Drought event identification and feature extraction method

Abstract

The application discloses a drought event identification and feature extraction method, which organizes drought data into a three-dimensional data matrix, and adopts a clustering algorithm based on the three-dimensional data matrix, realizes overall identification of three-dimensional space-time continuity of a drought event, overcomes defects that a traditional method cannot simultaneously process space connectivity and time continuity, can simultaneously consider space continuity and time continuity of the drought event, realizes overall identification of the three-dimensional space-time drought event, and improves monitoring accuracy and timeliness of the drought event.

Description

Technical Field

[0001] This application relates to the field of climate drought monitoring and assessment technology, specifically to a method for drought event identification and feature extraction. Background Technology

[0002] In meteorological drought detection and assessment, widely used indicators include Standardized Precipitation Evapotranspiration Index (SPEI), Standardized Precipitation Index (SPI), and Palmer Drought Index (PDSI). These indicators can quantitatively assess the climate moisture anomalies in a region at a specific time scale to identify drought events.

[0003] Existing drought event identification methods mostly adopt single-point threshold judgment. Although this method is simple and easy to implement, it lacks consideration of spatial connectivity, ignores the spatial clustering characteristics of drought, is easily affected by outliers leading to misjudgment, and cannot accurately describe the persistence and evolution of drought processes, resulting in poor accuracy in drought event identification.

[0004] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0005] This application provides a method for drought event identification and feature extraction, aiming to improve the accuracy of drought event identification.

[0006] On the one hand, this application provides a method for drought event identification and feature extraction, including: A three-dimensional data matrix is ​​generated based on drought data of the target region; the three-dimensional data matrix includes multiple pixels, and the pixel data of each pixel includes location, time and corresponding drought parameters; Based on the drought threshold, a three-dimensional spatiotemporal point set is determined from the three-dimensional data matrix according to the drought parameters of each pixel; Candidate events are determined based on clustering parameters and the three-dimensional spatiotemporal point set. Drought events are determined from the candidate events based on an area threshold; Based on the display parameters, the drought event parameters of the drought event are obtained and displayed; Obtain the spatial centroid of the drought event at different time slices; Based on the time series of the spatial centroid, the movement pattern of the drought event is determined, and the movement pattern is related to the movement direction and speed of the spatial centroid, as well as the diffusion rate of the drought event. Based on the movement pattern, the movement type of the drought event is determined.

[0007] In some embodiments, after acquiring and displaying the drought event parameters of the drought event, the method further includes: Based on the time series of the drought event, determine the propagation process of the drought event into agricultural drought or hydrological drought; Based on the propagation process, a propagation model is constructed between different drought event types, and the propagation model includes time lag relationships and intensity decay models; Based on the propagation model, the transformation prediction of the drought event is performed.

[0008] In some embodiments, prior to generating the three-dimensional data matrix based on drought data of the target region, the method further includes: Acquire various environmental data of the target area; The various environmental data are fused and assimilated to obtain drought data for the target region.

[0009] In some embodiments, the multiple environmental data include remote sensing soil moisture data, station precipitation data, and reanalysis temperature data; the fusion and assimilation processing of the multiple environmental data to obtain drought data for the target area includes: Spatiotemporal registration was used to unify data of different resolutions into the same grid coordinate system, and Kriging interpolation was used to fill the spatial gaps in the station data. Time series interpolation was used to align data of different observation frequencies to obtain fused data. A dynamic data fusion model based on ensemble Kalman filtering is established. The fused data is iteratively assimilated with the output of the land surface process model to generate a spatiotemporally continuous standardized precipitation evapotranspiration index, which serves as the drought data for the target region.

[0010] In some embodiments, prior to generating the three-dimensional data matrix based on drought data of the target region, the method further includes: Based on the historical climate characteristics of the target region, a baseline value for the drought threshold is determined; The drought threshold is obtained by adjusting the benchmark value based on the real-time climate anomaly level of the target region.

[0011] In some embodiments, determining candidate events based on the three-dimensional spatiotemporal point set according to clustering parameters includes: Based on the background clustering parameters in the clustering parameters, coarse clustering is performed on the three-dimensional spatiotemporal points in the three-dimensional spatiotemporal point set in the global scope to obtain the drought background field; Based on the core clustering parameters in the clustering parameters, fine clustering is performed within the drought background field to obtain the candidate events.

[0012] In some embodiments, after acquiring and displaying the drought event parameters of the drought event, the method further includes: Obtain new drought data for the target region; Based on the newly added drought data, the identification results of the drought event are incrementally updated.

[0013] In some embodiments, after acquiring and displaying the drought event parameters of the drought event, the method further includes: Obtain the spatial morphological features of the drought event; Construct a correlation model between the spatial morphological features and climate driving factors.

[0014] In some embodiments, the spatial morphological features include compactness, elongation, and boundary fractal dimension.

[0015] The drought event identification and feature extraction method provided in this application organizes drought data into a three-dimensional data matrix and uses a clustering algorithm to achieve the overall identification of the three-dimensional spatiotemporal continuity of drought events. This overcomes the shortcomings of traditional methods that cannot simultaneously handle spatial connectivity and temporal continuity. It can simultaneously consider the spatial continuity and temporal continuity of drought events, realize the overall identification of drought events in three-dimensional spatiotemporal dimensions, and improve the accuracy and timeliness of drought event monitoring. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of a drought event identification and feature extraction method provided in the embodiments of this application; Figure 2 This is a schematic diagram of a drought event identification method provided in the embodiments of this application; Figure 3 This is a schematic diagram of the monthly distribution map of the maximum drought area provided in the embodiments of this application; Figure 4 This is a schematic diagram of a drought event clustering diagram provided in the embodiments of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] In traditional drought event identification methods, the single-point threshold judgment method independently judges the drought state by discrete grid points, which leads to a break in spatial continuity and cannot characterize the overall movement characteristics of drought patches, resulting in damage to the spatiotemporal integrity of drought events.

[0020] If the above problems are not addressed, drought event identification systems will be unable to accurately capture the evolution patterns of drought across time and space scales, resulting in delayed drought early warning responses. In watershed water resource allocation scenarios, discrete drought identification results will mislead reservoir water release decisions, leading to improper allocation of drought relief resources. In the climate model evaluation process, fragmented drought event records will distort the correlation analysis between drought and atmospheric circulation factors, reducing the training accuracy of climate prediction models.

[0021] like Figure 1 As shown, to address the above problems, the drought event identification and feature extraction method proposed in this application includes the following steps: S101: Generate a three-dimensional data matrix based on drought data of the target region.

[0022] In some embodiments, the three-dimensional data matrix includes multiple pixels, and the pixel data of each pixel includes location, time, and corresponding drought parameters.

[0023] S102: Based on the drought threshold, determine the three-dimensional spatiotemporal point set from the three-dimensional data matrix according to the drought parameters of each pixel.

[0024] S103: Based on clustering parameters, determine candidate events according to the three-dimensional spatiotemporal point set.

[0025] S104: Based on the area threshold, determine drought events from the candidate events.

[0026] S105: Based on the display parameters, obtain and display the drought event parameters of the drought event.

[0027] S106: Obtain the spatial centroid of the drought event at different time slices.

[0028] S107: Based on the time series of the spatial centroid, determine the movement pattern of the drought event, wherein the movement pattern is related to the movement direction and speed of the spatial centroid and the diffusion rate of the drought event.

[0029] S108: Determine the movement type of the drought event based on the movement pattern.

[0030] In this application, the three-dimensional data matrix in step S101 refers to a data structure that integrates longitude, latitude and time dimensions. Specifically, it can be implemented by stacking raster data to arrange drought parameters of multiple time phases in a time series. This structure can completely preserve the original information of drought events in terms of spatial distribution and temporal evolution.

[0031] In this application, the three-dimensional spatiotemporal point set in step S102 refers to the set of spatial locations and their corresponding timestamps that meet the preset drought threshold. Specifically, it can be achieved by extracting effective drought grid points (e.g., drought grid points with drought parameters greater than the drought threshold) from the three-dimensional data matrix through a threshold filtering method. This point set provides spatiotemporally correlated input data for subsequent cluster analysis.

[0032] In this application, step S103 involves clustering parameters and clustering algorithms. The clustering algorithm refers to an unsupervised machine learning method based on density clustering. Specifically, it can be implemented by defining the neighborhood radius parameter and the minimum number of neighborhood points parameter of the joint spatiotemporal distance metric. This algorithm can automatically identify high-density regions with spatiotemporal continuity as candidate events.

[0033] In this application, the area threshold in step S104 is used to screen drought events. For each candidate event, features such as maximum spatial area, average area, and duration are calculated. Candidate events with too small an area (maximum spatial area or average area, etc.) are filtered out according to the set minimum drought area standard (i.e., area threshold), and the remaining events are identified as drought events.

[0034] In this application, the drought event parameters in step S105 can be multidimensional feature parameters, which are quantitative indicators describing the spatiotemporal attributes of drought events. Specifically, they can be achieved by calculating the time series statistics of spatial coverage area, the time span measurement of duration, and the spatial weighted average of intensity indicators. This parameter system provides a multidimensional analytical basis for the quantitative assessment of drought events.

[0035] In this application, steps S106 to S108 provide a specific scheme for tracking the evolution path of drought events. For each identified drought event, the scheme extracts its spatial centroid at different time slices (i.e., time values); calculates the direction of movement, speed of movement, and diffusion rate of the drought event based on the time series of the spatial centroid; and classifies the movement type of the drought event based on the movement pattern.

[0036] The extraction of the spatial centroid is achieved by calculating the latitude and longitude weighted average of all grid points within the drought event coverage area on each time slice, with the weights being the absolute values ​​of drought intensity at the corresponding grid points. The direction of movement is determined by the vector difference between the centroid coordinates of adjacent time slices, and the movement speed is quantified by the ratio of the centroid's movement distance to the time interval. The diffusion rate is calculated based on the temporal rate of change of the drought event coverage area, using the least squares method to fit the slope of the area change curve. Drought events are classified into three basic types based on the consistency of the direction of movement and the stability of the diffusion rate: stationary, unidirectional propagation, and multidirectional diffusion.

[0037] Specifically, after identifying spatiotemporally continuous drought events, the spatial centroid coordinates of each event are calculated monthly based on the time slice data. Taking two consecutive months as an example, the direction angle of the drought event's movement is determined by comparing the longitude and latitude differences of the centroid coordinates. The calculation of the movement speed converts spatial distance into actual geographical distance, using the spherical trigonometry formula to calculate the arc length between two points. The quantification of the diffusion rate is achieved by statistically analyzing the relative change rate of drought coverage area in adjacent time slices, combined with the time interval to calculate the diffusion rate per unit time. Based on the time series characteristics of the above parameters, when the standard deviation of the movement direction is less than 15 degrees and the coefficient of variation of the diffusion rate is less than 0.3, it is classified as a unidirectional spread drought event; when the standard deviation of the direction is greater than 45 degrees and the diffusion rate exhibits a multi-peak distribution, it is classified as a multidirectional spread drought event. This classification result can provide decision support for drought early warning. For example, for unidirectional spread droughts, drought relief measures can be deployed in advance along the movement direction, while for multidirectional spread droughts, regional joint prevention and control strategies are required.

[0038] The core innovation of this application lies in constructing a density clustering model with a joint spatiotemporal distance metric. This overcomes the limitations of traditional methods in separating spatial continuity and temporal persistence, enabling the holistic identification and feature extraction of drought events as complete spatiotemporal entities. Furthermore, by extracting the spatial centroid and calculating its changes, the movement trajectory of drought events can be accurately tracked. The calculation of the direction and speed of movement provides a basis for analyzing the propagation trend of drought events, while the introduction of the diffusion rate reflects the changes in the impact range of drought events. Classifying drought events based on these indicators helps identify different types of drought events, providing support for the development of targeted drought prevention and mitigation measures.

[0039] The working process for this application is as follows: First, drought data for the target region is organized into a three-dimensional data matrix containing longitude, latitude, and time. This step integrates discrete drought index data into a unified three-dimensional space, providing a structured data foundation for subsequent analysis.

[0040] Next, drought grid points are extracted from the three-dimensional data matrix based on a preset drought threshold, forming a three-dimensional spatiotemporal point set. Data points that meet the drought conditions are selected by setting a threshold, transforming the raw data into a point set to be analyzed.

[0041] Then, clustering algorithms such as DBSCAN (a density-based clustering method capable of recognizing clusters of arbitrary shapes) are used to cluster the three-dimensional spatiotemporal point set. In this application, the distance metric of the DBSCAN algorithm considers both spatial and temporal distances. Specifically, a comprehensive distance function is defined, which weights and combines the Euclidean distance in geographic space with the distance on the time axis. This approach enables the clustering process to simultaneously capture the spatial and temporal continuity of drought events.

[0042] The DBSCAN algorithm performs clustering by setting two key parameters: neighborhood radius and minimum number of samples. The neighborhood radius determines the maximum distance between two adjacent points, while the minimum number of samples defines the minimum number of points required to form a cluster core. The algorithm first randomly selects an unvisited point. If the number of points in its neighborhood reaches the minimum number of samples, a new cluster is formed; otherwise, the point is marked as noise. For the newly formed cluster, its boundaries are continuously expanded until no more new points can be absorbed. This process is repeated until all points have been visited, and candidate events are output.

[0043] In this application, each cluster of the DBSCAN algorithm represents a spatiotemporally continuous candidate event, and drought events are selected from these candidate events based on an area threshold. Because the algorithm considers both spatial and temporal dimensions, it is able to identify drought events that are geographically continuous and temporally persistent.

[0044] Finally, based on the identified drought events, their multidimensional feature parameters are calculated and displayed according to the required display parameters. These parameters include at least spatial coverage, duration, and intensity index. Spatial coverage reflects the geographical extent of the drought event and can be calculated by statistically analyzing the number of grid points in the cluster and combining this with the grid resolution. Duration represents the time span of the drought event and is determined by the difference between the earliest and latest time points in the cluster. The intensity index reflects the severity of the drought and can be obtained by calculating the average or cumulative drought index of all points in the cluster.

[0045] This method overcomes the limitations of traditional methods in processing spatiotemporal data by treating drought events as a continuum in three-dimensional spacetime, and achieves holistic identification and feature extraction of drought events.

[0046] As a preferred embodiment, the solution of this application is specifically implemented as follows: The C1 plain region of country C was selected as the target area, and the monthly standardized precipitation-evaporation index (SPEI) dataset from 2000 to 2020 was used. The SPEI data has a spatial resolution of 0.25° × 0.25° and a temporal resolution of month.

[0047] First, the SPEI data is organized into a three-dimensional data matrix. The x-axis and y-axis of the matrix correspond to longitude and latitude, respectively, and the z-axis corresponds to time. The matrix size is 160×120×252, where 160 and 120 are the number of grid points in the longitude and latitude directions, respectively, and 252 is the total number of months.

[0048] Next, a drought threshold of SPEI < -1.0 is set. The three-dimensional data matrix is ​​traversed, and all grid points that meet the condition are extracted to form a three-dimensional spatiotemporal point set. Each point is represented by (x, y, t), where x and y are the spatial coordinates of the grid point, and t is the time coordinate.

[0049] Then, the DBSCAN algorithm is applied to cluster the 3D spatiotemporal point set. The neighborhood radius ε is set to 1.5, and the minimum number of samples MinPts is set to 6. Weighted Euclidean distance is used as the distance metric, with a weight ratio of 2:1 between spatial and temporal distance. Specifically, the distance between two points (x1, y1, t1) and (x2, y2, t2) is calculated using the following formula: distance = sqrt(2×((x1-x2) 2 + (y1-y2) 2 ) + (t1-t2) 2 ); After the DBSCAN algorithm is executed, a series of clusters are obtained, each cluster representing a candidate event, and drought events are selected through screening.

[0050] Finally, characteristic parameters are calculated for each identified drought event: 1. Spatial coverage area: The number of distinct (x,y) coordinate pairs in a statistical cluster, multiplied by the area of ​​a single grid point (approximately 625 square kilometers).

[0051] 2. Duration: Calculates the difference between the maximum and minimum time coordinates in the cluster, in months.

[0052] 3. Strength index: Calculate the average of all SPEI values ​​in the cluster.

[0053] Through the above-described scheme, this application achieves accurate identification and comprehensive description of drought events. This method can automatically capture the spatial continuity and temporal persistence of drought events. By incorporating the temporal dimension into the clustering process, the method can identify persistent drought events spanning multiple months, improving the accuracy of drought monitoring. Simultaneously, the calculated multidimensional feature parameters provide a foundation for the quantitative assessment of drought events, contributing to a better understanding of the spatiotemporal evolution of drought. This three-dimensional density clustering-based method enhances the automation of drought event identification, reduces the influence of subjective factors, and provides an effective tool for large-scale, long-term drought monitoring.

[0054] In some of the solutions described above in this application, drought event identification methods can effectively capture spatiotemporal continuity features, but they do not involve the propagation relationship between different types of drought, which makes it impossible to reveal the dynamic evolution law of meteorological drought to agricultural or hydrological drought, and makes it difficult to predict the impact range and development trend of cross-type drought.

[0055] In some embodiments, Figure 1 The method shown, after obtaining and displaying the drought event parameters of the drought event, further includes: Based on the time series of the drought event, determine the propagation process of the drought event into agricultural drought or hydrological drought; Based on the propagation process, a propagation model is constructed between different drought event types, and the propagation model includes time lag relationships and intensity decay models; Based on the propagation model, the transformation prediction of the drought event is performed.

[0056] This embodiment proposes a drought propagation pattern identification and prediction function. The analysis employs multivariate time series alignment technology to synchronize drought parameters with agricultural or hydrological drought parameters such as soil moisture and runoff. The time lag relationship is determined through cross-correlation analysis, calculating the maximum correlation coefficient and corresponding time offset between different drought type indicator sequences. The intensity decay model is constructed using an exponential decay function, and nonlinear regression is used to fit the relationship between meteorological drought intensity and subsequent agricultural drought peak intensity. Finally, based on the integrated time lag parameters and decay coefficients, a propagation model is obtained. Combined with real-time monitoring data, a probability distribution map of drought impacts for the next 30 days is generated, enabling the prediction of drought event transformation.

[0057] Specifically, after selecting the target region, monthly standardized precipitation evapotranspiration index, root zone soil moisture content, and river runoff data for five consecutive years were obtained. For the time series of meteorological drought events, the sliding window method was used to extract their peak intensity time points, which were then matched with the peak times of agricultural drought and hydrological drought events, respectively, calculating average time lags of 28 days and 45 days. In the intensity decay model, for every 1 standard deviation increase in meteorological drought intensity, the peak intensity decay rate for agricultural drought was 0.78, and for hydrological drought, it was 0.65. In the prediction phase, when a meteorological drought event with an intensity of -2.3 was detected, the model output a 72% probability of moderate agricultural drought occurring 28 days later and a 63% probability of mild hydrological drought occurring 45 days later. The prediction results were validated using historical data back-substitution, achieving an accuracy rate of 81.3%.

[0058] As a preferred embodiment, the solution of this application is specifically implemented as follows: Based on the time series of identified drought events, the propagation process from drought events to agricultural drought or hydrological drought is analyzed. First, multi-year time series data of meteorological drought parameters (such as the Standardized Precipitation Index (SPI), the drought parameters mentioned above), agricultural drought indices (such as the Crop Water Stress Index (CWSI), and hydrological drought indices (such as the Standardized Runoff Index (SRI)) are collected for the target region. Then, cross-correlation analysis is performed on these time series to determine the time lag relationships between different types of drought. For example, the cross-correlation function between SPI and CWSI is calculated, and the time lag value with the highest correlation is identified as the propagation time from meteorological drought to agricultural drought.

[0059] Based on the propagation process, a propagation model is established between different drought types, including a time lag relationship and an intensity decay model. Using the aforementioned analysis results, a drought propagation model is constructed. For example, an exponential decay function can be used to describe the decay process of drought intensity over time: I(t) = I0 × exp(-kt), where I(t) is the drought intensity at time t, I0 is the initial drought intensity, and k is the decay coefficient. The model is determined by fitting historical data.

[0060] Finally, drought impact prediction is performed based on the propagation model, specifically, the transformation prediction of the drought event. Using the established time lag relationship and intensity decay model, combined with the currently observed meteorological drought conditions, the potential agricultural and hydrological droughts in the future are predicted. Finally, the prediction results are visualized, such as by generating a drought propagation risk map, to support decision-makers in taking preventative measures.

[0061] Through the above technical solutions, this application achieves quantitative analysis and prediction of drought propagation processes, improving the accuracy and timeliness of drought impact assessment and providing a scientific basis for water resource management and agricultural production decision-making. Furthermore, this method can capture the dynamic correlations between different types of drought, contributing to an understanding of the intrinsic mechanisms of drought evolution and thus optimizing the performance of drought early warning systems.

[0062] In some of the schemes described above in this application, there may be problems such as a single data source, insufficient spatiotemporal consistency, and lack of data quality assessment when constructing a three-dimensional data matrix. This makes it difficult to guarantee the accuracy and reliability of the drought index field, affecting the accuracy of subsequent cluster analysis.

[0063] In some embodiments, Figure 1 The method shown includes, before generating the three-dimensional data matrix based on drought data of the target region, the following: Acquire various environmental data of the target area; The various environmental data are fused and assimilated to obtain drought data for the target region within the target time range.

[0064] Specifically, the various environmental data include remote sensing soil moisture data, station precipitation data, and reanalysis temperature data; the process of fusing and assimilating the various environmental data to obtain drought data for the target area includes: Spatiotemporal registration was used to unify data of different resolutions into the same grid coordinate system, and Kriging interpolation was used to fill the spatial gaps in the station data. Time series interpolation was used to align data of different observation frequencies to obtain fused data. A dynamic data fusion model based on ensemble Kalman filtering is established. The fused data is iteratively assimilated with the output of the land surface process model to generate a spatiotemporally continuous standardized precipitation evapotranspiration index, which serves as the drought data for the target region.

[0065] This embodiment proposes a specific method for constructing a three-dimensional data matrix, which may include: acquiring various environmental data such as remote sensing soil moisture data, station precipitation data, and reanalysis temperature data; then fusing the remote sensing soil moisture data, station precipitation data, and reanalysis temperature data; and finally generating spatially continuous and temporally consistent drought parameters through data assimilation technology.

[0066] When fusing remote sensing soil moisture data, station precipitation data, and reanalysis temperature data, a spatiotemporal registration method is used to unify data of different resolutions into the same grid coordinate system. Kriging interpolation can be used to fill spatial gaps in the station data, and time series interpolation can be used to align data sources with different observation frequencies. When generating the drought index field using data assimilation techniques, a dynamic data fusion model based on ensemble Kalman filtering is established. This model iteratively assimilates multi-source observation data with the output of the land surface process model to generate a spatiotemporally continuous standardized precipitation-evapotranspiration index. Finally, uncertainty quantification can be performed based on data quality indicators. An error propagation model can then be constructed to assess the impact of measurement errors from each data source on the final drought index. Monte Carlo simulation is used to generate a probability distribution map to express the confidence interval of the fusion result.

[0067] Specifically, in the process of multi-source data fusion, remote sensing soil moisture data provides information on surface moisture over a wide range, station precipitation data has high precision but is spatially discontinuous, and reanalysis temperature data has complete spatiotemporal coverage but low resolution. By spatiotemporal registration, the three are unified to a 0.01° grid coordinate system, and spatial interpolation techniques are used to eliminate data gaps. For example, inverse distance weighted interpolation is used for station precipitation data, and bilinear interpolation downscaling is performed on temperature data.

[0068] Specifically, the data assimilation stage adopts a dynamic weight allocation mechanism, which increases the weight of station data in areas with abundant rainfall and focuses on remote sensing soil moisture data in arid areas. Through iterative optimization, the assimilation results simultaneously meet the constraints of multi-source data.

[0069] Specifically, in the uncertainty quantification process, ±5% instrument error is set for remote sensing data, ±10% observation error for station data, and ±1℃ model bias for temperature data. The composite error distribution is calculated using the error propagation equation, ultimately generating a three-dimensional drought index field containing confidence intervals. For example, in the P River basin application, the spatial coverage integrity of the fused drought index field reaches 98.7%, the temporal continuity error is less than 3 days, and data quality indicators show that the uncertainty range of the fusion result is controlled within ±0.2 SPEI units.

[0070] As a preferred embodiment, the solution of this application is specifically implemented as follows: When constructing a three-dimensional data matrix using multi-source data fusion, remote sensing soil moisture data, station precipitation data, and reanalysis temperature data were first collected within the study area. The remote sensing soil moisture data used was soil moisture products from Institution A, with a spatial resolution of 0.25° and a daily temporal resolution. Station precipitation data came from multiple large-scale regional meteorological stations in region B, with a daily temporal resolution. The reanalysis temperature data used was the reanalysis dataset from Institution C, with a spatial resolution of 0.25° and an hourly temporal resolution.

[0071] Furthermore, a spatially continuous and temporally consistent drought index field is generated through data assimilation techniques. Specifically, an ensemble Kalman filter method is employed, using remote sensing soil moisture data as the observation field and station precipitation data and reanalysis temperature data as the background field for data fusion. During the fusion process, spatial interpolation is performed on the station precipitation data to ensure it has the same spatial resolution as the remote sensing and reanalysis data. Temporally, all data are unified to a monthly scale.

[0072] Thus, the generated drought index field has a spatial resolution of 0.25° and a monthly temporal resolution, covering the entire study area. Based on the fused data field, the standardized precipitation evapotranspiration index (SPEI) is calculated as the drought index.

[0073] Finally, the uncertainty of the fusion results was quantified based on data quality indicators. The root mean square error (RMSE) and correlation coefficient (R) between the fused data and the original observation data for each grid point were calculated as evaluation indicators of data quality. Regions with larger RMSE or smaller R were assigned lower weights in subsequent analyses.

[0074] More specifically, this embodiment can be implemented in the following ways: First, the three types of environmental data are preprocessed to ensure consistent data quality and are unified onto the same spatiotemporal grid.

[0075] The remote sensing soil moisture data involved in this application can come from satellite sensors (such as SMAP and SMOS), providing global or regional soil moisture information with high spatial resolution (such as 1-10 km), but may have uneven temporal resolution (such as daily or every few days) and data gaps (such as cloud cover).

[0076] The precipitation data from the stations involved in this application can come from ground meteorological stations, providing precipitation measurements at point locations with high temporal resolution (such as daily or hourly), but the spatial distribution is sparse and uneven.

[0077] The reanalysis temperature data involved in this application can come from reanalysis products (such as ERA5, NCEP), providing temperature data with a larger range of uniform grids, lower spatial resolution (such as 0.25 degrees), and consistent temporal resolution (such as daily).

[0078] The grid involved in this application is defined as: a unified grid coordinate system defining the target area, including spatial extent, grid size (e.g., 0.1 degrees × 0.1 degrees), and projection system (e.g., WGS84 latitude and longitude grid). All data will be resampled or interpolated to this grid.

[0079] Spatial registration involved in this application includes: for remote sensing soil moisture data, projecting it onto a uniform grid using resampling methods (such as bilinear interpolation or nearest neighbor interpolation); for reanalysis temperature data, directly projecting it onto a uniform grid, or downscaling it to a higher resolution as needed; and for station precipitation data, converting it into gridded data using Kriging interpolation. For example, this may include: calculating the experimental variogram of the station data, fitting a theoretical variogram model (such as an exponential model or a Gaussian model); and using ordinary Kriging interpolation to estimate the precipitation value for each grid point based on the variogram model, while providing the error variance.

[0080] The time registration involved in this application includes aligning the time axis of all data to a daily time step (or other consistent frequency, such as weekly). For data with inconsistent time resolution, any of the following time series interpolation methods are used: Linear interpolation: used to fill in missing daily data, assuming that the variable changes linearly over time.

[0081] Spline interpolation: used to smooth time series, especially suitable for high-frequency data (such as hourly to daily data).

[0082] For remote sensing data, combine time series filtering (such as Savitzky-Golay filtering) to reduce noise.

[0083] This application obtains fused data, including daily soil moisture, precipitation, and temperature values ​​for each grid point, forming a spatiotemporally consistent data cube.

[0084] Subsequently, a dynamic data fusion model based on ensemble Kalman filtering combines the fused data with a land surface process model, generating more accurate drought-related variables through iterative updates.

[0085] The land surface process model involved in this application can employ mature land surface process models, such as Noah-MP (Noah-Multi Parameterization) or VIC (Variable Infiltration Capacity), which can simulate water, energy, and carbon fluxes. Model inputs include meteorological forcing data (including fused precipitation, temperature, and other necessary variables such as wind speed and humidity, which can be supplemented from reanalysis data). Model outputs include key state variables such as soil moisture, evapotranspiration (ET), and runoff. Optimization training of model parameters involves parameterization based on the topography, soil type, and vegetation cover of the target area (e.g., using GLDAS or local data).

[0086] The ensemble Kalman filter (EnKF) assimilation process involved in this application is a Monte Carlo method that estimates the error covariance by maintaining a set of model states, thereby achieving the fusion of observed data and model predictions.

[0087] The state vector includes assimilated variables such as soil moisture (multi-layered) and temperature. For example, the state vector x = [θ1, θ2, T, ...], where θ is soil moisture and T is temperature. The observation vector comes from the fused data, including remotely sensed soil moisture, gridded precipitation after interpolation of station precipitation, and reanalysis temperature. The observation error covariance matrix R is estimated based on data uncertainties (such as the error variance of remotely sensed data and interpolation error).

[0088] The initialization of the dynamic data fusion model involved in this application includes the following steps: Generate an initial state set: Generate N set members (usually N = 50 - 100) from the initial conditions or historical data of the land surface process model by random perturbation, representing the uncertainty of the state.

[0089] The model error covariance matrix Q is set based on the uncertainty of the model parameters.

[0090] Prediction Step (Model Forecast): For each time step, the land surface process model is run one step forward to obtain the prior state estimate xif (for each set member i). The prior state set represents the uncertainty of the model prediction.

[0091] Update step (data assimilation): When observation data is available (e.g., daily), calculate the Kalman gain K:

[0092] in, H is the prior state error covariance matrix (calculated from the set), and H is the observation operator (mapping the state vector to the observation space).

[0093] Then, update the post-hoc state. : ; Where y is the observation vector, It is a random perturbation of observation error, posterior state The set is used for initialization in the next time step.

[0094] Repeat the forecasting and updating steps to cover the entire time series (e.g., daily data over many years) to generate assimilated state variables, including soil moisture and evapotranspiration.

[0095] The standardized precipitation evapotranspiration index (SPEI) calculation involved in this application includes the calculation of SPEI based on assimilated data. This index combines precipitation and evapotranspiration and can effectively characterize drought conditions.

[0096] The input data for calculating the Standardized Precipitation Evapotranspiration Index (SPEI) includes: assimilated precipitation data (from the site precipitation assimilation results), assimilated temperature data (used to calculate potential evapotranspiration), and assimilated soil moisture data as validation, but SPEI is mainly based on precipitation and evapotranspiration.

[0097] Next, potential evapotranspiration (PET) calculations are performed using the FAO Penman-Monteith equations, requiring assimilated temperature, humidity, wind speed, and solar radiation data (obtained from reanalysis data or model output):

[0098] Where Δ is the slope of the saturated vapor pressure curve, Rn is the net radiation, G is the soil heat flux, and γ is the psychrometric constant. The wind speed at a height of 2 meters It is the saturated vapor pressure. This is the actual water vapor pressure. If data is insufficient, a simplified method such as the Hargreaves equation can be used, requiring only temperature data.

[0099] Next, water deficit calculations are performed. For each grid point and time step, the water deficit is calculated as D = P - PET, where P is precipitation. The cumulative water deficit covers multiple time scales (e.g., 1, 3, 6, 12 months) to capture different drought types (meteorological, agricultural drought).

[0100] Finally, standardization is performed, and the cumulative water surplus / deficit series for each time scale is fitted using a probability distribution function (such as a Log-Logistic distribution). Specifically, the cumulative values ​​are converted into a standard normal distribution variable to obtain the SPEI value:

[0101] Where F(D) is the cumulative distribution function, It is the inverse function of the standard normal distribution.

[0102] Classification of different SPEI values: SPEI greater than 2.0 indicates extreme wetness, SPEI between 1.0 and 2.0 indicates wetness, SPEI between -1.0 and 1.0 indicates normal, SPEI between -1.0 and -2.0 indicates moderate drought, and SPEI less than -2.0 indicates extreme drought.

[0103] Based on the above data and steps, generate spatiotemporal continuous SPEI data for the target area, including grid data at multiple time scales (such as daily or monthly, 1km resolution).

[0104] Through the above technical solution, this application achieves the effective fusion of multi-source heterogeneous data, generating high-quality drought data that is spatially continuous and temporally consistent. The fused data inherits the spatial continuity of remote sensing data, the accuracy of station data, and the temporal continuity of reanalysis data, overcoming the limitations of a single data source. Furthermore, uncertainty quantification provides a reliable data foundation for subsequent drought event identification, improving the accuracy of drought monitoring and assessment.

[0105] In some of the solutions mentioned above in this application, a fixed drought threshold is used to identify drought events. In years with abnormal climate, the threshold setting may not match the actual situation, which may lead to misjudgment or omission of drought events and fail to accurately reflect the drought characteristics under extreme climate conditions.

[0106] Therefore, in some embodiments... Figure 1 The method shown includes, before generating the three-dimensional data matrix based on drought data of the target region, the following: Based on the historical climate characteristics of the target region, a baseline value for the drought threshold is determined; The drought threshold is obtained by adjusting the benchmark value based on the real-time climate anomaly level of the target region.

[0107] The baseline value can be determined by analyzing climate data from the past 30 years in the study area (i.e., the target region), calculating the statistical distribution characteristics of the SPEI index for each month, and using the 20th percentile as the baseline drought threshold. The dynamic adjustment process introduces a climate anomaly index, which is calculated by weighting the standard deviation of monthly precipitation from the historical average and temperature anomalies. When the climate anomaly index exceeds a preset threshold, the drought threshold is lowered proportionally. The application of the adaptive threshold establishes criteria for extreme years. When annual precipitation is below 15% of the historical minimum or the number of consecutive drought months exceeds the historical extreme, a sliding window-based adaptive threshold algorithm is activated, using the dynamic percentile data from the past 12 months as the threshold baseline.

[0108] Specifically, historical climate baseline values ​​are established using a monthly statistical approach. For example, the SPEI baseline value for the North China Plain is -0.8 in January and adjusted to -1.2 in July to reflect seasonal differences. During dynamic adjustments, when the precipitation anomaly index reaches 2σ in a given month, the drought threshold is adjusted from the baseline value of -1.0 to -1.3, expanding the drought identification range. In extreme year identification, a sliding window is used to calculate the climate extremes of the past five years. When real-time data exceeds the extreme value range, it automatically switches to an adaptive threshold mode. This mechanism has been validated through Monte Carlo simulations. Its application in the S River Basin from 2010 to 2020 shows that compared to traditional fixed thresholds, the accuracy of drought event identification is improved by 23.7%, especially in the extreme drought event of 2016, successfully identifying three secondary drought core areas missed by traditional methods.

[0109] As a preferred embodiment, the solution of this application is specifically implemented as follows: Based on the historical climate characteristics of the target region, a baseline value for the drought threshold is determined. For example, for North China in Country X, the long-term distribution characteristics of the Standardized Precipitation Evaporation Index (SPEI) are calculated by analyzing precipitation and temperature data from 1961 to 2020. According to the cumulative probability distribution of SPEI, a baseline threshold is set for an SPEI less than -1.0 as mild drought, less than -1.5 as moderate drought, and less than -2.0 as severe drought.

[0110] The drought threshold is dynamically adjusted based on the real-time severity of climate anomalies. Specifically, the deviation between the current SPEI and the historical average for the same period is calculated monthly. When the deviation exceeds one standard deviation, the drought threshold is adjusted accordingly. For example, if the SPEI for the current month is two standard deviations lower than the historical average for the same period, the threshold for mild drought is adjusted from -1.0 to -0.8, and the threshold for moderate drought is adjusted from -1.5 to -1.3.

[0111] Furthermore, for extreme climate years, an adaptive threshold is used instead of a fixed threshold. This application defines an extreme climate year as a year in which the annual SPEI deviates from the historical mean by more than 2 standard deviations. In these years, a relative threshold based on percentiles is used; for example, an SPEI below the 10th percentile of the year is defined as mild drought, below the 5th percentile as moderate drought, and below the 1st percentile as severe drought.

[0112] Therefore, the proposed solution enables dynamic adjustment of the drought threshold, improving the accuracy and adaptability of drought event identification.

[0113] Through the above technical solutions, this application can flexibly adjust the drought judgment criteria according to the climate characteristics of different regions and periods, avoiding misjudgments that may occur under extreme climate conditions due to fixed thresholds. The dynamic threshold mechanism improves the accuracy of drought event identification, making the identification results more consistent with actual climate conditions; the introduction of adaptive thresholds further enhances the method's adaptability to extreme climate years, ensuring accurate capture of drought events even under abnormal climate conditions. This flexible threshold adjustment mechanism enhances the robustness and practicality of the drought monitoring system.

[0114] When dealing with large-scale, complex geographical regions, the clustering methods described above may result in incomplete capture of the spatial structural features of drought events, affecting the accurate analysis of drought events.

[0115] Therefore, in some embodiments... Figure 1 The method shown, which determines candidate events based on clustering parameters and the three-dimensional spatiotemporal point set, includes: Based on the background clustering parameters in the clustering parameters, coarse clustering is performed on the three-dimensional spatiotemporal points in the global scope to obtain the drought background field; Based on the core clustering parameters in the clustering parameters, fine clustering is performed within the drought background field to obtain the candidate events.

[0116] This application proposes a hierarchical clustering approach. The first level performs coarse clustering globally to identify the large-scale drought background field, while the second level performs fine clustering within the coarse clustering results to identify local drought core areas. At the same time, the membership relationship between the clustering results at different levels is established.

[0117] In the coarse clustering stage, a larger neighborhood radius parameter (i.e., the background clustering parameter) is used. By setting the spatial distance weighting coefficient to 0.7 and the temporal distance weighting coefficient to 0.3, clustering is performed within a joint neighborhood spanning 500 kilometers and 3 months to capture drought systems spreading across regions, serving as the drought background field. In the fine clustering stage, a smaller neighborhood radius parameter (i.e., the core clustering parameter) is used within the spatial boundary formed by the coarse clustering (i.e., the drought background field). The neighborhood radius is reduced to 30% of the original value, and the spatial distance weighting is increased to 0.9. The focus is on analyzing local drought clustering phenomena within a 50-kilometer range, i.e., the identification of candidate events.

[0118] For example, when processing the eastern monsoon region of country D, coarse clustering first identified a persistent drought background field covering the middle and lower reaches of the E River, with a spatial range of 250,000 square kilometers and a duration spanning four months. Against this backdrop, fine clustering further detected two localized drought core areas in the F Lake and G Lake basins, each with different peak intensity times and spatial expansion patterns. Furthermore, the development trajectory of these localized drought core areas within the larger drought background field can be traced; for instance, the intensity peak of the drought core area in G Lake was found to lag behind the overall peak of the background field by 15 days. This hierarchical processing approach allows the drought event identification system to simultaneously retain macro-level drought patterns and micro-level drought hotspot information, providing a multi-scale analytical basis for regional drought mitigation decision-making.

[0119] Specifically, a hierarchical clustering architecture is first employed for drought event identification. First, coarse clustering is performed globally to identify large-scale drought background fields. Then, the DBSCAN algorithm is used to cluster the three-dimensional spatiotemporal point set of the entire study area, setting a large neighborhood radius ε1 and a small minimum sample size MinPts1, for example, ε1=5 and MinPts1=10. This step can identify large-scale, long-term persistent drought background fields.

[0120] Furthermore, fine clustering is performed within the coarse clustering results to identify localized drought core regions. For each drought background field obtained from the coarse clustering, the DBSCAN algorithm is applied again, but with a smaller neighborhood radius ε2 and a larger minimum sample size MinPts2, for example, ε2=2 and MinPts2=25. This step enables the identification of more intense and concentrated localized drought core regions within the drought background field.

[0121] Finally, a hierarchical relationship can be established between clustering results at different levels, with each local arid core area belonging to a specific large-scale arid background field. This hierarchical relationship is achieved by recording the coarse cluster label of each fine clustering result. For example, a two-dimensional array can be used to store this relationship, where the first column is the coarse cluster label and the second column is the corresponding fine cluster label.

[0122] Through the above technical solutions, this application achieves multi-scale identification of drought events, capable of simultaneously capturing both large-scale drought background and localized drought core areas, thus improving the accuracy and comprehensiveness of drought event identification. The hierarchical clustering structure helps in understanding the spatial structure and evolution of drought events, providing more detailed and accurate information support for drought monitoring and early warning.

[0123] Traditional clustering methods require reprocessing all historical data each time new data arrives, resulting in wasted computing resources and an inability to capture the dynamic evolution of drought events in a timely manner.

[0124] Therefore, in some embodiments Figure 1 The method shown, after obtaining and displaying the drought event parameters of the drought event, further includes: Obtain new drought data for the target region; Based on the newly added drought data, the identification results of the drought event are incrementally updated.

[0125] Incremental updates are achieved by maintaining a dynamic set of core cluster objects. New data points are matched with historical clusters in a neighborhood-based manner, and density connectivity is recalculated only for affected areas. Incremental updates can include new event identification using a boundary point detection mechanism. This mechanism verifies the density reachability of newly added drought points that are not connected to existing clusters, triggering new cluster generation when the core point condition is met. Incremental updates can also include extinction process monitoring implemented through a time decay function. This involves counting time windows where no drought points appear consecutively, and marking them as extinction events when the duration exceeds a preset threshold.

[0126] For example, incremental clustering, such as incremental updates, can establish a cache queue of historical cluster core points. When new data arrives, only the spatiotemporal distance between it and the core points in the cache queue needs to be calculated. For new points that meet the neighborhood radius requirement, they are directly merged into existing clusters and the cluster feature parameters are updated. For isolated point sets, local density calculations are used to determine whether a new cluster has formed. Furthermore, in the emerging event identification stage, a sliding time window is used to analyze the spatiotemporal expansion pattern of new clusters, and the event development status is determined by calculating the spatial overlap rate of adjacent time slices. Correspondingly, in the extinction process monitoring, the last update timestamp is recorded for each active cluster. When no new member point is detected after a set time threshold is exceeded, a cluster split detection algorithm is triggered, and the event termination time is confirmed by tracing back historical data. This scheme reduces the computational complexity to a linear level through a local update strategy, while ensuring the continuous expression of event evolution through a time window mechanism.

[0127] As a preferred embodiment, the specific implementation of this application is as follows: When constructing a real-time drought monitoring system in the X River Basin, after initial clustering based on historical data, the latest meteorological observation data is received every hour. The data preprocessing module aligns the newly arrived data with the historical three-dimensional data matrix in time and space, generating an incremental dataset that includes time dimension extension. The incremental clustering module determines whether a newly added data point belongs to the extended part of an existing drought event or the starting point of a new event by calculating the spatiotemporal distance between the new data point and the core area of ​​the existing cluster. When a new drought point with a spatial density reaching a threshold is detected within three consecutive time slices, a new event early warning mechanism is triggered. For existing event clusters that have not seen new drought points, the system automatically calculates the interval between its most recent active time and the current time. When the interval exceeds a preset threshold, it is marked as a dying event and a decline analysis report is generated.

[0128] Through the above technical solutions, this application effectively addresses the shortcomings of traditional clustering methods in adapting to real-time data streams, achieving dynamic monitoring of the entire lifecycle of drought events. The incremental update mechanism reduces the overhead of redundant computation, ensuring the timeliness of monitoring results. By identifying emerging events, it can capture the spatial clustering characteristics in the early stages of drought, providing early warnings of potential major drought events. The drought extinction monitoring function can accurately determine drought mitigation trends, providing real-time decision support for water resource allocation.

[0129] In some of the schemes mentioned above in this application, there is a lack of quantitative description of the spatial morphological characteristics of drought events, which makes it impossible to accurately distinguish the types of drought events under different development patterns, and also makes it difficult to reveal the intrinsic relationship between drought morphological characteristics and climate driving factors.

[0130] In some embodiments, Figure 1 The method shown, after obtaining and displaying the drought event parameters of the drought event, further includes: Obtain the spatial morphological features of the drought event; Construct a correlation model between the spatial morphological features and climate driving factors.

[0131] The spatial shape features in this application, including compactness, elongation, and boundary fractal dimension, can be further analyzed based on a correlation model between morphological features and climate driving factors.

[0132] Among these methods, compactness is characterized by the ratio of the perimeter to the area of ​​drought patches to represent the degree of spatial aggregation; elongation is determined by principal component analysis to reflect the spatial extension direction by the ratio of the major and minor axes; and the boundary fractal dimension is quantified by box counting. K-means clustering is used to group drought events with similar morphological characteristics into the same category, with each category corresponding to a specific development pattern.

[0133] For example, after completing the three-dimensional spatiotemporal clustering, for the three-dimensional point cloud data of each drought event, the two-dimensional spatial distribution on each time slice is first extracted. The compactness calculation uses a modified Graves formula to eliminate the influence of grid resolution on the calculation results; the elongation calculation obtains the principal axis direction of the spatial distribution through eigenvalue decomposition of the covariance matrix; the boundary fractal dimension calculation uses a multi-grid covering method to obtain the logarithmic linear relationship between the fractal dimension and the grid scale. When constructing the correlation model, a stepwise regression method is used to screen significantly correlated climate factors, and correlation models are established between spatial morphological characteristic parameters and climate driving factors such as the ENSO index and the intensity of the Western Pacific subtropical high, revealing the evolutionary patterns of drought morphology under different climate backgrounds.

[0134] As a preferred embodiment, the specific implementation of this application is as follows: After completing the three-dimensional spatiotemporal clustering identification of drought events, spatial morphological analysis is performed on each cluster. Specific steps include: projecting the spatial distribution of drought events onto a geographic grid to construct polygonal boundaries; calculating spatial compactness using the ratio of area to the square of perimeter, where a ratio close to 1 indicates an approximately circular distribution; determining the elongation length using the ratio of the major and minor axes of the minimum circumscribed ellipse, where a ratio greater than 3 indicates a strip-like distribution; and applying box counting to calculate the boundary fractal dimension, iteratively calculating with grid sizes ranging from 10 km to 100 km, where a fractal dimension greater than 1.35 indicates an irregular boundary. Furthermore, correlation analysis between spatial morphological characteristics and concurrent sea surface temperature index and atmospheric circulation index can reveal the quantitative relationship between El Niño events and the spatial expansion of the drought core area.

[0135] Through the above technical solutions, this application effectively solves the problem of insufficient quantification of the spatial morphological characteristics of drought by traditional methods, realizes the objective classification of drought event types, and, based on the correlation model between morphological characteristics and climate factors, can analyze the driving mechanism of atmospheric fluctuations on the spatial structure of drought, providing a physical basis for drought evolution prediction.

[0136] The functions of drought event identification and display involved in the drought event identification and feature extraction method of this application will now be described in detail with reference to the embodiments.

[0137] This embodiment aims to provide a three-dimensional spatiotemporal drought event identification method based on the density clustering (DBSCAN) algorithm. It can automatically detect continuous drought events from multi-temporal SPEI data and extract key feature parameters such as duration, area, intensity, and severity of continuous drought, thereby enabling refined analysis of regional drought processes.

[0138] This application treats drought events as high-density clusters of points within a three-dimensional spatiotemporal volume (spatial dimensions: longitude and latitude; temporal dimension: months). The DBSCAN algorithm, by defining spatiotemporal neighborhoods (the parameter epsilon controls the joint spatial and temporal distance range, and MinPts controls the minimum number of neighborhood points required for the core point), can automatically discover these densely connected clusters. Each cluster represents a spatiotemporally continuous drought event entity. Noise points correspond to discrete, transient drought signals and can be effectively filtered out. This density-based clustering method overcomes the limitation of existing two-dimensional methods that require post-hoc merging of patches in the temporal dimension, achieving direct and holistic identification of drought events in three-dimensional spatiotemporal space.

[0139] This method can be implemented using Matlab programming. In this case, as follows: Figure 2 As shown, the identification method provided in this embodiment is as follows: S201: Data reading and initialization.

[0140] First, relevant parameters (such as the number of grid rows and columns, drought criteria, minimum drought area percentage, etc.) are set, and SPEI data for the study area are read and stored in a three-dimensional matrix. Data from each month is saved as a slice, forming a complete SPEI time series dataset.

[0141] The Matlab code involved in this step is as follows: data_folder = 'Enter the path to the file you want to open'; list = dir([data_folder, 'xx.xlsx']); SPEI = zeros(nrows, ncols, num_months); for i = 1:num_months spei_data = xlsread([data_folder, list(i).name]); SPEI(:,:,i) = spei_data; end S202: Extraction from arid sites.

[0142] Based on the SPEI value being less than the predetermined drought standard (e.g., X0=-1) and the judgment of data validity (e.g., not -9999 and not NaN), all SPEI data points are traversed to extract the three-dimensional coordinates of the drought points and their corresponding SPEI values.

[0143] The Matlab code involved in this step is as follows: drought_points = []; for t = 1:num_months for row = 1:nrows for col = 1:ncols if SPEI(row, col, t) <X0&&SPEI(row, col, t) ~= -9999&&~isnan(SPEI(row,col, t)) drought_points = [drought_points; row, col, t]; end end end end S203: DBSCAN cluster analysis.

[0144] The DBSCAN clustering algorithm was applied to analyze drought points and identify different drought events. DBSCAN uses density-based clustering to effectively identify spatially independent drought regions. Noise points (marked as -1) in the clustering results were filtered out.

[0145] The Matlab code involved in this step is as follows: labels = DBSCAN(drought_points, epsilon, MinPts).

[0146] S204: Drought event filtering and feature extraction.

[0147] For each candidate drought event, its maximum spatial area, average area, duration, and other characteristics are calculated. Based on the set minimum drought area standard, candidate drought events with too small an area are filtered out, leaving only valid drought events.

[0148] The Matlab code involved in this step is as follows: for idx = 1:length(valid_clusters) k = valid_clusters(idx); [rows, cols, times] = ind2sub(size(bankuai_cl), find(bankuai_cl ==k)); event_spei = SPEI(rows, cols, times); max_area = max(monthly_areas); intensity = mean(event_spei); severity = sum(abs(event_spei - X0)); end S205: Data Visualization.

[0149] The spatial distribution of drought points is displayed through three-dimensional visualization, while a two-dimensional heat map is presented based on the month with the largest drought area, which helps to understand the temporal and spatial evolution of drought events more intuitively.

[0150] The Matlab code involved in this step is as follows: scatter3(drought_points(:,2), drought_points(:,1), drought_points(:,3), 30, labels, 'filled'); imagesc(bankuai_cl(:,:,max_time)); This could have been achieved using conventional computer hardware and the MATLAB software environment, for example: Processor: Modern x86 architecture processor (such as Intel Core i5 and above); Memory: At least 8GB of memory, 16GB or more is recommended to meet the needs of large-scale data processing; Storage space: at least 1TB of hard disk space for storing raw SPEI data and processing results; Operating systems: Windows, Linux, macOS, and other mainstream operating systems; Software: MATLAB 2017b and above, supporting data processing, computation and visualization.

[0151] Taking the X River Basin, a typical river basin in Country X, as the target area, we use SPEI index raster data with a spatial resolution of 0.01° from March to October 2000 for eight months in this region to illustrate the complete process of the automatic drought event identification and feature extraction method proposed in this embodiment.

[0152] Input data includes: 1. Target area: With 108.3556°E and 29.032°N as the lower left corner coordinates, the grid is 21 × 18, totaling 378 pixels; 2.2. Time range: March 2000 to October 2000 (a total of 8 months); 3. Drought Indicator: Standardized Precipitation Evapotranspiration Index (SPEI), stored monthly as an .xlsx file; 4. Data reading path: local directory D:\matlab.DBSCAN\t2000\data\; missing marker value: -9999.

[0153] Based on this, the method provided in this embodiment includes: Step 1: Parameter settings.

[0154] Set the regional grid parameters and discrimination thresholds: drought threshold: SPEI < -1; minimum drought area proportion: 15%; clustering parameters: neighborhood radius ε = 2, minimum number of samples MinPts = 25. Based on the number of effective pixels in the samples, automatically set the lower limit of drought area A0 ≈ 56 pixels (i.e., 378 × 0.15).

[0155] The clustering parameters include two parameters: neighborhood radius and minimum number of samples, which can be adjusted automatically by the total number of samples.

[0156] Step 2: Data reading and 3D matrix construction.

[0157] Read eight months of SPEI data sequentially from a specified folder and construct a three-dimensional data matrix, where each two-dimensional slice represents a month's data.

[0158] Step 3: Arid point extraction and cluster analysis.

[0159] All spatiotemporal pixels were traversed, and effective drought pixels with SPEI less than -1 were selected. Their spatial coordinates and time labels were recorded to form the point set drought_points to be clustered. A total of 361 drought points were extracted.

[0160] The DBSCAN algorithm was used to perform three-dimensional spatial clustering of the point set, with the following settings: ε = 2 (spatial-temporal joint distance radius); MinPts = 25 (minimum number of points in the cluster). The results identified one initial drought event cluster (cluster label > 0), and removed some discrete noise points (cluster label = -1).

[0161] Step 4: Drought event screening and feature extraction.

[0162] For each drought event: calculate its monthly area over 8 months, filter out events with a maximum area of ​​less than 56 pixels, and treat the remaining events as drought events; extract the data of each indicator from the indicator list shown in Table 1 below for the retained drought events.

[0163]

[0164] Table 1 List of Indicators Finally, one valid drought event was selected and its information was stored in the structure array drought_events.

[0165] Step 5: Results Output and Visualization.

[0166] This step can display the monthly distribution map of the maximum drought area, drought event clustering results, and other content.

[0167] For the monthly distribution map of maximum drought area, the time point with the largest drought coverage (e.g., August 2000) is automatically extracted and displayed spatially as a two-dimensional heatmap. The display results can be shown as follows: Figure 3 As shown, the yellow area indicates drought, while the blue area indicates no drought.

[0168] For the drought event clustering diagram, the horizontal axis represents the longitude grid; the vertical axis represents the latitude grid; the Z-axis represents the month; different colors represent different drought events, such as... Figure 4 As shown, two drought events are illustrated.

[0169] The results of parameter analysis for drought events can be: Drought Event ID: 1; Duration: 2 months (February–March); Maximum Area: 364 pixels; Average Intensity (SPEI): -1.5323; Cumulative Intensity: 382.7367.

[0170] This embodiment fully demonstrates the operability and accuracy of this application in actual climate drought analysis scenarios. The method can automatically identify drought events without relying on prior knowledge, and at the same time provides detailed spatiotemporal feature information. It is applicable to various fields such as agricultural disaster assessment, water resource scheduling and early warning, and climate change monitoring.

[0171] In summary, compared with the prior art, this application has the following advantages and beneficial effects: 1. Automatically identify drought events, reducing interference from manually set time periods and regions: the drought event detection rate is increased by 32%; the event boundary identification error is reduced by ±0.5 grids.

[0172] 2. Feature quantification capability, supporting comprehensive evaluation: It can simultaneously output four types of core drought feature parameters; and supports three-dimensional dynamic visualization of the spatial location and temporal evolution of drought events.

[0173] 3. Engineering application advantages: The parameters are adjustable, adaptable to multi-regional and multi-temporal data analysis tasks; for grid data with a spatial resolution of about 1km, the processing time for 10 years of data is less than 5 minutes.

[0174] 4. It has strong compatibility with existing remote sensing or meteorological data and is easy to integrate.

[0175] To better implement the methods in the embodiments of this application, an apparatus is also provided in the embodiments of this application, the apparatus comprising: The preprocessing module is used to generate a three-dimensional data matrix based on drought data of the target area; the three-dimensional data matrix includes multiple pixels, and the pixel data of each pixel includes location, time and corresponding drought parameters; The comparison module is used to determine the three-dimensional spatiotemporal point set from the three-dimensional data matrix based on the drought threshold and the drought parameters of each pixel; The clustering module is used to determine candidate events based on the three-dimensional spatiotemporal point set according to the clustering parameters; A filtering module is used to determine drought events from the candidate events based on an area threshold; The display module is used to obtain and display the drought event parameters of the drought event based on the display parameters.

[0176] This application also provides a computer device that integrates the apparatus provided in this application.

[0177] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0178] Therefore, embodiments of this application provide a computer-readable storage medium, which may include: read-only memory (ROM), random access memory (RAM), a magnetic disk, or an optical disk, etc. A computer program is stored thereon, which is loaded by a processor to execute the steps of the method provided in embodiments of this application.

[0179] Embodiments of this application also provide a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the method described in any of the preceding claims.

[0180] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the detailed descriptions of other embodiments above, which will not be repeated here.

[0181] In practice, each of the above units or structures can be implemented as an independent entity or can be arbitrarily combined to be implemented as the same or several entities. For the specific implementation of each of the above units or structures, please refer to the previous method embodiments, which will not be repeated here.

[0182] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0183] The above provides a detailed description of a drought event identification and feature extraction method and related equipment provided in the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for drought event identification and feature extraction, characterized in that, include: A three-dimensional data matrix is ​​generated based on drought data of the target region; the three-dimensional data matrix includes multiple pixels, and the pixel data of each pixel includes location, time and corresponding drought parameters; Based on the drought threshold, a three-dimensional spatiotemporal point set is determined from the three-dimensional data matrix according to the drought parameters of each pixel; Candidate events are determined based on clustering parameters and the three-dimensional spatiotemporal point set. Drought events are determined from the candidate events based on an area threshold; Based on the display parameters, the drought event parameters of the drought event are obtained and displayed; Obtain the spatial centroid of the drought event at different time slices; Based on the time series of the spatial centroid, the movement pattern of the drought event is determined, and the movement pattern is related to the movement direction and speed of the spatial centroid, as well as the diffusion rate of the drought event. Based on the consistency of the direction of movement and the stability of the diffusion rate, drought events are classified into three types of movement: stationary, unidirectional propagation, and multidirectional diffusion. The method further includes, after acquiring and displaying the drought event parameters, determining the propagation process of the drought event to agricultural drought or hydrological drought based on the time series of the drought event; constructing a propagation model between different drought event types based on the propagation process, the propagation model including time lag relationship and intensity decay model; and predicting the transformation of the drought event based on the propagation model. Before generating a three-dimensional data matrix based on drought data of the target region, the method further includes: determining a baseline value for the drought threshold based on the historical climate characteristics of the target region; and adjusting the baseline value to obtain the drought threshold based on the real-time climate anomaly level of the target region. After acquiring and displaying the drought event parameters, the method further includes: acquiring the spatial morphological features of the drought event; and constructing a correlation model between the spatial morphological features and climate driving factors.

2. The drought event identification and feature extraction method as described in claim 1, characterized in that, Before generating the three-dimensional data matrix based on drought data from the target region, the method further includes: Acquire various environmental data of the target area; The various environmental data are fused and assimilated to obtain drought data for the target region.

3. The drought event identification and feature extraction method as described in claim 2, characterized in that, The various environmental data include remote sensing soil moisture data, station precipitation data, and reanalysis temperature data; the fusion and assimilation processing of the various environmental data to obtain drought data for the target area includes: Spatiotemporal registration was used to unify data of different resolutions into the same grid coordinate system, and Kriging interpolation was used to fill the spatial gaps in the station data. Time series interpolation was used to align data of different observation frequencies to obtain fused data. A dynamic data fusion model based on ensemble Kalman filtering is established. The fused data is iteratively assimilated with the output of the land surface process model to generate a spatiotemporally continuous standardized precipitation evapotranspiration index, which serves as the drought data for the target region.

4. The drought event identification and feature extraction method as described in claim 1, characterized in that, The process of determining candidate events based on clustering parameters and the three-dimensional spatiotemporal point set includes: Based on the background clustering parameters in the clustering parameters, coarse clustering is performed on the three-dimensional spatiotemporal points in the three-dimensional spatiotemporal point set in the global scope to obtain the drought background field; Based on the core clustering parameters in the clustering parameters, fine clustering is performed within the drought background field to obtain the candidate events.

5. The drought event identification and feature extraction method as described in claim 1, characterized in that, After acquiring and displaying the drought event parameters, the method further includes: Obtain new drought data for the target region; Based on the newly added drought data, the identification results of the drought event are incrementally updated.

6. The drought event identification and feature extraction method according to any one of claims 1 to 5, characterized in that, The spatial morphological features include compactness, elongation, and boundary fractal dimension.

Images