A flood forecasting method, device and equipment based on a rain measurement radar

Abstract

The present application relates to the field of electric digital data processing, in particular to a flood prediction method, device and equipment based on a rain measuring radar. According to the quality evaluation and dynamic calibration of the rain measuring radar data, the calibrated radar precipitation field is obtained; according to the calibrated radar precipitation field, combined with the radar echo motion field, the short-term quantitative precipitation forecast (QPF) sequence of the future N hours is generated; the preset macro hydrological model and micro hydrological model are driven to carry out hydrological simulation respectively, and the macro hydrological state variable and micro hydrological state variable are obtained; and the real-time exchange and coupling of the state variable are carried out, and the coupled multi-scale model state is obtained; according to the multi-scale model state and the real-time obtained ground observation data, the data assimilation is carried out through the gradient optimization nonlinear least square method, and the updated hydrological prediction result is obtained; according to the updated hydrological prediction result, the flood warning information is generated, and the precision of the flood prediction is improved.

Description

Technical Field

[0001] This invention relates to the field of electronic digital data processing, and specifically to a flood forecasting method, apparatus, and equipment based on rainfall radar. Background Technology

[0002] Floods are among the most severe natural disasters globally, posing a significant threat to people's lives and property, as well as to socio-economic development. Accurate and timely flood forecasting is a crucial element in disaster prevention and mitigation. Traditional flood forecasting methods primarily rely on surface rainfall monitoring networks and conceptual hydrological models, which have the following significant limitations:

[0003] Insufficient precision in precipitation monitoring: The sparse density of surface rain gauge networks makes it difficult to capture dramatic spatial changes in precipitation, especially in areas with complex topography and diverse watershed scales. This results in significant errors in the precipitation data input into hydrological models, directly impacting the accuracy of flood forecasts.

[0004] Rigid Model Structure and Parameters: Traditional conceptual hydrological models (such as the Xin'anjiang Model and TOPMODEL) have fixed structures and their parameter calibration relies on historical data. They are difficult to adapt to the impacts of climate change and underlying surface changes (such as urbanization and land use change), and their generalization ability and adaptability are poor.

[0005] Poor forecast timeliness: Traditional hydrological models have long calculation cycles and rely on fixed time steps (such as hours or days), making it difficult to achieve minute-level real-time early warnings for sudden floods (such as flash floods and urban waterlogging) caused by short-duration heavy rainfall.

[0006] Poor scale adaptability: A single model cannot simultaneously meet the needs of flood simulation and prediction from small and medium-sized watersheds (such as urban waterlogging and mountain torrent gullies) to large watersheds (such as the Yangtze River and Yellow River tributaries), resulting in inconsistent application effects at different scales.

[0007] In recent years, precipitation radar technology has been widely used in precipitation monitoring due to its high spatiotemporal resolution. However, simply inputting radar-estimated precipitation (QPE) products directly into traditional hydrological models also faces the following problems:

[0008] Radar data suffers from systematic biases, attenuation, and ground clutter. Unprocessed radar data will directly transmit errors to the hydrological model.

[0009] Traditional models lack an effective mechanism to dynamically incorporate real-time observation data (including radar and rain gauge data) into the model state in order to correct the model's initial conditions and forecast trajectory.

[0010] Therefore, there is an urgent need for a flood simulation and prediction system that can overcome all the above-mentioned shortcomings and achieve high accuracy, high timeliness, and strong adaptability. Summary of the Invention

[0011] The purpose of this invention is to provide a flood forecasting method, device, and equipment based on rainfall radar, which solves the problems in the prior art.

[0012] This invention is achieved through the following technical solution:

[0013] In a first aspect, embodiments of the present invention provide a flood prediction method based on rainfall radar, comprising:

[0014] The quality assessment and dynamic calibration are performed based on the rainfall radar data to obtain the calibrated radar precipitation field.

[0015] Based on the calibrated radar precipitation field and combined with the radar echo motion field, a short-term quantitative precipitation forecast (QPF) sequence for the next N hours is generated.

[0016] Hydrological simulations are performed using the preset macro-hydrological model and micro-hydrological model driven by the QPF sequence, to obtain the macro-hydrological state variables corresponding to the macro-hydrological model and the micro-hydrological state variables corresponding to the micro-hydrological model; wherein, the macro-hydrological model covers the entire basin and operates at a first spatial resolution, and the micro-hydrological model covers key areas within the basin and operates at a second spatial resolution higher than the first spatial resolution.

[0017] Based on the macro-hydrological state variables and the micro-hydrological state variables, the state variables are exchanged and coupled in real time to obtain the coupled multi-scale model state.

[0018] Based on the state of the multi-scale model and the real-time ground observation data, the data is assimilated using a gradient-optimized nonlinear least squares method to obtain the updated hydrological forecast results.

[0019] Flood warning information is generated based on the updated hydrological forecast results.

[0020] Preferably, the step of performing quality assessment and dynamic calibration based on rainfall radar data to obtain the calibrated radar precipitation field includes:

[0021] Based on the rain measurement radar data, a multi-level quality assessment is performed to generate a comprehensive quality score for each grid point in the rain measurement radar data.

[0022] Based on the comprehensive quality score of each grid point, all grid points are divided into high-quality areas, medium-quality areas, and low-quality areas.

[0023] Based on the regional division results, dynamic calibration is performed using calibration strategies corresponding to each quality region to obtain the radar precipitation field.

[0024] Preferably, the step of performing multi-level quality assessment based on the rain-measuring radar data to generate a comprehensive quality score for each grid point in the rain-measuring radar data includes:

[0025] Based on the terrain obstruction information, non-meteorological clutter characteristics, and original signal quality in the rainfall radar data, the terrain obstruction score, clutter identification score, and signal quality score for each grid point are calculated respectively.

[0026] Based on the radar reflectivity data of each grid point, the consistency between the grid point and its neighboring grid points in the horizontal and vertical directions is analyzed, and the spatial consistency score and vertical consistency score are calculated respectively.

[0027] Based on radar reflectivity data from consecutive time intervals, the characteristics of reflectivity variation at each grid point over time are analyzed, and a time consistency score is calculated.

[0028] Based on the physical relationship between radar reflectivity and precipitation rate at each grid point, and using available ground rain gauges or satellite precipitation data for cross-validation, physical consistency scores and cross-validation scores are calculated respectively.

[0029] Based on the meteorological-hydrological joint scoring standard, a comprehensive quality score is generated by integrating various basic scores.

[0030] Preferably, the step of dynamically calibrating according to the regional division results and using calibration strategies corresponding to each mass region to obtain the radar precipitation field includes:

[0031] For the high-quality areas, an optimized averaging method based on the statistical relationship of local climate characteristics is used for calibration;

[0032] For the medium-quality region, a weighted calibration method combining spatial correlation and distance weight is used for calibration, wherein the calibration weight is related to the comprehensive quality score of the grid point and the distance between the grid point and the high-quality region;

[0033] For the low-quality area, a fusion calibration method that integrates multi-source auxiliary data is used for calibration. The multi-source auxiliary data includes ground rain gauge data and satellite precipitation estimation data.

[0034] Preferably, the step of performing real-time exchange and coupling of state variables based on the macro-hydrological state variables and the micro-hydrological state variables to obtain the coupled multi-scale model state includes:

[0035] Based on preset exchange triggering conditions, the timing and content of state variable exchange are determined. The exchange triggering conditions include forced variable exchange and state variable exchange. The forced variable exchange must be performed in each coupled calculation cycle, and the exchanged variables include at least the upstream boundary water level and the lateral inflow flow. The state variable exchange is performed conditionally and is triggered when the difference in soil moisture between the micro-region and the corresponding macro-region exceeds a first preset threshold or the rate of change of groundwater level exceeds a second preset threshold.

[0036] When the exchange triggering condition is met, the macroscopic hydrological state variables are mapped to the grid of the microscopic hydrological model through a spatial interpolation algorithm to obtain the multi-scale model state.

[0037] Preferably, when the exchange triggering condition is met, the macroscopic hydrological state variables are mapped to the grid of the microscopic hydrological model using a spatial interpolation algorithm to obtain the multi-scale model state, including:

[0038] Identify the micro-hydrological model region where state variable exchange is required, and define all computational grid points of the micro-hydrological model within the micro-hydrological model region as target grid points;

[0039] When the exchange triggering condition is met, the macroscopic hydrological state variables to be exchanged are obtained, and the three-dimensional spatial coordinates of the target grid point are obtained.

[0040] A three-dimensional bilinear interpolation algorithm is used to calculate the interpolation of the macro-hydrological state variables at each target grid point, based on the values ​​of the macro-hydrological state variables at the grid points of the macro-hydrological model.

[0041] The calculated interpolation result is assigned to the corresponding target grid point as the boundary condition of the target grid point;

[0042] If the data from the macro-hydrological model and the data from the micro-hydrological model are not synchronized in time, a time interpolation method is used to match the time series of the macro-hydrological state variables to the time step calculated by the micro-hydrological model.

[0043] The variable set after spatial and temporal matching is integrated with the unexchanged state variables within the micro-hydrological model, and the output is a consistent multi-scale model state.

[0044] Preferably, the step of assimilating data based on the multi-scale model state and real-time acquired ground observation data using a gradient-optimized nonlinear least squares method to obtain updated hydrological forecast results includes:

[0045] Using the coupled multi-scale model state as the background field, a cost function is constructed to assimilate multiple observation times within a time window. The cost function includes a deviation constraint term between the background field and the state variable increment, and a residual constraint term between the observed value and the model simulation value at each observation time.

[0046] The gradient information of the cost function with respect to the increment of the state variables is calculated using adjoint model techniques or perturbation methods.

[0047] A nonlinear optimization algorithm is used to iteratively optimize the cost function based on the gradient information until the convergence condition is met, thereby obtaining the optimal state analysis increment that minimizes the cost function.

[0048] The optimal state analysis increment is added to the background field to obtain the updated hydrological state analysis field, which is used as the updated hydrological forecast result.

[0049] Preferably, the method further includes:

[0050] Based on the system's real-time resource utilization rate, task calculation timeout status, real-time precipitation intensity, and key water level rise rate, the calculation priority of macro-hydrological simulation, micro-hydrological simulation, state variable exchange, and data assimilation is dynamically determined.

[0051] Based on the computation priority, computing resources are allocated to each task;

[0052] When the system's real-time resource utilization rate is continuously higher than the first resource threshold, or the actual processing time of a task exceeds the preset maximum allowable processing time, adaptive degradation control is triggered and executed. The adaptive degradation control implements one or more measures in stages according to the severity of resource shortage or timeout, including reducing the spatial resolution of micro-models in non-key areas, reducing the size of data assimilation sets, suspending the calculation of micro-hydrological models in non-key areas, and reducing the forecast update frequency.

[0053] Secondly, embodiments of the present invention provide a flood prediction device based on rainfall radar, comprising:

[0054] The calibration module is used to perform quality assessment and dynamic calibration based on the rain measurement radar data to obtain the calibrated radar precipitation field.

[0055] The QPF sequence module is used to generate a short-term quantitative precipitation forecast QPF sequence for the next N hours based on the calibrated radar precipitation field and the radar echo motion field.

[0056] The hydrological simulation module is used to drive preset macro-hydrological models and micro-hydrological models to perform hydrological simulations according to the QPF sequence, thereby obtaining macro-hydrological state variables corresponding to the macro-hydrological model and micro-hydrological state variables corresponding to the micro-hydrological model. The macro-hydrological model covers the entire watershed and operates at a first spatial resolution, while the micro-hydrological model covers key areas within the watershed and operates at a second spatial resolution higher than the first spatial resolution. The macro-hydrological state variables include upstream boundary water level, lateral inflow discharge, soil moisture, and groundwater level, while the micro-hydrological state variables include grid water level, discharge, and surface water depth.

[0057] The coupling module is used to perform real-time exchange and coupling of state variables based on the macro-hydrological state variables and the micro-hydrological state variables to obtain the coupled multi-scale model state.

[0058] The data assimilation module is used to assimilate the data based on the state of the multi-scale model and the real-time acquired ground observation data using a nonlinear least squares method based on gradient optimization, so as to obtain updated hydrological forecast results.

[0059] The early warning module is used to generate flood early warning information based on the updated hydrological forecast results.

[0060] Thirdly, embodiments of the present invention provide an electronic device, including: at least one processor, at least one memory, and computer program instructions stored in the memory, which, when executed by the processor, implement the method of the first aspect described above.

[0061] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0062] This scheme significantly improves the spatial resolution and quantitative accuracy of the input precipitation field by establishing a preprocessing link for radar data quality assessment and hierarchical dynamic calibration, providing highly reliable driving data for subsequent hydrological simulations. Through the coupling of macroscopic and microscopic dual models and real-time state exchange, it achieves synchronization and coordination between the overall watershed hydrological response and local refined inundation simulation, enhancing the physical rationality and spatial consistency of multi-scale prediction results. The introduction of a gradient-optimized nonlinear least squares method for data assimilation, continuously utilizing real-time observation data to correct the model state, effectively suppresses error accumulation and improves the timeliness accuracy and uncertainty quantification capability of forecasts. Attached Figure Description

[0063] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0064] Figure 1 This is a flowchart illustrating the flood prediction method based on rainfall radar provided by the present invention.

[0065] Figure 2 A schematic diagram of the structure of the flood prediction device based on rainfall radar provided by the present invention;

[0066] Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0067] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0068] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0069] It should be noted that all actions involving the acquisition of signals, information, or data in this invention are carried out in compliance with the relevant data protection laws and regulations of the locality and with authorization from the owner of the relevant device.

[0070] Example 1

[0071] Please see Figure 1 This invention provides a flood prediction method based on rainfall radar, comprising:

[0072] S1. Based on the rainfall radar data, perform quality assessment and dynamic calibration to obtain the calibrated radar precipitation field;

[0073] This step aims to transform raw rainfall radar data into a reliable spatial precipitation input field. Rainfall radar data refers to data obtained by fusing hydrological rainfall radar and meteorological radar. Quality assessment involves identifying various errors and unreliable components in the data through a series of quantitative and qualitative analyses. These errors may originate from terrain obstruction, electromagnetic wave attenuation, non-meteorological clutter interference, and instrument noise. Dynamic calibration is a technique that, instead of using fixed parameters, adaptively selects and adjusts calibration algorithms and parameters based on real-time data quality assessments, terrain features, and precipitation types to correct the aforementioned errors. The calibrated radar precipitation field is the output of this step; it is a spatially continuous, temporally serialized, and systematically error-corrected quantitative precipitation intensity field, serving as the mandatory input driver for subsequent hydrological models.

[0074] During implementation, the system first performs a comprehensive error diagnosis on the input radar data, generating quantitative indicators reflecting the reliability of each spatial data point. Subsequently, based on these indicators, the system divides the entire detection area into sub-regions with different quality levels. For each sub-region of different quality levels, the system calls upon a matching calibration algorithm library. For example, for high-reliability regions, a parameter optimization method based on climatological statistical laws is used, while for low-reliability regions, a multi-source data fusion mechanism is activated to compensate for missing information. This process, by dynamically linking the quality assessment results with calibration decisions, ensures that, under complex terrain and weather conditions, the final generated precipitation field retains the high spatiotemporal resolution advantage of the radar data while significantly suppressing systematic biases, providing more accurate and reliable precipitation input for subsequent hydrological simulations.

[0075] Furthermore, rainfall radar data can be processed by time synchronization and spatial coordinate unification of raw base data from different radar networks; a fusion algorithm using radar detection capability, terrain obstruction degree, and distance to the target area as weights can be employed to generate a spatiotemporally consistent fused reflectivity factor field covering the entire basin. Meteorological radar data: typically S- or C-band, with wide coverage (e.g., radius 250-460 km), long volume scan cycles (e.g., 6-10 minutes), and data products include base data (reflectivity, radial velocity, etc.) and some derived products. Water conservancy rainfall radar data: typically C- or X-band, station deployment targets key flood control areas of the basin, with high spatial density and potentially more flexible scanning strategies (e.g., sector scanning of key areas), but relatively smaller coverage.

[0076] In some implementations, S1, quality assessment and dynamic calibration are performed based on rainfall radar data to obtain a calibrated radar precipitation field, including:

[0077] S11. Based on the rain measurement radar data, perform multi-level quality assessment and generate a comprehensive quality score for each grid point in the rain measurement radar data;

[0078] The comprehensive quality scoring standard used for the quality assessment is a specialized evaluation system for multi-source fusion precipitation radar data, jointly developed by meteorological and hydrological departments to address common needs in flood forecasting. This system comprehensively considers the meteorological department's requirements for radar data observation accuracy and stability, as well as the hydrological department's requirements for the spatiotemporal continuity and consistency of precipitation input. Key scoring factors include: data acquisition stability and completeness, spatiotemporal consistency, and quantitative accuracy, while incorporating physical factors such as spatial representativeness differences and radar range attenuation into the specific scoring. This step is crucial for refined reliability diagnosis of the raw radar data. The comprehensive quality score is a quantitative indicator used to comprehensively characterize the overall reliability level of a single radar data grid point under the influence of multi-dimensional error sources. Its value is typically normalized to between 0 and 1, with higher values ​​indicating higher data reliability. The purpose of generating this score is to provide an objective basis for subsequent classification and differentiated calibration, avoiding a "one-size-fits-all" approach to all data. Multi-level quality assessment refers to the systematic analysis of radar data quality from different levels and perspectives. This typically encompasses the analysis of the characteristics of the original signal itself, the verification of the rationality of the spatial distribution of the data, and the validation of the consistency with the meteorological and physical laws it reflects. Each level further includes several specific evaluation factors, such as signal strength, spatial continuity, vertical structure characteristics, and the degree of matching with climate statistics.

[0079] In its implementation, the system first executes multiple quality assessment algorithms in parallel, each targeting a specific type of error source or unreliable factor. For example, one algorithm focuses on identifying missing or underestimated data due to terrain occlusion or beam attenuation; another analyzes the spatial texture characteristics of echoes to filter out non-meteorological clutter; and yet another checks whether the conversion relationship between reflectance and precipitation rate conforms to local climatic characteristics. Each algorithm outputs one or more basic quality scores for each grid point. Subsequently, the system integrates these basic scores from different dimensions, reflecting different aspects of quality, according to pre-defined weighted fusion rules. The weighting may consider the terrain characteristics of the region; for example, in mountainous areas, scores for terrain occlusion and attenuation correction are given more weight, while in plains, scores for clutter identification and spatial consistency are given more weight. Finally, through weighted calculation, a unique comprehensive quality score is generated for each grid point. This score serves as a "data quality ID," comprehensively and quantitatively reflecting the reliability of the data at that point.

[0080] In some embodiments, S11 involves performing a multi-level quality assessment based on the rain-measuring radar data, generating a comprehensive quality score for each grid point in the rain-measuring radar data, including:

[0081] S111. Based on the terrain obstruction information, non-meteorological clutter characteristics, and original signal quality in the rain measurement radar data, the terrain obstruction score, clutter identification score, and signal quality score for each grid point are calculated respectively.

[0082] This step focuses on a fundamental assessment of the physical signal reliability of the radar data itself. The terrain obstruction score quantifies the risk of missing or severely underestimated echo data due to terrain obstacles (such as mountains) blocking the radar beam. This score is typically calculated using a ray tracing algorithm based on a high-precision digital elevation model. The clutter identification score assesses the proportion of interference components originating from non-precipitating targets (such as ground objects, birds, and anomalous propagation) in the echo signal. This score is generated by analyzing the differences between the echo's spatial texture, Doppler velocity, and polarization characteristics (such as differential reflectivity and correlation coefficient) and typical meteorological echoes. The signal quality score characterizes the original radar signal's signal-to-noise ratio, the presence of velocity ambiguity, and other basic properties. These three scores, starting from the physical level of data acquisition, collectively constitute the core indicators for assessing the reliability of the data source.

[0083] S112. Based on the radar reflectivity data of each grid point, analyze the consistency between the grid point and its neighboring grid points in the horizontal and vertical directions, and calculate the spatial consistency score and the vertical consistency score respectively.

[0084] This step evaluates the data from the perspectives of statistical plausibility and physical structure in its spatial distribution. The spatial consistency score measures the horizontal similarity between the radar reflectivity of a target grid point and the reflectivity of its neighboring grid points, typically achieved by calculating local variance or gradient. A low score may indicate that the point is an isolated anomaly or is subject to local interference. The vertical consistency score examines whether the vertical distribution of reflectivity exhibits typical characteristics of continuous meteorological echoes, such as vertical continuity of the reflectivity core with increasing altitude, or the presence of anomalous discontinuities. This score is generated by analyzing the vertical profiles of scan data at different elevation angles. Together, these two scores verify the inherent plausibility and physical coherence of the data's three-dimensional spatial distribution, effectively identifying data points that appear spatially or physically anomalous or potentially unreliable.

[0085] S113. Based on the radar reflectivity data of consecutive time intervals, analyze the characteristics of reflectivity variation of each grid point over time, and calculate the time consistency score.

[0086] This step aims to assess the continuity and physical plausibility of radar data over time, which is particularly crucial for fusing multi-source, multi-mode radar data (such as hydraulic and meteorological radar). The temporal consistency score quantifies the smoothness and physical interpretability of the reflectivity variation of target grid points across consecutive observations. Abnormally sharp jumps often indicate noise, mode-switching interference, or systematic bias between different radar sources.

[0087] Set target grid points In continuous Each time period ( The reflectance sequence of ) is ,in .

[0088] a) Evaluation of local temporal gradient:

[0089] Calculate the absolute value of the reflectance difference between adjacent time intervals:

[0090] ;

[0091] Given a maximum reasonable change threshold based on climate statistics or physical constraints (For example, the reflectance variation per minute of stratiform cloud precipitation is typically less than 1 dBZ, while convective cells may allow for larger but limited variations.) Temporal gradient scoring It can be calculated using the following formula:

[0092] ;

[0093] in, This is a scaling parameter used to control the rate of score decay. The closer it is to 1, the smoother and more reasonable the change over time.

[0094] b) Time alignment and consistency evaluation of multiple radar sources:

[0095] When integrating water conservancy radar (volume scan cycle of 1 minute) and meteorological radar (volume scan cycle of 6 minutes), time alignment is required. Assume the data originates from the radar source. and The observations, when interpolated to the same standard time grid, The reflectances after are respectively and .

[0096] First, the difference between observations from different radar sources at the same standard time is calculated as follows: :

[0097] ;

[0098] Considering the differences in wavelength, calibration, and attenuation characteristics among different radars, an allowable cross-source deviation threshold is set. .

[0099] Calculate the mean absolute deviation of all available standard times within the evaluation time window. and the standard deviation of the mean absolute deviation :

[0100] ;

[0101] ;

[0102] Time consistency score can be defined as:

[0103] ;

[0104] in, and To adjust the parameters, This prevents division by zero for small constants. The score penalizes both excessive mean deviation (systematic bias) and excessive deviation fluctuation (time mismatch or noise).

[0105] c) Final time consistency score:

[0106] For a single radar source, time consistency score .

[0107] For multi-radar fusion regions, temporal consistency scoring can combine gradient scoring and cross-source consistency scoring:

[0108] ;

[0109] Among them, weight It can be dynamically adjusted based on whether the grid point is covered by multiple radars (e.g., single radar coverage area). Multiple radar overlapping areas ).

[0110] This scoring system can effectively identify data jumps and inconsistencies caused by asynchronous radar volume scan modes, incorrect self-reported timestamps, or poor alignment between different radar systems, providing crucial temporal quality information for subsequent dynamic calibration (especially multi-source fusion calibration in low-quality areas).

[0111] S114. Based on the physical relationship between radar reflectivity and precipitation rate at each grid point, and using available ground rain gauges or satellite precipitation data for cross-validation, calculate the physical consistency score and cross-validation score respectively.

[0112] This step provides final confirmation from the perspective of meteorological physical laws and external verification of multi-source data. The physical consistency score assesses whether the conversion relationship between radar reflectivity factor and estimated precipitation rate conforms to the physical and statistical characteristics of typical precipitation in the region (such as stratiform cloud precipitation and convective precipitation), i.e., assessing the physical rationality of the ZR relationship used. The cross-validation score directly compares the radar-estimated precipitation with measured precipitation data from ground rain gauges or independent satellite precipitation estimation products from the same time and region. It objectively measures the external consistency accuracy of the radar-estimated precipitation by calculating the differences between the two (such as average error and correlation coefficient). These two scores elevate the evaluation from purely signal and data spatial analysis to a level of verification with actual precipitation physics and independent observations.

[0113] S115. Based on the meteorological-hydrological joint scoring standard, integrate various basic scores to generate a comprehensive quality score.

[0114] This step, based on a standardized fusion formula jointly developed by meteorological and hydrological departments, integrates the multiple basic quality scores generated in the preceding steps. (in Representing rating categories, such as: Terrain occlusion score Clutter identification scoring Signal quality score Spatial consistency score Vertical consistency score Time consistency score Physical consistency score (including cross-validation scores, etc.) are combined into a comprehensive quality score. .

[0115] Considering the varying contributions of different scoring factors to the reliability of hydrological applications and meteorological observations, and the significant impact of spatial location (such as distance from the radar station and terrain) on radar data quality, the fusion formula employs dynamic weight allocation. Comprehensive Quality Score The calculation formula is as follows:

[0116] ;

[0117] in, The weighting of basic industries, jointly determined by meteorological and hydrological experts, reflects the first... The scoring factors have fixed weights based on their importance in general business scenarios. For example:

[0118] Data stability and integrity-related scores (signal quality, time consistency) may be given a higher base weight to ensure the continuous reliability of business systems.

[0119] Quantitative accuracy-related scores (cross-validation, physical consistency) are also given high weight, directly determining the accuracy of precipitation input.

[0120] Spatiotemporal consistency score is an important indicator for internal error checks, and its weight is appropriate.

[0121] This is a spatially adaptive modulation factor, which is related to the grid points. Geographic location-related factors are used to dynamically adjust the effective weights of each score based on spatial differences and distance scale factors. For example:

[0122] Terrain occlusion score In areas with complex terrain (such as mountainous regions), its modulation factor The influence of topography should be emphasized and increased.

[0123] For scores related to range attenuation: in the long-range region of radar, the modulation factor should decrease with increasing range, or the score result itself should be corrected for range.

[0124] For multi-radar fusion areas: the modulation factor of the spatiotemporal consistency score can be increased to rigorously check the consistency of data from different sources.

[0125] For the basic score of grid points, the basic quality score values ​​of each item calculated in the previous steps are normalized to the interval [0,1], where 1 represents the best.

[0126] S12. Based on the comprehensive quality score of each grid point, divide all grid points into high-quality areas, medium-quality areas, and low-quality areas.

[0127] This step aims to spatially cluster and classify the radar coverage area based on the comprehensive quality score generated in the previous step, forming structured processing partitions. High-quality, medium-quality, and low-quality areas are three spatial subdomains defined by thresholds based on the comprehensive quality score. Grid point data within high-quality areas are considered least affected by various errors and best reflect actual precipitation; data in medium-quality areas contains a certain degree of uncertainty or error; and data in low-quality areas has the lowest reliability and may contain serious biases or missing information. The purpose of this type of grid point division is to provide a clear spatial framework for implementing differentiated and targeted calibration strategies, optimize the allocation of computational resources, and avoid over-processing of high-confidence data or insufficient correction of low-confidence data.

[0128] In implementation, the system first sets scoring thresholds to distinguish between the three quality levels. These thresholds may be determined based on statistical analysis of historical data, error propagation theory, or expert experience. Subsequently, the system iterates through all grid points, classifying them into the corresponding initial quality level based on their comprehensive quality score. Since radar data errors often exhibit spatial continuity, simple point-based classification may generate numerous scattered points, hindering regional processing. Therefore, the system typically employs spatial clustering or region growing algorithms to aggregate spatially adjacent grid points belonging to the same quality level into continuous physical regions with well-defined boundaries. For example, on the leeward slope of a mountain affected by terrain shadows, most grid points may be aggregated into a contiguous low-quality area; while in open plains with good radar detection conditions, a large area of ​​high-quality data may be formed. This regional division based on scoring and spatial adjacency not only clarifies the data quality status of different regions but also generates geographically meaningful processing unit maps that are easy to call in subsequent processes, making it possible to implement customized calibration strategies for regions of different quality levels.

[0129] S13. Based on the regional division results, dynamic calibration is performed using calibration strategies corresponding to each quality region to obtain the radar precipitation field.

[0130] This step is the execution phase of data quality control. Its core is to apply calibration interventions of varying intensity and method based on the inherent differences in data reliability. The calibration strategies corresponding to each quality zone refer to a set of pre-defined parameters, algorithms, or data processing procedures for high-quality, medium-quality, and low-quality zones. These strategies typically exhibit gradient differences in principle complexity and data dependency. Their design principle is: to apply minimal necessary intervention to high-reliability data to preserve its original information, while employing stronger, multi-source information-assisted correction methods for low-reliability data to maximally repair or replace unreliable information. Dynamic calibration emphasizes that the calibration process is not static; its specific execution details (such as fusion weights and reference data selection) can be fine-tuned based on real-time quality scores. The radar precipitation field is the final output of this step. It is a spatially continuous quantitative precipitation estimation field based on the original radar data, processed through this quality-zone-based, differentiated processing.

[0131] In practice, the system uses the quality area map generated in the previous step to call the preset calibration algorithm module for each area. For high-quality areas, the calibration strategy may be relatively simple, such as using optimized localized parameters obtained from a large number of historical samples to convert reflectance, or performing slight smoothing and denoising, with the goal of maintaining the core information of high-precision data. For medium-quality areas, the strategy is more complex and may introduce spatial constraints, such as using data from surrounding high-quality areas to constrain and correct the current area's estimate through spatial interpolation or statistical extrapolation, giving higher weight to neighboring high-quality data during calibration. For low-quality areas, the most complex calibration strategy is used, which usually actively integrates information from other independent data sources, such as point observations from ground rain gauges, precipitation products retrieved from satellites, and even the output of numerical weather prediction models. Through data fusion algorithms (such as optimal interpolation, variational assimilation, etc.), an optimal estimate under the joint constraints of multi-source information is generated to compensate for the severe deficiencies or biases of radar data in that area. Through this hierarchical and targeted calibration process, the system can achieve a reasonable match between calibration intensity and data reliability across the entire region, thereby producing a more accurate and reasonable radar precipitation field than that produced by homogenization, providing higher-quality driving data for downstream hydrological simulation.

[0132] In some implementations, S13, based on the region division results, dynamic calibration is performed using calibration strategies corresponding to each mass region to obtain the radar precipitation field, including:

[0133] S131. For the high-quality area, calibration is performed using an optimized averaging method based on the statistical relationship of local climate characteristics;

[0134] This step involves the simplest and most efficient calibration for the data area with the highest reliability. The optimized averaging method based on local climate characteristics utilizes the empirical conversion relationship between radar reflectivity and precipitation rate—the ZR relationship—obtained through statistical analysis of long-term historical observation data and best suited to the region. The parameters in this relationship are locally optimized to effectively reflect the average microphysical characteristics of regional precipitation. The calibration process primarily applies this relationship to directly convert reflectivity to precipitation rate, and then performs appropriate spatiotemporal smoothing or averaging on the conversion results to suppress random noise while preserving the true precipitation structure information inherent in the original high-precision data to the greatest extent possible. This method avoids introducing complex calculations or external uncertainties, achieving accurate quantification of high-quality basic data with minimal intervention while ensuring physical rationality.

[0135] S132. For the medium-quality region, a weighted calibration method combining spatial correlation and distance weight is used for calibration, wherein the calibration weight is related to the comprehensive quality score of the grid point and the distance between the grid point and the high-quality region.

[0136] This step aims to constrain and correct uncertainties in medium-quality data by utilizing reliable information from high-quality regions. The weighted calibration method, combining spatial correlation and distance weighting, is a spatial statistical interpolation method. Its principle is based on the assumption that the precipitation field is spatially continuous, and that data from neighboring high-quality regions are meaningful references for the true values ​​in medium-quality regions. The calibration weights are determined by two core factors: first, the data reliability of the target medium-quality grid point itself, i.e., its comprehensive quality score; the lower the score, the less reliable the data is, and the greater the need to rely on external references; second, the spatial distance between the grid point and surrounding high-quality reference grid points; the greater the distance, the lower the reference value, and the weight decreases according to the distance decay function. This calibration process replaces or corrects the original estimated value of the medium-quality point by calculating the weighted average of the calibrated precipitation values ​​of surrounding high-quality grid points, thereby "propagating" high-reliability information spatially to adjacent regions and effectively suppressing outliers in medium-quality regions caused by local errors or uncertainties.

[0137] S133. For the low-quality area, a fusion calibration method integrating multi-source auxiliary data is used for calibration, wherein the multi-source auxiliary data includes ground rain gauge data and satellite precipitation estimation data.

[0138] This step targets areas with the lowest data reliability, employing external independent information sources for powerful data reconstruction or replacement. The core of the multi-source auxiliary data fusion calibration method is to fuse precipitation estimation products from other sources as primary information when radar data itself is severely unreliable. Ground-based rain gauge data provides accurate point-based measurements, while satellite precipitation estimation data provides independent, relatively accurate isometric distribution information. The fusion process typically employs data integration algorithms such as optimal interpolation, statistical fusion, or variational assimilation. These algorithms consider not only the individual estimates from different data sources in the target area but also their known error statistics and spatial correlations. By solving a statistically optimal estimation problem, a final precipitation field is calculated that simultaneously fits the information from all data sources and satisfies certain spatial smoothing constraints. Essentially, this method constructs a synthetic precipitation field in areas where radar data is ineffective, anchored by the ground station network and spatially framed by satellite data, thereby maximally compensating for the lack or severe bias of radar information.

[0139] S2. Based on the calibrated radar precipitation field and combined with the radar echo motion field, generate a short-term quantitative precipitation forecast (QPF) sequence for the next N hours. This step aims to use the calibrated radar precipitation field and radar echo motion information to forecast precipitation in the short term, providing input for flood lead-ahead simulation. The short-term quantitative precipitation forecast (QPF) refers to a quantitative precipitation prediction sequence for the next few hours, generated based on radar echo extrapolation or fusion with numerical models. The radar echo motion field is extracted from radar reflectivity data over consecutive time intervals using optical flow, semi-Lagrange advection algorithms, or deep learning models, and is used to describe the movement and evolution trend of the precipitation system. The QPF sequence is the output of this step; it is a temporally continuous and spatially consistent sequence of future precipitation intensity fields, serving as the core driving data for subsequent hydrological models to simulate lead-ahead floods.

[0140] In implementation, the system first generates an echo motion vector field based on multi-time-series radar echo data using a motion estimation algorithm. Then, using the calibrated current radar precipitation field as the initial field, advection extrapolation is performed along the motion vector to generate hourly precipitation forecasts. To further improve forecast accuracy, the system can integrate precipitation forecast products from high-resolution numerical weather prediction models, correcting and optimizing the simple extrapolation QPF through variational assimilation or statistical fusion methods. The final generated QPF sequence retains the high spatiotemporal resolution advantage of radar data while enhancing the physical rationality and accuracy of short-term precipitation forecasts through the fusion of forecast information, providing reliable future precipitation input for hydrological models.

[0141] S3. Based on the QPF sequence, perform hydrological simulations using the preset macro-hydrological model and micro-hydrological model respectively to obtain the macro-hydrological state variables corresponding to the macro-hydrological model and the micro-hydrological state variables corresponding to the micro-hydrological model; wherein, the macro-hydrological model covers the entire basin and operates at a first spatial resolution, and the micro-hydrological model covers key areas within the basin and operates at a second spatial resolution higher than the first spatial resolution.

[0142] This step utilizes the previously obtained QPF sequence to drive a multi-resolution hydrological model system for parallel hydrological process simulation. The macro-hydrological model is a distributed numerical model that focuses on a large-scale watershed, employs a relatively coarse spatial grid, and primarily simulates the main processes of the hydrological cycle, such as runoff generation and confluence. The micro-hydrological model, on the other hand, focuses on key local areas within the watershed, employs a fine spatial grid, and can characterize detailed processes such as surface runoff, drainage networks, and water obstruction by structures. The first and second spatial resolutions refer to the computational grid sizes used in the two models, respectively. The second spatial resolution is smaller than the first, indicating that the micro-model has a more refined spatial description capability. Hydrological simulation refers to the model's deduction of the spatiotemporal changes in water bodies within a watershed by solving a series of physical equations or empirical relationships describing water movement and transformation, given precipitation input and initial conditions.

[0143] In implementation, the system downscales and resamples a unified radar precipitation field according to the different spatial grid structures of the two models, forming the driving data required for each. The macroscopic model runs across the entire watershed, rapidly simulating large-scale runoff generation, river network confluence, and channel evolution, outputting a series of macroscopic state variables reflecting the overall hydrological response of the watershed. Simultaneously, the microscopic model, within pre-defined key areas, utilizes detailed topographic and feature data to simulate detailed processes such as urban flooding and localized inundation, outputting high-resolution microscopic state variables. This dual-model parallel architecture achieves a balance between rapid understanding of the overall hydrological situation of the watershed and detailed characterization of inundation risks in key areas. By keeping the computational load of large-scale simulations within a reasonable range, while precisely allocating high-computational-cost hydrodynamic calculations to the areas of greatest interest, a balance between breadth and depth of prediction is achieved within limited computational resources.

[0144] Furthermore, the macro-hydrological model is a basin-wide lumped hydrological model with a first spatial resolution between 500 meters and 5 kilometers and a simulation time step between 5 minutes and 1 hour.

[0145] The micro-hydrological model is a Buergy hydrological model with a second spatial resolution between 5 meters and 100 meters and a simulation time step between 1 second and 5 minutes.

[0146] In a preferred embodiment of the present invention, the macro-hydrological model adopts a grid-based Xin'anjiang model or a TOPKAPI model. The first spatial resolution is preferably 1 km, and the simulation time step is preferably 15 minutes. The parameters required for this model include, but are not limited to: a watershed digital elevation model (DEM), land use / cover data, soil type data, and river network data. Key state variables include, but are not limited to: soil moisture content, surface water depth, groundwater level, and water level and flow rate at river cross-sections for each grid cell.

[0147] The micro-hydrological model employs a flood evolution model based on two-dimensional shallow water equations (such as a model solved using the finite volume method). The second spatial resolution is preferably 10 meters, and the simulation time step is dynamically adjusted according to CFL conditions, typically ranging from 1 to 3 seconds. The required input data for this model includes: high-precision DEM (accuracy better than 5 meters), structure outlines, river cross-sections, and hydraulic engineering facility data. Key state variables include, but are not limited to: water depth and two-dimensional velocity vectors for each computational grid.

[0148] S4. Based on the macro-hydrological state variables and the micro-hydrological state variables, perform real-time exchange and coupling of state variables to obtain the coupled multi-scale model state.

[0149] This step is crucial for enabling collaborative work between macroscopic and microscopic models, aiming to facilitate information exchange and state coordination between the two models at different scales during simulation. Real-time exchange and coupling of state variables refers to establishing a two-way data transfer and coordination mechanism. This allows the large-scale boundary conditions and driving forces provided by the macroscopic model to be transmitted to the microscopic model, while the local results simulated by the microscopic model can also be fed back appropriately and potentially influence the evolution of the macroscopic model. The exchange is triggered by preset rules, such as when the hydraulic conditions at the microscopic boundary change drastically, or when there are significant differences in the simulation results of the two models at the same spatial location. The coupled multi-scale model state is the output of this process. It represents a logically self-consistent and unified system state description at both the overall and local levels, achieved through inter-scale information coordination and consistency processing.

[0150] To achieve the aforementioned coupling, the system first defines the set of key state variables to be exchanged between the two models and their transmission direction based on physical correlation. During the simulation time step, the system continuously monitors the pre-set coupling trigger criteria. Once the conditions are met, the exchange process is initiated: the macro model transforms its variables such as water level and flow rate at the micro-region boundary into boundary condition inputs that conform to the mesh accuracy of the micro model through spatial interpolation methods; simultaneously, the micro model may also summarize its simulated total outflow and other information and feed it back to the upstream boundary of the macro model. Subsequently, the system compares the states of the relevant regions before and after the exchange, evaluates the consistency of the coupling, and initiates corresponding parameter fine-tuning or weight allocation based on the degree of deviation to ensure that the simulations at the two scales do not produce physically contradictory discrepancies. Through this dynamic, bidirectional information exchange and state coordination, the two originally independent models are integrated into an organic whole, enabling large-scale hydrological evolution to constrain local fine-grained simulations, while local fine-grained results can correct deviations in large-scale simulations, thereby significantly improving the physical rationality and predictive consistency of the overall simulation of complex watershed systems.

[0151] In some implementations, S4 involves real-time exchange and coupling of state variables based on the macro-hydrological state variables and the micro-hydrological state variables to obtain the coupled multi-scale model state, including:

[0152] S41. Based on the preset exchange trigger conditions, determine the timing and content of state variable exchange. The exchange trigger conditions include forced variable exchange and state variable exchange. The forced variable exchange must be performed in each coupling calculation cycle. The exchanged variables include at least the upstream boundary water level and the lateral inflow flow. The state variable exchange is performed conditionally and is triggered when the difference in soil moisture between the micro-region and the corresponding macro-region exceeds a first preset threshold or the rate of change of groundwater level exceeds a second preset threshold.

[0153] Exchange triggering conditions are a set of predefined logical rules used to determine when and what type of state variables should be exchanged. The purpose is to make data transfer between models intelligently selective, avoiding unnecessary frequent exchanges and balancing computational cost with coupling accuracy. It is designed to include two basic types. Mandatory variable exchange defines the exchange content that must be executed unconditionally in each preset coupling computation cycle, ensuring the most basic collaborative operation of the coupled system. Its exchange content is limited to driving forces that have a global and decisive impact on the hydrological response of the downstream or water-receiving area, typically including at least the upstream boundary water level process line and lateral inflows, providing critical boundary inputs for the downstream model. State variable exchange, on the other hand, is conditionally triggered, its execution depending on whether the difference or change between the two models in a specific state reaches a level requiring intervention. Specific triggering criteria include when the cumulative difference in soil moisture content between the micro-region and its corresponding macro-region exceeds a set tolerance limit, or when an abnormally rapid rise and fall rate of groundwater level is detected, exceeding a set rate of change threshold.

[0154] During system operation, the macroscopic and microscopic models integrate independently according to their respective step sizes. The coupling control module continuously monitors the state of the two models and performs real-time comparisons. On one hand, it triggers forced exchanges at fixed time intervals to ensure synchronization of the hydrological driving force basis. On the other hand, it calculates key indicators such as soil moisture differences and groundwater level change rates. Once these indicators exceed their set thresholds, it indicates a significant deviation in the two scale models' description of the same physical process. In this case, state variable exchange is immediately triggered to transmit the latest state information and reconcile this inconsistency. This hybrid triggering mechanism combines the reliability of time-driven systems with the sensitivity of event-driven systems. It ensures that the coupled system can maintain synchronized operation with low overhead during stable periods, while intervening promptly when critical states undergo drastic changes or significant divergences occur. Data exchange prevents error accumulation and model drift, thus achieving a dynamic balance between computational efficiency and coupling accuracy.

[0155] S42. When the exchange triggering condition is met, the macroscopic hydrological state variables are mapped to the grid of the microscopic hydrological model through a spatial interpolation algorithm to obtain the multi-scale model state.

[0156] Spatial interpolation is a mathematical method used to estimate data from one set of source grid points to another set of target grid points. In coupled modeling, because macroscopic and microscopic models use computational grids with different spatial resolutions, the state variables output by the macroscopic model cannot be directly assigned to the microscopic model's grid. Spatial interpolation is a technique to solve this scale mismatch problem. Based on spatial proximity or geometric relationships, it constructs a continuous function estimate from coarse-resolution source data to fine-resolution target data, thereby downscaling and transferring macroscopic information to the microscopic grid. The output obtained through this process, namely the multi-scale model state, is a comprehensive product after data fusion. It includes both the transferred and corrected macroscopic information (as boundaries or initial conditions) and the states independently calculated within the microscopic model. It represents a snapshot of a physically consistent model system at the current time step after inter-scale information coordination.

[0157] After determining in step S41 that an exchange is needed, the system enters the mapping phase. At this point, the macroscopic state variables to be exchanged are located on the coarse grid nodes of the macroscopic model. The system first identifies the specific regions in the microscopic model that need to receive this information (usually the coupling boundary region or the entire computational domain) and obtains the precise spatial coordinates of all its grid points. Next, the system calls a preset spatial interpolation algorithm. This algorithm uses the variable values ​​on the macroscopic grid nodes as known sample points and calculates an estimated value based on the spatial relative position of the microscopic target point and these macroscopic nodes. For example, for a microscopic viewpoint located between four macroscopic grid points, the algorithm may calculate the value of the microscopic viewpoint by weighting the values ​​of these four points based on an inverse distance ratio or a bilinear relationship. After completing the interpolation calculation for all target points, these newly generated values ​​are assigned to the corresponding grid of the microscopic model as boundary conditions or initial fields for its next calculation. Finally, the system integrates the grid states inside the microscopic model that are not affected by the exchange with these newly input boundary states to form a new, unified snapshot of the system state. This snapshot represents the current state of the multi-scale model, enabling the micro-model to conduct fine-scale physical process simulations under the reasonable constraints provided by the macro environment. This ensures the consistency between local simulations and the overall hydrological situation, and is a concrete manifestation of macro-level guidance for micro-level simulations and local simulations being controlled by global evolution.

[0158] In some implementations, S42, when the exchange triggering condition is met, the macroscopic hydrological state variables are mapped to the grid of the microscopic hydrological model using a spatial interpolation algorithm to obtain the multi-scale model state, including:

[0159] S421. Identify the micro-hydrological model region where state variable exchange is required, and define all computational grid points of the micro-hydrological model within the micro-hydrological model region as target grid points.

[0160] Target grid points are the set of grid points in a micro-hydrological model that are designated to receive state variable information from the macro-hydrological model. Identifying the micro-hydrological model region is the process of determining the information receiving range. This region is usually predefined based on physical connections, such as a key micro-region that is a sub-unit of the macro-watershed, or a region whose boundaries are directly affected by macro-flow. All micro-model computational grid points located within this region are classified as target grid points, forming the receiver set for subsequent spatial interpolation operations.

[0161] S422. When the exchange triggering condition is met, obtain the macroscopic hydrological state variable to be exchanged, and obtain the three-dimensional spatial coordinates of the target grid point;

[0162] This step is the initialization preparation for data mapping. The macroscopic hydrological state variables to be exchanged refer to the specific data items that need to be transferred according to the aforementioned exchange triggering conditions, such as water level or soil moisture. Obtaining the three-dimensional spatial coordinates of the target grid points means extracting the position information of all target grid points defined in S421 in the horizontal and vertical directions from the grid system of the microscopic hydrological model. This coordinate information is the geometric basis for connecting the macroscopic data positions with the microscopic grid positions and performing spatial transformation.

[0163] S423. Using a three-dimensional bilinear interpolation algorithm, based on the values ​​of the macro-hydrological state variables at the grid points of the macro-hydrological model, the interpolation of the macro-hydrological state variables at each target grid point is calculated.

[0164] The three-dimensional bilinear interpolation algorithm is a spatial interpolation method for three-dimensional regular grid data. This method uses the eight nearest macroscopic grid points around a target point as a reference and estimates the value of the target point through cubic linear interpolation calculations. Specifically, it performs linear interpolation in each spatial dimension and finally synthesizes a continuous estimate in three-dimensional space. Interpolation is the mathematical process of estimating the value of an unknown point using the values ​​of known discrete points. In this step, the state variable values ​​of the macroscopic grid points are known, and the coordinates of the target grid point are known. Through the three-dimensional bilinear interpolation algorithm, a spatially smoothed estimate derived from macroscopic data but conforming to its microscopic location can be generated for each target grid point.

[0165] S424. Assign the calculated interpolation result to the corresponding target grid point as the boundary condition of the target grid point;

[0166] This step completes the physical assignment of the data. Assignment refers to writing or associating the calculated values ​​into the storage cells or variables of specific grid points in the micro-hydrological model. As boundary conditions for the target grid points, these interpolated data from the macro-model will be used as constraints to define the boundary states or drive the internal calculations of the micro-model in subsequent calculations, such as water level or flux as inflow boundaries.

[0167] S425. If the data from the macro-hydrological model and the data from the micro-hydrological model are not synchronized in time, a time interpolation method shall be used to match the time series of the macro-hydrological state variables to the time step calculated by the micro-hydrological model.

[0168] Temporal interpolation is a technique used to handle differences in time scales. Time synchronization is necessary when the time step or moment output by the macroscopic model is inconsistent with the moment when the microscopic model needs to receive data. A time series is a discrete data sequence in which macroscopic state variables change over time. Temporal matching refers to the process of interpolating or resampling macroscopic data from one time series to another. This step uses a temporal interpolation algorithm to calculate the corresponding value of the macroscopic variable at the current calculation moment of the microscopic model based on the value of the macroscopic variable at a known time point, ensuring that the boundary conditions input to the microscopic model are also continuous and appropriate in the time dimension.

[0169] S426. Integrate the variable set after spatial and temporal matching with the unexchanged state variables within the micro-hydrological model, and output a consistent multi-scale model state.

[0170] This step is the final stage in completing the model state synthesis. The variable set refers to the macroscopic state variables that have been processed by S423 and S425 and are synchronized spatially and temporally. The unexchanged state variables within the micro-hydrological model refer to all other grid points and state variables in the micro-model that have not been affected by this exchange, such as water depth and flow velocity within the model's internal grids. Integration involves merging the exchanged variables from the outside with the model's self-sustaining variables to form the complete and consistent internal state of the micro-model at that moment. The output, coordinated and consistent multi-scale model state, refers to a state representation that is coordinated and unified in space, time, and variable relationships, obtained after all the above matching and integration processes. It serves as the basis for further integration of the micro-model or as the output of the entire coupled system, reflecting the latest state of the micro-system under macroscopic information constraints.

[0171] S5. Based on the state of the multi-scale model and the real-time ground observation data, the data is assimilated using the gradient-optimized nonlinear least squares method to obtain the updated hydrological forecast results.

[0172] The core of this step is to transform the data assimilation problem into a nonlinear least squares optimization problem. The goal is to find an optimal system state that, considering the model background field and observation errors, is numerically neither too far removed from the background field nor too far from the observation sequence over a period of time.

[0173] In some implementations, S5 involves assimilating the data based on the multi-scale model state and real-time acquired ground observation data using a gradient-optimized nonlinear least squares method to obtain updated hydrological forecast results, including:

[0174] S51. Using the coupled multi-scale model state as the background field, construct a cost function for assimilating multiple observation times within the time window. The cost function includes a deviation constraint term between the background field and the state variable increment, and a residual constraint term between the observed value and the model simulation value at each observation time.

[0175] S52. Calculate the gradient information of the cost function with respect to the increment of the state variables using the adjoint model technique or perturbation method;

[0176] S53. Using a nonlinear optimization algorithm, the cost function is iteratively optimized based on the gradient information until the convergence condition is met, and the optimal state analysis increment that minimizes the cost function is obtained.

[0177] S54. Add the optimal state analysis increment to the background field to obtain the updated hydrological state analysis field, which is used as the updated hydrological forecast result.

[0178] Specifically, set The background scene is from S4. Within the time window The observation vector at time step. Define the state analysis increment. Construct the cost function:

[0179] ;

[0180] in, The background error covariance matrix, for Nonlinear observation operator at time step, Let be the observation error covariance matrix. The first term represents the background constraint, and the second term represents the observation constraint.

[0181] To efficiently solve this large-scale optimization problem, it is necessary to calculate the cost function. Relative to increment gradient The gradient expression is:

[0182] ;

[0183] In the formula Let be the Jacobian matrix of the observation operator. In practical calculations, the gradient vector can be calculated efficiently and in one step using the adjoint model of the inverse integral hydrological model, without the need to explicitly construct and store the large Jacobian matrix.

[0184] Optimization is performed using gradient-based iterative algorithms (such as the nonlinear conjugate gradient method or the quasi-Newton method L-BFGS). Starting from the initial guess (which could be...),... Beginning with: In each iteration:

[0185] Run the model and observation operator in the forward direction to calculate the cost function value under the current increment. .

[0186] Run the adjoint model in reverse and calculate the gradient. .

[0187] Based on the optimization algorithm rules, determine the search direction and step size, and update the state increment. .

[0188] Loop until When the norm of the change or gradient is less than a preset threshold, the optimal analytical increment is obtained. .

[0189] Optimal analysis increment This represents the statistically optimal correction to the background field. Adding it to the background field yields the assimilated optimal state estimate:

[0190] ;

[0191] This analysis field This is an updated initial field for hydrological forecasting that incorporates the latest observational information, and can serve as the starting point for deterministic forecasts.

[0192] To further quantify forecast uncertainty, in a preferred embodiment of the present invention, the ensemble forecasting concept is combined with the aforementioned variational framework. Specifically: before S51, the background field... Generate a set of states to estimate the background error covariance matrix of the flow dependency. This makes it more consistent with the error structure under current hydrological and meteorological conditions. After S54, ensemble transformation techniques (such as ETKF) are used to transform the analysis increments. The information is reasonably allocated to each set member, thereby generating an analytical set consistent with the statistics of the analytical field, which is used for subsequent probability forecasting and to provide an updated error covariance estimate for the next assimilation cycle.

[0193] S6. Generate flood warning information based on the updated hydrological forecast results.

[0194] This step transforms numerical weather prediction results into disaster risk information that decision-makers can understand. Flood warning information is a structured risk notification product. Its core is to compare and interpret predicted hydrological indicators with pre-set risk thresholds to form qualitative or semi-quantitative conclusions about the timing, location, severity, and impact of potential floods. The process of generating warning information involves not only simple threshold judgments but also a comprehensive analysis of forecast uncertainties, risk propagation paths, and possible consequences.

[0195] At the implementation level, the system receives updated hydrological forecasts and extracts key indicators, such as the future water level process line, peak flow, and arrival time at specific cross-sections. These indicators are compared with multi-level early warning thresholds pre-set based on historical disaster data, topography, and protection standards, such as warning water levels and guaranteed water levels. Through comparison, the system automatically determines the warning level for each area of ​​concern. Subsequently, the system integrates information such as the spatiotemporal range of the affected area, the distribution of different risk levels, the expected peak time window, and the confidence level of the forecast itself. Based on these elements, the system automatically generates structured early warning text or graphic products according to standardized templates. The content includes a clear warning level, specific affected area, key expected time nodes, descriptions of main risk characteristics, and suggestions for response measures based on the current forecast. This step completes the transformation from complex numerical forecasts to intuitive risk information, enabling forecast results to directly serve emergency command and public notification for disaster prevention and mitigation. It is a key link connecting forecasting technology with practical applications. Through an automated and standardized early warning information generation process, the timeliness, accuracy, and consistency of risk information transmission are ensured.

[0196] In some embodiments, the method further includes:

[0197] Based on the system's real-time resource utilization rate, task calculation timeout status, real-time precipitation intensity, and key water level rise rate, the calculation priority of macro-hydrological simulation, micro-hydrological simulation, state variable exchange, and data assimilation is dynamically determined.

[0198] Based on the computation priority, computing resources are allocated to each task;

[0199] When the system's real-time resource utilization rate is continuously higher than the first resource threshold, or the actual processing time of a task exceeds the preset maximum allowable processing time, adaptive degradation control is triggered and executed. The adaptive degradation control implements one or more measures in stages according to the severity of resource shortage or timeout, including reducing the spatial resolution of micro-models in non-key areas, reducing the size of data assimilation sets, suspending the calculation of micro-hydrological models in non-key areas, and reducing the forecast update frequency.

[0200] The core function of this step is to ensure that the complex, multi-module hydrological forecasting system can operate stably and efficiently under limited and potentially fluctuating computing resources. Real-time system resource utilization reflects the current proportion of hardware resources such as computing cores and memory being used, serving as a fundamental indicator for judging system load. Task computation timeout status indicates whether specific computational tasks such as macro / micro simulation, variable exchange, and data assimilation have been completed within the specified time limit. Real-time precipitation intensity and key water level rise rates are direct inputs from external hydrological conditions, used to assess the urgency and importance of forecasting tasks. Dynamically determining computational priorities is a real-time decision-making process that ranks the importance and urgency of parallel computational tasks in the system based on the aforementioned indicators. Resource allocation involves scheduling available physical resources such as computing cores and memory blocks to the corresponding tasks according to priority. Adaptive degradation control is an intelligent strategy that proactively reduces the computational accuracy or execution frequency of some non-core functions when the system faces resource bottlenecks or performance degradation risks, in order to ensure the stability of core functions and the overall system.

[0201] During implementation, an independent timing control unit continuously monitors the system's operational status. It first collects resource utilization data from the hardware monitoring module, completion time reports from each task scheduler, and the latest hydrological situation data from the data input module. Based on a pre-defined weighted scoring rule, this unit fuses and analyzes this multi-source information, calculating the priority score for each task (macroscopic simulation, microscopic simulation, variable exchange, and data assimilation) in real time. For example, when real-time precipitation intensity increases sharply or key water levels rise rapidly, the priority scores of the directly related microscopic model simulation and data assimilation tasks will increase accordingly. Subsequently, the resource scheduler, based on this priority score queue, prioritizes the allocation of newly released computing cores, memory, and other resources to the highest-scoring tasks, ensuring that the most critical computations can obtain resources and start promptly.

[0202] To address the risk of system overload caused by sudden high loads or complex calculations, this unit also features two levels of safety triggers. The first level trigger is based on average resource utilization; when this value consistently exceeds a preset safety threshold, it indicates the system may be under prolonged high load. The second level trigger is based on task execution time; when the actual time taken for any task exceeds its preset maximum allowable time, it indicates an abnormal delay in a single calculation. Once either trigger is activated, the system determines the current state as "resource stress" or "performance timeout" and automatically enters an adaptive degradation control mode. Degradation control does not employ a single measure but rather selects and executes measures in stages from a predefined library of degradation measures based on the triggering factors and their severity. These measures aim to reduce computational overhead, and their selection follows the principle of "periphery first, then core; accuracy first, then functionality." For example, in cases of mild resource stress, the spatial resolution of micro-models in non-critical areas, which have a smaller impact on overall forecasting but higher computational costs, may be reduced first. If the stress intensifies or task timeouts occur, the number of data assimilation ensemble members may be further reduced to shorten assimilation computation time. In more severe cases, the system may temporarily suspend the entire micro-simulation calculation for non-critical areas, retaining only the detailed forecasts for the core areas, while simultaneously reducing the overall system's result update frequency to allow for longer buffer calculation time. Through these progressive and targeted degradation operations, the system can proactively sacrifice some secondary performance when computing resources are insufficient to support full-function, full-accuracy operation, prioritizing ensuring that core early warning functions are not interrupted and forecasts for key areas are not missing. This maintains the service availability and basic reliability of forecasting operations even in extreme situations, achieving an effective balance between the robustness of the forecasting system and the elasticity of computing resource requirements.

[0203] Example 2

[0204] Please see Figure 2 This invention provides a flood prediction device based on rainfall radar, comprising:

[0205] The calibration module 201 is used to perform quality assessment and dynamic calibration based on the rain measurement radar data to obtain the calibrated radar precipitation field.

[0206] QPF sequence module 202 is used to generate a short-term quantitative precipitation forecast QPF sequence for the next N hours based on the calibrated radar precipitation field and the radar echo motion field.

[0207] The hydrological simulation module 203 is used to drive preset macro-hydrological models and micro-hydrological models to perform hydrological simulations according to the QPF sequence, thereby obtaining macro-hydrological state variables corresponding to the macro-hydrological model and micro-hydrological state variables corresponding to the micro-hydrological model. The macro-hydrological model covers the entire basin and operates at a first spatial resolution, while the micro-hydrological model covers key areas within the basin and operates at a second spatial resolution higher than the first spatial resolution. The macro-hydrological state variables include upstream boundary water level, lateral inflow discharge, soil moisture, and groundwater level, while the micro-hydrological state variables include grid water level, discharge, and surface water depth.

[0208] The coupling module 204 is used to perform real-time exchange and coupling of state variables based on the macro-hydrological state variables and the micro-hydrological state variables to obtain the coupled multi-scale model state.

[0209] The data assimilation module 205 is used to assimilate the data based on the state of the multi-scale model and the real-time acquired ground observation data by using a nonlinear least squares method based on gradient optimization to obtain updated hydrological forecast results.

[0210] The early warning module 206 is used to generate flood early warning information based on the updated hydrological forecast results.

[0211] It should be noted that each module and unit in the flood prediction device based on rainfall radar in this embodiment corresponds one-to-one with each step in the flood prediction method based on rainfall radar in the aforementioned embodiment. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned flood prediction method based on rainfall radar, and will not be repeated here.

[0212] Example 3

[0213] Please see Figure 3 This embodiment provides an electronic device, including at least one processor 301 and a memory 302. Optionally, the device further includes a communication component 303. The processor 301, memory 302, and communication component 303 are connected via a bus 304.

[0214] In a specific implementation, at least one processor 301 executes computer execution instructions stored in memory 302, causing at least one processor 301 to perform the above-described method.

[0215] The specific implementation process of processor 301 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0216] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0217] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0218] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0219] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0220] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0221] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0222] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0223] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0224] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0225] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0226] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0227] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0228] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A flood forecasting method based on rainfall radar, characterized in that, include: The quality assessment and dynamic calibration are performed based on the rainfall radar data to obtain the calibrated radar precipitation field. Based on the calibrated radar precipitation field and combined with the radar echo motion field, a short-term quantitative precipitation forecast (QPF) sequence for the next N hours is generated. Hydrological simulations are performed using a preset macro-hydrological model and a micro-hydrological model driven by the QPF sequence, resulting in macro-hydrological state variables corresponding to the macro-hydrological model and micro-hydrological state variables corresponding to the micro-hydrological model. The macro-hydrological model covers the entire watershed and operates at a first spatial resolution, while the micro-hydrological model covers key areas within the watershed and operates at a second spatial resolution higher than the first spatial resolution. The macro-hydrological state variables include upstream boundary water level, lateral inflow, soil moisture, and groundwater level, while the micro-hydrological state variables include grid water level, flow rate, and surface water depth. Based on the macro-hydrological state variables and the micro-hydrological state variables, the state variables are exchanged and coupled in real time to obtain the coupled multi-scale model state. Based on the state of the multi-scale model and the real-time ground observation data, the data is assimilated using a gradient-optimized nonlinear least squares method to obtain the updated hydrological forecast results. Flood warning information is generated based on the updated hydrological forecast results; The step of performing real-time exchange and coupling of state variables based on the macro-hydrological state variables and the micro-hydrological state variables to obtain the coupled multi-scale model state includes: Based on preset exchange triggering conditions, the timing and content of state variable exchange are determined. The exchange triggering conditions include forced variable exchange and state variable exchange. The forced variable exchange must be performed in each coupled calculation cycle, and the exchanged variables include at least the upstream boundary water level and the lateral inflow flow. The state variable exchange is performed conditionally and is triggered when the difference in soil moisture between the micro-region and the corresponding macro-region exceeds a first preset threshold or the rate of change of groundwater level exceeds a second preset threshold. When the exchange triggering condition is met, the macroscopic hydrological state variables are mapped to the grid of the microscopic hydrological model through a spatial interpolation algorithm to obtain the multi-scale model state. When the exchange triggering condition is met, the macroscopic hydrological state variables are mapped to the grid of the microscopic hydrological model using a spatial interpolation algorithm to obtain the multi-scale model state, including: Identify the micro-hydrological model region where state variable exchange is required, and define all computational grid points of the micro-hydrological model within the micro-hydrological model region as target grid points; When the exchange triggering condition is met, the macroscopic hydrological state variables to be exchanged are obtained, and the three-dimensional spatial coordinates of the target grid point are obtained. A three-dimensional bilinear interpolation algorithm is used to calculate the interpolation of the macro-hydrological state variables at each target grid point, based on the values ​​of the macro-hydrological state variables at the grid points of the macro-hydrological model. The calculated interpolation result is assigned to the corresponding target grid point as the boundary condition of the target grid point; If the data from the macro-hydrological model and the data from the micro-hydrological model are not synchronized in time, a time interpolation method is used to match the time series of the macro-hydrological state variables to the time step calculated by the micro-hydrological model. The variable set after spatial and temporal matching is integrated with the unexchanged state variables within the micro-hydrological model, and the output is a consistent multi-scale model state.

2. The method according to claim 1, characterized in that, The process of quality assessment and dynamic calibration based on rainfall radar data to obtain the calibrated radar precipitation field includes: Based on the rain measurement radar data, a multi-level quality assessment is performed to generate a comprehensive quality score for each grid point in the rain measurement radar data. Based on the comprehensive quality score of each grid point, all grid points are divided into high-quality areas, medium-quality areas, and low-quality areas. Based on the regional division results, dynamic calibration is performed using calibration strategies corresponding to each quality region to obtain the radar precipitation field.

3. The method according to claim 2, characterized in that, The step involves performing a multi-level quality assessment based on the rainfall radar data, generating a comprehensive quality score for each grid point in the rainfall radar data, including: Based on the terrain obstruction information, non-meteorological clutter characteristics, and original signal quality in the rainfall radar data, the terrain obstruction score, clutter identification score, and signal quality score for each grid point are calculated respectively. Based on the radar reflectivity data of each grid point, the consistency between the grid point and its neighboring grid points in the horizontal and vertical directions is analyzed, and the spatial consistency score and vertical consistency score are calculated respectively. Based on radar reflectivity data from consecutive time intervals, the characteristics of reflectivity variation at each grid point over time are analyzed, and a time consistency score is calculated. Based on the physical relationship between radar reflectivity and precipitation rate at each grid point, and using available ground rain gauges or satellite precipitation data for cross-validation, physical consistency scores and cross-validation scores are calculated respectively. Based on the meteorological-hydrological joint scoring standard, a comprehensive quality score is generated by integrating various basic scores.

4. The method according to claim 2, characterized in that, The process involves dynamically calibrating the radar precipitation field based on the regional division results, using calibration strategies corresponding to each quality region, to obtain the radar precipitation field, including: For the high-quality areas, an optimized averaging method based on the statistical relationship of local climate characteristics is used for calibration; For the medium-quality region, a weighted calibration method combining spatial correlation and distance weight is used for calibration, wherein the calibration weight is related to the comprehensive quality score of the grid point and the distance between the grid point and the high-quality region; For the low-quality area, a fusion calibration method that integrates multi-source auxiliary data is used for calibration. The multi-source auxiliary data includes ground rain gauge data and satellite precipitation estimation data.

5. The method according to claim 1, characterized in that, The updated hydrological forecast results are obtained by assimilating data based on the multi-scale model state and real-time acquired ground observation data using a gradient-optimized nonlinear least squares method, including: Using the coupled multi-scale model state as the background field, a cost function is constructed to assimilate multiple observation times within a time window. The cost function includes a deviation constraint term between the background field and the state variable increment, and a residual constraint term between the observed value and the model simulation value at each observation time. The gradient information of the cost function with respect to the increment of the state variables is calculated using adjoint model techniques or perturbation methods. A nonlinear optimization algorithm is used to iteratively optimize the cost function based on the gradient information until the convergence condition is met, thereby obtaining the optimal state analysis increment that minimizes the cost function. The optimal state analysis increment is added to the background field to obtain the updated hydrological state analysis field, which is used as the updated hydrological forecast result.

6. The method according to claim 1, characterized in that, The method further includes: Based on the system's real-time resource utilization rate, task calculation timeout status, real-time precipitation intensity, and key water level rise rate, the calculation priority of macro-hydrological simulation, micro-hydrological simulation, state variable exchange, and data assimilation is dynamically determined. Based on the computation priority, computing resources are allocated to each task; When the system's real-time resource utilization rate is continuously higher than the first resource threshold, or the actual processing time of a task exceeds the preset maximum allowable processing time, adaptive degradation control is triggered and executed. The adaptive degradation control implements one or more measures in stages according to the severity of resource shortage or timeout, including reducing the spatial resolution of micro-models in non-key areas, reducing the size of data assimilation sets, suspending the calculation of micro-hydrological models in non-key areas, and reducing the forecast update frequency.

7. A flood forecasting device based on rainfall radar, characterized in that, include: The calibration module is used to perform quality assessment and dynamic calibration based on the rain measurement radar data to obtain the calibrated radar precipitation field. The QPF sequence module is used to generate a short-term quantitative precipitation forecast QPF sequence for the next N hours based on the calibrated radar precipitation field and the radar echo motion field. The hydrological simulation module is used to drive preset macro-hydrological models and micro-hydrological models to perform hydrological simulations according to the QPF sequence, thereby obtaining macro-hydrological state variables corresponding to the macro-hydrological model and micro-hydrological state variables corresponding to the micro-hydrological model. The macro-hydrological model covers the entire watershed and operates at a first spatial resolution, while the micro-hydrological model covers key areas within the watershed and operates at a second spatial resolution higher than the first spatial resolution. The macro-hydrological state variables include upstream boundary water level, lateral inflow discharge, soil moisture, and groundwater level, while the micro-hydrological state variables include grid water level, discharge, and surface water depth. The coupling module is used to perform real-time exchange and coupling of state variables based on the macro-hydrological state variables and the micro-hydrological state variables to obtain the coupled multi-scale model state. The data assimilation module is used to assimilate the data based on the state of the multi-scale model and the real-time acquired ground observation data using a nonlinear least squares method based on gradient optimization, so as to obtain updated hydrological forecast results. The early warning module is used to generate flood early warning information based on the updated hydrological forecast results; The coupling module is also used for: Based on preset exchange triggering conditions, the timing and content of state variable exchange are determined. The exchange triggering conditions include forced variable exchange and state variable exchange. The forced variable exchange must be performed in each coupled calculation cycle, and the exchanged variables include at least the upstream boundary water level and the lateral inflow flow. The state variable exchange is performed conditionally and is triggered when the difference in soil moisture between the micro-region and the corresponding macro-region exceeds a first preset threshold or the rate of change of groundwater level exceeds a second preset threshold. When the exchange triggering condition is met, the macroscopic hydrological state variables are mapped to the grid of the microscopic hydrological model through a spatial interpolation algorithm to obtain the multi-scale model state. When the exchange triggering condition is met, the macroscopic hydrological state variables are mapped to the grid of the microscopic hydrological model using a spatial interpolation algorithm to obtain the multi-scale model state, including: Identify the micro-hydrological model region where state variable exchange is required, and define all computational grid points of the micro-hydrological model within the micro-hydrological model region as target grid points; When the exchange triggering condition is met, the macroscopic hydrological state variables to be exchanged are obtained, and the three-dimensional spatial coordinates of the target grid point are obtained. A three-dimensional bilinear interpolation algorithm is used to calculate the interpolation of the macro-hydrological state variables at each target grid point, based on the values ​​of the macro-hydrological state variables at the grid points of the macro-hydrological model. The calculated interpolation result is assigned to the corresponding target grid point as the boundary condition of the target grid point; If the data from the macro-hydrological model and the data from the micro-hydrological model are not synchronized in time, a time interpolation method is used to match the time series of the macro-hydrological state variables to the time step calculated by the micro-hydrological model. The variable set after spatial and temporal matching is integrated with the unexchanged state variables within the micro-hydrological model, and the output is a consistent multi-scale model state.

8. An electronic device, characterized in that, include: At least one processor, at least one memory, and computer program instructions stored in the memory, which, when executed by the processor, implement the method as described in any one of claims 1-6.

Images