conservancy operations at the watershed level and urban watershed.

Description

Technical Field

[0001] This invention relates to the field of hydrological and meteorological monitoring and information processing technology, and in particular to a method for quantitative precipitation estimation and error correction using X-band phased array radar. Background Technology

[0002] Over the past decade, short-duration heavy rainfall events in small watersheds have become more frequent and extreme. Water conservancy operations have rapidly upgraded their spatiotemporal indicators for rainfall monitoring from "hourly and kilometer-level" to "minute-level and hundred-meter-level," requiring stable output and reliable peak value characterization within the rapid cycle of strong convection development and dissipation. While traditional ground-based rain gauges offer high precision and traceability in point measurements, their limitations in network density and spatial representativeness make them unable to capture small- to medium-scale convective systems with a scale of only a few kilometers and a lifespan of less than one hour. S-band weather radars offer wide coverage and are insensitive to attenuation, making them suitable for regional monitoring. However, their limited scanning cycle and low-altitude blind spot coverage make them unsuitable for high-frequency requirements such as flash flood warnings and urban flooding, which demand data on near-surface water condensate structure and time-varying intensity.

[0003] X-band phased array dual-polarization radar, relying on electronic scanning and a narrow beam, can achieve sub-minute volume scanning and high near-surface coverage under a low-elevation, multi-surface strategy. , , , , Isopolarization parameters provide a basis for droplet spectrum and phase state discrimination, and are therefore considered an important means of "near-range blind spot compensation" at the watershed level. However, the path attenuation caused by the short wavelength of the X-band is particularly significant under moderate to heavy rain conditions, and the antenna wet cover and system bias will further amplify the absolute quantitative error; the phased array is composed of a large number of T / R modules, and the gain and phase drift accumulate with time and temperature. System constants and Zero-bias stability has become a key constraint for long-term operation. The reduction in pulse samples during rapid volumetric scanning makes... More susceptible to noise, which in turn affects Stable differential estimation; complex terrain leads to widespread partial beam obstruction and partial filling, and the elevation difference caused by beam rise with distance, near-surface evaporation and wind shear introduce systematic differences between radar estimation and rainfall station measurements. Ground clutter, sea clutter, biological echoes and anomalous propagation (AP) frequently occur in nearshore, piedmont and urban agglomeration environments, forcing quality control to evolve from threshold-based to textured and polarization consistency constraints.

[0004] Existing technologies for quantitative precipitation estimation (QPE) can be summarized into three categories: First, single-parameter or fixed-type relationships (such as ZR or ZI) are simple to implement and have good real-time performance, but they are highly sensitive to droplet spectra and weather patterns, and their errors are significant under stratigraphic-convective transition or typhoon backgrounds; Second, dual-polarization hybrid algorithms typically use... Strong and stable rank Weighting is based on the principle of weak precipitation advantage, but the weights are mostly static experience or threshold switching, which is difficult to adapt to spatiotemporal heterogeneity and process evolution; thirdly, physical correction and geometric correction (such as ZPHI path attenuation correction, Zero-bias correction, VPR bright band correction, and partial occlusion and filling compensation based on DEM are relatively mature in engineering. However, they are prone to overcorrection or undercorrection in high-gradient X-band, wet mask, and bright band crossing scenarios, requiring gating strategies and sequential control linked to polarization reliability. To reduce scenario dependence, the industry often introduces bias correction and multi-source fusion (Kriging with External Drift, Bayesian fusion, etc.) in conjunction with rain gauges. However, most of them use single-layer multiplicative bias, which makes it difficult to simultaneously take into account minute-level transients and hour-level mean conservation; extreme quantiles (such as above P90) often have residual systematic underestimation. At the same time, many products only output the "best estimate" and lack pixel-level uncertainty and quality masking, making it difficult to serve threshold-sensitive early warning decisions.

[0005] The water conservancy project also raised two types of engineering constraints: first, end-to-end low latency and high throughput, requiring 1-5 minute updates and sub-minute processing links; second, cross-radar and cross-time period consistency, requiring products to be splicable and comparable, and to maintain aperture stability under equipment drift and environmental changes when multiple X-band and upstream S-band systems are networked. These constraints, combined with the physical characteristics of X-band phased arrays, make it difficult for the traditional paradigm of "segmented processing + experience-based weighting + single-layer deviation correction" to continuously meet the dual requirements of accuracy and stability under complex terrain and diverse weather conditions.

[0006] In summary, existing technologies generally lack a solution that couples and controls polarization consistency calibration, non-meteorological echo suppression, attenuation and vertical profile correction, terrain occlusion and filling compensation, multi-parameter fusion adaptively based on weather pattern and confidence level, hierarchical deviation correction with rainfall coordination, and multi-source fusion and uncertainty assessment within the same process. In particular, there is a lack of clear technical constraints on gating order, weighting mechanisms, and the hierarchical structure of the deviation field to ensure stable output of high spatiotemporal resolution precipitation products with reliable quality information even in scenarios with strong attenuation, significant bright bands, and complex terrain. Therefore, there is an urgent need to propose an adaptive QPE and error correction method for X-band phased array dual-polarization precipitation radar, realizing a closed loop of physical correction, statistical correction, multi-source fusion, and uncertainty assessment within a unified engineering framework, so as to meet the comprehensive requirements of accuracy, stability, and availability for watershed-level water conservancy monitoring and early warning. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a method for quantitative precipitation estimation and error correction using X-band phased array radar, which solves the technical problems of insufficient stability and accuracy of quantitative precipitation (QPE) in X-band phased array dual-polarization rain measurement radar under conditions of strong attenuation, bright band crossing, complex terrain, and rapid volume scanning.

[0008] To address the aforementioned technical problems, this invention provides a quantitative precipitation estimation and error correction method for X-band phased array radar. The method uses polarization-gated path attenuation correction as the core constraint, and adaptive weighting of multi-parameter fusion based on weather pattern, rainfall intensity segmentation, and cross-correlation coefficient as the main line. It achieves convergence through hierarchical deviation correction closed loop using a combination of multiplicative (hourly scale) and additive (minute scale) methods, ultimately outputting a high spatiotemporal resolution precipitation product with pixel uncertainty and quality masking. This method is suitable for real-time water conservancy operations at the basin level and in urban watersheds.

[0009] The method includes data access and gridding, polarization self-consistency calibration, non-meteorological echo suppression, attenuation and vertical structure correction, terrain shading and partial filling compensation, adaptive quantitative precipitation estimation, rainfall coordination hierarchical deviation correction, multi-source fusion and uncertainty assessment, and other processing steps, which are run online in this order. Specifically, it includes the following steps: S1. Data Access and Gridding: The dual-polarization volume scan observation parameters and spectral width of the X-band phased array radar are acquired and then gridded after temporal and spatial registration with ground rain gauge data, S-band weather radar precipitation products, and a digital elevation model (DEM). The dual-polarization volume scan observation parameters include horizontal reflectivity. Differential reflectivity Differential phase shift Differential phase shift rate Cross-correlation coefficient ; Preferably, the time step is no more than 60s, the horizontal grid resolution is 250–1000m, the vertical resolution is 100–250m, and the mapping of radar volume scan to CAPPI / 3D voxel grid is completed by using spherical-terrain geometry that takes into account the effective sphere radius.

[0010] It should be noted that this invention uniformly adopts the CAPPI height surface. The two-dimensional mesh is used as the working domain for processing and fusion, denoted as .in For 2D / 3D mesh domains, such as CAPPI height surfaces A 2D mesh domain. Pixel The spatial resolution is determined by the radial spacing / azimuth resolution / height plane spacing, for example, 500m × 500m, or consistent with the engineering settings. A pixel is denoted as... This represents a pixel (CAPPI or projected onto the ground, etc.) in a radar 2D / 3D grid, whose physical location can be represented as... (Polar coordinates), or (Cartesian projection).

[0011] in, Radar radial and azimuth, The target height (or CAPPI height surface); after geometric or refractive correction, mapped to Cartesian coordinates. ; It is a projection mapping after geometric or refractive correction, and the two correspond one-to-one.

[0012] Field quantity and mask: for any polarization parameter field Defined from the grid domain to the real number field The scalar field, namely: S2. Quality Control and System Calibration: In situations with high cross-correlation coefficients Furthermore, in stratiform, weak precipitation samples without strong convection, based on horizontal reflectance... Differential reflectivity With differential phase shift rate The polarization self-consistency relation is used to calibrate the radar system constants. That is, differential reflectivity is corrected using the polarization self-consistency relation. Zero bias and fine-tuning of horizontal reflectivity A system constant is used to ensure that the two parameters satisfy the consistency constraint under weak gradient conditions. Preferably, the reference differential reflectivity is used. The value is set to 0.0–0.2 dB, and a gradual correction is applied to the zero bias for different elevation angles or sectors.

[0013] S3. Non-meteorological echo suppression: Based on horizontal reflectivity With cross-correlation coefficient Texture quantity, differential phase shift The gradient, spectral width W, and terrain information are used to identify and shield ground clutter and anomalous propagation echoes, resulting in quality-controlled echo data.

[0014] Furthermore, non-meteorological echo suppression employs a supervised classification mask that integrates polarization and texture features, i.e., using pixel-based methods. The central window Calculate texture features , , Where N is an odd number and 3 ≤ N ≤ 11, that is: Together with spectral width and topographic elevation, they form a feature vector. ,Right now: in, For spectral width; This refers to the terrain elevation; Represents a cell Horizontal reflectance at a location (unit: dBZ); Represents a pixel Differential reflectance at a given location (unit: dB); Represents a cell The differential phase shift rate (unit: ° / km) at the differential phase shift point is determined by the differential phase shift rate. The derivative in the radial direction is obtained; Represents a cell The cross-correlation coefficient at the location (dimensionless, 0-1).

[0015] Furthermore, all observations (such as ) and derived texture features , , All at the pixel center definition. Refers to pixels Centered ( (Odd number of windows)

[0016] Preferably, the feature vector Input classifier Probability of obtaining meteorological targets and with threshold Generate a binary mask ,Right now: Based on this, pixel-level shielding is applied to the polarization parameters to obtain echo data with quality control. ,Right now: in, The polarization parameter field is defined.

[0017] In this embodiment, all derivatives or integrals involving "along a ray" (such as...) , All are in a fixed orientation and height. On the ray, with radial variable (i.e., radial distance) is calculated as the independent variable.

[0018] S4. Attenuation and Vertical Structure Correction: Using differential phase shift rate... With differential reflectivity Path attenuation backfilling was performed on the hybrid ZPHI model with common constraints. Zero-degree bright band (VPR) correction was implemented in combination with the height layer. The low-layer cover and shading backfilling were calculated based on the DEM to obtain a physically consistent reflectivity field.

[0019] Furthermore, the attenuation and vertical structure correction includes path attenuation correction and zero-degree layer bright band (VPR) correction, and is performed in a gated priority order. In step S4, the following is included: S41, along the fixed orientation and CAPPI height plane Radar rays, for differential phase shift First, perform a length of M Radial median filtering with (odd numbers) yields smooth phase shifts. Then, the differential phase shift rate is obtained by radial differentiation. ,Right now: in, Radial distance from the radar, in km; It is the azimuth angle; For CAPPI height surface.

[0020] S42. Define the path ratio attenuation coefficient. ,include: Only when gated through, at differential phase shift rate With differential reflectivity The combined constraint estimation, i.e.: in, Indicates the same ray The path integral variable on the radial distance Equivalence, that is ; The reference differential reflectance is set at 0.0–0.2 dB. , , All are calibrated constants (path ratio attenuation coefficient when dimensions and units are matched). (in dB / km), with an upper limit added if necessary. Inhibits HB dispersion.

[0021] In cross-correlation coefficient Or horizontal reflectance after quality control Gating regions below the threshold directly reduce the path ratio attenuation coefficient. The expression is: in, Cross-correlation coefficient The threshold value is 0.96-0.98; The threshold value for amplitude or signal-to-noise ratio, in dBZ, has a value of [value missing]. dBZ; S43, Based on the path ratio attenuation coefficient The path attenuation is integrally corrected to obtain the attenuation-corrected horizontal reflectance. And backfill to the pixel.

[0022] Preferably, the defined path ratio attenuation coefficient is... Substituting into the following formula completes the calculation of horizontal reflectivity. Path integral correction. That is: in, Indicates the position relative to the CAPPI elevation plane. rays Above, the horizontal reflectance was observed (without path attenuation correction); Indicates two journeys (i.e., round trip); Then on the CAPPI height surface Above using pixels The one-to-one correspondence is used to assign radial results to grid cells, expressed as: in, For pixels Horizontal reflectance; Two-way path attenuation; For pixels Observed horizontal reflectance; Radial distance from the radar (in km); It is the azimuth angle; For CAPPI height surface.

[0023] Furthermore, an upper limit can be set for the cumulative decay, and the differential phase shift rate can be updated in a single iteration. With differential reflectivity Then add points to prevent over-correction.

[0024] Specifically, path ratio attenuation coefficient From differential phase shift rate With differential reflectivity The constraints are defined and follow the order of gating first, then integration, i.e.: When the cross-correlation coefficient or signal amplitude is below a preset threshold, the path ratio attenuation coefficient is adjusted. Furthermore, it is not included in the integration of the above formula. The path ratio attenuation coefficient is only calculated based on the above constraints when passing through the gating point. Substitute them into the above formula.

[0025] S44. Based on the voxelized vertical reflectance profile, perform height segmentation correction on the zero-degree layer bright band (VPR) to obtain the corrected horizontal reflectance. Specifically, it includes the following steps: In a fixed geographical location And the radius is On the column ( km), discretized by height layer (layer thickness is) , (km), for the horizontal reflectivity field that has been corrected for path attenuation. By performing columnar statistics, the vertical reflectance profile is obtained. The expression is: in, Geometric height (km, altitude or terrain already taken into account; obtained by correction based on radar geometry and refraction); The thickness is the layer thickness.

[0026] Using the height of the 0℃ isothermal surface as a priori and combining it with the vertical reflectivity profile The peak shape is used to estimate the center height of the zero-degree bright band (VPR). With thickness (If half-width at half-height (FWHM) is used, then ). In the interval Internally, use continuous piecewise functions After subtracting the brightness band gain, the brightness band corrected profile is obtained. ,Right now: continuous piecewise functions (Taking Gaussian type as an example, note that the exponent is squared) The expression is: in, ; The peak subtraction amplitude (dB) represents the maximum overestimation caused by the zero-degree layer bright band; derived from the vertical reflectivity profile. The peak value is obtained by fitting or regressing the excess relative to the baseline (it is recommended to add an upper limit). (To prevent overcalibration) The shape dimension or half-width (km) controls the width of the subtracted curve. (If related to thickness...) Using FWHM alignment, the following options are available: It can also be fitted using least squares. and .

[0027] The outline after the highlight band is corrected Applied to pixels For any pixel ,make: but To avoid overcorrection, constraints need to be introduced, namely: in, The lower limit of allowable limits (e.g.) dBZ).

[0028] S45. Based on the DEM, generate partial beam blocking coefficients and partial filling coefficients to compensate for the system deviations caused by beam blocking and distant small-scale targets, and obtain the reflectivity field. .

[0029] Furthermore, the terrain occlusion and partial filling compensation is based on DEM estimation for each cell. Partial beam blocking coefficient With partial filling coefficient The reflectivity field was obtained with a small amount of additive compensation. ,Right now: in, This is the terrain correction factor; It is a constant.

[0030] Preferably, The upper limit is set to 1–2 dB and spatially smoothed to avoid triggering false alarms due to overcompensation.

[0031] In this embodiment, on the CAPPI height plane work grid Above, image is recorded as Horizontal reflectance after correction for the zero-degree layer bright band. Estimating Partial Beam Blocking Factor Using DEM and Beam Geometry With partial filling coefficient The reflectivity field was obtained with a small additive compensation. .in: (1) —Visibility ratio at lower levels Define the low-level reference height km, the lower edge height of the beam at the radial direction of the pixel is obtained from the radar altitude, elevation angle, or lower edge of the beam. ; Pixel surface elevation obtained from DEM The lower-level "visible" height is: Pick but, This indicates that the lower layers are fully visible; This indicates that the lower level is completely blocked.

[0032] (2) —Beam-filling ratio The lower layer of the cell cylinder Within, record the visible length. The length of the echo is recorded. For the above interval to satisfy ( For example, a threshold The total height (dBZ); if only 2DCAPPI is available, this length can be approximated using VPR or neighborhood cylinder statistics. Take: but, This indicates that the visible lower layers are "fully filled" by the target. This indicates that targets in the visible lower layers are sparse or much smaller than the beam.

[0033] (3) Compensation range Application scope and constraints For small additive compensation (dB), only used for low-level and visible pixels. Take a constant. (For example ), and set an upper limit. dB, spatially smoothed using a 3×3 or 5×5 Gaussian or mean kernel, i.e.: Only when the aforementioned quality gate passes ( Compensation should only be applied when [the specific condition is met]; otherwise... .

[0034] (4) Variables and unit scope 、 dBZ; dB (can be directly added to dBZ); Dimensionless ; km (or m, units must be consistent); It is the CAPPI height surface The pixels on the radar; the geometric calculation uses the same radar altitude or elevation angle or beamwidth and refraction correction as steps S41–S44.

[0035] S5. Adaptive Quantitative Precipitation Estimation: Classifies weather patterns for each grid cell and estimates precipitation intensity using a weighted fusion method. .

[0036] Furthermore, the adaptive quantitative precipitation estimation is achieved through multi-parameter weighted fusion. Precipitation intensity The estimation formula is: in, In the formula, For the sake of reflectivity With differential reflectivity Constructed deterministic function ; Fusion weights Based on weather type T and polarization confidence Sure; To determine and The coefficient; The weights in the weighted fusion are given by a deterministic function W, namely: Where T∈{Conv, Strat, TC, Mix}, represents the weather type label of the cell; Rainfall intensity is segmented; , where is the polarization confidence; W is a pre-defined deterministic function obtained by lookup table or training to ensure the interpretability and numerical stability of the weights, and the domain of W is {T}×{q}×[0,1], and the range is a two-dimensional simplex. .

[0037] S6. Rainfall Coordination Hierarchical Bias Correction: A multiplicative bias model is established in the logarithmic domain based on rain gauge observations to correct for rainfall intensity. Corrections were performed to obtain the corrected radar precipitation data. .

[0038] Preferably, the rainfall coordination hierarchical deviation correction adopts a two-layer structure combining hourly-scale multiplicative deviation and minute-scale additive residual, that is: By establishing a multiplicative bias model in the logarithmic field, a spatially slowly varying multiplicative bias field is obtained. The expression for the multiplicative bias model is: Residual in the linear domain on a minute scale Spatiotemporal low-pass filtering is performed to obtain the additive residual field. Finally, the corrected radar precipitation data is given. ,Right now: in, This indicates the intensity of precipitation observed by the rain gauge; This indicates the intensity of precipitation estimated by radar fusion. Undefined residuals; This indicates the radar precipitation intensity after bias correction; Hierarchical deviation correction output; Preferably, in order to suppress the underestimation of extreme segments, quantile mapping can be used for tail recorrection of quantiles above P90.

[0039] S7. Multi-source fusion and uncertainty output: Incorporating radar precipitation data... Spatial statistical fusion was performed with rainfall data from rain gauge stations and precipitation products from S-band weather radar to obtain fused precipitation. and pixel-level uncertainty Specifically, in the CAPPI height aspect... grid Above, for each time step Fusion estimation is performed (on a minute or hourly basis) to obtain the fused precipitation. and pixel-level uncertainty .

[0040] Preferably, the fusion employs Kriging with External Drift (KED) or an equivalent Bayesian method, and the external drift includes at least one of the following: terrain height. Near-surface wind shear Total precipitation capacity of the entire floor, PW(x).

[0041] Multi-source fusion estimation and uncertainty satisfy: in, For time steps The total amount; For time steps External drift regression coefficients (with time) It is variable, but applies to all cells at that moment. Consistent), estimated on the site or validation sample by ordinary least squares or generalized least squares; The Kriging weights are derived from the residual semivariogram and spatial location. The solution is obtained, which satisfies the condition. (The corresponding constraints apply in cases involving drift). The output for hierarchical deviation correction, in mm / h, is the "first approximation" of the fusion. The unit is the pixel terrain height or surface elevation (m or km, which must be consistent with the unit of the regression). It refers to the near-surface wind shear intensity or wind field index (or other external drift, such as total precipitable water volume PW).

[0042] The residual is defined as follows: Choose either an exponential or isomeric variogram. Parameters are estimated using leave-one-out cross-validation (LOOCV) or maximum likelihood estimation. If necessary, anisotropic or spherical distances may be used.

[0043] pixel-level uncertainty The Kriging variance or posterior variance is used for estimation, and a quality mask is generated accordingly for threshold discrimination and risk weighting in the business side. ,Right now: in, The threshold is determined by historical error or quantile.

[0044] As one implementation method, the present invention can be configured with an online learning and parameter management mechanism: within a 7–30 day rolling window, {A,b,c,d} and W are updated in small steps with the constraint of non-degradation (e.g., CC / RMSE does not deteriorate); better, independent parameter buffers are maintained for different weather types and rollback strategies are set to ensure consistency and controllability across processes, seasons and radars.

[0045] This invention specifically discloses a quantitative precipitation estimation and error correction method for X-band phased array rain-measuring radar. Addressing issues such as strong close-range attenuation, a bright band at the zero-degree layer, terrain obstruction, and partial beam congestion in this band, physical correction is first performed on the observation side: using... and Path attenuation backfilling was performed on the restricted ratio attenuation model for independent variables. VPR bright band correction was implemented in conjunction with the zero-degree layer height field. Low-level cover and shading backfilling were calculated based on the DEM to obtain a physically consistent reflectivity field.

[0046] Then, three estimated components were constructed. 、 、 According to a fixed gating priority order An adaptive fusion was performed with W(T, q, γ), which consists of weather type, intensity classification, and derived quality indices, to obtain the baseline precipitation field. Three statistical corrections were applied sequentially to the results: logarithmic domain multiplicative correction to eliminate scale bias, station-by-station short-memory additive correction to smooth local systematic errors, and external drift regression (including at least one of elevation, wind shear, and impermeability, with amplitude limiting) to suppress systematic errors introduced by the underlying surface and environmental field. Finally, a lightweight proportional normalization was used to stabilize the release bias.

[0047] This method can improve correlation and significantly reduce RMSE under various operating conditions such as stratification, mixing and strong convection. The system bias can be stably controlled within a small range. It supports minute-level generation of high spatiotemporal resolution (≤1 km and ≤5 min) products and is suitable for business scenarios such as refined monitoring of water conservancy basins and early warning of urban waterlogging.

[0048] By employing the above technical solution, the present invention provides a method for quantitative precipitation estimation and error correction using X-band phased array radar, which has at least the following beneficial effects: 1. This invention employs a hybrid ZPHI model with gating followed by integration, further utilizing differential phase shift rate. With differential reflectivity Dual constraints suppress overcorrection and undercorrection, significantly mitigating the underestimation of heavy precipitation segments; based on weather type T and rainfall intensity segmentation. With polarization confidence The deterministic function W is preferably defined in the case of increased convective precipitation. Weighting, in the context of layered weak precipitation Weighting increases the correlation coefficient and reduces false alarms.

[0049] 2. This invention employs hierarchical correction that combines logarithmic multiplication with linear additiveness to further control minute-level transients and hour-level mean values ​​simultaneously, thereby improving RMAE / RMSE and enhancing consistency at extreme quantiles (above P90).

[0050] 3. This invention introduces a partial beam obstruction coefficient and a partial filling coefficient generated by DEM, preferably reducing system bias in the obstructed area and improving product spatial continuity; it employs Kriging or an equivalent Bayesian method with external drift to fuse and output pixel-level uncertainty. With quality mask This further supports threshold-based early warning decisions and meets the needs of low-latency online services. Attached Figure Description

[0051] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart of the quantitative precipitation estimation and error correction method using X-band phased array radar in this invention; Figure 2 This is a schematic diagram of path attenuation correction in this invention; Figure 3 This is a scatter plot of precipitation over 5 minutes in Embodiment 1 of the present invention. Figure 4 This is a scatter plot comparing the cumulative precipitation over 1 hour in Embodiment 1 of the present invention; Figure 5 This is a scatter plot of precipitation over 5 minutes in Example 2 of the present invention. Figure 6 This is a scatter plot comparing the cumulative precipitation over 1 hour in Embodiment 2 of the present invention; Figure 7 This is a scatter plot comparing 5-minute precipitation data in Example 3 of the present invention. Figure 8 This is a scatter plot comparing the cumulative precipitation over 1 hour in Embodiment 3 of the present invention. Detailed Implementation

[0052] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. This will allow for a full understanding of how the present application uses technical means to solve technical problems and achieve technical effects, and to facilitate its implementation.

[0053] Example 1: Integrated QPE processing of X-band phased array radar in the Yihe River Basin scenario.

[0054] This embodiment illustrates the engineering deployment and verification process in the Yi River Basin to demonstrate the feasibility, parameter configuration, runtime sequence, and performance improvements of the present invention. This embodiment does not limit the scope of the claims in any way.

[0055] I. Test Area and Data Scope

[0056] Station network and timing: 10 automatic rain gauges were deployed at a distance of 8–30 km from the radar; the time duration was 1 hour, with a step size of 5 minutes (12 time steps).

[0057] External factors: Each station is assigned a topographic height H (20–600m) and a near-surface wind shear S (2–15, dimensional index) for subsequent external drift.

[0058] Radar polarization parameters: and spectral width; among which It is written in dBZ.

[0059] Assessment criteria: two levels, 5-minute and 1-hour, with comparison to rain gauges at different station scales.

[0060] II. Data Generation and “Observation” Construction.

[0061] 2.1 Actual Precipitation Field (mm / h); The superposition structure of "moving convection core + layered background" is adopted: the peak of the convection core is about 60 mm / h, and it slowly moves northeastward over time; the layered background fluctuates slightly in time and space to ensure the continuity of weak precipitation.

[0062] 2.2, Based on actual precipitation fields Inversely solving the radar polarization parameters, i.e.: True horizontal reflectance (linear, mm) 6 / m 3 ): Specific phase shift : To approximate real-world conditions, the phase shift rate was compared. Multiplied by a multiplicative noise on the order of 10%.

[0063] Differential reflectivity : Suppose and Weak positive correlation, with 0.08 dB Gaussian noise added, and amplitude limited to 0.05–2.5 dB; Cross-correlation coefficient The value for stratified precipitation is around 0.985, while that for convection and a few "abnormal" points is reduced to the range of 0.90–0.98.

[0064] 2.3 X-band path attenuation and observed horizontal reflectivity ; For each radial path, discretize it into 10 segments, calculate the path ratio attenuation coefficient using a gated hybrid model, and integrate it, i.e.: , , , like or Then the gate is set to zero. ,in , dBZ. Let the two-way integral attenuation be 2∫αds, and construct the observed horizontal reflectance. .

[0065] III. Processing flow and parameters.

[0066] S3.1 Path attenuation correction (gating first, then integration). Recalculate 2∫αds based on the above formula, and obtain the attenuation-corrected horizontal reflectivity using the following formula. ,Right now: in, , representing the observed (uncorrected for path attenuation) horizontal reflectance (in dBZ) at a fixed azimuth. With CAPPI height surface Above, as radial distance The function. Its correction yields: If in the grid cell The above statement, then After correction .

[0067] S3.2, Bright band / VPR correction; Using columnar statistical detection of the bright band at zero degree, approximately 25% of the samples fell into the bright band region, according to a Gaussian piecewise function. Deduction, that is: S3.3, DEM occlusion and partial filling compensation; occlusion rate With filling coefficient A small amount of additive compensation was obtained ,Right now: , in, , , ≤2dB.

[0068] S4. Adaptive quantitative precipitation estimation; Will Convert to linear Construct three components, namely: Segmented by weather type T (convective, stratiform, or mixed) and rainfall intensity With polarization confidence Take weight And with polarization confidence Smooth scaling to obtain precipitation intensity ,Right now: S5, Rainfall Coordination "Hierarchical Deviation Correction".

[0069] Assume the rainfall intensity observed by the rain gauge is First, perform multiplication in the logarithmic field to obtain the multiplicative bias field. Finally, the corrected radar precipitation data is given. ,Right now: in, This indicates the selected time window (usually an hourly window). Within this range, the spatial expectation or smoothing operator of the logarithmic residuals of rain gauges is used to obtain the multiplicative bias field. (Logarithmic field, dimensionless). There are two possible implementation methods: 1. Constant multiplication (changes only with time).

[0070] in, The number of rain gauge stations; It is a rain gauge station; ,prevent .

[0071] Subsequently obtained corrected radar precipitation data ,Right now: 2. Spatial weighting (optimized results) field).

[0072] Multiplicative bias field for: in, For spatial smoothing, kriging, or LOESS weights, Subsequently, corrected radar precipitation data was obtained. ,Right now: After obtaining the corrected radar precipitation Then, perform a 3-point moving average of the minute-level linear additive residuals, i.e.: External drift includes at least H or S (both are used in this example). Pixel-level variance uncertainty. This will not be elaborated upon in this small sample.

[0073] IV. Evaluation of design and calculation methods.

[0074] Rain gauge construction: based on real precipitation field Multiply by 5% Gaussian noise to obtain the precipitation intensity observed by the rain gauge. (Zero mean, controlled variance) is used as the verification benchmark.

[0075] Indicators: Correlation coefficient (CC), Relative deviation (RB), Relative mean absolute error (RMAE), Root mean square error (RMSE), i.e.: in, For the first Precipitation estimated by method for each sample location (if 5-minute caliber, it is rainfall intensity in mm / h; if 1-hour cumulative, it is rainfall in mm). For the first Rainfall observed at each sample location by rain gauge (caliber and...) (Keep the units consistent; same as above). This represents the number of samples included in the statistics. Where: Minute caliber: Counted by “all pixels × time steps” for those included at the threshold (e.g., obs≥0.5 mm / h); Hourly caliber: Counted by "cumulative per station per hour" (or you can write...) ).

[0076] Summary criteria: (i) 5min: all stations × time step; (ii) 1h: the sum of precipitation depths at each station over 12 steps (equivalent to hourly accumulation), as shown in Table 1.

[0077] Table 1 Summary Results According to the results shown in Table 1, the CC value approaches 0.98 at a 5-minute aperture, indicating a good fit between the minute-level phase and amplitude. The RMAE is significantly affected by transient fluctuations and strong segments, which is consistent with actual characteristics. At a 1-hour cumulative level, the RMAE and RMSE converge significantly, while the RB value approaches 0, indicating that the hierarchical bias correction effectively balances the conservation of hourly averages with the suppression of minute-level transient errors. Under the synthesized conditions of occlusion / filling and bright band perturbation, the linkage of "gating before integration + VPR segmentation + small DEM compensation + ternary adaptive weighting + hierarchical bias + external drift" stably reduces systematic errors.

[0078] This embodiment fully demonstrates the engineered processing chain of the present invention on a small sample of data with sufficient physical constraints. For example... Figure 3 and Figure 4 As shown in the 5-minute scatter plot and the 1-hour cumulative scatter plot, the results show that under the combined perturbation of "strong attenuation + bright band + occlusion / filling", through the integrated process of gate-priority hybrid ZPHI, VPR segmented correction, DEM small-amplitude compensation, adaptive three-component fusion, hierarchical bias correction and external drift fusion, robust quantitative precipitation estimation at both minute and hourly levels can be achieved, while maintaining low bias and high correlation.

[0079] Example 2: Application of complex terrain layered-convective mixing process (outer rainband of typhoon).

[0080] I. Application Scenarios and Data Sources.

[0081] This embodiment selects the outer rainband process of a typhoon in the Yihe River Basin as the verification object. The precipitation is mainly in the form of strata and mixed with short-term convective echoes. The zero-degree layer is obvious, and the topographic relief causes local shading. The data and equipment used are as follows: (1) Radar data: One X-band phased array rain-measuring radar, using electronic scanning, volume scan period of 1 min, azimuth 0–360°, elevation 0.5°–10°, range resolution 250 m; output parameters include (dBZ) (dB) (° / km) .

[0082] (2) Reference external parameters: reflectivity of adjacent S-band weather radar mosaic and height field of 0℃ isothermal surface (updated every 5 minutes), used for VPR prior.

[0083] (3) Ground station network: 68 automatic rain gauges, sampling for 1 minute (in this embodiment, the data is summarized in 5 minutes); (4) Auxiliary data: 30mDEM and near-surface wind field (reanalysis product), used for terrain shading determination and external drift.

[0084] II. Data quality control and physical correction.

[0085] Perform the following steps in order to ensure dimensional consistency and physical plausibility: (1) Non-meteorological echo gating: with and Simultaneous threshold (preferred) ≥0.90 (≥−3dBZ) to eliminate clutter and anomalous propagation.

[0086] (2) Path attenuation correction: The path ratio attenuation coefficient α is estimated radially and integrated to obtain the cumulative attenuation. Backfill to ,Right now: Preferred parameters , , , .

[0087] (3) Zero-degree bright band (VPR) correction: Combine the S-band 0° surface height field to identify the bright band region and apply a correction amount. , to obtain horizontal reflectance .

[0088] in, The horizontal reflectance (dBZ) is the path attenuation corrected value, compared to the uncorrected observed horizontal reflectance. Plus two-way path attenuation get.

[0089] Horizontal reflectance (dBZ) corrected for the zero-degree bright band (VPR). Within the identified bright band height range. Internal deduction of correction amount The result obtained later is: outside the interval Therefore .

[0090] (4) Terrain shading and backfilling (DEM): Calculate the shading rate With backfill ratio To form terrain correction coefficients ,Right now: Preferred , The corrected reflectivity field is obtained. .

[0091] III. Integration of quantitative estimation components and gating.

[0092] based on , , Construct three QPE components, namely: The preferred values ​​are A=0.017 and b=0.714. Preferred ; Preferred =0.011, =0.72, =0.11.

[0093] Furthermore, an adaptive weight W(T,q,γ) is calculated using a three-dimensional mapping of weather type T, intensity level q, and quality index γ. Where: when the intensity... ≥22mm / h is judged as convection type; when the intensity <3mm / h and ≥0.982 indicates a layered structure; the rest are mixed types; the preferred intensity grading is [0,2), [2,10), [10,30), [30,∞); the polarization confidence γ is determined by the cross-correlation coefficient. Normalized to [0,1]. Gating priority is fixed as follows: Based on this, a basic weight table is given for each class (T,q), and then normalized after linear amplification with polarization confidence γ. The final fusion estimate is: IV. Error Correction and External Drift.

[0094] exist Implement sequentially based on the above: (1) Multiplicative correction (logarithmic domain): estimating the multiplicative bias field of the global bias based on a 5-minute rain gauge. ,make: (2) Additive short memory correction: The additive residual field is obtained by performing a 3-point moving average of the residuals at each station. ,Right now: (3) External drift term: Introduce at least one external variable (preferably the site elevation H and the near-surface wind shear intensity S) to the The residuals are subjected to least squares regression with amplitude limiting (preferably ≤ ±15%), i.e.: in, The topography or station elevation (m or km) of a pixel; Indicates the current moment Sample set used for regression superior The spatial average (constant), that is: If weighted regression or kriging is used, the following can be employed: in, For fusion weights; For rain gauge The quantity.

[0095] The near-surface wind shear intensity of the pixel (or the external variable you choose, in units such as...) (or dimensionless index); Similarly, it is the spatial average over the same regression sample set at that time, i.e.: V. Evaluation criteria and indicators.

[0096] The evaluation criteria include: a 5-minute criterion (only samples with observations ≥0.5 mm / h are counted) and a 1-hour cumulative criterion (only stations with cumulative observations ≥1.0 mm are counted). The evaluation indicators are the correlation coefficient (CC), relative deviation (RB), relative mean absolute error (RMAE), and root mean square error (RMSE), the definitions of which are consistent with those described above in this manual.

[0097] VI. Results like Figure 5 and Figure 6 The 5-minute scatter plot and the 1-hour cumulative scatter plot are shown. Over a process of 12 5-minute steps (1 hour, 68 stations), the method of this invention (fusion + full correction) yields: 5-minute caliber: CC=0.995, RB=+0.60%, RMAE=7.66%, RMSE=0.73mm / h; 1-hour cumulative: CC=1.000, RB=+0.60%, RMAE=0.69%, RMSE=0.08mm.

[0098] Compared with the control method (local Z-R or KDP single parameter), the correlation is improved and the systematic bias and root mean square error are significantly reduced; VPR and DEM correction have a significant suppressive effect on spurious enhancement near the layered peak and partial clogging of the windward slope beam.

[0099] VII. Project Implementation In a single-node 8-core CPU environment, the average inference latency is about 35–45 seconds when processed in 1 minute rolling mode. Through weight table pre-compilation and incremental VPR updates, it can be compressed to ≤30 seconds, meeting the needs of minute-level updates.

[0100] Example 3: Application of urban rainstorms (strong convection, strong attenuation, KDP-dominated).

[0101] I. Application Scenarios and Data Sources.

[0102] A short-duration, intense convective rainstorm event in a coastal plain city was selected as the verification object. The peak rainfall intensity exceeded 100 mm / h in 5 minutes. The strong echoes from nearby locations caused significant path attenuation, and the cross-correlation coefficient was high. A slight decrease was observed in the strong core region. Data and apparatus are as follows: (1) X-band phased array rain radar: electronic scanning, volume scan period 1 min, azimuth 0–360°, elevation 0.5°–12°, range resolution 250 m; output (dBZ) (dB) (° / km) .

[0103] (2) Ground automatic rain gauges: 128 stations, sampling for 1 minute (summarized in 5-minute intervals).

[0104] (3) Auxiliary data: 30mDEM (mainly plains), urban impermeability grid (with the cell where the station is located representing station I), near-surface wind shear intensity S.

[0105] II. Data quality control and physical correction.

[0106] Perform quality control and correction in the following order: (1) Non-meteorological echo gating: preferred ≥0.90 ≥−2dBZ combined thresholds to eliminate ground clutter and anomaly propagation; (2) Path attenuation correction: The radial segmentation estimates the attenuation α, and the cumulative attenuation is obtained by integration and backfilling. and Preferred: in , , , And apply an upper limit suppression to the average α of the segments to prevent HB from diverging; (3) Zero-degree layer bright band (VPR) correction: Convection is dominant, the bright band is weak, and the correction amount is only applied to the identified local area. To obtain the horizontal reflectance ; (4) Terrain shading and backfilling (DEM): Plains have weaker shading, based on the shading rate With backfill ratio Give terrain correction coefficients Forming a reflectivity field .

[0107] III. Integration of quantitative estimation components and gating.

[0108] based on , , Construct three QPE components: The preferred values ​​are A=0.017 and b=0.714. Preferred ; Preferred =0.0105, =0.73, =0.10.

[0109] Furthermore, an adaptive weight W(T,q,γ) is calculated using a three-dimensional mapping of weather type T, intensity level q, and quality index γ; when the intensity is ≥30 mm / h, it is judged as convective; when it is <3 mm / h, it is judged as convective. ≥0.985 is judged as layered, the rest are mixed type; the preferred strength grades are [0,3), [3,15), [15,45), [45,∞).

[0110] The gating priority is fixed as follows: .

[0111] Based on this, a basic weight table is given for each class (T,q), which is then normalized after linear amplification with polarization confidence γ, and finally fused for estimation, i.e.: IV. Error Correction and External Drift.

[0112] exist Implement sequentially based on the above: (1) Multiplicative correction (logarithmic domain): estimating the multiplicative bias field of the global bias based on a 5-minute rain gauge. ,make: (2) Additive short memory correction: Residual G− by station The additive residual field is obtained by performing a 3-point moving average. ,Right now: (3) External drift term: Introduce at least one external variable for linear regression and limit the amplitude (preferably within ±15%). In this embodiment, a subset of the station elevation H, near-surface wind shear intensity S, and impermeability I is selected to obtain: Preferably, during the release process... The ratio is normalized to keep the relative deviation (RB) within ±1%.

[0113] V. Evaluation criteria and indicators.

[0114] To emphasize the reliability of strong convection, the evaluation criteria are set as follows: 5-minute caliber: Only samples with a flow rate ≥1.0 mm / h are statistically analyzed; 1-hour cumulative caliber: Only stations with cumulative observations ≥5.0 mm are counted; and a high rainfall intensity quantile test is added: CC, RB, RMAE, and RMSE of subsets P90 and P95 are calculated in the 5-minute caliber sample.

[0115] VI. Results like Figure 7 and Figure 8 The 5-minute scatter plot and the 1-hour cumulative scatter plot are shown. Over a process of 12 five-minute steps (1 hour, 128 stations), the method of this invention (fusion + full correction) yields: 5-minute caliber: CC=0.993, RB= +0.60% (released normalized value), RMAE=7.51%, RMSE =1.96mm / h (sample size N=1534).

[0116] 1-hour cumulative: CC=1.000, RB=+0.75%, RMAE=0.92%, RMSE=0.20mm (Number of stations N=48).

[0117] High rainfall intensity distribution (5 minutes): P90 subset: CC = 0.928, RB = +0.88%, RMAE = 7.72%, RMSE = 5.85 mm / h (N = 154); P95 subset: CC=0.867, RB=+0.33%, RMAE=8.36%, RMSE=7.28mm / h (N=77).

[0118] The comparison and ablation showed that without attenuation correction or with only a single parameter (such as Z-R or KDP only), the correlation decreased significantly and the systematic bias and RMSE were significantly amplified. The above results corroborate that the combination of "fixed gating priority + strong attenuation correction + external drift" has a stable advantage under short-duration strong convection conditions.

[0119] VII. Project Implementation Preferably, in a single-node environment with an 8-core CPU, the average inference latency, including segmented decay integral and weighted table lookup, is ≤35s when the processing is performed in 1-minute rolling mode. Furthermore, by incrementally updating Apath and W, the processing latency can be stabilized at around 30s, meeting the minute-level business requirements.

[0120] The core of this invention lies in proposing an "integrated QPE" method for X-band phased array rain measurement, which fixes the gating priority order at the algorithm level. , build The three components are grouped by W(T,q,γ) (based on weather type, intensity, and...). The derived quality metrics are used to achieve adaptive fusion; path attenuation correction is performed sequentially on the observation side. Constraints to prevent HB divergence, zero-degree layer / VPR correction, and DEM low-layer coverage / occlusion backfilling are used to obtain physically consistent input.

[0121] On the results side, a three-stage statistical correction is performed: multiplicative (logarithmic domain) – additive short memory – external drift (at least one external variable, such as altitude / wind shear / impermeability), and the RB is stabilized by lightweight proportional normalization (≈±1%). This link simultaneously improves correlation and significantly reduces RMSE / bias under various operating conditions such as stratified, mixed, and strong convection, with controllable computational load, meeting the requirements for minute-level real-time business deployment.

[0122] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0123] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. Since the above embodiments are substantially similar to the method embodiments, their descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0124] The above embodiments provide a detailed description of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for quantitative precipitation estimation and error correction using X-band phased array radar, characterized in that, The method includes the following steps: S1. Data Access and Gridding: Collect the dual-polarization volume scan observation parameters and spectral width of the X-band phased array radar, and then grid them after time and space registration with ground rain gauge data, S-band weather radar precipitation products and digital elevation model. The dual-polarization volume scan observation parameters include reflectivity. Differential reflectivity Differential phase shift Differential phase shift rate Cross-correlation coefficient ; S2. Quality Control and System Calibration: In situations with high cross-correlation coefficients Furthermore, based on reflectivity, in stratiform, weak precipitation samples without strong convection, Differential reflectivity With differential phase shift rate The polarization self-consistency relationship is used to calibrate the radar system constants; S3. Non-meteorological echo suppression: based on reflectivity With cross-correlation coefficient Texture quantity, differential phase shift Gradient, spectral width and terrain information are used to identify and shield ground clutter and anomalous propagation echoes to obtain quality-controlled echo data. S4. Attenuation and Vertical Structure Correction: Using differential phase shift rate... With differential reflectivity Path attenuation backfilling is performed on the hybrid ZPHI model with common constraints, zero-degree layer brightness band correction is implemented in combination with the height layer, and low-layer coverage and occlusion backfilling are calculated based on DEM to obtain a physically consistent reflectivity field. S5. Adaptive Quantitative Precipitation Estimation: Classifies weather patterns for each grid cell and estimates precipitation intensity using a weighted fusion method. ; S6. Rainfall Coordination Hierarchical Bias Correction: A multiplicative bias model is established in the logarithmic domain based on rain gauge observations to correct for rainfall intensity. Corrections were performed to obtain the corrected radar precipitation data. ; S7. Multi-source fusion and uncertainty output: Incorporating radar precipitation data... Spatial statistical fusion was performed with rainfall data from rain gauge stations and precipitation products from S-band weather radar to obtain fused precipitation. and pixel-level uncertainty .

2. The method for quantitative precipitation estimation and error correction using X-band phased array radar according to claim 1, characterized in that, Step S2 includes: Perform differential reflectance correction using polarization self-consistency relation Zero bias and fine-tuning of reflectivity The system constant ensures that the two parameters satisfy the consistency constraint under weak gradient conditions.

3. The method for quantitative precipitation estimation and error correction using X-band phased array radar according to claim 1, characterized in that, Step S2 includes: With pixels The central window Calculate texture features , , Where N is an odd number and 3 ≤ N ≤ 11, that is: Together with spectral width and topographic elevation, they form a feature vector. ,Right now: in, For spectral width; This refers to the terrain elevation; Represents a cell Horizontal reflectance at that location; Represents a pixel Differential reflectivity at that location; Represents a pixel The differential phase shift rate at that point is determined by the differential phase shift. The derivative in the radial direction is obtained; Represents a cell Cross-correlation coefficient at; Feature vector Input classifier Probability of obtaining meteorological targets and with threshold Generate a binary mask ,Right now: Based on this, pixel-level shielding is applied to the polarization parameters to obtain echo data with quality control. ,Right now: in, The polarization parameter field is defined.

4. The method for quantitative precipitation estimation and error correction using X-band phased array radar according to claim 1, characterized in that, The attenuation and vertical structure correction includes path attenuation correction and brightness band / VPR correction, and is performed in a gating priority order, including: S41, along the fixed orientation and CAPPI height plane Radar rays, for differential phase shift First, perform a length of M Radial median filtering yields smooth phase shift. Then, the differential phase shift rate is obtained by radial differentiation. ,Right now: in, This represents the radial distance from the radar. It is the azimuth angle; For CAPPI height surface; S42. Define the path ratio attenuation coefficient. ,include: Only during gate passage, at differential phase shift rate With differential reflectivity The combined constraint estimation, i.e.: in, Indicates the same ray The path integral variable on the radial distance Equivalence, that is ; Reference differential reflectance; , , All are calibrated constants; In cross-correlation coefficient Or horizontal reflectance after quality control Gating regions below the threshold directly reduce the path ratio attenuation coefficient. The expression is: in, Cross-correlation coefficient The gating threshold; The threshold for amplitude or signal-to-noise ratio; S43, Based on the path ratio attenuation coefficient The path attenuation is integrally corrected to obtain the attenuation-corrected horizontal reflectance. And backfill to the pixels; S44. Based on the voxelized vertical reflectance profile, the brightness band of the zero-degree layer is segmented for height correction to obtain the corrected horizontal reflectance. ; S45. Based on the DEM, generate partial beam blocking coefficients and partial filling coefficients to compensate for the system deviations caused by beam blocking and distant small-scale targets, and obtain the reflectivity field. .

5. The method for quantitative precipitation estimation and error correction using X-band phased array radar according to claim 4, characterized in that, Step S43 includes: Define the path ratio attenuation coefficient Substituting into the following formula completes the calculation of horizontal reflectivity. The path integral correction is as follows: in, Indicates the position relative to the CAPPI elevation plane. rays Above, the observed horizontal reflectance; Then on the CAPPI height surface Above using pixels The one-to-one correspondence is used to assign radial results to grid cells, expressed as: in, For pixels Horizontal reflectance; Two-way path attenuation; For pixels Observed horizontal reflectance; This represents the radial distance from the radar. It is the azimuth angle; For CAPPI height surface.

6. The method for quantitative precipitation estimation and error correction using X-band phased array radar according to claim 4, characterized in that, Step S44 includes: In a fixed geographical location And the radius is On the column, it is discretized according to height layers. Horizontal reflectivity field with path attenuation correction By performing columnar statistics, the vertical reflectance profile is obtained. The expression is: in, Geometric height; For layer thickness; Using the height of the 0℃ isothermal surface as a priori and combining it with the vertical reflectivity profile The peak shape is used to estimate the center height of the bright band in the zero-degree layer. With thickness In the interval Internally, a continuous piecewise function is used. After subtracting the brightness band gain, the brightness band corrected profile is obtained. ,Right now: continuous piecewise functions The expression is: in, ; The peak value is the deduction margin, representing the maximum overestimation caused by the zero-degree layer bright band. The shape dimension or half-width controls the width of the subtracted curve; The outline after the highlight band is corrected Applied to pixels For any pixel ,make: but ; To avoid overcorrection, constraints are introduced, namely: in, This is the lower limit that is allowed.

7. The method for quantitative precipitation estimation and error correction using X-band phased array radar according to claim 4, characterized in that, In step S45, the terrain occlusion and partial filling compensation is based on DEM estimation for each cell. Partial beam blocking coefficient With partial filling coefficient The reflectivity field was obtained with a small amount of additive compensation. ,Right now: in, This is the terrain correction factor; It is a constant.

8. The method for quantitative precipitation estimation and error correction using X-band phased array radar according to claim 1, characterized in that, Step S5 includes: The adaptive quantitative precipitation estimation is achieved through multi-parameter weighted fusion, and the precipitation intensity... The estimation formula is: in, In the formula, For the sake of reflectivity With differential reflectivity Constructed deterministic function ; Fusion weights Based on weather type T and polarization confidence Sure; To determine and The coefficient; The weights in the weighted fusion are given by a deterministic function W, namely: Where T represents the weather type label for the pixel; Rainfall intensity is segmented; is the polarization confidence level; W is a pre-defined lookup table or a deterministic function obtained through training.

9. The method for quantitative precipitation estimation and error correction using X-band phased array radar according to claim 1, characterized in that, Step S6 includes: By establishing a multiplicative bias model in the logarithmic field, a spatially slowly varying multiplicative bias field is obtained. The expression for the multiplicative bias model is: Residual in the linear domain on a minute scale Spatiotemporal low-pass filtering is performed to obtain the additive residual field. Finally, the corrected radar precipitation data is given. ,Right now: in, This indicates the intensity of precipitation observed by the rain gauge; This indicates the intensity of precipitation estimated by radar fusion. Undefined residuals; This indicates the radar precipitation intensity after bias correction; Hierarchical deviation correction output.

10. The method for quantitative precipitation estimation and error correction using X-band phased array radar according to claim 1, characterized in that, Step S7 includes: Employ Kriging or an equivalent Bayesian method with external drift, where the external drift includes at least one of the following: terrain height. Near-surface wind shear Total precipitation per unit floor: PW(x); Multi-source fusion estimation and uncertainty satisfy: in, For time steps The total amount; For time steps The external drift regression coefficient; Kriging weights; Output for hierarchical deviation correction; This refers to the pixel terrain height or surface elevation. This refers to near-surface wind shear intensity or wind field indicators. pixel-level uncertainty The mass mask is generated based on the estimation of Kriging variance or posterior variance. ,Right now: in, The threshold value is used.

Images