A three-dimensional identification method and system for regional drought events

By dividing regional drought events into time scales and identifying drought patches, and combining bidirectional correlation features and spatiotemporal correlations, the problems of spatial morphological complexity and grid data perturbation in drought event identification were solved, thereby improving the stability and spatiotemporal representation of drought events.

CN122286355BActive Publication Date: 2026-11-10HEFEI UNIV OF TECH
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202610773192.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-06-01
Publication Date
2026-11-10
Estimated Expiration
2046-06-01

AI Technical Summary

Technical Problem

Existing methods for identifying regional drought events are inaccurate in complex spatial morphologies, grid data perturbations affect identification stability, and temporal evolution processes are difficult to accurately characterize, affecting the reliable acquisition of indicators such as the duration, affected area, severity, and evolution path of drought events.

Method used

By dividing drought-related grid data in the target area into time scales, drought patches are identified and patch label data is generated. The bidirectional correlation characteristics and spatiotemporal correlation between drought patches are determined and merged into three-dimensional drought events. The spatiotemporal distribution information of drought events is characterized by combining preset dynamic evolution rules.

Benefits of technology

It improves the stability and spatiotemporal characterization of drought event identification, enabling more accurate depiction of the duration, development, and branching evolution of drought events, and reducing the impact of grid data perturbation on the identification results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122286355B_ABST
    Figure CN122286355B_ABST
Patent Text Reader

Abstract

This application provides a three-dimensional identification method and system for regional drought events, relating to the field of meteorological and hydrological disaster monitoring and assessment technology. The method includes: determining drought status data for multiple time steps based on drought-related grid data of a target area within a preset time period; identifying drought patches and generating label data based on the spatial clustering relationship of drought grid points in each time step; determining bidirectional correlation features based on the label data of adjacent time steps, and determining spatiotemporal correlation relationships in conjunction with preset dynamic evolution rules; merging drought patches from different time steps based on spatiotemporal correlation relationships to obtain a three-dimensional drought event; and determining event characterization information based on its spatiotemporal distribution information. The method provided in this application can improve the stability and completeness of drought event identification under complex spatial morphology and address the problem of inaccurate characterization of the temporal evolution process of drought events.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application belongs to the field of meteorological and hydrological disaster monitoring and assessment technology, and in particular relates to a three-dimensional identification method and system for regional drought events. Background Technology

[0002] Drought is a significant natural disaster affecting agricultural production, water resource allocation, ecological environment stability, and regional economic and social operations. It is typically characterized by long duration, wide impact, slow development, and significant cumulative effects. With the continuous development of remote sensing observation, meteorological reanalysis, hydrological simulation, and ground monitoring data, monitoring and assessing regional drought conditions based on gridded spatiotemporal data has become an important technical means for drought disaster identification, risk assessment, and early warning analysis. Analyzing drought conditions at different times and spatial locations within a region can provide a data foundation for assessing the scope, duration, and impact of drought events.

[0003] In related technologies, drought status is typically determined for grid points or pixels within a target area based on precipitation, soil moisture content, runoff, or corresponding drought indices. The determination results are then subjected to spatial statistical or temporal continuity analysis to obtain relevant information about regional drought events. However, actual regional drought is influenced by multiple factors, including climate conditions, topography, underlying surface type, soil properties, and spatial differences in precipitation. Its spatial distribution is often not a regular, continuous pattern, but may exhibit characteristics such as unclear boundaries, local fragmentation, multi-center distribution, or phased expansion and contraction. Furthermore, remote sensing inversion data, reanalysis data, or gridded monitoring data may contain local anomalies, short-term fluctuations, and scattered drought grid points, making drought area identification results susceptible to data perturbation. Existing drought event identification methods often struggle to balance the accuracy of spatial extent identification, the continuity of temporal evolution analysis, and the stability of identification results when facing such complex regional scenarios, thus affecting the reliable acquisition of indicators such as drought event duration, affected area, severity, and evolution path.

[0004] Therefore, in the identification and evolution analysis of regional drought events, the complex spatial distribution of drought, the easy impact of grid data perturbation on identification stability, and the difficulty in accurately representing the temporal evolution of drought events have become urgent problems to be solved. Summary of the Invention

[0005] This application provides a three-dimensional identification method and system for regional drought events, aiming to address the technical problems of insufficient accuracy in existing regional drought event identification methods in terms of complex spatial morphology identification, data perturbation suppression, and spatiotemporal evolution characterization.

[0006] In a first aspect, this application provides a three-dimensional identification method for regional drought events, the method comprising: Based on drought-related grid data of the target area within a preset time period, drought status data corresponding to multiple time steps are determined. The multiple time steps are multiple time units obtained by dividing the preset time period according to a preset time scale. For each time step, based on the spatial clustering relationship between drought grid points in the drought state data corresponding to the time step, drought patches corresponding to the time step are identified, and drought patch label data corresponding to the time step is generated; Based on the drought patch label data corresponding to adjacent time steps, determine the bidirectional association features between drought patches in adjacent time steps; Based on the bidirectional correlation features and the preset dynamic evolution rules, the spatiotemporal correlation between drought patches at adjacent time steps is determined. The preset dynamic evolution rules are used to characterize the continuity and branching evolution of drought patches at adjacent time steps. Based on the spatiotemporal correlations corresponding to multiple adjacent time steps, drought patches in different time steps are merged to obtain the three-dimensional drought event corresponding to the target region. Based on the spatiotemporal distribution information of the three-dimensional drought event, the event characterization information corresponding to the three-dimensional drought event is determined.

[0007] In one possible design, determining drought status data corresponding to multiple time steps based on drought-related grid data of the target area within a preset time period includes: Obtain the drought-related grid data of the target area within the preset time period; Outlier detection and imputation are performed on the drought-related grid data to obtain corrected grid data; The modified grid data is aggregated over time according to the preset time scale to obtain scaled grid data corresponding to multiple time steps. A probability distribution model is constructed based on historical grid data of the target area; Based on the probability distribution model, the scaled grid data is subjected to probability transformation to obtain standardized drought index data corresponding to multiple time steps. Based on the standardized drought index data and the preset drought threshold, the drought status data corresponding to multiple time steps are determined.

[0008] In one possible design, identifying drought patches corresponding to the time step based on the spatial clustering relationships between drought grid points in the drought state data corresponding to the time step, and generating drought patch label data corresponding to the time step, includes: Obtain the clustering parameters corresponding to the time step, the clustering parameters including the neighborhood search range and the core point determination conditions; Extract the grid coordinates of grid points in a drought state from the drought state data corresponding to the time step to obtain a set of drought grid coordinates; Based on the spatial distance between each drought grid point in the drought grid point coordinate set and the neighborhood search range, determine the neighborhood grid point set corresponding to each drought grid point; Based on the neighborhood grid set corresponding to each drought grid point and the core point determination condition, the drought grid points are labeled with types to obtain core drought grid points, boundary drought grid points and noise grid points; Starting with unassigned patch identifiers, core drought grid points are clustered and assigned cluster identifiers based on the density reachability relationships between different core drought grid points and between the core drought grid points and the boundary drought grid points, thus obtaining candidate drought patches. Based on the candidate drought patches and the noise grid points, drought patch label data corresponding to the time step is generated.

[0009] In one possible design, generating drought patch label data corresponding to the time step based on the candidate drought patches and the noise grid points includes: The noisy grid points are removed from the drought grid points indicated by the drought state data corresponding to the time step to obtain the denoised drought state data; Based on the spatial location and grid area of ​​each drought grid point in the candidate drought patch, the patch morphology parameters corresponding to the candidate drought patch are determined; wherein, the patch morphology parameters include the patch area, or the patch morphology parameters include the patch area and the patch centroid; Based on the patch morphology parameters and preset patch screening conditions, target drought patches are determined from the candidate drought patches; wherein, the preset patch screening conditions include the patch area being greater than or equal to a preset area threshold; Based on the patch affiliation relationship of each drought grid point in the target drought patch, the denoised drought state data is labeled to obtain the drought patch label data corresponding to the time step.

[0010] In one possible design, determining the bidirectional association features between drought patches at adjacent time steps based on the drought patch label data corresponding to adjacent time steps includes: Based on the drought patch label data corresponding to adjacent time steps, determine the spatial overlap information between the drought patches in the previous time step and the drought patches in the next time step. Based on the spatial overlap information and the patch area of ​​the drought patch in the previous time step, a first association feature is determined. The first association feature is used to characterize the degree to which the drought patch in the previous time step continues to the drought patch in the subsequent time step. Based on the spatial overlap information and the patch area of ​​the drought patch in the next time step, a second correlation feature is determined. The second correlation feature is used to characterize the degree to which the drought patch in the next time step originates from the drought patch in the previous time step. The bidirectional association feature is obtained based on the first association feature and the second association feature.

[0011] In one possible design, determining the spatiotemporal correlation between drought patches at adjacent time steps based on the bidirectional correlation characteristics and preset dynamic evolution rules includes: Based on the first associated feature and the preset strong inheritance conditions, the continuation relationship is determined, and the strong inheritance determination result is obtained. When the strong inheritance determination result indicates that there is no strong inheritance relationship between drought patches in adjacent time steps, the branch relationship is determined according to the first association feature, the second association feature and the preset branch continuation condition to obtain the branch continuation determination result. Based on the strong inheritance determination result, or, if the branch continuation determination result is obtained, based on the branch continuation determination result, it is determined that drought patches in adjacent time steps have a spatiotemporal correlation relationship for merging into the same three-dimensional drought event; Specifically, if the strong inheritance determination result indicates that there is a strong inheritance relationship between drought patches in adjacent time steps, or if the branch continuation determination result is obtained and the branch continuation determination result indicates that there is a branch continuation relationship between drought patches in adjacent time steps, then it is determined that drought patches in adjacent time steps have a spatiotemporal correlation relationship for merging into the same three-dimensional drought event.

[0012] In one possible design, the merging of drought patches in different time steps based on the spatiotemporal correlations corresponding to multiple adjacent time steps to obtain the three-dimensional drought event corresponding to the target region includes: Based on the drought patch label data corresponding to multiple time steps, patch node identifiers are assigned to drought patches in different time steps; Based on the spatiotemporal correlations corresponding to multiple adjacent time steps, the patch node identifiers with correlations are subjected to equivalent merging processing to obtain multiple spatiotemporal equivalence classes; Generate a global event identifier based on the multiple spatiotemporal equivalence classes; Based on the global event identifier, drought patches in different time steps are re-labeled to obtain the initial three-dimensional drought event; Based on the spatiotemporal voxel scale and preset spatiotemporal voxel threshold of each event in the initial three-dimensional drought event, the initial three-dimensional drought event is filtered to obtain the three-dimensional drought event corresponding to the target region.

[0013] In one possible design, determining the event characterization information corresponding to the three-dimensional drought event based on the spatiotemporal distribution information of the three-dimensional drought event includes: Based on the distribution range of the three-dimensional drought event in the time dimension, determine the time dimension characteristics corresponding to the three-dimensional drought event; Based on the spatial coverage of the three-dimensional drought event, determine the spatial coverage characteristics corresponding to the three-dimensional drought event; Based on the standardized drought index data corresponding to the three-dimensional drought event and the spatial cover characteristics, the intensity characteristics corresponding to the three-dimensional drought event are determined; Based on the spatial location changes of the three-dimensional drought event at different time steps, the evolution path characteristics corresponding to the three-dimensional drought event are determined; Based on the patch morphology changes of the three-dimensional drought event at different time steps, the morphological features corresponding to the three-dimensional drought event are determined; The event representation information is generated based on at least one of the time dimension feature, the spatial coverage feature, the intensity feature, the evolution path feature, and the morphological feature.

[0014] In one possible design, the method further includes: Based on the event characterization information, the event type of the three-dimensional drought event is determined to obtain drought event type information; Based on the spatiotemporal distribution information of the three-dimensional drought event within the target area and the event characterization information, drought risk area information is determined; Based on at least one of the three-dimensional drought event, the drought event type information, and the drought risk area information, generate drought event application output information; The drought event application output information includes at least one of the following: drought event database records, drought evolution visualization information, drought monitoring reports, and drought early warning decision information.

[0015] Secondly, this application provides a three-dimensional identification method system for regional drought events, the system comprising: The drought status determination module is used to determine the drought status data corresponding to multiple time steps based on the drought-related grid data of the target area within a preset time period. The multiple time steps are multiple time units obtained by dividing the preset time period according to a preset time scale. A drought patch identification module is used to identify drought patches corresponding to each time step based on the spatial clustering relationship between drought grid points in the drought state data corresponding to the time step, and generate drought patch label data corresponding to the time step. The association feature determination module is used to determine the bidirectional association features between drought patches in adjacent time steps based on the drought patch label data corresponding to adjacent time steps. The spatiotemporal correlation determination module is used to determine the spatiotemporal correlation between drought patches at adjacent time steps based on the bidirectional correlation features and the preset dynamic evolution rules. The preset dynamic evolution rules are used to characterize the continuity and branching evolution relationships of drought patches at adjacent time steps. The three-dimensional event generation module is used to merge drought patches in different time steps based on the spatiotemporal correlations corresponding to multiple adjacent time steps to obtain three-dimensional drought events corresponding to the target region. The event characterization determination module is used to determine the event characterization information corresponding to the three-dimensional drought event based on the spatiotemporal distribution information of the three-dimensional drought event.

[0016] Thirdly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed, implement the method described in the first aspect or various possible designs of the first aspect.

[0017] Fourthly, this application provides a computer program product, which includes computer program code that, when run on a computer, causes the computer to implement the method described in the first aspect or various possible designs of the first aspect.

[0018] This application provides a method and system for three-dimensional identification of regional drought events. The method determines drought status data corresponding to multiple time steps based on drought-related grid data of a target region within a preset time period, providing a unified time series analysis basis for regional drought identification. Furthermore, for each time step, drought patches are identified based on the spatial clustering relationships between drought grid points, and drought patch label data is generated. This allows drought regions to no longer rely solely on simple grid point determination, but rather to form regionally significant patch results based on spatial clustering characteristics, thereby improving the stability of drought extent identification under complex spatial distribution patterns. Simultaneously, drought patches from adjacent time steps are identified... Label data is used to determine bidirectional correlation features, and combined with preset dynamic evolution rules used to characterize continuity and branching evolution relationships, spatiotemporal correlation relationships are determined. This allows drought patches at different time steps to be associated according to their evolutionary relationships. Then, based on the spatiotemporal correlation relationships of multiple adjacent time steps, a three-dimensional drought event is obtained. Event characterization information is determined based on its spatiotemporal distribution information. This can more accurately depict the continuity, development, and branching evolution process of drought events, and solve the problems of complex spatial distribution of drought, easy perturbation of grid data affecting identification stability, and difficulty in accurately characterizing the temporal evolution process of drought events in existing regional drought event identification and evolution analysis. Attached Figure Description

[0019] Figure 1 A flowchart illustrating a three-dimensional identification method for regional drought events provided in this application embodiment; Figure 2 A schematic diagram of another three-dimensional identification method for regional drought events provided in this application embodiment; Figure 3 This is a schematic diagram of the structure of a three-dimensional recognition system for regional drought events provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, 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, 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.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims and drawings of this application are intended to cover non-exclusive inclusion.

[0022] The term "embodiment" as used herein means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of the phrase "embodiment" in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0023] In this article, the term "and / or" simply describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists, A and B can exist simultaneously, and B exists. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0024] Furthermore, the terms "first," "second," etc., in the specification and claims of this application or in the aforementioned drawings are used to distinguish different objects rather than to describe a specific order, and may explicitly or implicitly include one or more of the features.

[0025] In the description of this application, unless otherwise stated, "multiple" and "at least two" mean two or more (including two), and similarly, "multiple groups" and "at least two groups" mean two or more (including two groups).

[0026] In the description of this application, it should be noted that, unless otherwise explicitly specified and limited, the terms "connected" and "linked" should be interpreted broadly. For example, "connected" or "linked" can refer not only to a physical connection, but also to an electrical connection or a signal connection. For instance, it can be a direct connection, i.e., a physical connection, or an indirect connection through at least one intermediate component, as long as the circuit is connected. It can also refer to the internal connection between two components. A signal connection can refer not only to a signal connection through a circuit, but also to a signal connection through a medium, such as radio waves. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0027] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. It should be noted that, unless otherwise specified, different technical features in this application can be combined with each other.

[0028] Drought events are typically characterized by their long duration, wide impact, slow development, and significant cumulative effects. Regional drought monitoring and assessment usually require identifying drought conditions at different spatial locations and time stages within a target area based on data such as precipitation, soil moisture content, runoff, or corresponding drought indices. This further involves obtaining information on the duration, affected area, severity, and spatial evolution of drought events. With the development of remote sensing data, meteorological reanalysis data, and land surface hydrological simulation data, identifying regional drought events based on gridded spatiotemporal data has become a common technique in drought disaster monitoring, risk assessment, and early warning analysis.

[0029] In related technologies, regional drought event identification typically begins by determining whether grid points within a target area are in a drought state at a given time based on a preset drought threshold. Subsequently, spatial connectivity analysis or region merging is performed on grid points in a drought state at the same time to obtain drought regions on a single time slice. Then, based on the spatial overlap of drought regions in adjacent time slices, it is determined whether drought regions in consecutive time slices belong to the same drought event, thus obtaining drought event identification results with temporal continuity. This type of method can, to a certain extent, extract regional drought events and provide a foundation for subsequent statistical analysis of drought duration, area, and intensity.

[0030] However, in real-world regional drought monitoring scenarios, the spatial distribution of drought is typically influenced by a combination of factors, including topography, spatial variations in precipitation, soil properties, underlying surface conditions, and regional hydrological processes. Drought-stricken areas do not always exhibit a regular, compact, or continuous spatial form; instead, they may display characteristics such as fragmented boundaries, local dispersion, multi-center distribution, strip-like expansion, or irregular contraction. If relevant technologies rely solely on fixed connectivity rules, regular geometric assumptions, or simple regional merging methods to determine drought patches, they may struggle to adapt to these complex spatial morphologies. This could easily lead to the segmentation of a truly relevant drought area into multiple independent regions, or inaccurate merging decisions for regions with significant spatial structural differences, thereby affecting the accuracy of drought impact range and spatial structure identification.

[0031] Furthermore, the remote sensing inversion data, reanalysis data, or gridded monitoring data relied upon for regional drought identification are typically affected by factors such as observation errors, inversion errors, spatial resolution differences, and local meteorological fluctuations. The data may contain scattered anomalous grid points or short-term local fluctuations. If related technologies primarily rely on a single threshold to determine the drought status of grid points, these local anomalous grid points may be misclassified as part of a drought-stricken area, thus interfering with the area, centroid, intensity, and spatial boundaries of drought patches. In some cases, scattered anomalous grid points may even cause spatially independent drought areas to appear connected in the identification results, leading to deviations in the number, extent, and statistical characteristics of drought events.

[0032] Meanwhile, drought events are not static spatial objects, but rather spatiotemporal processes that continuously change over time. A regional drought event may undergo processes such as occurrence, development, expansion, weakening, splitting, merging, and disappearance within its life cycle. Related technologies, when tracking the temporal continuity of drought events, typically rely on the area overlap of drought regions in adjacent time slices. While this method is simple to implement, it may fail to accurately reflect the true evolutionary relationship between drought regions in successive time slices when significant expansion, contraction, splitting, or merging of drought regions occur. For example, when a large drought region disperses into multiple smaller regions in a later time stage, judging solely based on a single overlap ratio may misclassify regions that still exhibit evolutionary continuity as new drought events. Similarly, when multiple regions gradually approach or merge in a later time stage, it may also lead to inaccurate judgments about event attribution, thus affecting the complete characterization of the drought event's life cycle, evolutionary path, and duration.

[0033] Therefore, when dealing with drought events in complex regions, related technologies suffer from several problems, including inaccurate identification of drought spatial morphology, the impact of grid data perturbation on identification stability, and incomplete characterization of the temporal evolution of drought events. These issues further affect the reliability of subsequent analysis results regarding drought event duration, spatial coverage, severity, evolution path, and risk areas. Therefore, it is necessary to improve the three-dimensional spatiotemporal identification process for regional drought events to enhance the stability and spatiotemporal characterization capabilities of the identification results.

[0034] To address the problems existing in the aforementioned related technologies, this application provides a method and system for three-dimensional identification of regional drought events. This method divides drought-related grid data of a target region within a preset time period into time-scale segments and determines the drought state, obtaining drought state data corresponding to multiple time steps. Within each time step, drought patches are identified based on the spatial clustering relationships between drought grid points, and drought patch label data is generated. This eliminates the reliance on regular geometric shapes or fixed connectivity for drought patch identification, thus better adapting to situations such as irregular spatial distribution, fragmented boundaries, or multi-center distribution of regional drought.

[0035] Furthermore, in identifying drought patches corresponding to each time step, this embodiment of the application can perform clustering and labeling assignment on drought grid points based on the neighborhood relationships and density reachability relationships between drought grid points, and identify and remove noisy grid points. This reduces the impact of scattered abnormal grid points or local fluctuations on the drought patch identification results, lowers the possibility of isolated drought grid points being misidentified as valid drought patches or causing different drought regions to be incorrectly connected, and makes the obtained drought patches have more stable spatial clustering characteristics.

[0036] Furthermore, this embodiment of the application determines the bidirectional correlation characteristics between drought patches in adjacent time steps based on drought patch label data from adjacent time steps, and determines the spatiotemporal correlation between drought patches in adjacent time steps in conjunction with preset dynamic evolution rules. By simultaneously considering the degree of continuation of drought patches from the previous time step to the next time step, and the degree to which drought patches in the next time step originate from drought patches in the previous time step, the continuation and branching evolution relationships of drought patches between adjacent time steps can be more accurately characterized, reducing the possibility of drought events being incorrectly cut off or incorrectly merged during splitting, merging, expansion, or decline.

[0037] Based on the above processing, the embodiments of this application can merge drought patches with spatiotemporal correlations at different time steps into a three-dimensional drought event, and further determine event characterization information based on the spatiotemporal distribution information of the three-dimensional drought event. Thus, regional drought events can be characterized from aspects such as temporal duration, spatial coverage, intensity changes, evolution paths, and morphological changes, improving the completeness and stability of the three-dimensional drought event identification results, and providing a data foundation for subsequent drought event classification, risk area identification, monitoring report generation, and early warning decision-making.

[0038] Figure 1 This is a flowchart illustrating a three-dimensional identification method for regional drought events provided in an embodiment of this application. Figure 1 As shown, the three-dimensional identification method for regional drought events provided in this application specifically includes S101 to S106, and S101 to S106 will be described in detail below.

[0039] It should be noted that the method provided in this application embodiment can be executed by a regional drought event three-dimensional recognition system, or by a server, computer equipment, meteorological and hydrological disaster monitoring platform, or other electronic equipment with data processing capabilities. This application embodiment does not limit this.

[0040] S101. Based on the drought-related grid data of the target area within a preset time period, determine the drought status data corresponding to multiple time steps.

[0041] The target area is the geographical area where drought events need to be identified, such as a watershed area, agricultural planting area, administrative area, or spatial area within a preset latitude and longitude range.

[0042] The preset time period is a historical period or monitoring period used for drought identification, such as one year, many years or multiple continuous monitoring cycles.

[0043] Among them, multiple time steps are multiple time units obtained by dividing a preset time period according to a preset time scale.

[0044] It should be noted that the preset time scale can be a daily scale, a phenological scale, a weekly scale, a monthly scale, or other time scales suitable for characterizing the duration of drought. Multiple time steps are multiple time units obtained by discretizing the preset time period according to the preset time scale, and each time step corresponds to a set of drought status data for the target area within that time unit.

[0045] The drought-related grid data includes hydrological and meteorological element data of multiple grid points within the target area at different times. The hydrological and meteorological element data may include at least one of precipitation, runoff, soil moisture content, evapotranspiration, and drought index data.

[0046] In one possible embodiment, outlier detection, missing value imputation, time aggregation, and standardized drought index calculation can be performed on drought-related grid data first. Then, based on a preset drought threshold, it can be determined whether each grid point in each time step is in a drought state, thereby obtaining drought state data corresponding to multiple time steps.

[0047] Drought status data can be represented as binary drought mask data, or as a grid status matrix that includes grid location, drought status identifier, and drought severity information.

[0048] This embodiment can convert drought-related gridded data over a continuous time range into drought state data organized by time step, providing a unified data foundation for subsequent drought patch identification and spatiotemporal evolution analysis.

[0049] S102. For each time step, based on the spatial clustering relationship between drought grid points in the drought state data corresponding to the time step, identify the drought patches corresponding to the time step and generate drought patch label data corresponding to the time step.

[0050] Among them, drought grid points refer to grid points in the drought state data at the corresponding time step that are determined to be in a drought state.

[0051] Spatial clustering relationships are used to characterize the spatial proximity, clustering degree, or density distribution relationship of multiple drought grid points.

[0052] Arid patches refer to arid regions formed by multiple arid grid points with spatial clustering relationships at the same time step.

[0053] Drought patch label data is used to indicate the drought patch to which each drought grid point belongs in the time step. Different drought patches can correspond to different patch labels or patch identifiers.

[0054] This embodiment can convert grid-level drought status data into patch-level drought region results within a single time step, enabling drought identification results to reflect not only whether an individual grid point is arid, but also the spatial aggregation structure of the arid region. Compared to methods that rely solely on statistics based on a single grid point or fixed connectivity rules, this processing improves the stability of drought patch identification under complex spatial distribution conditions.

[0055] S103. Based on the drought patch label data corresponding to adjacent time steps, determine the bidirectional association features between drought patches in adjacent time steps.

[0056] Adjacent time steps can include the previous time step and the next time step. There may be evolutionary relationships between drought patches in the previous time step and drought patches in the next time step, such as continuation, expansion, contraction, branching, merging or disappearance.

[0057] Bidirectional correlation features are used to characterize the degree of correlation between drought patches at adjacent time steps from two directions.

[0058] This embodiment avoids judging the relationship between drought patches in adjacent time steps from only a single direction. Instead, it considers the degree of continuation of the patch from the previous time step to the patch in the next time step, as well as the degree to which the patch in the next time step originates from the patch in the previous time step, thereby providing more sufficient feature basis for subsequent dynamic evolution relationship determination.

[0059] S104. Based on the bidirectional correlation characteristics and the preset dynamic evolution rules, determine the spatiotemporal correlation between drought patches in adjacent time steps. The preset dynamic evolution rules are used to characterize the continuity and branching evolution of drought patches in adjacent time steps.

[0060] The preset dynamic evolution rules may include continuation rules for determining whether drought patches continue stably, or branching evolution rules for determining whether drought patches continue to branch.

[0061] Spatiotemporal correlation is used to indicate whether two drought patches in adjacent time steps should be considered as components of the same drought event in different time steps.

[0062] This embodiment can determine whether there is a continuous evolution relationship between drought patches in adjacent time steps when drought events undergo complex changes such as expansion, contraction, splitting or merging, thereby reducing event misjudgment or misconnection caused by relying solely on a single overlap condition.

[0063] S105. Based on the spatiotemporal correlation of multiple adjacent time steps, drought patches in different time steps are merged to obtain the three-dimensional drought event corresponding to the target area.

[0064] Among them, a three-dimensional drought event refers to a drought event entity composed of both spatial and temporal dimensions.

[0065] Specifically, the same three-dimensional drought event may include one or more drought patches in multiple time steps, and these drought patches form a continuous event chain through spatiotemporal correlation.

[0066] This embodiment can organize two-dimensional drought patches that are related to each other in multiple time steps into a continuous three-dimensional drought event, thereby further forming a spatiotemporal identification result that can reflect the life cycle of the drought event from the spatial identification result of a single time step.

[0067] S106. Based on the spatiotemporal distribution information of three-dimensional drought events, determine the event characterization information corresponding to the three-dimensional drought events.

[0068] The spatiotemporal distribution information of the three-dimensional drought event includes the start and end times and duration of the three-dimensional drought event in the time dimension, the coverage and area changes in the spatial dimension, the spatial location changes in different time steps, the changes in drought intensity and morphological changes, etc.

[0069] Event characterization information is used to describe the spatiotemporal and physical characteristics of three-dimensional drought events.

[0070] This embodiment can further convert the identified three-dimensional drought events into event-level characterization results that are easy to statistically analyze and apply, so that drought events not only have spatial and temporal identification boundaries, but also have quantifiable information such as duration, scope of impact, severity and evolution path.

[0071] In this embodiment, drought state data corresponding to multiple time steps is first determined based on drought-related grid data. Then, drought patches are identified and drought patch label data is generated based on the spatial clustering relationship between drought grid points within each time step, enabling the drought spatial identification process to better adapt to irregular, fragmented, or multi-center distributed drought spatial morphologies. Furthermore, bidirectional correlation features are determined through drought patch label data from adjacent time steps, and spatiotemporal correlations are determined in conjunction with preset dynamic evolution rules, allowing the continuity and branching evolution relationships between drought patches in adjacent time steps to be effectively characterized. Based on this, a three-dimensional drought event is obtained by merging the spatiotemporal correlations of multiple adjacent time steps, and event characterization information is determined based on the spatiotemporal distribution information of the three-dimensional drought event. Therefore, the stability and completeness of the three-dimensional identification results of regional drought events can be improved, the impact of grid data disturbances on the identification results can be reduced, and the problem of accurately representing the temporal evolution process of drought events can be addressed.

[0072] In one possible embodiment, to analyze the spatiotemporal dynamic evolution of regional drought at the event scale, a spatial patch identification method based on density clustering and an event connection method based on spatiotemporal overlap can be used to perform three-dimensional identification of drought events within the target area. Specifically, based on standardized drought index data and a preset drought threshold, the grid data corresponding to each time step within a preset time period can be converted into a binary drought mask, where grid points in a drought state are marked as 1, and grid points in a non-drought state are marked as 0. The binary drought masks corresponding to multiple time steps can form a three-dimensional drought mask matrix, with each time slice corresponding to a two-dimensional grid plane, used to represent the drought state of each grid point within the target area at that time step.

[0073] In the embodiments of this application, a time slice refers to two-dimensional spatial grid data extracted from three-dimensional drought state data at a time step; each time step corresponds to a time slice.

[0074] For each time slice, the coordinates of grid points with a value of 1 in the drought mask can be extracted, and all grid points in a drought state within that time step can be used as the input sample point set for independent density-based spatial clustering. For example, the DBSCAN algorithm can be used to perform spatial clustering on the input sample point set. The DBSCAN algorithm clusters the spatial point set using the neighborhood search range eps and the core point determination condition MinPts, where eps represents the neighborhood search radius and MinPts represents the minimum number of neighborhood samples corresponding to the core point. This process can identify sufficiently densely distributed drought grid points in space and aggregate them into the same drought patch, while spatially isolated grid points, those with insufficient density, or those that do not meet the region size requirements are marked as noise grid points. Compared to identification methods that rely solely on fixed connectivity rules, density-based clustering can adapt to irregular, non-convex, or fragmented drought patches and reduce the interference of scattered drought grid points on the drought patch identification results.

[0075] In one possible embodiment, the neighborhood search range eps and the core point determination criterion MinPts can be set according to the grid resolution of the target area, the minimum identification scale of drought events, and operational monitoring requirements. For example, the neighborhood search range eps can be set to 15 grid points, and the core point determination criterion MinPts can be set to 560 drought grid points. Under these parameters, if the spatial distance between two grid points is less than or equal to 15 grid points, they can be considered as candidate neighbors in the neighborhood relationship. If the number of drought grid points contained in the neighborhood of a certain drought grid point reaches MinPts, then the drought grid point can be marked as a core drought grid point. The above parameter settings can be used to filter spatially isolated, scattered, and excessively small drought grid points or drought patches, thereby retaining contiguous drought areas with regional impact. It should be noted that the specific values ​​of eps and MinPts mentioned above are only examples, and in other embodiments, they can be adjusted according to the spatial resolution of the target area and drought identification requirements.

[0076] After identifying drought patches at each time step, the spatiotemporal overlap of drought patches between adjacent time steps can be used to determine whether drought patches in different time steps belong to the same three-dimensional drought event. Specifically, let a drought patch in a previous time step be the parent patch P, and a drought patch in a later time step be the offspring patch C, with their spatial overlap region being I = P ∩ C. The proportion of the overlap area to the parent patch area Rp = Area(I) / Area(P) and the proportion of the overlap area to the offspring patch area Rc = Area(I) / Area(C) can be calculated respectively. Here, Rp represents how much area of ​​the parent patch extends to the offspring patch, and Rc represents how much area of ​​the offspring patch originates from the parent patch.

[0077] In one possible implementation, the spatiotemporal relationship between drought patches in adjacent time steps can be determined based on strong inheritance and branch continuation conditions. If Rp ≥ 80%, meaning that most of the area of ​​the parent patch is inherited by the offspring patch, a strong inheritance relationship can be determined between the parent and offspring patches, and they possess a spatiotemporal relationship suitable for merging into the same three-dimensional drought event. This situation can correspond to the stable continuation, local expansion, or slight morphological changes of a drought event between adjacent time steps. If the strong inheritance condition is not met, but 0.5 ≤ Rp < 0.8 and Rc ≥ 0.8 are satisfied, the offspring patch can be determined as a branch continuation patch of the parent patch, and they still possess a spatiotemporal relationship suitable for merging into the same three-dimensional drought event. This condition is mainly used to handle the case where a drought event splits, that is, when a large drought patch splits into multiple sub-patterns in a later time step, as long as the main area of ​​a sub-pattern still originates from the parent patch, it can be regarded as a continuation branch of the original drought event. If neither the strong inheritance condition nor the branch continuation condition mentioned above is met, it can be considered that there is no direct inheritance relationship between the offspring patch and the parent patch, and the offspring patch can be used as a new starting point for drought events to participate in subsequent identification.

[0078] In one possible implementation, drought patch identification and spatiotemporal connectivity determination can be performed step-by-step in chronological order, starting from the first time step, and an equivalence class relationship between drought patches can be maintained using a disjoint-set data structure. For a drought patch in the current time step, if it satisfies a strong inheritance condition or a branch continuation condition with a drought patch in the previous time step, the two drought patches can be merged into the same event label using a disjoint-set union operation; if the connectivity condition is not met, a new event label can be assigned to the drought patch. After traversing all time steps, all drought patches connected by spatiotemporal relationships share the same root node and can be further normalized into a unified global event label. Thus, a complete three-dimensional drought event can be represented as a set of drought patches with spatiotemporal relationships within a continuous time interval.

[0079] In one possible embodiment, to ensure that the 3D drought events have actual regional impact, a minimum voxel scale filtering can be performed on the constructed 3D drought events. For example, a preset spatiotemporal voxel threshold can be set to 800 spatiotemporal grids. If the number of spatiotemporal voxels covered by a 3D drought event is less than the preset threshold, the 3D drought event can be determined to be a transient disturbance event with a small spatial range or a short lifespan, and it can be discarded. If the number of spatiotemporal voxels covered by a 3D drought event is greater than or equal to the preset threshold, the 3D drought event can be retained. Through this filtering process, small-scale disturbances without significant regional impact can be eliminated, making the final 3D drought events more reflective of drought processes with continuous and cumulative effects within the target region.

[0080] In one possible embodiment, the method steps shown in S101 can be implemented by S1011 to S1016, which are described in detail below.

[0081] S1011. Obtain drought-related grid data for the target area within a preset time period.

[0082] In one possible embodiment, the geographical boundaries of the area to be identified and the study period can be determined first, and the area to be identified can be designated as the target area, while the study period can be designated as a preset time period. The geographical boundaries of the target area can be determined by latitude and longitude ranges, administrative boundaries, watershed boundaries, agricultural planting area boundaries, or other preset spatial boundaries; the preset time period can be a continuous historical time period or a real-time monitoring time period.

[0083] After determining the target area and the preset time period, hydrological and meteorological element data of the target area within the preset time period can be obtained from multi-source reanalysis datasets, remote sensing inversion datasets, ground observation interpolation datasets, or land surface hydrological simulation datasets, and the hydrological and meteorological element data can be used as drought-related grid data.

[0084] The hydrological and meteorological data may include at least one of precipitation P, runoff R, and soil moisture θ, wherein precipitation P can be used to characterize meteorological drought, runoff R can be used to characterize hydrological drought, and soil moisture θ can be used to characterize agricultural drought.

[0085] In other embodiments, hydrological and meteorological data may also include evapotranspiration, surface temperature, vegetation status parameters, or other gridded variables related to drought conditions; however, this application does not limit such data.

[0086] Furthermore, the drought-related grid data can be organized based on the geographical coordinates of each grid point within the target area and the time dimension within a preset time period.

[0087] Specifically, if V represents hydrological and meteorological data, x and y represent the spatial coordinates of grid points, and t represents time, then a dataset Draw = V(x, y, t) can be constructed, where... ∈Ω, t∈T, where Ω is the spatial domain and T is the time domain. In other words, the dataset Draw is used to represent the hydrological and meteorological element values ​​corresponding to each spatial grid point within the target area at different times. When multiple hydrological and meteorological elements are included, the variable type dimension can be organized together with the spatial and temporal dimensions into a multidimensional tensor structure, thereby forming drought-related grid data suitable for subsequent outlier detection, temporal aggregation, probability transformation, and drought status determination.

[0088] Through the above processing, the multi-source hydrological and meteorological data of the target area within a preset time period can be uniformly organized into a gridded data structure with spatial coordinates and time identifiers, providing a data foundation for subsequently determining the drought status data corresponding to multiple time steps.

[0089] S1012. Perform outlier detection and imputation on drought-related grid data to obtain corrected grid data.

[0090] In one possible embodiment, after acquiring drought-related grid data of the target area within a preset time period, spatiotemporal outlier detection is performed on the drought-related grid data to identify missing values, invalid values, and physical outliers in the drought-related grid data.

[0091] Missing values ​​can include grid data where no valid values ​​were collected, grid data that is empty, or invalid grid data formed due to failures in data transmission, data fusion, or remote sensing inversion. Physical outliers can include data values ​​that exceed the reasonable physical range of the corresponding hydrological and meteorological elements, such as precipitation less than zero, soil moisture content exceeding the preset moisture content range, runoff showing unreasonable abrupt changes, or drought-related variables deviating significantly from the historical range for the same period.

[0092] Specifically, drought-related gridded data can be traversed, and the hydrological and meteorological element values ​​corresponding to each grid point in the spatial domain Ω at each time location in the time domain T can be detected to obtain a set of grid points to be corrected. The set of grid points to be corrected includes at least one grid point, and each grid point can correspond to a spatial location and a time location. For multivariate drought-related gridded data, outlier detection can be performed on different hydrological and meteorological elements such as precipitation P, runoff R, and soil moisture θ separately, or multiple hydrological and meteorological elements can be jointly detected under unified data quality control rules.

[0093] After identifying the grid points to be corrected, a filling method based on spatiotemporal neighborhood weights can be used to repair them. Specifically, for any grid point to be corrected, a set of spatial neighborhood grid points corresponding to that grid point can be determined from drought-related grid point data based on its spatial location. The set of spatial neighborhood grid points can include multiple neighborhood grid points located within a preset spatial distance range of the grid point to be corrected and having valid data values ​​at the corresponding time positions. Further, the spatial weights corresponding to each spatial neighborhood grid point can be determined based on the spatial distance between each spatial neighborhood grid point and the grid point to be corrected; wherein, the spatial weights can be determined using an inverse distance weighting method, that is, the smaller the spatial distance, the greater the spatial weight of the neighborhood grid point.

[0094] In one possible embodiment, the spatial imputation value corresponding to the grid point to be corrected can be determined based on the valid data values ​​and corresponding spatial weights of each spatial neighbor grid point in the spatial neighborhood grid point set. The spatial imputation value is used to represent the spatial interpolation result estimated based on the valid grid point data around the grid point to be corrected. To avoid the correction result being affected by local anomalies due to relying solely on the spatial neighborhood, a time reference value corresponding to the grid point to be corrected can also be determined based on the grid point data of the grid point to be corrected in the same historical period. The grid point data of the same historical period can be valid data of the grid point to be corrected in the same season, the same month, the same phenological period, or the same statistical period corresponding to the current time position in the multi-year historical data; the time reference value can be the historical average, the historical median, or the historical climatological statistical value of the same period.

[0095] In one possible embodiment, the correction value corresponding to the grid point to be corrected can be determined based on the space filling value and the time reference value.

[0096] For example, you can follow the formula Determine the correction value for the grid points to be corrected; where, This represents the variable value after grid point correction; This represents the set of spatial neighborhood grid points corresponding to the grid point to be corrected; Represents the set of spatial neighborhood grid points The original variable value of the i-th spatial neighborhood grid point at time t; This represents the spatial weight corresponding to the i-th spatial neighborhood grid point; This represents the average value of the grid points to be corrected over the same historical period. The smoothing regularization coefficient represents the time dimension and is used to balance the contribution between the spatial interpolation result and the historical average value for the same period.

[0097] Furthermore, the grid points in the drought-related grid data that need correction can be updated based on the correction values ​​to obtain corrected grid data. For data values ​​without anomalies, their original data values ​​can be retained; for data values ​​identified as missing values, invalid values, or physical anomalies, the corresponding correction values ​​can be used to replace them. Thus, the corrected grid data not only preserves the spatial distribution structure and temporal series structure of the original drought-related grid data, but also reduces the impact of missing values, physical anomalies, and local anomaly fluctuations on subsequent temporal aggregation, probability distribution modeling, and drought status determination.

[0098] The method in this embodiment can perform data quality control and spatiotemporal consistency repair on drought-related gridded data before entering the time aggregation process, so that the obtained corrected gridded data has spatial continuity and temporal conformity to the historical variation pattern of the same period, thereby providing a more stable data foundation for subsequent determination of scaled gridded data and standardized drought index data.

[0099] S1013. Perform time aggregation processing on the corrected grid data according to the preset time scale to obtain scaled grid data corresponding to multiple time steps.

[0100] In one possible embodiment, after obtaining the corrected gridded data, the corrected gridded data can be time-aggregated according to a preset time scale to convert the gridded data at the original time resolution into scaled gridded data suitable for drought event identification. Since drought events typically exhibit persistence, cumulativeity, and slow evolution, directly using daily data or data with higher time resolution for identification is easily affected by short-term precipitation fluctuations, local observation errors, or random meteorological disturbances. Therefore, time aggregation processing can eliminate or reduce high-frequency random noise, making the obtained scaled gridded data more suitable for characterizing the continuous changing process of drought conditions.

[0101] In one possible embodiment, the preset time scale may include a phenological scale, a weekly scale, a ten-day scale, a monthly scale, or other pre-defined time scales. Specifically, a phenological scale may represent a five-day consecutive period as a time unit, a weekly scale may represent a seven-day consecutive period as a time unit, and a monthly scale may represent a calendar month or a preset monthly statistical period as a time unit. The preset time scale can be set according to the climate characteristics of the target area, drought monitoring needs, data temporal resolution, and subsequent drought event identification accuracy requirements.

[0102] Specifically, a preset time period can be divided into multiple time aggregation windows based on a preset time scale, with each time aggregation window corresponding to a time step. For each time step, grid data corresponding to multiple original time positions falling within that time aggregation window can be extracted from the corrected grid data, and these grid data are statistically aggregated to obtain the scaled grid data corresponding to that time step. Thus, the corrected grid data can be transformed from a data sequence at the original time resolution into a scaled grid data sequence corresponding to multiple time steps.

[0103] In one possible embodiment, when the corrected grid data is daily data, the daily corrected grid data Vdaily can be resampled into standardized grid data Vscale according to a preset time scale.

[0104] Taking the pentad scale as an example, if the k-th time step corresponds to 5 consecutive days, then the daily data of the same spatial grid point within those 5 consecutive days can be aggregated to obtain the scaled variable value corresponding to that spatial grid point at the k-th time step.

[0105] In one possible embodiment, when the preset time scale is a pentad scale, five consecutive days can be used as a time step to perform time averaging on the daily corrected grid data, resulting in scaled grid data at the pentad scale. Specifically, the scaled grid data corresponding to the k-th time step can be obtained according to the formula... Determined; among them, Indicates spatial location as The scaled variable value corresponding to the grid point at the k-th time step; Indicates spatial location as The grid point represents the corrected variable value on day d; d represents the daily time sequence number; k represents the time step number at the phenological scale. Through the above processing, the daily corrected grid data can be converted into phenological scale grid data, thereby reducing the impact of high-frequency random fluctuations in the daily data on subsequent drought status determination.

[0106] It should be noted that for different types of drought-related variables, the time aggregation method can be determined based on the physical meaning of the variables. For example, for state-related variables such as soil moisture content and surface temperature, the above-mentioned averaging method can be used; for cumulative variables such as precipitation and runoff, the summation method can be used to obtain the cumulative variable values ​​within the corresponding time step. This application does not limit this approach.

[0107] Different time aggregation methods can be used for different types of hydrological and meteorological elements. For cumulative variables such as precipitation and runoff, the cumulative variable value corresponding to that time step can be obtained by summing multiple daily values ​​within the time aggregation window. For state variables such as soil moisture content, surface temperature, and vegetation state parameters, the state variable value corresponding to that time step can be obtained by averaging, medianing, or weighted averaging multiple daily values ​​within the time aggregation window. For example, for the time aggregation window Tk corresponding to the k-th time step, the formula can be used... Determine the scaled grid data; among which, Indicates spatial location as The scaled variable value corresponding to the grid point at the k-th time step; This represents the corrected grid data obtained in S1012; Tk represents the time aggregation window corresponding to the k-th time step; This represents a time aggregation function, which can be a summation function, an average function, a median function, or a weighted average function.

[0108] In one possible implementation, if the hydrometeorological element is precipitation P, the daily precipitation within the time aggregation window can be summed to obtain the cumulative precipitation corresponding to the k-th time step; if the hydrometeorological element is runoff R, summation or averaging can be used according to the physical meaning of the runoff data to obtain the scaled runoff corresponding to the k-th time step; if the hydrometeorological element is soil moisture θ, the daily soil moisture within the time aggregation window can be averaged to obtain the scaled soil moisture corresponding to the k-th time step. By using appropriate time aggregation functions for different variable types, the consistency between the scaled grid data and the physical meaning of the corresponding hydrometeorological elements can be ensured.

[0109] In one possible embodiment, time aggregation processing can employ either a fixed-window aggregation method or a sliding-window aggregation method. Fixed-window aggregation involves dividing a preset time period into multiple non-overlapping time aggregation windows according to a preset time scale, and then aggregating the data within each time aggregation window separately. Sliding-window aggregation involves performing continuous window statistics on the corrected grid data according to a preset window length and a preset sliding step size. Fixed-window aggregation is characterized by its simple calculation process and clear time step boundaries; sliding-window aggregation can improve the continuity of time series changes. The specific time aggregation method used can be set according to the drought event monitoring requirements.

[0110] The method in this embodiment can convert corrected grid data into scaled grid data corresponding to multiple time steps, enabling subsequent probability distribution model construction, standardized drought index calculation, and drought status determination to be processed based on a unified time scale. This reduces the impact of short-term random fluctuations on drought identification results and makes the subsequently identified drought patches and three-dimensional drought events more consistent with the actual characteristics of the continuous evolution of drought events.

[0111] S1014. Construct a probability distribution model based on historical grid data of the target area.

[0112] In one possible embodiment, after obtaining scaled grid data corresponding to multiple time steps, a probability distribution model can be constructed based on historical grid data of the target area. The historical grid data can be grid data of hydrological and meteorological elements collected or acquired in the target area during historical periods, and the historical grid data can use the same or corresponding time scale as the scaled grid data obtained in S1013.

[0113] For example, when the phenological scale is used to aggregate the corrected grid data over time in S1013, the historical grid data can also be organized according to the phenological scale so that the probability distribution model can reflect the historical climate state at the same time scale.

[0114] Specifically, a probability distribution model can be constructed for each spatial grid point within the target area. For spatial locations... For any grid point, a sequence of similar hydrological and meteorological variables over multiple historical years or periods can be extracted from historical grid data, and a probability distribution model can be constructed based on this historical variable sequence. The hydrological and meteorological variables may include at least one of precipitation P, runoff R, and soil moisture θ. When using multiple hydrological and meteorological variables, corresponding probability distribution models can be constructed for different variables to subsequently obtain standardized drought index data to characterize meteorological drought, hydrological drought, or agricultural drought.

[0115] In one possible implementation, to reduce the impact of seasonal climate differences on the probability transformation results, a probability distribution model can be constructed based on historical samples from the same time period corresponding to each time step. For example, when the preset time scale is a phenological scale, multi-year historical samples can be extracted for the same phenological period for each grid point, and a probability distribution model corresponding to that grid point in that phenological period can be constructed based on these multi-year historical samples; when the preset time scale is a monthly scale, multi-year historical samples can be extracted for the same month for each grid point, and a probability distribution model corresponding to that grid point in that month can be constructed based on these multi-year historical samples. Thus, the probability distribution model can reflect the historical climate statistical characteristics of different spatial locations and different time stages within the target area.

[0116] In one possible embodiment, the probability distribution model can be a non-parametric probability distribution model. That is, when constructing the probability distribution model, it is not necessary to pre-assuming that the historical grid data follows a normal distribution or other fixed parameter distribution. Instead, the probability density function or cumulative distribution function is determined based on the distribution characteristics of the historical samples themselves. For example, empirical distribution functions, kernel density estimation methods, or other non-parametric statistical methods can be used to fit the historical variable sequence for each spatial grid point to obtain the corresponding probability density function f(v) or cumulative distribution function F(v). Here, v represents the hydrological and meteorological variable value, f(v) is used to characterize the probability density of the variable value, and F(v) is used to characterize the cumulative probability that the variable value is not greater than v.

[0117] In one possible embodiment, the probability distribution models corresponding to each grid point within the target area can be organized into a set of probability distribution models. This set of probability distribution models may include multiple grid point probability distribution models, each corresponding to a spatial grid point, a hydrological and meteorological variable, and a concurrent statistical unit at a given time scale. For example, for a grid point... For the variable value in the same period statistical unit at the k-th time step, the probability distribution model corresponding to the grid point, the variable, and the same period statistical unit can be called to perform probability transformation processing on the scaled grid data in S1015.

[0118] In one possible implementation, for variables such as precipitation, runoff, and soil moisture content, lower values ​​generally indicate a higher degree of water deficit, so the cumulative probability of low values ​​can be determined directly based on the corresponding probability distribution model. For other drought-related variables, if their physical meanings correspond to the drought severity in different directions, the direction of the variables can be uniformly processed before constructing the probability distribution model or before performing probability transformation, so that different variables can participate in the subsequent standardized drought index calculation with consistent probability meanings.

[0119] The method in this embodiment can establish a probability distribution model that reflects the historical climate statistics characteristics for different spatial grid points within the target area. This allows the subsequent probabilistic transformation of scaled grid point data to be performed without relying on fixed distribution assumptions. It can better adapt to the distribution differences of different regions, different grid points, and different hydrological and meteorological variables, thereby improving the accuracy of extreme drought conditions and providing a probabilistic basis for generating standardized drought index data in S1015.

[0120] S1015. Based on the probability distribution model, the scaled grid data is subjected to probability transformation to obtain standardized drought index data corresponding to multiple time steps.

[0121] In one possible embodiment, after constructing a probability distribution model based on historical grid data of the target region, the scaled grid data can be probabilistically transformed using the probability distribution model. Specifically, for spatial locations within the target region... For any grid point, the scaled variable value corresponding to the grid point at time step t can be input into the probability distribution model corresponding to the grid point to obtain the cumulative probability corresponding to the scaled variable value; then, the cumulative probability is mapped through the inverse function of the standard normal distribution function to obtain the standardized drought index corresponding to the grid point at time step t.

[0122] Specifically, the standardized drought index can be calculated using the formula... It is confirmed that, among them, Indicates spatial location as The grid point corresponds to the standardized drought index at time step t; Indicates spatial location as The scaled variable values ​​corresponding to the grid points at time step t; (·) represents the cumulative distribution function obtained based on the probability distribution model corresponding to this grid point; It represents the inverse function of the standard normal distribution function.

[0123] In one possible implementation, when the hydrological and meteorological element corresponding to the scaled grid data is precipitation, the obtained standardized drought index can be the standardized precipitation index; when the hydrological and meteorological element corresponding to the scaled grid data is runoff, the obtained standardized drought index can be the standardized runoff index; and when the hydrological and meteorological element corresponding to the scaled grid data is soil moisture content, the obtained standardized drought index can be the standardized soil moisture index. This eliminates the differences in dimensions and numerical ranges between different hydrological and meteorological elements, enabling drought-related grid data of different types to be used for drought status determination at a unified scale.

[0124] The method in this embodiment can convert scaled grid data into standardized drought index data corresponding to multiple time steps, providing a unified and comparable data foundation for subsequent determination of drought status data based on preset drought thresholds.

[0125] S1016. Based on the standardized drought index data and the preset drought threshold, determine the drought status data corresponding to multiple time steps.

[0126] In one possible embodiment, after obtaining standardized drought index data corresponding to multiple time steps, a physical threshold for drought identification can be set based on runs theory, and this physical threshold can be used as a preset drought threshold. The preset drought threshold can be denoted as... , Used to determine whether a grid point enters a drought state at the corresponding time step.

[0127] For example, It can be set to -1.0 or -1.5, where -1.0 corresponds to the moderate drought trigger threshold and -1.5 corresponds to the severe drought trigger threshold. The specific threshold can be set according to the drought monitoring level, the climate characteristics of the target area, or the needs of operational applications.

[0128] In one possible embodiment, for a spatial location of For the grid points, with a standardized drought index of SI(x, y, t) corresponding to time step t, the drought index can be determined by comparing SI(x, y, t) with a preset drought threshold. Based on the magnitude relationship between them, construct a binary discriminant function B(x, y, t). ;in, Indicates spatial location as The grid points represent the drought state at time step t; when When the value is 1, it indicates that the grid point is in a dry state at time step t; when When the value is 0, it indicates that the grid point is in a non-dry state at time step t.

[0129] Therefore, the drought status of all spatial grid points within the target area at multiple time steps can be binarized and determined, generating a three-dimensional drought mask matrix Mmask. The three-dimensional drought mask matrix Mmask uses spatial location and time step as indices to record the drought status of each grid point at each time step. This three-dimensional drought mask matrix Mmask can serve as the drought status data itself, or as a part of the drought status data, providing an input basis for subsequent extraction of drought grid points from each time step, identification of drought patches, and construction of three-dimensional drought events.

[0130] The method in this embodiment can convert standardized drought index data into a three-dimensional drought mask matrix Mmask, so that the continuous numerical drought index can be converted into drought / non-drought state results that can be used for spatial patch identification, thereby providing clear data input for subsequent identification of drought patches based on spatial clustering relationships.

[0131] In one possible embodiment, the method steps shown in S102 can be implemented by S1021 to S1026, which are described in detail below.

[0132] S1021. Obtain the clustering parameters corresponding to the time step. The clustering parameters include the neighborhood search range and the core point determination conditions.

[0133] In one possible embodiment, for each time step, clustering parameters corresponding to that time step can be obtained first. These clustering parameters are used to identify grid points in arid states within that time step based on spatial clustering relationships. The clustering parameters may include a neighborhood search range and core point determination criteria. The neighborhood search range is used to limit the spatial range of each arid grid point during its neighborhood search, and the core point determination criteria are used to determine whether an arid grid point has sufficient local spatial clustering.

[0134] In one possible embodiment, the neighborhood search range can correspond to the neighborhood search radius ε in the density clustering algorithm, and the core point determination condition can correspond to the minimum number of neighbors MinPts of the core point. Specifically, the neighborhood search radius ε can be set according to the spatial resolution of the target area, the grid spacing, or the drought patch identification accuracy requirements; the minimum number of neighbors MinPts of the core point can be set according to the minimum spatial size of the effective drought patch in the target area, the grid density, and the noise suppression requirements. Optionally, the neighborhood search range can also be adaptively determined according to the spatial resolution of the target area, so that drought grid points at different spatial resolutions can be clustered and identified according to the matching neighborhood range.

[0135] The method described in this embodiment can provide a parameter basis for subsequent determination of neighborhood grid set, labeling of core drought grid points, and execution of density reachable clustering.

[0136] S1022. Extract the grid coordinates of grid points in a drought state from the drought state data corresponding to the time step to obtain the set of drought grid coordinates.

[0137] In one possible embodiment, after obtaining the drought state data corresponding to a certain time step, grid points in a drought state can be extracted from the drought state data corresponding to that time step based on the drought state identifier. If the drought state data corresponding to that time step is a two-dimensional slice of a three-dimensional drought mask matrix at that time step, grid points in the two-dimensional slice whose values ​​represent the drought state can be extracted, and the spatial coordinates of these grid points can be obtained to form a drought grid point coordinate set.

[0138] Specifically, for time step t, the following can be satisfied: Grid points with coordinates equal to 1 are identified as those in a drought state, and their coordinates are grouped into a drought grid point coordinate set Pt. Pt can be represented as Pt = {p1, p2, ..., pm}, where... , This represents the coordinates of the i-th grid point that is in a drought state. This represents the latitude coordinates of the i-th arid grid point. Let represent the longitude coordinates of the i-th arid grid point, and m represent the number of grid points in the arid state at time step t.

[0139] In other embodiments, Projected coordinates, row and column index coordinates, or other coordinate forms that can characterize the spatial position of grid points can also be used, and this application does not limit this.

[0140] The method in this embodiment can convert the drought state data corresponding to the time step into a set of drought grid coordinates for spatial clustering, so that subsequent clustering is performed only on the foreground grid points that are in a drought state.

[0141] S1023. Based on the spatial distance between each drought grid point in the drought grid point coordinate set and the neighborhood search range, determine the neighborhood grid point set corresponding to each drought grid point.

[0142] In one possible implementation, after obtaining the set of arid grid coordinates Pt, a spatial distance affinity matrix between arid grid points can be constructed. For any two arid grid points in the set of arid grid coordinates Pt... and This allows us to calculate the weighted Euclidean distance or geodesic distance Dij between the two.

[0143] For example, when , At that time, you can follow the formula Determine the spatial distance between the two ;in, Represents the i-th arid grid point With the j-th arid grid point Spatial distance between them; Represents the latitude coordinates of the j-th arid grid point; This represents the longitude coordinates of the j-th arid grid point. The cosine term in the above formula is used to correct for longitude distortion errors in high-latitude regions, thereby improving the geographical accuracy of the spatial distance calculation results.

[0144] Furthermore, the neighborhood search range can be denoted as... For any drought grid point The coordinates of the arid grid points in set Pt can be compared with... The spatial distance between them is less than or equal to The drought grid was determined as The corresponding neighborhood grid points are thus obtained The corresponding neighborhood grid set It can be represented as ;in, Representing drought grid points The corresponding set of neighborhood grid points; Pt represents the set of dry grid point coordinates corresponding to the current time step; This represents any dry grid point in Pt.

[0145] Through the above processing, the set of neighborhood grid points corresponding to each drought grid point can be obtained, which provides a basis for the subsequent type labeling of core drought grid points, boundary drought grid points, and noise grid points.

[0146] S1024. Based on the neighborhood grid set and core point determination conditions corresponding to each drought grid point, drought grid points are labeled with types to obtain core drought grid points, boundary drought grid points and noise grid points.

[0147] In one possible implementation, core drought grid points can be determined based on neighborhood density.

[0148] Specifically, iterate through each arid grid point in the arid grid point coordinate set Pt. Calculate its neighborhood grid set The number of grid points in the neighborhood. If the set of neighborhood grid points... If the number of grid points in the data meets the criteria for determining a core point, then the drought grid point will be... Marked as core drought grid points.

[0149] For example, the core point determination condition can be that the number of neighboring grid points is greater than or equal to the core point number threshold μ, and the determination method can be expressed as follows: ;in, Representing drought grid points The type labeling result; CORE represents the core drought grid point; Represents the set of neighborhood grid points The number of grid points in the data; This represents the threshold for the number of core points, which can also be understood as the minimum number of neighbors for a core point.

[0150] In one possible embodiment, for drought grid points that do not meet the core point determination criteria, it can be further determined whether they are located within the neighborhood of a core drought grid point. If a drought grid point... It does not meet the criteria for core point determination, but core drought grid points exist. , making and Spatial distance between Less than or equal to the neighborhood search range Then you can The grid point marked as the boundary drought point can be represented as: =BORDER; where BORDER represents a boundary drought grid. Although the neighborhood density of a boundary drought grid is insufficient to become a core drought grid, it is located within the neighborhood of a core drought grid and can serve as a boundary component of a candidate drought patch.

[0151] If a drought grid point pi neither meets the criteria for a core point nor is located within the neighborhood of any core drought grid point, then pi can be marked as a noise grid point, which can be represented as: =NOISE; where NOISE represents a noise grid. Noise grids are typically spatially discrete, locally insufficient in density, or drought-prone grids that do not exhibit significant spatial clustering characteristics.

[0152] Through the above type labeling process, the drought grid points in the current time step can be divided into core drought grid points, boundary drought grid points, and noise grid points, providing a basis for subsequent clustering label allocation based on density reachability relations.

[0153] S1025. Taking the unassigned patch identifiers of the core drought grid points as the starting point for clustering, and based on the density reachability relationships between different core drought grid points and between core drought grid points and boundary drought grid points, cluster identifiers are assigned to the core drought grid points and boundary drought grid points in the drought grid point coordinate set to obtain candidate drought patches.

[0154] In one possible embodiment, an unvisited point set U can be established, and the set of arid grid coordinates Pt corresponding to the current time step can be used as the initial set of the unvisited point set U, i.e., U = Pt. Then, a core arid grid point p without assigned patch identifiers is selected from the unvisited point set U as the starting point for clustering, and a new patch identifier Cid is assigned to the core arid grid point p.

[0155] After determining the starting point for clustering, a breadth-first search strategy can be used to find core drought grids and boundary drought grids that are density-reachable from the core drought grid p. These found core and boundary drought grids are added to the grid set corresponding to the patch identifier Cid and removed from the unvisited point set U. Further, for each core drought grid added to the grid set corresponding to the patch identifier Cid, other density-reachable core and boundary drought grids can be searched, and these newly found grids are added to the grid set corresponding to the patch identifier Cid until no new core or boundary drought grids that can be added to the patch identifier Cid through density reachability are found. This yields a candidate drought patch.

[0156] After obtaining a candidate drought patch, the unassigned core drought grid points in the unvisited point set U can be selected as new clustering starting points, and the above clustering label assignment process can be repeated until the unvisited point set U no longer contains any unassigned core drought grid points. Finally, a set of candidate drought patches corresponding to the current time step t can be obtained, which can be represented as... , ... , where k represents the number of candidate drought patches corresponding to the current time step t.

[0157] The method in this embodiment can utilize the density reachability relationships between core drought grid points and between core drought grid points and boundary drought grid points to group drought grid points with spatial clustering characteristics into candidate drought patches, thereby providing patch-level identification results for subsequent generation of drought patch label data.

[0158] S1026. Based on candidate drought patches and noise grid points, generate drought patch label data corresponding to the time step.

[0159] In one possible embodiment, after obtaining a set of candidate drought patches , ... After the noise grid points, drought patch label data corresponding to the current time step t can be generated based on the patch identifiers corresponding to each candidate drought patch. Specifically, core drought grid points and boundary drought grid points belonging to the same candidate drought patch can be assigned the same patch label, drought grid points belonging to different candidate drought patches can be assigned different patch labels, and noise grid points can be marked as empty labels, non-patch labels, or preset invalid labels.

[0160] In one possible embodiment, the drought patch label data corresponding to the current time step t can be represented as: .in, The value of is used to represent the spatial location. The grid point at time step t is identified as the candidate drought patch; if the grid point does not belong to any candidate drought patch, then... The label can be null, 0, or other preset invalid labels. This drought patch label data can be used to record the affiliation between each drought grid point and candidate drought patches within the current time step.

[0161] In one possible embodiment, steps S1021 to S1026 can be executed repeatedly in chronological order until all time steps within a preset time period are traversed, thereby obtaining drought patch label data corresponding to multiple time steps. The drought patch label data corresponding to multiple time steps can form a label sequence L1, L2, ..., Lend, which is used to subsequently determine the bidirectional correlation features between drought patches at adjacent time steps.

[0162] The method in this embodiment can convert candidate drought patches in each time step into drought patch label data with spatial location index and patch affiliation identifier, so that the spatial overlap relationship between patches can be calculated based on the drought patch label data of adjacent time steps, and the spatiotemporal correlation relationship between drought patches can be further determined.

[0163] In one possible embodiment, the method steps shown in S1026 can be implemented by Sa1 to Sa4, which are described in detail below.

[0164] Sa1. Remove noisy grid points from the drought grid points indicated by the drought state data corresponding to the time step to obtain denoised drought state data.

[0165] In one possible embodiment, after obtaining candidate drought patches and noisy grid points, spatial filtering can be performed on the drought state data corresponding to the current time step. Specifically, noisy grid points marked as NOISE in the aforementioned steps can be removed from the drought grid points indicated by the drought state data corresponding to the current time step, or the state value of the noisy grid point in the drought state data can be updated to an ineffective drought state, for example, set to 0, thereby obtaining denoised drought state data.

[0166] Noisy grid points can be understood as spatially discrete drought grid points with insufficient local density or those that fail to belong to candidate drought patches. By removing noisy grid points, the interference of isolated drought grid points on the drought patch identification results can be reduced, and noisy grid points can be avoided from being mistakenly included as part of valid drought patches.

[0167] Sa2. Based on the spatial location and grid area of ​​each drought grid point in the candidate drought patch, determine the patch morphology parameters corresponding to the candidate drought patch.

[0168] Among them, the plaque morphology parameters include plaque area, or the plaque morphology parameters include plaque area and plaque centroid.

[0169] In one possible embodiment, for each candidate drought patch identified at the current time step t, the patch morphology parameters corresponding to the candidate drought patch can be calculated based on the spatial location and area of ​​each drought grid point in the candidate drought patch. The patch morphology parameters may include the patch area, or they may include the patch area and the patch centroid.

[0170] Specifically, for the k-th candidate drought patch in the current time step t The area of ​​a candidate drought patch can be determined by summing the areas of the grid points corresponding to each grid point in the candidate drought patch.

[0171] For example, the patch area can be calculated using the formula Determined; among them, This represents the patch area corresponding to the k-th candidate drought patch in time step t; This represents the k-th candidate drought patch at time step t; Indicates spatial location as The actual area corresponding to each grid point.

[0172] In one possible embodiment, the centroid of a candidate drought patch can also be determined based on the spatial location of each drought grid point in the candidate drought patch.

[0173] For example, the plaque centroid can be determined according to the formula Determined; among them, This represents the centroid of the k-th candidate drought patch at time step t; This represents the sum of the lateral coordinates of all drought grid points in the candidate drought patch; This represents the sum of the vertical coordinates of all drought grid points in the candidate drought patch; This indicates the number of drought grid points in the candidate drought patch.

[0174] The method in this embodiment can obtain morphological description information such as the area and centroid of candidate drought patches, providing a basis for subsequent patch screening and label assignment.

[0175] Sa3. Based on the patch morphology parameters and preset patch screening conditions, the target drought patch is determined from the candidate drought patches.

[0176] Among them, the preset patch screening conditions include that the patch area is greater than or equal to a preset area threshold.

[0177] In one possible embodiment, candidate drought patches can be initially screened based on the relationship between the patch area corresponding to the candidate drought patch and a preset area threshold. The preset area threshold can be set according to the spatial resolution of the target area, the scale of drought event identification, and operational monitoring requirements. For example, the preset area threshold can be set as the minimum area threshold Areamin used to filter out insignificant drought patches.

[0178] Specifically, if candidate drought patches Corresponding patch area If the area is smaller than the preset area threshold Areamin, the candidate drought patch can be considered a small, insignificant drought patch and removed; if the candidate drought patch... Corresponding patch area If the area is greater than or equal to the preset area threshold Areamin, then the candidate drought patch can be identified as the target drought patch.

[0179] The method in this embodiment can further eliminate candidate drought patches with small spatial coverage, so that the remaining target drought patches have more obvious regional drought significance.

[0180] Sa4. Based on the patch affiliation relationship of each drought grid point in the target drought patch, label the denoised drought state data to obtain the drought patch label data corresponding to the time step.

[0181] In one possible embodiment, after identifying the target drought patch, the denoised drought status data can be labeled according to the patch affiliation relationships of each drought grid point within the target drought patch. Specifically, drought grid points belonging to the same target drought patch can be assigned the same patch label, while drought grid points belonging to different target drought patches can be assigned different patch labels. For noisy grid points and grid points corresponding to removed non-significant drought patches, invalid labels, empty labels, or non-drought patch labels can be assigned.

[0182] In one possible implementation, the final identification result at the current time step t can be saved as a two-dimensional label matrix. ,in, The spatial location corresponding to the value is The grid point is identified as belonging to the target arid patch. If the grid point does not belong to any target arid patch, then... It can take 0, null, or other preset invalid values.

[0183] Furthermore, the time steps can be advanced sequentially, for example, by setting t=t+1, and repeatedly performing drought patch identification, noise removal, patch morphology parameter calculation, patch selection, and label assignment until all time steps within the preset time period have been traversed. This yields drought patch label data corresponding to multiple time steps, forming a label sequence L1, L2, ..., Lend.

[0184] In this embodiment, noisy grid points and insignificant drought patches can be further removed from the candidate drought patches, and the remaining target drought patches can be transformed into a two-dimensional label matrix. This allows the drought patch identification results corresponding to each time step to have a clear spatial location and patch ownership identifier, providing a data foundation for subsequent drought patch association analysis between adjacent time steps.

[0185] In one possible embodiment, the method steps shown in S103 can be implemented by S1031 to S1034, which are described in detail below.

[0186] S1031. Based on the drought patch label data corresponding to adjacent time steps, determine the spatial overlap information between the drought patches in the previous time step and the drought patches in the next time step.

[0187] In one possible implementation, drought patch label data corresponding to adjacent time steps can be extracted. For example, the drought patch label matrix corresponding to the previous time step t can be extracted. And the drought patch label matrix corresponding to the next time step t+1. .according to It is possible to determine each drought patch Pi in the previous time step t, based on Each drought patch Cj in the next time step t+1 can be determined.

[0188] For any drought patch Pi in the previous time step t and any drought patch Cj in the next time step t+1, the spatial intersection Iij between them can be calculated, which can be expressed as: Iij=Pi∩Cj.

[0189] Here, Iij represents the spatial overlap region between drought patch Pi and drought patch Cj. Further, the overlap area Area(Iij) corresponding to this spatial intersection Iij can be calculated, and the spatial intersection Iij and the overlap area Area(Iij) can be used as spatial overlap information between drought patches in the previous time step and drought patches in the next time step.

[0190] S1032. Based on the spatial overlap information and the patch area of ​​the drought patch in the previous time step, determine the first association feature. The first association feature is used to characterize the degree to which the drought patch in the previous time step continues to the drought patch in the next time step.

[0191] In one possible implementation, the first association feature can be set as the forward overlap rate, also known as the parent inheritance rate. This first association feature is used to characterize how much area of ​​the drought patch Pi in the previous time step t extends to the drought patch Cj in the next time step t+1.

[0192] Specifically, the first association feature can be determined according to the formula Rparent(i,j) = Area(Pi∩Cj) / Area(Pi); where Rparent(i,j) represents the first association feature between drought patch Pi and drought patch Cj; Area(Pi∩Cj) represents the spatial overlap area between drought patch Pi and drought patch Cj; and Area(Pi) represents the patch area of ​​drought patch Pi in the previous time step t. The larger the value of Rparent(i,j), the higher the degree to which drought patch Pi extends to drought patch Cj.

[0193] S1033. Based on the spatial overlap information and the patch area of ​​the drought patch in the next time step, determine the second association feature. The second association feature is used to characterize the degree to which the drought patch in the next time step originates from the drought patch in the previous time step.

[0194] In one possible embodiment, the second association feature can be set as the backward overlap rate, also known as the offspring sourcing rate. This second association feature is used to characterize how much area of ​​drought patch Cj in the later time step t+1 originates from drought patch Pi in the previous time step t.

[0195] Specifically, the second association feature can be determined according to the formula Rchild(i,j) = Area(Pi∩Cj) / Area(Cj); where Rchild(i,j) represents the second association feature between drought patch Pi and drought patch Cj; Area(Pi∩Cj) represents the spatial overlap area between drought patch Pi and drought patch Cj; and Area(Cj) represents the patch area of ​​drought patch Cj in the next time step t+1. The larger the value of Rchild(i,j), the higher the degree to which drought patch Cj originates from drought patch Pi.

[0196] S1034. Based on the first and second association features, bidirectional association features are obtained.

[0197] In one possible embodiment, for each patch pair formed by drought patch Pi in the previous time step and drought patch Cj in the next time step, the first association feature Rparent(i,j) and the second association feature Rchild(i,j) can be used together as the bidirectional association feature of the patch pair. That is, the bidirectional association feature can be expressed as: Rij={Rparent(i,j),Rchild(i,j)}; where Rij represents the bidirectional association feature between drought patch Pi and drought patch Cj.

[0198] This bidirectional correlation feature can simultaneously describe the degree of continuity of drought patches from the previous time step to the next time step, as well as the degree to which drought patches in the next time step originate from drought patches in the previous time step, providing a basis for determining the spatiotemporal correlation between drought patches in adjacent time steps according to preset dynamic evolution rules.

[0199] In this embodiment, based on drought patch label data from adjacent time steps, the spatial overlap region, parent inheritance rate, and offspring sourcing rate between drought patches at different time steps can be calculated, thereby obtaining correlation features that characterize the bidirectional evolutionary relationship of drought patches. This processing can provide a data foundation for subsequent identification of drought patch continuity relationships, branching evolution relationships, and three-dimensional drought event merging.

[0200] In one possible embodiment, the method steps shown in S104 can be implemented by S1041 to S1043, which are described in detail below.

[0201] S1041. Determine the continuation relationship based on the first association feature and the preset strong inheritance conditions to obtain the strong inheritance determination result.

[0202] In one possible embodiment, a dynamic threshold set can be first established to determine whether drought patches in adjacent time steps belong to the same three-dimensional drought event. This dynamic threshold set may include a strong inheritance threshold αstrong, a branch lower bound threshold αlower, and a source purity threshold βpure. The strong inheritance threshold αstrong is used to determine whether a drought patch from the previous time step continues to a drought patch in the next time step at a relatively high proportion; the branch lower bound threshold αlower and the source purity threshold βpure are used to subsequently determine the branch continuation relationship.

[0203] For example, αstrong can be set to 0.8, αlower can be set to 0.5, and βpure can be set to 0.8. The specific values ​​can be set according to the drought evolution characteristics and identification accuracy requirements of the target area.

[0204] In one possible embodiment, the first association feature can be the forward overlap rate Rparent(i,j) between drought patch Pi in the previous time step and drought patch Cj in the next time step. Rparent(i,j) can be compared with the strong inheritance threshold αstrong to obtain the strong inheritance determination result.

[0205] Specifically, when Rparent(i,j) ≥ αstrong, drought patch Cj can be determined to be a direct strong successor of drought patch Pi. The strong inheritance determination result indicates that there is a strong inheritance relationship between drought patches in adjacent time steps. This situation can correspond to the stable persistence or simple expansion process of drought events between adjacent time steps. If Rparent(i,j) does not reach αstrong, the strong inheritance determination result indicates that there is no strong inheritance relationship between drought patches in adjacent time steps.

[0206] S1042. If the strong inheritance determination result indicates that there is no strong inheritance relationship between drought patches in adjacent time steps, the branch relationship is determined according to the first association feature, the second association feature and the preset branch continuation condition to obtain the branch continuation determination result.

[0207] In one possible embodiment, when the strong inheritance condition is not met between drought patches in adjacent time steps, it can be further determined whether a branch continuation relationship exists. The second association feature can be the backward overlap rate Rchild(i,j) between drought patch Pi in the previous time step and drought patch Cj in the next time step.

[0208] Rparent(i,j) is used to characterize how much area of ​​drought patch Pi extends to drought patch Cj, and Rchild(i,j) is used to characterize how much area of ​​drought patch Cj originates from drought patch Pi.

[0209] In one possible embodiment, the preset branch continuation condition may include: a first associated feature is greater than or equal to the branch lower limit threshold αlower and less than the strong inheritance threshold αstrong, and a second associated feature is greater than or equal to the source purity threshold βpure. This condition can be expressed as: αlower≤Rparent(i,j)<αstrong, and Rchild(i,j)≥βpure.

[0210] If the above conditions are met, drought patch Cj can be determined as a major branch continuation patch of drought patch Pi. The branch continuation determination result indicates that there is a branch continuation relationship between drought patches in adjacent time steps. This situation can correspond to the case where a large-scale drought patch splits in a later time step, and one of the sub-patterns still mainly originates from the original drought patch. Through this determination, the possibility of drought events being incorrectly segmented into multiple independent events when they split can be reduced.

[0211] S1043. Based on the strong inheritance determination result, or, if the branch continuation determination result is obtained, determine that drought patches in adjacent time steps have a spatiotemporal correlation relationship for merging into the same three-dimensional drought event.

[0212] Specifically, if the strong inheritance determination result indicates that there is a strong inheritance relationship between drought patches in adjacent time steps, or if the branch continuation determination result is obtained and the branch continuation determination result indicates that there is a branch continuation relationship between drought patches in adjacent time steps, then it is determined that drought patches in adjacent time steps have a spatiotemporal correlation relationship for merging into the same three-dimensional drought event.

[0213] In one possible embodiment, when the strong inheritance determination result indicates that there is a strong inheritance relationship between drought patches in adjacent time steps, it can be determined that the drought patches in adjacent time steps have a spatiotemporal correlation for merging into the same three-dimensional drought event. When the strong inheritance determination result indicates that there is no strong inheritance relationship, but the branch continuation determination result indicates that there is a branch continuation relationship between drought patches in adjacent time steps, it can also be determined that the drought patches in adjacent time steps have a spatiotemporal correlation for merging into the same three-dimensional drought event.

[0214] In one possible embodiment, a patch connectivity record can be generated based on the above determination result. This record records the spatiotemporal relationship between drought patch Pi in the previous time step and drought patch Cj in the next time step. Furthermore, when merging three-dimensional drought events subsequently, equivalent merging processing can be performed on drought patches Pi and Cj based on this connectivity record. For example, the patch identifier ID (Pi) of drought patch Pi and the patch identifier ID (Cj) of drought patch Cj can be used as merging objects so that they can be subsequently merged into the same spatiotemporal event tree.

[0215] Specifically, if the strong inheritance determination result indicates that there is a strong inheritance relationship between drought patches in adjacent time steps, or if the branch continuation determination result is obtained and the branch continuation determination result indicates that there is a branch continuation relationship between drought patches in adjacent time steps, then it is determined that drought patches in adjacent time steps have a spatiotemporal correlation relationship for merging into the same three-dimensional drought event.

[0216] In this embodiment, strong inheritance and branch continuation relationships between drought patches at adjacent time steps can be determined based on forward overlap rate and backward overlap rate, respectively, thereby obtaining the spatiotemporal correlation relationship used to construct a three-dimensional drought event. This process can more accurately determine the event attribution relationship between drought patches at different time steps when drought events are stable, expanding, or splitting, reducing event misjudgment or misconnection caused by a single overlap threshold judgment.

[0217] In one possible embodiment, the method steps shown in S105 can be implemented by S1051 to S1055, which are described in detail below.

[0218] S1051. Assign patch node identifiers to drought patches in different time steps based on drought patch label data corresponding to multiple time steps.

[0219] In one possible embodiment, after obtaining the drought patch label data corresponding to multiple time steps, each drought patch in each time step can be treated as an independent patch node, and a patch node identifier can be assigned to that patch node. Specifically, for the i-th drought patch in time step t, its patch node identifier can be represented as IDt,i; for the j-th drought patch in time step t+1, its patch node identifier can be represented as IDt+1,j. Through this processing, two-dimensional drought patches in different time steps can be converted into node objects that can be used for spatiotemporal merging processing.

[0220] S1052. Based on the spatiotemporal correlations of multiple adjacent time steps, perform equivalent merging on the patch node identifiers with correlations to obtain multiple spatiotemporal equivalence classes.

[0221] In one possible embodiment, a disjoint-set data structure can be initialized, and based on the spatiotemporal correlation between drought patches in adjacent time steps determined in the aforementioned steps, equivalent merging processing can be performed on the node identifiers of patches with correlation. Specifically, for two drought patches in adjacent time steps that are determined to be suitable for merging into the same three-dimensional drought event, a connection pair (IDt,i,IDt+1,j) can be formed, and a Union(IDt,i,IDt+1,j) operation can be performed in the disjoint-set data structure to merge the two patches into the same spatiotemporal event tree.

[0222] In one possible implementation, the disjoint-set data structure can employ path compression and rank-based merging for equivalence class merging. By traversing the patch connectivity relationships corresponding to all adjacent time steps, all two-dimensional drought patches linked by strong inheritance or branch continuation relationships can be logically stitched together into a continuous three-dimensional spatiotemporal tubular structure. After merging, multiple spatiotemporal equivalence classes can be obtained by querying the root node identifier corresponding to each patch node identifier, with each spatiotemporal equivalence class corresponding to a candidate three-dimensional drought event.

[0223] S1053. Generate a global event identifier based on multiple spatiotemporal equivalence classes.

[0224] In one possible implementation, since the root node identifiers obtained after merging disjoint-set data are typically discontinuous and sparse, global event identifiers can be generated based on multiple spatiotemporal equivalence classes to facilitate subsequent storage, indexing, and statistical analysis. Specifically, all valid root node identifiers can be extracted, and a global event identifier mapping table can be established, mapping the root node identifiers to a continuously increasing sequence of positive integers starting from 1: E1, E2, ..., EN, where N represents the number of identified candidate 3D drought events. Through this mapping process, each spatiotemporal equivalence class can correspond to a globally unique and continuous event number.

[0225] S1054. Based on the global event identifier, the drought patches in different time steps are re-labeled to obtain the initial three-dimensional drought event.

[0226] In one possible implementation, the original 3D label matrix can be re-encoded using a global event identifier mapping table. Specifically, for any arid patch at any time step, the patch node identifier corresponding to the arid patch can be determined first. Then, the root node identifier corresponding to the patch node identifier is queried using a disjoint-set data structure, and the corresponding global event identifier is determined using the global event identifier mapping table. Then, the values ​​of all grid points covered by the arid patch in the 3D label matrix can be updated to the corresponding global event identifiers.

[0227] In one possible embodiment, the relabeled three-dimensional label matrix can be represented as follows: ,in, The value of is used to represent the spatial location. The grid points are identified by the 3D drought event identifier at time step t. After relabeling, drought patches belonging to the same spatiotemporal equivalence class at different time steps have the same global event identifier, thus obtaining the initial 3D drought event.

[0228] S1055. Based on the spatiotemporal voxel scale and preset spatiotemporal voxel threshold of each event in the initial three-dimensional drought event, the initial three-dimensional drought event is filtered to obtain the three-dimensional drought event corresponding to the target region.

[0229] In one possible embodiment, although area screening has been performed during the aforementioned two-dimensional drought patch identification process, transient disturbance events with short lifecycles or small total cumulative impact ranges may still exist after the three-dimensional event is constructed. Therefore, the initial three-dimensional drought events can be screened a second time based on the spatiotemporal voxel scale corresponding to each event in the initial three-dimensional drought event.

[0230] Specifically, for the k-th initial three-dimensional drought event Ek, the total number of spatiotemporal voxels it contains can be calculated. , According to the formula Determined; among them, This represents the spatiotemporal voxel scale corresponding to the k-th initial three-dimensional drought event; This represents the relabeled 3D label matrix; (·) denotes an indicator function, when The value is 1 when the value is equal to k, otherwise the value is 0.

[0231] In one possible embodiment, a preset spatiotemporal voxel threshold, Volmin, can be set, for example, 800 spatiotemporal grids. If the value is less than Volmin, then the initial three-dimensional drought event can be determined as non-significant climate noise, and it can be removed from [the list of events]. Delete or mark as invalid events; if If the value is greater than or equal to Volmin, the initial 3D drought event is retained. Through the above filtering process, the 3D drought events corresponding to the target region can be obtained.

[0232] In the embodiments of this application, two-dimensional drought patches in multiple time steps can be merged into three-dimensional drought events with unified event identifiers according to spatiotemporal correlation. Transient disturbance events with extremely short lifecycles or small cumulative impact ranges can be removed by spatiotemporal voxel-scale screening, so that the final retained three-dimensional drought events have more significant spatiotemporal continuity and cumulative impact characteristics.

[0233] In one possible embodiment, the method steps shown in S106 can be implemented by S1061 to S1066, which are described in detail below.

[0234] S1061. Based on the distribution range of three-dimensional drought events in the time dimension, determine the time dimension characteristics corresponding to the three-dimensional drought events.

[0235] In one possible implementation, for any three-dimensional drought event Ek, a set of spatiotemporal coordinates corresponding to the three-dimensional drought event, Coordsk, can be extracted. Coordsk can include the spatial coordinates and time step coordinates corresponding to all voxels in the three-dimensional drought event. Based on the time step coordinates in Coordsk, the start time Tstart, end time Tend, and duration Dk of the three-dimensional drought event can be determined.

[0236] In this context, the start time Tstart is the earliest occurrence time step of the three-dimensional drought event, the end time Tend is the last disappearance time step of the three-dimensional drought event, and the duration Dk can be determined using the formula Dk = (Tend - Tstart + 1) × Δt; where Δt represents the temporal resolution of the data, for example, when the time step is a phenological scale, Δt can be 5 days. Therefore, the start time, end time, and duration can be considered as the temporal dimension features corresponding to the three-dimensional drought event.

[0237] S1062. Based on the spatial coverage of the three-dimensional drought event, determine the spatial coverage characteristics corresponding to the three-dimensional drought event.

[0238] In one possible embodiment, the spatial coverage characteristics of a three-dimensional drought event Ek can be determined based on its spatial coverage range at different time steps. Specifically, the actual physical area covered by the three-dimensional drought event at each time step can be calculated, and the maximum instantaneous area and the average impact area can be further determined. The maximum instantaneous area can be expressed as the maximum physical area covered by the three-dimensional drought event at any time step within its lifecycle, and the average impact area can be expressed as the average of the coverage areas of the three-dimensional drought event at each time step within its lifecycle.

[0239] For example, the maximum instantaneous area Amax can be calculated using the formula Determined; among them, This represents the set of spatial grid points corresponding to a three-dimensional drought event at time step t. Indicates spatial location as The actual physical area corresponding to the grid points.

[0240] S1063. Based on the standardized drought index data and spatial cover characteristics corresponding to the three-dimensional drought events, determine the intensity characteristics corresponding to the three-dimensional drought events.

[0241] In one possible embodiment, the cumulative severity of the three-dimensional drought event over its lifecycle can be calculated as an intensity feature based on the spatiotemporal coordinate set Coordsk corresponding to the three-dimensional drought event Ek and the standardized drought index data corresponding to each voxel.

[0242] Specifically, the cumulative severity Sk can be calculated using the formula... Determined; among them, This indicates the preset drought threshold. Indicates spatial location as The grid point corresponds to the standardized drought index at time step t. This represents the actual physical area corresponding to the grid point. This intensity feature comprehensively reflects the degree of drought and water scarcity, as well as the spatiotemporal extent of its impact.

[0243] In one possible embodiment, the spatiotemporal point with the minimum standardized drought index can be searched throughout the entire lifecycle of the three-dimensional drought event Ek, and this spatiotemporal point can be identified as the peak intensity point (xpeak, ypeak, tpeak). Here, xpeak and ypeak represent the spatial location of the peak intensity point, and tpeak represents the time step in which the peak intensity occurs. This peak intensity point can be used to characterize the time and location of the most intense development of the drought event.

[0244] S1064. Based on the spatial location changes of three-dimensional drought events at different time steps, determine the evolution path characteristics corresponding to the three-dimensional drought events.

[0245] In one possible implementation, a weighted geometric centroid can be calculated for the spatial distribution of the three-dimensional drought event Ek at each time step. Specifically, drought intensity can be used as the weight to make the centroid location closer to the region with higher drought intensity.

[0246] For time step t, the centroid coordinates of this three-dimensional drought event can be determined using the formula... Determined; among them, and This represents the spatial coordinates of the i-th grid point in time step t. This indicates the drought intensity weight corresponding to that grid point. = - Therefore, the time-varying centroid trajectory sequence of a three-dimensional drought event can be obtained. .

[0247] Furthermore, the migration path, migration direction, and total migration distance of drought events can be determined based on the centroid trajectory sequence. According to the formula Determined; among them, This represents the spatial distance between the centroids of adjacent time steps.

[0248] Furthermore, it can be based on the total migration distance and duration Determine the average migration speed (Speed) mig Speedmig= Therefore, the centroid trajectory, migration direction, total migration distance, and average migration speed can be used as the evolution path features corresponding to three-dimensional drought events.

[0249] S1065. Based on the changes in patch morphology at different time steps of a three-dimensional drought event, determine the corresponding morphological characteristics of the three-dimensional drought event.

[0250] In one possible embodiment, a patch compactness index can be calculated for the drought patch morphology at each time step of the three-dimensional drought event Ek to characterize whether the spatial morphology of the drought patch is regular.

[0251] For time step t, the patch compaction Compt can be calculated using the formula Compt = 4π·Areat / (Perimetert). 2 Determined; where Aret represents the patch area corresponding to time step t of the three-dimensional drought event, and Perimetert represents the perimeter of the patch corresponding to that time step. The closer Compt is to 1, the closer the drought patch shape is to a circle at that time step; the smaller Compt is, the more elongated, fragmented, or irregular the drought patch shape is.

[0252] In one possible embodiment, the average compactness of a three-dimensional drought event over its entire life cycle can be calculated, and this average compactness can be used as a morphological feature to characterize the regularity or fragmentation of the overall spatial morphology of the three-dimensional drought event.

[0253] S1066. Generate event representation information based on at least one of the following: time dimension features, spatial coverage features, intensity features, evolution path features, and morphological features.

[0254] In one possible embodiment, the aforementioned temporal dimension features, spatial coverage features, intensity features, evolution path features, and morphological features can be combined to generate event characterization information corresponding to a three-dimensional drought event. The event characterization information may include at least one of the following: event start time, end time, duration, maximum instantaneous area, average affected area, cumulative severity, peak intensity time and location, centroid migration path, total migration distance, average migration velocity, and morphological compactness.

[0255] In the embodiments of this application, three-dimensional drought events can be quantitatively characterized from multiple dimensions such as duration, spatial coverage, drought severity, migration process, and morphological changes, so that the identified three-dimensional drought events not only have spatiotemporal boundaries, but also have event-level features that can be used for subsequent classification, risk assessment, and early warning analysis.

[0256] Figure 2 This is a flowchart illustrating another method for three-dimensional identification of regional drought events provided in an embodiment of this application. Figure 2As shown, in one possible embodiment, the method further includes S201 to S203, which are described in detail below.

[0257] S201. Based on event characterization information, determine the event type of three-dimensional drought events to obtain drought event type information.

[0258] In one possible embodiment, the event type of a three-dimensional drought event can be determined based on the event characterization information corresponding to that event. Specifically, drought event classification rules can be established based on the duration Dk in the event characterization information, and drought event type information can be determined based on the classification rules.

[0259] For example, if the duration Dk is less than or equal to 30 days, the three-dimensional drought event can be labeled as an acute drought; if the duration Dk is greater than 30 days and less than or equal to 180 days, the three-dimensional drought event can be labeled as a seasonal drought; and if the duration Dk is greater than 180 days, the three-dimensional drought event can be labeled as a multi-year persistent drought. Thus, based on the temporal dimension characteristics of three-dimensional drought events, information on drought event types with physical meaning can be obtained.

[0260] In one possible embodiment, when the input data includes both meteorological and hydrological variables, composite drought identification can be performed based on the spatiotemporal overlap between different types of three-dimensional drought events. For example, the spatiotemporal overlap of meteorological drought events and hydrological drought events can be compared. If the spatiotemporal overlap meets preset composite determination conditions, the corresponding event is identified as a meteorological-hydrological composite drought event, and the corresponding lag response time is determined. This allows for more refined drought event type information.

[0261] S202. Based on the spatiotemporal distribution information and event characterization information of three-dimensional drought events in the target area, determine the drought risk area information.

[0262] In one possible implementation, drought risk statistics can be performed on each spatial grid point within the target area based on multiple three-dimensional drought events identified over a long time series. Specifically, multiple three-dimensional drought events can be superimposed in the spatial dimension, and the number of times each grid point is covered by the three-dimensional drought events can be counted to obtain the cumulative drought frequency; alternatively, the drought severity corresponding to each grid point can be accumulated based on the intensity features in the event characterization information to obtain the cumulative drought severity.

[0263] Furthermore, drought risk area information corresponding to the target area can be generated based on the cumulative drought frequency and cumulative drought severity. This drought risk area information may include drought risk zoning results, high-frequency drought areas, high-intensity drought areas, and drought risk hotspots. Through this processing, spatial areas within the target area that are more prone to drought or have a higher degree of drought impact can be identified.

[0264] S203. Generate drought event application output information based on at least one of the following: three-dimensional drought event, drought event type information, and drought risk area information.

[0265] The output information of drought event applications includes at least one of the following: drought event database records, drought evolution visualization information, drought monitoring reports, and drought early warning decision information.

[0266] In one possible embodiment, drought event application output information can be generated based on at least one of three-dimensional drought events, drought event type information, and drought risk area information. The drought event application output information may include at least one of drought event database records, drought evolution visualization information, drought monitoring reports, and drought early warning decision information.

[0267] In one possible implementation, a spatiotemporal database of drought events can be constructed. Specifically, a relational database or a NoSQL database can be used to establish an event attribute table and a related trajectory table, and the event identifier, start and end time, duration, total severity, maximum area, average area, migration path, event type, and other information of the three-dimensional drought event can be structured and stored in the database to support subsequent queries, retrieval, and statistical analysis.

[0268] In one possible implementation, drought evolution visualization information can be generated based on three-dimensional drought events. For example, computer graphics methods can be used to construct isosurface models of drought events based on spatiotemporal voxel data of three-dimensional drought events, and then rendered and displayed in a three-dimensional geographic information system to intuitively demonstrate the occurrence, development, migration, and disappearance of drought events in space and time.

[0269] In one possible implementation, a drought monitoring report can be generated based on the three-dimensional drought event identification results within the current monitoring period. Specifically, according to a preset report template, key indicators such as the number of newly added drought events this month, the current location of the largest drought center, duration, affected area, severity, and risk areas can be automatically extracted, and a drought monitoring briefing containing statistical charts and textual descriptions can be generated.

[0270] In one possible embodiment, drought early warning decision information can also be output through an application programming interface (API) or a human-computer interaction interface (HCI). The drought early warning decision information may include a list of currently developing drought events, predictions of possible future migration directions, statistics on affected administrative regions, and quantitative decision-making basis related to water resource allocation and drought relief.

[0271] In one possible embodiment, a feedback optimization mechanism for the identification results can also be established. Specifically, the three-dimensional drought event identification results can be compared and corrected based on historical typical drought event records, and genetic algorithms, Bayesian optimization methods, or other parameter optimization methods can be used to fine-tune the neighborhood search range, core point determination conditions, and overlap rate thresholds in the spatiotemporal connectivity rules during the spatial clustering process, so that the three-dimensional drought event identification results are more adapted to the drought evolution characteristics of the target area.

[0272] Based on the obtained three-dimensional drought events and their characterization information, the embodiments of this application can further form drought event types, drought risk areas and application output information, so that the three-dimensional drought event identification results can be used for event classification, risk assessment, visualization, monitoring report generation and early warning decision support.

[0273] Compared with related technologies, the embodiments of this application have at least the following beneficial effects: This application's embodiments improve the accuracy of drought patch identification under complex spatial morphologies. Related technologies typically rely on fixed connectivity rules or regular geometric shapes to identify drought regions, which are easily affected by factors such as fragmented drought region boundaries, irregular spatial morphology, and multi-center distribution, leading to incorrect segmentation or merging of drought patches. This application's embodiments identify drought patches corresponding to each time step based on the spatial aggregation relationships between drought grid points. It can determine patch affiliation according to the spatial proximity and aggregation degree of drought grid points, making drought patch identification no longer solely dependent on fixed geometric connectivity relationships. This allows for better adaptation to strip-shaped, ring-shaped, edge-fragmented, or loosely aggregated drought spatial morphologies, improving the accuracy of drought impact range identification.

[0274] This application's embodiments improve the completeness of tracking the spatiotemporal evolution of drought events. Related technologies, when determining whether drought patches at adjacent time steps belong to the same event, typically rely solely on the single area overlap ratio, which is insufficient to accurately handle the complex evolutionary situations such as splitting, merging, expansion, and decline that occur during the development of drought events. This application's embodiments determine bidirectional correlation features through drought patch label data from adjacent time steps and combine this with preset dynamic evolution rules to determine spatiotemporal relationships. This can simultaneously characterize the degree of continuation of drought patches from the previous time step to the next time step, as well as the degree to which drought patches in the next time step originate from drought patches in the previous time step. Therefore, even when drought events undergo structural splitting or local merging, the event attribution relationship between patches can still be determined more accurately, reducing the possibility of long-duration drought events being incorrectly segmented into multiple short events, and improving the continuity and accuracy of drought event lifecycle identification.

[0275] The embodiments of this application enhance the noise resistance and physical saliency of drought identification results. Drought-related grid data may contain scattered anomalous grid points, local short-term fluctuations, or isolated drought pixels. Direct identification based on a single threshold or simple connectivity relationships can easily lead to false drought events or incorrectly connect originally independent drought regions through noisy grid points. In the drought patch identification process, the embodiments of this application can distinguish core drought grid points, boundary drought grid points, and noisy grid points based on neighborhood clustering relationships. Furthermore, during the construction of 3D drought events, spatiotemporal voxel scale is used for filtering, thereby eliminating perturbation results with small spatial scale, short lifecycles, or lacking significant regional influence. This reduces the impact of random climate noise and local data fluctuations on drought patch identification, centroid location, and intensity calculation, making the final identified 3D drought events more regionally representative and physically meaningful.

[0276] This application's embodiments enable refined characterization of the three-dimensional spatiotemporal features of drought events. By merging correlated two-dimensional drought patches across multiple time steps into a three-dimensional drought event, this application's embodiments can perform a holistic analysis of the drought process at the event scale, rather than being limited to the state judgment of a single time slice or grid point. Furthermore, based on the spatiotemporal distribution information of the three-dimensional drought event, event characterization information such as temporal dimension features, spatial coverage features, intensity features, evolution path features, and morphological features can be determined. Thus, it is possible not only to identify whether a drought event has occurred, but also to quantify the duration, impact range, severity, migration path, migration speed, and morphological changes of the drought event, providing more complete data support for drought risk zoning, drought evolution mechanism analysis, monitoring report generation, and drought relief and disaster reduction decision-making.

[0277] In summary, the embodiments of this application, through drought state data determination, single-time-step drought patch identification, bidirectional correlation feature calculation between adjacent time steps, dynamic evolution rule determination, and three-dimensional drought event merging and characterization, can improve the accuracy, stability, and completeness of the three-dimensional identification results of regional drought events under complex spatial morphology, data noise interference, and complex spatiotemporal evolution relationships.

[0278] Figure 3 This is a schematic diagram of the structure of a three-dimensional recognition system for regional drought events provided in an embodiment of this application. Figure 3 As shown, the regional drought event three-dimensional recognition system 300 includes a drought state determination module 301, a drought patch recognition module 302, a correlation feature determination module 303, a spatiotemporal correlation determination module 304, a three-dimensional event generation module 305, and an event characterization determination module 306.

[0279] The drought status determination module 301 is used to determine drought status data corresponding to multiple time steps based on drought-related grid data of the target area within a preset time period. The multiple time steps are multiple time units obtained by dividing the preset time period according to a preset time scale.

[0280] The drought patch identification module 302 is used to identify drought patches corresponding to each time step based on the spatial clustering relationship between drought grid points in the drought state data corresponding to the time step, and generate drought patch label data corresponding to the time step.

[0281] The association feature determination module 303 is used to determine the bidirectional association features between drought patches in adjacent time steps based on the drought patch label data corresponding to adjacent time steps.

[0282] The spatiotemporal correlation determination module 304 is used to determine the spatiotemporal correlation between drought patches at adjacent time steps based on the bidirectional correlation features and the preset dynamic evolution rules. The preset dynamic evolution rules are used to characterize the continuity and branching evolution relationships of drought patches at adjacent time steps.

[0283] The three-dimensional event generation module 305 is used to merge drought patches in different time steps based on the spatiotemporal correlations corresponding to multiple adjacent time steps to obtain the three-dimensional drought event corresponding to the target region.

[0284] The event characterization determination module 306 is used to determine the event characterization information corresponding to the three-dimensional drought event based on the spatiotemporal distribution information of the three-dimensional drought event.

[0285] It should be noted that the specific process of each module in the regional drought event 3D recognition system executing the above method has been described in detail in the above embodiments, and this embodiment does not make specific limitations on it.

[0286] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 4 As shown, the electronic device 400 provided in this embodiment includes a memory 401 and a processor 402.

[0287] The memory 401 can be a separate physical unit, connected to the processor 402 via a bus 403. Alternatively, the memory 401 and processor 402 can be integrated and implemented in hardware. The memory 401 stores program instructions, which the processor 402 calls to execute the operations performed by the regional drought event three-dimensional recognition system in any of the above method embodiments.

[0288] Optionally, when some or all of the methods in the above embodiments are implemented by software, the electronic device 400 may also include only the processor 402. A memory 401 for storing programs is located outside the electronic device 400, and the processor 402 is connected to the memory via circuits / wires to read and execute the programs stored in the memory. The processor 402 may be a central processing unit (CPU), a network processor (NP), or a combination of a CPU and an NP. The processor 402 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof.

[0289] Memory 401 may include volatile memory, such as random-access memory (RAM); memory may also include non-volatile memory, such as flash memory, hard disk drive (HDD) or solid-state drive (SSD); memory may also include combinations of the above types of memory.

[0290] For example, this application provides a chip including: an interface circuit and a logic circuit. The interface circuit is used to receive signals from other chips outside the chip and transmit them to the logic circuit, or to send signals from the logic circuit to other chips outside the chip. The logic circuit is used to perform the operations performed by the regional drought event three-dimensional recognition system in the above method embodiments.

[0291] For example, this application provides a computer-readable storage medium storing computer program instructions thereon, which are executed by the processor of an electronic device to cause the electronic device to perform the operations performed by the regional drought event three-dimensional recognition system in the above method embodiments.

[0292] For example, this application provides a computer program product that, when run on an electronic device, causes the electronic device to perform the operations performed by the regional drought event three-dimensional recognition system in the above method embodiments.

[0293] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments described herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A three-dimensional identification method for regional drought events, characterized in that, The method includes: Based on drought-related grid data of the target area within a preset time period, drought status data corresponding to multiple time steps are determined. The multiple time steps are multiple time units obtained by dividing the preset time period according to a preset time scale. For each time step, based on the spatial clustering relationship between drought grid points in the drought state data corresponding to the time step, drought patches corresponding to the time step are identified, and drought patch label data corresponding to the time step is generated; Based on the drought patch label data corresponding to adjacent time steps, determine the bidirectional association features between drought patches in adjacent time steps; Based on the bidirectional correlation features and the preset dynamic evolution rules, the spatiotemporal correlation between drought patches at adjacent time steps is determined. The preset dynamic evolution rules are used to characterize the continuity and branching evolution of drought patches at adjacent time steps. Based on the spatiotemporal correlations corresponding to multiple adjacent time steps, drought patches in different time steps are merged to obtain the three-dimensional drought event corresponding to the target region. Based on the spatiotemporal distribution information of the three-dimensional drought event, the event characterization information corresponding to the three-dimensional drought event is determined; The step of determining the bidirectional association features between drought patches at adjacent time steps based on the drought patch label data corresponding to adjacent time steps includes: Based on the drought patch label data corresponding to adjacent time steps, determine the spatial overlap information between the drought patches in the previous time step and the drought patches in the next time step. Based on the spatial overlap information and the patch area of ​​the drought patch in the previous time step, a first association feature is determined. The first association feature is used to characterize the degree to which the drought patch in the previous time step continues to the drought patch in the subsequent time step. Based on the spatial overlap information and the patch area of ​​the drought patch in the next time step, a second correlation feature is determined. The second correlation feature is used to characterize the degree to which the drought patch in the next time step originates from the drought patch in the previous time step. Based on the first association feature and the second association feature, the bidirectional association feature is obtained; The step of determining the spatiotemporal correlation between drought patches at adjacent time steps based on the bidirectional correlation characteristics and preset dynamic evolution rules includes: Based on the first associated feature and the preset strong inheritance conditions, the continuation relationship is determined, and the strong inheritance determination result is obtained. When the strong inheritance determination result indicates that there is no strong inheritance relationship between drought patches in adjacent time steps, the branch relationship is determined according to the first association feature, the second association feature and the preset branch continuation condition to obtain the branch continuation determination result. Based on the strong inheritance determination result, or, if the branch continuation determination result is obtained, based on the branch continuation determination result, it is determined that drought patches in adjacent time steps have a spatiotemporal correlation relationship for merging into the same three-dimensional drought event; Specifically, if the strong inheritance determination result indicates that there is a strong inheritance relationship between drought patches in adjacent time steps, or if the branch continuation determination result is obtained and the branch continuation determination result indicates that there is a branch continuation relationship between drought patches in adjacent time steps, then it is determined that drought patches in adjacent time steps have a spatiotemporal correlation relationship for merging into the same three-dimensional drought event.

2. The method according to claim 1, characterized in that, The step of determining drought status data corresponding to multiple time steps based on drought-related grid data of the target area within a preset time period includes: Obtain the drought-related grid data of the target area within the preset time period; Outlier detection and imputation are performed on the drought-related grid data to obtain corrected grid data; The modified grid data is time-aggregated according to the preset time scale to obtain scaled grid data corresponding to multiple time steps. A probability distribution model is constructed based on historical grid data of the target area; Based on the probability distribution model, the scaled grid data is subjected to probability transformation to obtain standardized drought index data corresponding to multiple time steps. Based on the standardized drought index data and the preset drought threshold, the drought status data corresponding to multiple time steps are determined.

3. The method according to claim 1, characterized in that, The process of identifying drought patches corresponding to a given time step based on the spatial clustering relationships between drought grid points in the drought state data corresponding to that time step, and generating drought patch label data corresponding to that time step, includes: Obtain the clustering parameters corresponding to the time step, the clustering parameters including the neighborhood search range and the core point determination conditions; Extract the grid coordinates of grid points in a drought state from the drought state data corresponding to the time step to obtain a set of drought grid coordinates; Based on the spatial distance between each drought grid point in the drought grid point coordinate set and the neighborhood search range, determine the neighborhood grid point set corresponding to each drought grid point; Based on the neighborhood grid set corresponding to each drought grid point and the core point determination condition, the drought grid points are labeled with types to obtain core drought grid points, boundary drought grid points and noise grid points; Starting with unassigned patch identifiers, core drought grid points are clustered and assigned cluster identifiers based on the density reachability relationships between different core drought grid points and between the core drought grid points and the boundary drought grid points, thus obtaining candidate drought patches. Based on the candidate drought patches and the noise grid points, drought patch label data corresponding to the time step is generated.

4. The method according to claim 3, characterized in that, The step of generating drought patch label data corresponding to the time step based on the candidate drought patches and the noise grid points includes: The noisy grid points are removed from the drought grid points indicated by the drought state data corresponding to the time step to obtain the denoised drought state data; Based on the spatial location and grid area of ​​each drought grid point in the candidate drought patch, the patch morphology parameters corresponding to the candidate drought patch are determined; wherein, the patch morphology parameters include the patch area, or the patch morphology parameters include the patch area and the patch centroid; Based on the patch morphology parameters and preset patch screening conditions, target drought patches are determined from the candidate drought patches; wherein, the preset patch screening conditions include the patch area being greater than or equal to a preset area threshold; Based on the patch affiliation relationship of each drought grid point in the target drought patch, the denoised drought state data is labeled to obtain the drought patch label data corresponding to the time step.

5. The method according to claim 1, characterized in that, The process of merging drought patches in different time steps based on the spatiotemporal correlations of multiple adjacent time steps to obtain the three-dimensional drought event corresponding to the target region includes: Based on the drought patch label data corresponding to multiple time steps, patch node identifiers are assigned to drought patches in different time steps; Based on the spatiotemporal correlations corresponding to multiple adjacent time steps, the patch node identifiers with correlations are subjected to equivalent merging processing to obtain multiple spatiotemporal equivalence classes; Generate a global event identifier based on the multiple spatiotemporal equivalence classes; Based on the global event identifier, drought patches in different time steps are re-labeled to obtain the initial three-dimensional drought event; Based on the spatiotemporal voxel scale and preset spatiotemporal voxel threshold of each event in the initial three-dimensional drought event, the initial three-dimensional drought event is filtered to obtain the three-dimensional drought event corresponding to the target region.

6. The method according to claim 5, characterized in that, The determination of event characterization information corresponding to the three-dimensional drought event based on the spatiotemporal distribution information of the three-dimensional drought event includes: Based on the distribution range of the three-dimensional drought event in the time dimension, determine the time dimension characteristics corresponding to the three-dimensional drought event; Based on the spatial coverage of the three-dimensional drought event, determine the spatial coverage characteristics corresponding to the three-dimensional drought event; Based on the standardized drought index data corresponding to the three-dimensional drought event and the spatial cover characteristics, the intensity characteristics corresponding to the three-dimensional drought event are determined; Based on the spatial location changes of the three-dimensional drought event at different time steps, the evolution path characteristics corresponding to the three-dimensional drought event are determined; Based on the patch morphology changes of the three-dimensional drought event at different time steps, the morphological features corresponding to the three-dimensional drought event are determined; The event representation information is generated based on at least one of the time dimension feature, the spatial coverage feature, the intensity feature, the evolution path feature, and the morphological feature.

7. The method according to claim 6, characterized in that, The method further includes: Based on the event characterization information, the event type of the three-dimensional drought event is determined to obtain drought event type information; Based on the spatiotemporal distribution information of the three-dimensional drought event within the target area and the event characterization information, drought risk area information is determined; Based on at least one of the three-dimensional drought event, the drought event type information, and the drought risk area information, generate drought event application output information; The drought event application output information includes at least one of the following: drought event database records, drought evolution visualization information, drought monitoring reports, and drought early warning decision information.

8. A three-dimensional recognition system for regional drought events, characterized in that, The system includes: The drought status determination module is used to determine the drought status data corresponding to multiple time steps based on the drought-related grid data of the target area within a preset time period. The multiple time steps are multiple time units obtained by dividing the preset time period according to a preset time scale. A drought patch identification module is used to identify drought patches corresponding to each time step based on the spatial clustering relationship between drought grid points in the drought state data corresponding to the time step, and generate drought patch label data corresponding to the time step. The association feature determination module is used to determine the bidirectional association features between drought patches in adjacent time steps based on the drought patch label data corresponding to adjacent time steps. The spatiotemporal correlation determination module is used to determine the spatiotemporal correlation between drought patches at adjacent time steps based on the bidirectional correlation features and the preset dynamic evolution rules. The preset dynamic evolution rules are used to characterize the continuity and branching evolution relationships of drought patches at adjacent time steps. The three-dimensional event generation module is used to merge drought patches in different time steps based on the spatiotemporal correlations corresponding to multiple adjacent time steps to obtain three-dimensional drought events corresponding to the target region. The event characterization determination module is used to determine the event characterization information corresponding to the three-dimensional drought event based on the spatiotemporal distribution information of the three-dimensional drought event; The association feature determination module is specifically used for: determining spatial overlap information between drought patches in the previous time step and drought patches in the next time step based on the drought patch label data corresponding to adjacent time steps; determining a first association feature based on the spatial overlap information and the patch area of ​​the drought patch in the previous time step, the first association feature being used to characterize the degree to which the drought patch in the previous time step extends to the drought patch in the next time step; determining a second association feature based on the spatial overlap information and the patch area of ​​the drought patch in the next time step, the second association feature being used to characterize the degree to which the drought patch in the next time step originates from the drought patch in the previous time step; and obtaining the bidirectional association feature based on the first association feature and the second association feature. The spatiotemporal correlation determination module is specifically used for: determining a continuation relationship based on the first correlation feature and a preset strong inheritance condition to obtain a strong inheritance determination result; when the strong inheritance determination result indicates that there is no strong inheritance relationship between drought patches in adjacent time steps, determining a branch relationship based on the first correlation feature, the second correlation feature, and a preset branch continuation condition to obtain a branch continuation determination result; determining that drought patches in adjacent time steps have a spatiotemporal correlation relationship for merging into the same three-dimensional drought event based on the strong inheritance determination result, or, when the branch continuation determination result is obtained, determining that drought patches in adjacent time steps have a spatiotemporal correlation relationship for merging into the same three-dimensional drought event. Specifically, when the strong inheritance determination result indicates that there is a strong inheritance relationship between drought patches in adjacent time steps, or when the branch continuation determination result is obtained and the branch continuation determination result indicates that there is a branch continuation relationship between drought patches in adjacent time steps, determining that drought patches in adjacent time steps have a spatiotemporal correlation relationship for merging into the same three-dimensional drought event.

Citation Information

Patent Citations

  • Physiological health characteristic data monitoring method, device and equipment and storage medium

    CN112037888A

  • High-temperature disaster forecasting method based on directed graph neural network

    CN114626512A