Big data-based disaster chain hidden danger identification method in snow melting and temperature rising process

Abstract

The invention discloses a snow melting and temperature rising process disaster chain hidden danger recognition method based on big data. The method comprises the following steps that remote sensing snow covers, radar wet snow lines, daily temperature, flow and ground surface deformation data of a research area are obtained and preprocessed; an improved Bayesian online change point detection algorithm is adopted to determine a real-time node triggered by snow melting temperature rise; calculating a snow cover area change rate and a wet snow line change rate to identify a snow melting sensitive area; analyzing the surface deformation data to determine a surface acceleration sensitive area; analyzing the flow data to determine a runoff mutation sensitive area; constructing a disaster chain hidden danger transmission space-time structure model of snow melting triggering, surface acceleration and runoff response; and identifying a disaster chain hidden danger level in real time based on a space-time structure model. According to the invention, real-time and quantitative recognition of regional scale disaster chain hidden dangers is realized, and the accuracy and timeliness of disaster early warning are improved.

Description

Technical Field

[0001] This invention relates to the field of geological disaster risk identification and early warning technology, and in particular to a method for identifying potential hazards in the disaster chain during snowmelt and warming processes based on big data. Background Technology

[0002] In recent years, global climate change has led to frequent regional extreme weather events, among which the disaster chain reaction triggered by spring snowmelt and warming poses a serious threat to the ecological environment and socio-economic development of mountainous areas. To effectively manage such disasters, it is necessary to conduct precise hazard identification and risk assessment of the accelerated surface deformation and rapid runoff response processes induced by snowmelt and warming.

[0003] Existing technologies typically analyze the abnormal trends of snowmelt warming or a specific stage using methods such as remote sensing, radar measurement, hydrological observation, and surface deformation monitoring. They then use static thresholding or statistical methods to initially delineate risk areas. These methods generally process various single-source data separately and then identify risk areas through simple spatial overlay analysis. However, they neglect the potential spatiotemporal dependencies between different stages of the disaster chain and fail to systematically reveal the triggering mechanisms of snowmelt warming on subsequent surface instability and runoff mutations. Consequently, they struggle to achieve accurate and real-time identification of potential hazards within the disaster chain.

[0004] Furthermore, most currently widely used disaster chain hazard identification methods rely on static or manually set thresholds to determine hazard levels. This makes it difficult to effectively guarantee the real-time nature and reliability of disaster early warnings, particularly in terms of efficient real-time risk assessment at the regional scale. Therefore, it is crucial to fully utilize multi-source long-term data to systematically analyze the triggering and transmission patterns between snowmelt warming processes and surface deformation and runoff response, and to establish analytical methods capable of real-time, dynamic hazard level identification. This is essential for improving disaster early warning capabilities and risk prevention efficiency.

[0005] Therefore, how to provide a method for identifying potential hazards in the disaster chain of the snow melting and warming process based on big data is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a method for identifying potential hazards in the snowmelt warming process based on big data. Addressing the challenge of existing technologies failing to achieve real-time and quantitative identification of potential hazards in the snowmelt warming process under a unified spatiotemporal benchmark, this invention proposes a method utilizing an improved Bayesian online variable point detection algorithm to systematically identify sensitive areas and real-time triggering nodes for snowmelt warming, accelerated surface deformation, and rapid runoff response. It then constructs a spatiotemporal structural model of hazard transmission in the hazard chain, enabling real-time, quantitative, and accurate identification of regional-scale hazard chains. This invention offers advantages such as high accuracy and real-time performance in disaster early warning.

[0007] A method for identifying potential hazards in a snowmelt and warming process based on big data, according to an embodiment of the present invention, includes:

[0008] Long-term remote sensing snow cover data, radar wet snow line observation data, temperature data, flow data from hydrological monitoring stations, and surface deformation data from synthetic aperture radar interferometry were acquired within the study area and preprocessed to obtain the initial input dataset.

[0009] Based on the temperature data and radar wet snow line observation data in the initial input dataset, an improved Bayesian online change point detection algorithm is used to determine the real-time trigger time point sequence.

[0010] Based on the real-time trigger time sequence, long-term remote sensing snow cover data and radar wet snow line observation data before and after each trigger time point are extracted and analyzed to obtain a spatial distribution map of the sensitive area of ​​the snow melting and warming process.

[0011] An improved Bayesian online change point detection algorithm was used to perform online analysis on surface deformation data to obtain a spatial distribution map of sensitive areas for accelerated surface deformation.

[0012] An improved Bayesian online change point detection algorithm was used to analyze the change point characteristics in flow data and obtain a spatial distribution map of runoff change-sensitive areas.

[0013] Using spatial distribution maps of sensitive areas during snowmelt warming, accelerated surface deformation, and runoff mutation, a chain-like disaster chain hazard transmission path map is constructed to obtain a spatiotemporal structure model of disaster chain hazard transmission.

[0014] Based on the spatiotemporal structure model of disaster chain hazard transmission, the disaster chain hazard level of each spatial region under snow melting and warming conditions is determined, and the disaster chain hazard during the snow melting and warming process is quantitatively identified.

[0015] Optionally, the preprocessing includes cloud removal, noise reduction, and interpolation processing for long-term remote sensing snow cover data; noise removal and smoothing filtering processing for radar wet snow line observation data; outlier detection and correction processing for temperature data; missing data interpolation and outlier data removal processing for flow data from hydrological monitoring stations; and phase unwrapping, noise reduction, and smoothing processing for surface deformation data from synthetic aperture radar interferometry.

[0016] Optionally, the step of determining the real-time trigger time point sequence based on the temperature data and radar wet snow line observation data in the initial input dataset using an improved Bayesian online change point detection algorithm is as follows:

[0017] The temperature data and radar wet snow line observation data of each study area in the initial input dataset are obtained and simultaneously input into the improved Bayesian online change point detection algorithm.

[0018] In the improved Bayesian online change point detection algorithm, the prior probability at the current time is dynamically adjusted based on the posterior probability of the previous time step, and the Bayesian change point probability is adaptively updated.

[0019] Calculate the joint change probability of temperature data and wet snow line observation data in each study area, and use the probability threshold to determine whether the joint change probability meets the triggering condition;

[0020] For the real-time triggering time point determined when the probability of the joint change point first exceeds the probability threshold, a dual dynamic sliding window mechanism is used to verify its effectiveness. The first sliding window is used to verify whether the probability of the change point after the real-time triggering time point continues to exceed the probability threshold for a preset duration. The second sliding window is used to verify whether the changing trends of the temperature data and the wet snow line observation data before and after the real-time triggering time point show statistically stable differences.

[0021] The cumulative probability increment is calculated for the real-time trigger time points that have passed the verification of the dual dynamic sliding window mechanism. The real-time trigger time points that reach the preset cumulative probability increment are selected to obtain the real-time trigger time point sequence of the start of the snow melting and warming process.

[0022] Optionally, based on the real-time trigger time point sequence, long-term remote sensing snow cover data and radar wet snow line observation data before and after each trigger time point are extracted and analyzed to obtain a spatial distribution map of the sensitive area of ​​the snowmelt and warming process, specifically:

[0023] Based on the real-time trigger time sequence, long-term remote sensing snow cover data and radar wet snow line observation data within a preset time period before and after each real-time trigger time are extracted.

[0024] Calculate the change in snow cover area per unit time in long-term remote sensing snow cover data before and after each real-time triggering time point, and calculate the change in height per unit time in radar wet snow line observation data before and after each real-time triggering time point;

[0025] The spatial grid division method is adopted to divide the study area into multiple regular grid cells with consistent spatial resolution. The rate of change of snow cover area and the rate of change of wet snow line of each grid cell are calculated based on the change of snow cover area per unit time and the change of height per unit time, respectively.

[0026] Based on the preset sensitive spatial region identification threshold, spatial grid cells that exceed the preset threshold for the rate of change of snow cover area and the preset threshold for the rate of change of wet snow line per unit time are identified respectively.

[0027] The identified spatial grid cells are subjected to spatial clustering to obtain their spatial location range;

[0028] Spatial overlay analysis was performed on the spatial location range to identify the spatial intersection area between the snow cover melting acceleration area and the wet snow line rising rate acceleration area, and to determine the spatial area sensitive to the snow melting and warming process.

[0029] Based on the sensitive spatial areas of the snowmelt warming process, draw a spatial distribution map of the sensitive areas of the snowmelt warming process.

[0030] Optionally, the improved Bayesian online change point detection algorithm is used to perform online analysis of surface deformation data to obtain a spatial distribution map of sensitive areas for accelerated surface deformation, specifically:

[0031] Using the spatial distribution map of sensitive areas during the snowmelt warming process, the spatial location range of sensitive areas during the snowmelt warming process is determined, and surface deformation data at each observation time within the spatial location range is extracted;

[0032] The improved Bayesian online change point detection algorithm is used to calculate the change point probability of surface deformation data at each observation time, and a dynamic probability threshold judgment is made on the change point probability at each observation time to identify candidate real-time starting nodes with abnormally increased surface deformation rate.

[0033] For the identified candidate real-time starting nodes, a dynamic sliding time window is set. The sliding time window continuously observes the changes in the surface deformation rate at multiple time points and calculates the cumulative change point probability increment at each time point within the window.

[0034] The validity of the candidate real-time starting node is determined by whether the cumulative change point probability increment at each moment within the sliding time window exceeds the preset statistical threshold, and the candidate real-time starting nodes that meet the statistical conditions are retained to obtain the valid real-time starting nodes.

[0035] Based on the retained effective real-time starting nodes, spatial grid cells with significantly increased surface deformation rates are identified through spatial grid division methods. Spatial cluster analysis is then performed on the identified spatial grid cells to obtain the spatial location range of the sensitive areas of accelerated surface deformation, and a spatial distribution map of the sensitive areas of accelerated surface deformation is drawn.

[0036] Optionally, the improved Bayesian online change point detection algorithm is used to analyze the change point characteristics in the flow data to obtain a spatial distribution map of runoff change-sensitive areas, specifically as follows:

[0037] Using the spatial distribution map of areas sensitive to accelerated surface deformation, the spatial location range of these areas is determined, and flow data from hydrological monitoring stations within this spatial range is extracted.

[0038] For the extracted traffic data, an improved Bayesian online change point detection algorithm is used to calculate the change point probability of traffic data at each observation time. The probability threshold is dynamically updated to determine whether the change point probability of traffic data exceeds the preset threshold, and to identify candidate sensitive periods when traffic occurs rapidly.

[0039] For the candidate sensitive period, set up a dynamic verification time window that is symmetrical before and after. Calculate the statistical mean and variance of the traffic data in the two time windows before and after the candidate sensitive period, and compare whether the difference in the statistical characteristic values ​​in the two windows meets the preset statistical conditions.

[0040] Candidate sensitive periods that meet the preset statistical conditions are retained as effective sensitive periods for rapid runoff response. The spatial location range of the river section corresponding to the rapid runoff response is determined by combining the spatial coordinates of each hydrological monitoring station within the spatial location range.

[0041] The spatial grid division method is adopted to divide the determined spatial location range of the river section into multiple spatial grid units, and the flow change rate in each spatial grid unit is calculated to filter out spatial grid units whose flow change rate exceeds a preset threshold.

[0042] Spatial clustering analysis is performed on the selected spatial grid cells whose flow change rate exceeds a preset threshold to determine the spatial location range of each spatial cluster region;

[0043] Based on the spatial location range of the runoff change sensitivity area, draw a spatial distribution map of the runoff change sensitivity area.

[0044] Optionally, by utilizing the spatial distribution maps of sensitive areas during snowmelt warming, accelerated surface deformation, and runoff mutation, a chain-like disaster chain hazard transmission path map is constructed to obtain a spatiotemporal structural model of disaster chain hazard transmission, specifically:

[0045] Extract the spatial location range and corresponding start trigger time of each sensitive area from the spatial distribution maps of sensitive areas for snowmelt warming process, sensitive areas for accelerated surface deformation, and sensitive areas for runoff change.

[0046] Based on the spatial location range of the sensitive areas of snowmelt warming process and the spatial location range of the sensitive areas of accelerated surface deformation, the spatial topological relationship analysis method is used to calculate the spatial intersection area between the two sensitive areas, and to identify the spatial triggering relationship of snowmelt warming process on accelerated surface instability.

[0047] Based on the spatial location range of the sensitive areas of accelerated surface deformation and the spatial location range of the sensitive areas of sudden runoff change, the spatial topological relationship analysis method is used to calculate the spatial intersection area between the two sensitive areas, and to identify the spatial triggering relationship of accelerated surface instability to rapid runoff response.

[0048] The temporal dependency between the sensitive areas is determined by calculating the initial triggering time interval of the sensitive areas for accelerated surface deformation triggered by the snowmelt warming process, and the initial triggering time interval of the sensitive areas for accelerated surface deformation triggered by the sensitive areas for sudden runoff.

[0049] Based on the established spatial triggering and temporal dependencies, a chain disaster hazard transmission path map is constructed, which involves snowmelt warming triggering, accelerated surface instability, and rapid runoff response.

[0050] Based on the chain-like disaster chain hazard transmission path map, the spatiotemporal dependencies between each link are determined by spatial fusion and temporal sequence registration of the spatial location range and trigger time information of all sensitive areas;

[0051] Based on the established spatiotemporal dependencies, a spatiotemporal structural model of the transmission of potential hazards in the disaster chain is constructed.

[0052] Optionally, based on the determined spatiotemporal dependencies, a spatiotemporal structural model for the transmission of disaster chain hazards is constructed, specifically as follows:

[0053] Based on the established chain disaster chain hazard transmission path map, the spatial location range and triggering time information of each node in the sensitive areas of snow melting and warming process, the sensitive areas of accelerated surface deformation, and the sensitive areas of runoff change in the disaster chain path map are extracted;

[0054] The extracted spatial location ranges of each node are uniformly projected onto the same spatial coordinate reference system, and the spatial location range data are standardized to eliminate the differences in spatial scale between different regions.

[0055] Using the triggering time of the sensitive area of ​​snowmelt and warming process as the starting reference time, the triggering time of the sensitive area of ​​accelerated surface deformation and sensitive area of ​​runoff change are time-registered with the starting reference time, and the standardized relative time difference between each node is calculated and determined.

[0056] Based on the standardized relative time difference, the time sequence relationship is used as the time dimension basis of the model, and the normalized spatial location range of each node is used as the spatial dimension basis of the model, thus constructing a spatiotemporal three-dimensional structure containing information in both time and space dimensions.

[0057] The spatial distribution characteristics of each node's spatial location range are represented by the spatial dimension, and the trigger sequence and time interval between nodes are represented by the temporal dimension, thus constructing a spatiotemporal structure model for the transmission of potential hazards in the disaster chain.

[0058] Optionally, the step of determining the hazard level of each spatial region under snowmelt and warming conditions based on the spatiotemporal structure model of hazard transmission in the hazard chain, and quantitatively identifying the hazard in the hazard chain during the snowmelt and warming process, specifically involves:

[0059] Based on the spatiotemporal structure model of disaster chain hazard transmission, the spatial location range of sensitive areas during snowmelt and warming processes, sensitive areas for accelerated surface deformation, and sensitive areas for sudden runoff changes, along with the corresponding spatiotemporal characteristic parameters of disaster chain hazard transmission, are extracted.

[0060] For each spatial region, the probability of snowmelt warming, the probability of accelerated surface deformation, and the probability of sudden runoff are calculated separately. Based on the preset weighting factors of each disaster link, the triggering probability, acceleration probability, and sudden runoff probability of each spatial region are weighted and fused for calculation.

[0061] Based on the weighted fusion calculation results, the comprehensive risk index of disaster chain hazards in each spatial region is determined, and the comprehensive risk index is divided into multiple disaster chain hazard levels according to the preset disaster chain hazard level thresholds;

[0062] For the identified hazard levels of the disaster chain, a mapping relationship between the hazard level of the disaster chain and the degree of disaster risk is established, and the hazard level of the disaster chain corresponding to each spatial region under the conditions of snow melting and warming is marked;

[0063] Based on the marked hazard level of the disaster chain, calculate the area of ​​spatial risk range and the number of affected spatial grid units for each spatial region under the hazard level of the disaster chain.

[0064] By using spatial interpolation and rasterization analysis, the hazard level, risk range area, and number of affected spatial grid cells calculated for each spatial region are spatially fused to generate quantitative identification results.

[0065] Optionally, the improved Bayesian online change point detection algorithm is specifically as follows:

[0066] The initial prior distribution and its probability threshold parameters are set, and the Bayesian online change point detection algorithm is initialized by setting the mean and variance parameters of the initial prior distribution.

[0067] The observation data is acquired time by time, the prior probability of the current time is dynamically adjusted using the posterior probability of the previous time, and the change point posterior probability of the current time is calculated based on Bayes' theorem.

[0068] Based on the posterior probability of the change point at the current time, the probability threshold at the current time is dynamically updated, and whether the current time is a candidate change point is determined based on whether the posterior probability of the current change point exceeds the probability threshold.

[0069] When a candidate change point occurs, a dual dynamic sliding window is constructed to verify the validity of the candidate change point. The first sliding window is used to verify whether the posterior probability of the change point at multiple consecutive time points after the candidate change point continuously exceeds the dynamic probability threshold. The second sliding window is used to verify whether there are statistically significant differences in the statistical characteristics of the observed data within the window before and after the candidate change point.

[0070] After a candidate change point passes the validity verification of a dual dynamic sliding window, the cumulative probability increment of the candidate change point and its subsequent verification window is calculated. It is then determined whether the cumulative probability increment reaches a preset threshold and whether the candidate change point is a true and valid change point.

[0071] The beneficial effects of this invention are:

[0072] (1) This invention adopts an improved Bayesian online change point detection algorithm and combines multi-source long-term remote sensing, radar wet snow line, air temperature, runoff flow and surface deformation data fusion analysis to achieve accurate identification of sensitive areas in the snowmelt and warming process and real-time determination of disaster chain triggering nodes. This effectively improves the accuracy and efficiency of real-time identification of disaster hazards and enhances the timeliness and reliability of regional-scale disaster chain hazard identification and early warning.

[0073] (2) Through specific technical measures such as spatial topology analysis, spatial gridded clustering analysis and dynamic sliding window verification, this invention realizes the spatiotemporal structured modeling of disaster chain hazards, significantly improves the identification accuracy and stability of disaster chain hazard transmission paths, and shows better adaptability and practical effect in the risk management of snow melting and warming disasters in complex mountainous areas.

[0074] (3) In terms of real-time quantitative identification of disaster chain hazards during snow melting and warming process, this invention effectively solves the problem of lack of real-time comprehensive quantitative analysis of multi-link disaster chain hazards in the existing technology by real-time weighted fusion calculation of disaster chain hazard levels and spatial gridded analysis, breaks through the bottleneck of dynamic judgment of disaster hazard levels, realizes specific and significant improvement in the ability of refined disaster risk assessment and early warning, and effectively enhances the decision support capability of regional disaster prevention and control. Attached Figure Description

[0075] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0076] Figure 1 This is a flowchart of a method for identifying potential hazards in the disaster chain during snow melting and warming processes based on big data, as proposed in this invention. Detailed Implementation

[0077] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0078] refer to Figure 1 A method for identifying potential hazards in the disaster chain of snowmelt and warming processes based on big data, comprising:

[0079] Long-term remote sensing snow cover data, radar wet snow line observation data, temperature data, flow data from hydrological monitoring stations, and surface deformation data from synthetic aperture radar interferometry were acquired within the study area and preprocessed to obtain the initial input dataset.

[0080] Based on the temperature data and radar wet snow line observation data in the initial input dataset, an improved Bayesian online change point detection algorithm is used to calculate the change point probability of temperature and wet snow line observation data in each study area, and to determine the real-time trigger time point sequence of the start of the snow melting and warming process.

[0081] Based on the real-time trigger time sequence, long-term remote sensing snow cover data and radar wet snow line observation data before and after each trigger time point are extracted and analyzed. The snow cover area change rate and wet snow line change rate are calculated respectively to determine the sensitive spatial areas where snow melting is accelerated within the study area and obtain the spatial distribution map of the sensitive areas of the snow melting and warming process.

[0082] Using the spatial distribution map of sensitive areas during the snowmelt warming process, combined with surface deformation data, an improved Bayesian online change point detection algorithm is used to analyze the surface deformation data online, determine the real-time starting node of the abnormal increase in surface deformation rate, and obtain the spatial distribution map of sensitive areas for accelerated surface deformation.

[0083] By using the spatial distribution map of the sensitive areas of accelerated surface deformation and combining it with flow data, an improved Bayesian online change point detection algorithm is used to analyze the change point characteristics in the flow data, determine the sensitive time period and corresponding river section area of ​​rapid runoff response in the watershed, and obtain the spatial distribution map of the sensitive areas of runoff change.

[0084] Using spatial distribution maps of sensitive areas during snowmelt warming, accelerated surface deformation, and runoff mutation, a chain disaster hazard transmission path map is constructed, which involves snowmelt warming triggering, accelerated surface instability, and rapid runoff response. The spatiotemporal dependencies between each link in the chain disaster hazard transmission path map are determined, and a spatiotemporal structural model of disaster hazard transmission is obtained.

[0085] Based on the spatiotemporal structure model of disaster chain hazard transmission, the disaster chain hazard level of each spatial region under snow melting and warming conditions is determined, and the disaster chain hazard during the snow melting and warming process is quantitatively identified based on the disaster chain hazard level, so as to obtain the real-time identification results of the disaster chain hazard during the snow melting and warming process.

[0086] In this embodiment, the preprocessing includes cloud removal, noise reduction, and interpolation processing of long-term remote sensing snow cover data; noise removal and smoothing filtering processing of radar wet snow line observation data; outlier detection and correction processing of temperature data; missing data interpolation and outlier data removal processing of flow data from hydrological monitoring stations; and phase unwrapping, noise reduction, and smoothing processing of surface deformation data from synthetic aperture radar interferometry.

[0087] The cloud removal, noise reduction, and interpolation processing of long-term remote sensing snow cover data includes:

[0088] Based on the spectral reflectance characteristics of clouds and remote sensing data, invalid data areas obscured by clouds in the image are identified and extracted, and the identified invalid data is removed pixel by pixel.

[0089] The median filtering method based on the image window is used to denoise the remaining valid data and remove random noise generated by sensor or atmospheric interference during the data acquisition process.

[0090] By using a time-series-based linear interpolation method, missing remote sensing data sequences are interpolated and filled in, thereby forming a continuous and complete remote sensing snow cover data sequence.

[0091] The noise removal and smoothing filtering process for radar wet snow line observation data includes:

[0092] The statistical threshold method is used to analyze the probability distribution characteristics of the data, identify and remove outlier noise data points in the observation sequence;

[0093] The Savitzky-Golay filtering method is adopted, with a preset sliding window as the basic unit. Through local polynomial regression fitting, the smoothing filtering process of radar wet snow line observation data is completed, which preserves the true trend change characteristics in radar wet snow line observation data and effectively suppresses random fluctuations.

[0094] The outlier detection and correction process for temperature data includes:

[0095] The probability distribution characteristics of the temperature data series were determined by statistical analysis methods, and abnormal data in the series were automatically detected by the three-standard-deviation criterion.

[0096] For the detected abnormal data, the spatial weighted interpolation correction method is used to correct the abnormal data points based on the valid data of the spatially neighboring meteorological stations on the same date, so as to obtain continuous, stable and abnormal temperature data.

[0097] The process of interpolating missing data and removing outlier data from hydrological monitoring station flow data includes:

[0098] The continuity of the traffic data sequence is checked to identify missing data points caused by site failure or data transmission loss. Then, the effective traffic data of adjacent time periods within the sliding window are used to perform moving average interpolation to fill in the missing data.

[0099] The interquartile range method is used to analyze the dispersion of the traffic data sequence, determine the upper and lower thresholds of abnormal data, identify and remove abnormal data exceeding the thresholds, and obtain continuous, stable traffic data without abnormal interference.

[0100] The phase unwrapping, denoising, and smoothing processing of the surface deformation data from synthetic aperture radar interferometry includes:

[0101] The minimum cost flow algorithm is used to perform pixel-by-pixel phase unwrapping on the interferometric phase data to eliminate periodic errors caused by phase ambiguity.

[0102] High-pass filtering is used to remove spatially distributed low-frequency noise from interferometric data and enhance surface deformation signals.

[0103] The Gaussian filtering method is used to smooth the surface deformation data in the spatial domain, suppressing random noise and preserving the true trend information of surface deformation, thus obtaining high-quality surface deformation data.

[0104] In this embodiment, the step of determining the real-time trigger time point sequence based on the temperature data and radar wet snow line observation data in the initial input dataset, using an improved Bayesian online change point detection algorithm, specifically involves:

[0105] The temperature data and radar wet snow line observation data of each study area in the initial input dataset are obtained and simultaneously input into the improved Bayesian online change point detection algorithm.

[0106] In the improved Bayesian online change point detection algorithm, the prior probability at the current time is dynamically adjusted based on the posterior probability of the previous time step, and the Bayesian change point probability is adaptively updated.

[0107] The adjustment of the prior probability at the current moment specifically refers to:

[0108] In the process of calculating the joint change point probability, the prior probability distribution of the initial observation time is determined, and the posterior probability of the initial observation time is calculated.

[0109] At each subsequent observation time, the stored posterior probability distribution result of the previous time is retrieved, and the probability density value and corresponding distribution parameters of the posterior probability distribution of the previous time are directly assigned to the initial parameters of the prior probability distribution at the current time.

[0110] After obtaining new temperature data and radar wet snow line observation data at the current observation time, the likelihood function value under the current observation data is calculated using the new data, and the posterior probability distribution parameters at the current time are updated by Bayes' theorem and combined with the prior probability distribution parameters at the current time.

[0111] The updated posterior probability distribution parameters at the current time are stored for dynamic adjustment of the prior probability distribution at the next observation time, so as to achieve the goal of real-time dynamic updating of the prior probability distribution parameters as the data changes at each observation time.

[0112] Calculate the joint change probability of temperature data and wet snow line observation data in each study area, and use the probability threshold to determine whether the joint change probability meets the triggering condition;

[0113] The joint change point probability is calculated as follows:

[0114] Based on the temperature data and radar wet snow line observation data obtained at the current observation time, calculate the joint probability density value of the current observation data under the assumption that no change point has occurred, and the joint probability density value of the current observation data under the assumption that a change point has occurred.

[0115] Using the prior probability distribution parameters updated in the previous time step and the likelihood function of the current observation data, the posterior probability is calculated using Bayes' theorem for the cases where no change point has occurred and the cases where a change point has occurred.

[0116] The posterior probabilities of changing points are accumulated to obtain the cumulative changing point probabilities of temperature data and wet snow line observation data in each study area at the current observation time.

[0117] The cumulative change point probability calculated at the current observation time is normalized to obtain a joint change point probability value with a value range between 0 and 1. This joint change point probability value is then output for subsequent determination of whether the triggering condition is met.

[0118] For the real-time triggering time point determined when the probability of the joint change point first exceeds the probability threshold, a dual dynamic sliding window mechanism is used to verify its effectiveness. The first sliding window is used to verify whether the probability of the change point after the real-time triggering time point continues to exceed the probability threshold for a preset duration. The second sliding window is used to verify whether the changing trends of the temperature data and the wet snow line observation data before and after the real-time triggering time point show statistically stable differences.

[0119] The validity verification process specifically involves:

[0120] For the calculated joint change point probability sequence, when the joint change point probability value at a certain observation time exceeds the preset probability threshold for the first time, the corresponding time is initially determined as the real-time trigger time point, and the first dynamic sliding window and the second dynamic sliding window verification mechanism are started.

[0121] The first dynamic sliding window moves backward from the determined real-time trigger time point and continuously observes the joint change point probability value within the window for no less than a preset duration. It checks in real time whether the joint change point probability at all times within the window continuously exceeds the probability threshold. If the joint change point probability is lower than the probability threshold at any time within the window, it is determined that the first dynamic sliding window verification has failed.

[0122] The second dynamic sliding window is centered on the real-time trigger time point and extends forward and backward by a preset symmetrical time length. It extracts temperature data and wet snow line observation data within the window and calculates the statistical mean, standard deviation, and slope of the data before and after the trigger time point. Then, it compares and analyzes whether the difference in the trend of the data before and after the trigger time point has statistical stability. If the trend change of the data within the window does not reach the statistical stability difference significance level, the second dynamic sliding window is judged to have failed the verification.

[0123] Only when the determined real-time trigger time point passes the verification of both the first dynamic sliding window and the second dynamic sliding window is it determined to be a valid real-time trigger time point and retained; otherwise, the trigger time point is discarded as a false trigger point.

[0124] The dual dynamic sliding window verification mechanism verifies the validity of each real-time trigger point when the probability of a joint change point exceeds the threshold by moving the window and performing verification, thereby improving the accuracy and stability of the real-time trigger point.

[0125] The cumulative probability increment is calculated for the real-time trigger time points that have passed the verification of the dual dynamic sliding window mechanism. The real-time trigger time points that reach the preset cumulative probability increment are selected to obtain the real-time trigger time point sequence of the snow melting and warming process.

[0126] The cumulative probability increment is calculated as follows:

[0127] For real-time triggering time points that have been verified by the dual dynamic sliding window mechanism, extract the joint change point probability with the corresponding triggering time point as the starting time.

[0128] The joint change point probability of the previous moment is used as the initial reference value for calculating the cumulative probability increment. The joint change point probability of each moment after the triggering time point is successively subtracted from the initial reference value to obtain the probability difference of each moment relative to the initial reference value.

[0129] The probability differences obtained at each time point are accumulated sequentially in chronological order to obtain the cumulative probability difference starting from the real-time trigger time point;

[0130] The largest cumulative probability difference is selected from the cumulative probability differences and used as the cumulative probability increment output for the corresponding real-time trigger time point.

[0131] In this embodiment, the step of extracting and analyzing long-term remote sensing snow cover data and radar wet snowline observation data before and after each trigger time point based on the real-time trigger time point sequence to obtain a spatial distribution map of the sensitive area of ​​the snowmelt and warming process is as follows:

[0132] Based on the real-time trigger time sequence, long-term remote sensing snow cover data and radar wet snow line observation data within a preset time period before and after each real-time trigger time are extracted.

[0133] Calculate the change in snow cover area per unit time in long-term remote sensing snow cover data before and after each real-time triggering time point, and calculate the change in height per unit time in radar wet snow line observation data before and after each real-time triggering time point;

[0134] The method for calculating the change in snow cover area per unit time is as follows:

[0135] Based on each real-time trigger time point, remote sensing snow cover data for each observation time within a preset time length before and after the corresponding real-time trigger time point are extracted, and the snow cover area of ​​the study area at each observation time is calculated.

[0136] Using each real-time triggering time point as a boundary, linear regression analysis was performed on the snow cover area values ​​calculated at each observation time before and after the real-time triggering time point to obtain the snow cover area change trend lines before and after the real-time triggering time point.

[0137] Calculate the slope of the trend line of snow cover area change before and after the real-time trigger time point, respectively, as the rate of change of snow cover area over time.

[0138] The calculated rate of change of snow cover area over time is determined as the change in snow cover area per unit time in long-term remote sensing snow cover data before and after the corresponding real-time trigger time point.

[0139] The method for calculating the change in height per unit time is as follows:

[0140] Based on each real-time trigger time point, extract the radar wet snow line height observation values ​​at each observation time within the preset time length before and after the corresponding real-time trigger time point;

[0141] The observed values ​​of wet snow line height before and after the real-time triggering time point were linearly fitted using the least squares method to obtain the trend lines of wet snow line height change before and after the real-time triggering time point, respectively.

[0142] Based on the trend lines of wet snow line height change before and after the real-time trigger time point obtained by fitting, the slope values ​​of the trend lines are calculated respectively to obtain the rate of change of wet snow line height with time.

[0143] The calculated rate of change of wet snow line height over time is determined as the unit time change of radar wet snow line observation data before and after the corresponding real-time trigger time point.

[0144] The spatial grid division method is adopted to divide the study area into multiple regular grid cells with consistent spatial resolution. The rate of change of snow cover area and the rate of change of wet snow line of each grid cell are calculated based on the change of snow cover area per unit time and the change of height per unit time, respectively.

[0145] Based on the preset sensitive spatial region identification threshold, spatial grid cells that exceed the preset threshold for the rate of change of snow cover area and the preset threshold for the rate of change of wet snow line per unit time are identified respectively.

[0146] Spatial clustering was performed on the identified spatial grid cells to obtain the spatial location range of areas where snow cover melting was significantly accelerated and areas where the rate of rise of the wet snow line was significantly accelerated within the study area.

[0147] The implementation method for spatial clustering is as follows:

[0148] Spatial grid cells whose snow cover area change rate exceeds the threshold and whose wet snow line change rate exceeds the threshold are extracted according to the preset threshold, and the spatial position coordinates of each spatial grid cell are determined.

[0149] Based on the extracted spatial coordinates of the spatial grid cells, a density-based spatial clustering method is used to cluster the spatial grid cells. By calculating the spatial distance between each grid cell and its corresponding spatially adjacent grid cells, the grid cells that are densely packed and spatially close are identified, forming multiple independent spatial clustering regions.

[0150] For each spatial cluster region formed, the spatial boundary coordinates of the cluster region are calculated to obtain the smallest outer polygon that can accurately represent the spatial range of the cluster region.

[0151] The smallest bounding polygon obtained is output as the spatial location range of the spatial clustering region;

[0152] Spatial overlay analysis was performed on the spatial location range to identify the spatial intersection area between the snow cover melting acceleration area and the wet snow line rising rate acceleration area, and to determine the spatial area sensitive to the snow melting and warming process.

[0153] The implementation method for performing spatial overlay analysis is as follows:

[0154] The minimum bounding polygons of the snow cover melting acceleration region and the wet snow line rise rate acceleration region obtained through spatial clustering are imported into the spatial analysis platform and unified to the same spatial coordinate reference system.

[0155] In a unified spatial coordinate reference system, spatial topological relationship analysis is performed on the minimum bounding polygons of the snow cover melting acceleration region and the wet snow line rising rate acceleration region to calculate the spatial intersection range between the two regions.

[0156] The calculated spatial intersection range is further verified by topological relationship, and fragmented areas with too small area or poor spatial continuity are removed to ensure that the determined spatial intersection area has spatial continuity and statistics.

[0157] The spatial intersection region after topological relationship verification will be used as the final range of the sensitive spatial region during the snow melting and warming process.

[0158] Based on the sensitive spatial areas of the snowmelt warming process, draw a spatial distribution map of the sensitive areas of the snowmelt warming process.

[0159] In this embodiment, the improved Bayesian online change point detection algorithm is used to perform online analysis of surface deformation data to obtain a spatial distribution map of sensitive areas for accelerated surface deformation. Specifically:

[0160] Using the spatial distribution map of sensitive areas during the snowmelt warming process, the spatial location range of sensitive areas during the snowmelt warming process is determined, and surface deformation data at each observation time within the spatial location range is extracted;

[0161] The improved Bayesian online change point detection algorithm is used to calculate the change point probability of surface deformation data at each observation time, and a dynamic probability threshold judgment is made on the change point probability at each observation time to identify candidate real-time starting nodes with abnormally increased surface deformation rate.

[0162] For the identified candidate real-time starting nodes, a dynamic sliding time window is set. The sliding time window continuously observes the changes in the surface deformation rate at multiple time points and calculates the cumulative change point probability increment at each time point within the window.

[0163] The validity of the candidate real-time starting node is determined by whether the cumulative change point probability increment at each moment within the sliding time window exceeds the preset statistical threshold, and the candidate real-time starting nodes that meet the statistical conditions are retained to obtain the valid real-time starting nodes.

[0164] Based on the retained effective real-time starting nodes, spatial grid cells with significantly increased surface deformation rates are determined by spatial grid division method. Spatial cluster analysis is then performed on the determined spatial grid cells to obtain the spatial location range of the sensitive area of ​​accelerated surface deformation. Based on the spatial location range of the sensitive area of ​​accelerated surface deformation, a spatial distribution map of the sensitive area of ​​accelerated surface deformation is drawn.

[0165] The implementation method for performing spatial clustering analysis on the determined spatial grid cells is as follows:

[0166] The spatial grid cells with significantly increased surface deformation rates, determined by the spatial grid division method, are extracted one by one, and the spatial coordinates of each spatial grid cell are recorded.

[0167] By calculating the Euclidean spatial distance between each spatial grid cell and its surrounding spatial grid cells, spatial grid cells that are spatially adjacent and densely distributed with a spatial distance less than a preset spatial clustering threshold are divided into the same clustering region.

[0168] For each cluster region containing all spatial grid cells, the vertex coordinates of the smallest enclosing polygon that can accurately enclose the entire cluster region are determined based on the grid cell position coordinates.

[0169] The smallest bounding polygon of each cluster region is used as the spatial location range of the sensitive area for accelerated surface deformation.

[0170] In this embodiment, the improved Bayesian online change point detection algorithm is used to analyze the change point characteristics in the flow data to obtain a spatial distribution map of runoff change-sensitive areas, specifically as follows:

[0171] Using the spatial distribution map of areas sensitive to accelerated surface deformation, the spatial location range of these areas is determined, and flow data from hydrological monitoring stations within this spatial range is extracted.

[0172] For the extracted traffic data, an improved Bayesian online change point detection algorithm is used to calculate the change point probability of traffic data at each observation time. The probability threshold is dynamically updated to determine whether the change point probability of traffic data exceeds the preset threshold, and to identify candidate sensitive periods when traffic occurs rapidly.

[0173] For the candidate sensitive period, set up a dynamic verification time window with symmetry before and after it. Calculate the statistical mean and variance of the traffic data in the two time windows before and after the candidate sensitive period, and compare whether the difference in the statistical characteristic values ​​in the two windows meets the preset statistical conditions to verify the effectiveness of the candidate sensitive period.

[0174] Candidate sensitive periods that meet the preset statistical conditions are retained as effective sensitive periods for rapid runoff response. The spatial location range of the river section corresponding to the rapid runoff response is determined by combining the spatial coordinates of each hydrological monitoring station within the spatial location range.

[0175] The method for determining the spatial location range of the river section corresponding to the rapid runoff response is as follows:

[0176] For the effective sensitive period confirmed by the dynamic verification time window, the flow observation data of each hydrological monitoring station within the spatial location range of the effective sensitive period are extracted, and the average rate of change of flow data of each hydrological monitoring station within the effective sensitive period is calculated.

[0177] Based on the spatial coordinates of hydrological monitoring stations, spatial interpolation is performed on the average flow rate of change calculated for each hydrological monitoring station within the effective sensitive period along the river flow direction to obtain the spatial distribution of flow rate of change within the continuous spatial range of the entire river.

[0178] Based on the spatial distribution of flow change rate, determine the upstream and downstream boundary coordinates of continuous river segments where the flow change rate exceeds a preset threshold, so as to form the preliminary spatial range of river segments corresponding to the rapid runoff response.

[0179] The spatial buffer analysis is further performed on the initial spatial location range of the river section. The preset width is extended to both sides of the riverbank based on the centerline of the river channel to obtain the spatial location range of the river section that can accurately represent the rapid response of runoff.

[0180] The spatial grid division method is adopted to divide the determined spatial location range of the river section into multiple spatial grid units, and the flow change rate in each spatial grid unit is calculated to filter out spatial grid units whose flow change rate exceeds a preset threshold.

[0181] Spatial clustering analysis is performed on the selected spatial grid cells whose flow change rate exceeds the preset threshold to determine the spatial location range of each spatial cluster region, and output as the spatial location range of the runoff change sensitive area.

[0182] Based on the spatial location range of the runoff change sensitivity area, draw a spatial distribution map of the runoff change sensitivity area.

[0183] In this embodiment, the spatial distribution map of sensitive areas during snowmelt warming, the spatial distribution map of sensitive areas for accelerated surface deformation, and the spatial distribution map of sensitive areas for sudden runoff changes are used to construct a chain-like disaster chain hazard transmission path map, thereby obtaining a spatiotemporal structure model of disaster chain hazard transmission. Specifically, this involves:

[0184] Extract the spatial location range and corresponding start trigger time of each sensitive area from the spatial distribution maps of sensitive areas for snowmelt warming process, sensitive areas for accelerated surface deformation, and sensitive areas for runoff change.

[0185] Based on the spatial location range of the sensitive areas of snowmelt warming process and the spatial location range of the sensitive areas of accelerated surface deformation, the spatial topological relationship analysis method is used to calculate the spatial intersection area between the two sensitive areas, and to identify the spatial triggering relationship of snowmelt warming process on accelerated surface instability.

[0186] The implementation method for calculating the spatial intersection region between two sensitive regions using spatial topological relationship analysis is as follows:

[0187] Extract the spatial location range of the two sensitive regions respectively, and import the spatial location range of the two sensitive regions into the same spatial coordinate reference system;

[0188] By using spatial topology analysis, spatial overlay operation is performed on the spatial location ranges of two sensitive areas to obtain the spatial overlap of the two sensitive areas.

[0189] Perform spatial topology verification on the spatial overlapping parts, remove small or isolated areas caused by spatial topology operations, and retain the overlapping parts whose spatial continuity meets the preset threshold.

[0190] The overlapping portion after spatial topology verification is identified as the spatial intersection region between two sensitive regions and output.

[0191] Based on the spatial location range of the sensitive areas of accelerated surface deformation and the spatial location range of the sensitive areas of sudden runoff change, the spatial topological relationship analysis method is used to calculate the spatial intersection area between the two sensitive areas, and to identify the spatial triggering relationship of accelerated surface instability to rapid runoff response.

[0192] The temporal dependency between the sensitive areas is determined by calculating the initial triggering time interval of the sensitive areas for accelerated surface deformation triggered by the snowmelt warming process, and the initial triggering time interval of the sensitive areas for accelerated surface deformation triggered by the sensitive areas for sudden runoff.

[0193] Based on the established spatial triggering and temporal dependencies, a chain disaster hazard transmission path map is constructed, which involves snowmelt warming triggering, accelerated surface instability, and rapid runoff response.

[0194] The method for constructing the chain-like disaster chain hazard transmission path diagram is as follows:

[0195] Based on the established spatial triggering relationship, the spatial intersection areas between the sensitive areas of snowmelt warming process and the sensitive areas of accelerated surface deformation, and between the sensitive areas of accelerated surface deformation and the sensitive areas of sudden runoff change, are extracted, and the specific spatial coordinate range of each spatial intersection area is recorded.

[0196] Based on the established time dependency relationship, calculate the initial trigger time difference between the sensitive area of ​​snowmelt warming process and the corresponding sensitive area of ​​accelerated surface deformation, as well as the initial trigger time difference between the sensitive area of ​​accelerated surface deformation and the corresponding sensitive area of ​​sudden runoff change, and mark the specific value of the time difference.

[0197] A graph structure representation method combining nodes and directed edges is adopted. Sensitive areas of snowmelt warming process, sensitive areas of accelerated surface deformation, and sensitive areas of runoff change are respectively used as nodes in the chain disaster chain hidden danger transmission path map. The extracted spatial intersection area is used as the basis for connecting spatial triggering relationships between nodes, and the calculated starting trigger time difference is used as the basis for connecting temporal dependency relationships between nodes. The directed connection relationships between nodes are drawn level by level.

[0198] The spatial triggering relationship and temporal dependency relationship between nodes are drawn step by step to obtain a complete chain disaster chain hazard transmission path map that fully expresses the snowmelt warming trigger, surface instability acceleration and runoff rapid response process as the output result;

[0199] Based on the chain-like disaster chain hazard transmission path map, the spatiotemporal dependencies between each link are determined by spatial fusion and temporal sequence registration of the spatial location range and trigger time information of all sensitive areas;

[0200] Based on the established spatiotemporal dependencies, a spatiotemporal structural model of the transmission of potential hazards in the disaster chain is constructed.

[0201] In this embodiment, the step of constructing a spatiotemporal structural model for the transmission of disaster chain hazards based on the determined spatiotemporal dependencies specifically involves:

[0202] Based on the established chain disaster chain hazard transmission path map, the spatial location range and triggering time information of each node in the sensitive areas of snow melting and warming process, the sensitive areas of accelerated surface deformation, and the sensitive areas of runoff change in the disaster chain path map are extracted;

[0203] The extracted spatial location ranges of each node are uniformly projected onto the same spatial coordinate reference system, and the spatial location range data are standardized to eliminate the differences in spatial scale between different regions.

[0204] Using the triggering time of the sensitive area of ​​snowmelt and warming process as the starting reference time, the triggering time of the sensitive area of ​​accelerated surface deformation and sensitive area of ​​runoff change are time-registered with the starting reference time, and the standardized relative time difference between each node is calculated and determined.

[0205] Based on the standardized relative time difference, the time sequence relationship is used as the time dimension basis of the model, and the normalized spatial location range of each node is used as the spatial dimension basis of the model, thus constructing a spatiotemporal three-dimensional structure containing information in both time and space dimensions.

[0206] The spatial distribution characteristics of each node's spatial location range are represented by the spatial dimension, and the trigger sequence and time interval between nodes are represented by the temporal dimension, thus constructing a spatiotemporal structure model for the transmission of potential hazards in the disaster chain.

[0207] In this embodiment, the step of determining the disaster chain hazard level of each spatial region under snowmelt and warming conditions based on the spatiotemporal structure model of disaster chain hazard transmission, and quantitatively identifying the disaster chain hazard during the snowmelt and warming process, specifically involves:

[0208] Based on the spatiotemporal structure model of disaster chain hazard transmission, the spatial location range of sensitive areas during snowmelt and warming processes, sensitive areas for accelerated surface deformation, and sensitive areas for sudden runoff changes, along with the corresponding spatiotemporal characteristic parameters of disaster chain hazard transmission, are extracted.

[0209] For each spatial region, the probability of snowmelt warming, the probability of accelerated surface deformation, and the probability of sudden runoff are calculated separately. Based on the preset weighting factors of each disaster link, the triggering probability, acceleration probability, and sudden runoff probability of each spatial region are weighted and fused for calculation.

[0210] Based on the weighted fusion calculation results, the comprehensive risk index of disaster chain hazards in each spatial region is determined, and the comprehensive risk index is divided into multiple disaster chain hazard levels according to the preset disaster chain hazard level thresholds;

[0211] For the identified hazard levels of the disaster chain, a mapping relationship between the hazard level of the disaster chain and the degree of disaster risk is established, and the hazard level of the disaster chain corresponding to each spatial region under the conditions of snow melting and warming is marked;

[0212] Based on the marked hazard level of the disaster chain, calculate the area of ​​spatial risk range and the number of affected spatial grid units for each spatial region under the hazard level of the disaster chain.

[0213] By using spatial interpolation and rasterization analysis, the disaster chain hazard level, risk range area and number of affected spatial grid units calculated for each spatial region are spatially fused to generate quantitative identification results that include spatial region, disaster chain hazard level and disaster risk spatial distribution characteristics.

[0214] The spatial interpolation and rasterization analysis are as follows:

[0215] For each spatial region, the hazard level, risk range area, and number of affected spatial grid units of the disaster chain are extracted, and the spatial location range coordinates of each spatial region are recorded.

[0216] Using spatial interpolation, the hazard level of the disaster chain, the area of ​​risk range, and the number of affected spatial grid cells in each spatial region are used as input parameters. Interpolation is performed on spatial grid cells in the study area that do not have clearly marked values ​​to obtain continuous spatial interpolation results of the hazard level of the disaster chain, the area of ​​risk range, and the number of affected spatial grid cells in the study area.

[0217] Using the rasterization analysis method, the continuous spatial interpolation results of the interpolated hazard chain hazard level, risk range area and number of affected spatial grid cells are converted into raster layers with consistent spatial resolution, and the spatial coordinate system and spatial range between the raster layers are completely consistent.

[0218] Spatial alignment and registration were performed on the disaster chain hazard level raster layer, the risk range area raster layer, and the number of affected spatial grid cells raster layer one by one to eliminate possible spatial deviations between different raster layers.

[0219] The spatially aligned and registered hazard chain hazard level raster layer, risk range area raster layer, and affected spatial grid cell number raster layer are superimposed and fused unit by unit according to spatial coordinates to obtain a unified rasterized spatial fusion result that includes hazard chain hazard level, risk range area, and affected spatial grid cell number.

[0220] In this embodiment, the improved Bayesian online change point detection algorithm is specifically as follows:

[0221] The initial prior distribution and its probability threshold parameters are set, and the Bayesian online change point detection algorithm is initialized by setting the mean and variance parameters of the initial prior distribution.

[0222] The observation data is acquired time by time, the prior probability of the current time is dynamically adjusted using the posterior probability of the previous time, and the change point posterior probability of the current time is calculated based on Bayes' theorem.

[0223] Based on the posterior probability of the change point at the current time, the probability threshold at the current time is dynamically updated, and whether the current time is a candidate change point is determined based on whether the posterior probability of the current change point exceeds the probability threshold.

[0224] When a candidate change point occurs, a dual dynamic sliding window is constructed to verify the validity of the candidate change point. The first sliding window is used to verify whether the posterior probability of the change point at multiple consecutive time points after the candidate change point continuously exceeds the dynamic probability threshold. The second sliding window is used to verify whether there are statistically significant differences in the statistical characteristics of the observed data within the window before and after the candidate change point.

[0225] After a candidate change point passes the validity verification of a dual dynamic sliding window, the cumulative probability increment of the candidate change point and its subsequent verification window is calculated. It is then determined whether the cumulative probability increment reaches a preset threshold and whether the candidate change point is a true and valid change point.

[0226] Example 1:

[0227] To verify the feasibility of this invention in practice, it was applied to the real-time identification of disaster chain hazards induced by snowmelt and warming processes in a key watershed in a mountainous area, effectively improving the accuracy and real-time performance of hazard identification. In this practical application scenario, traditional technologies often employ single or isolated data sources such as remote sensing snow cover monitoring, radar wet snow line observation, temperature observation, hydrological flow monitoring, and surface deformation monitoring for independent analysis. These data lack effective spatiotemporal correlation and fusion, and typically use static thresholds to determine risk areas. Therefore, their accuracy and efficiency are insufficient in the real-time, dynamic quantitative identification of disaster chain hazards at the regional scale, failing to meet the actual needs of disaster risk management.

[0228] In the specific implementation process, satellite remote sensing was first used to acquire 30 consecutive days of long-term remote sensing snow cover data for the study area. Simultaneously, radar monitoring equipment was used to acquire wet snow line observation data, and daily temperature data, surface deformation data, and hydrological monitoring station flow data were obtained from ground automatic observation stations. Different data were preprocessed separately. Specifically, remote sensing snow cover data was processed using methods such as cloud contamination pixel removal, noise filtering, and spatial interpolation; radar wet snow line data underwent noise removal and filtering smoothing; temperature data was processed through outlier detection and interpolation correction; surface deformation data underwent phase unwrapping and denoising smoothing; and hydrological flow data underwent missing data interpolation and outlier removal, thus constructing a unified and standardized initial input dataset.

[0229] Based on the preprocessed temperature data and wet snowline observation data, the improved Bayesian online change point detection algorithm of this invention is used to calculate the joint change point probability of temperature and wet snowline data in the study area in real time and on a daily basis. The real-time trigger time node for snowmelt and warming is determined by dynamically updating the probability threshold, and the effectiveness of the node is verified by using a dual dynamic sliding window mechanism. Finally, the effective real-time trigger time nodes that meet the cumulative probability increment condition are calculated and screened.

[0230] After the real-time trigger node is determined, remote sensing snow cover data and wet snow line data for the time periods before and after each node are extracted. The rate of change of snow cover area and the rate of change of wet snow line height are calculated. The areas where snow cover melting is significantly accelerated and the areas where the wet snow line rises significantly are determined in a spatial grid manner. The spatial intersection of the two areas is identified through spatial overlay analysis to form a sensitive spatial area for the snow melting and warming process.

[0231] By utilizing the spatial location range of identified sensitive areas and further combining it with surface deformation data, an improved Bayesian online change point detection algorithm is used to calculate the probability of surface deformation change points at each time point, thereby identifying real-time starting nodes where the surface deformation rate is abnormally increasing. Through a dynamic sliding window and cumulative probability increment mechanism, effective real-time starting nodes are determined, thus identifying sensitive areas of accelerated surface deformation.

[0232] By utilizing identified sensitive spatial areas of accelerated surface deformation and combining them with runoff data, an improved Bayesian online change point detection algorithm is used to analyze the flow data daily, calculate the probability of flow change points and dynamically determine the probability threshold, identify sensitive periods of rapid runoff response in the watershed, and further verify the effectiveness through statistical features to determine the sensitive river sections corresponding to rapid runoff response, ultimately forming sensitive spatial areas of runoff change.

[0233] Based on the spatial location range and triggering time nodes of the above three sensitive areas, the spatial triggering relationship between the snowmelt warming process and the acceleration of surface deformation, as well as the spatial triggering relationship between the acceleration of surface deformation and the sudden change in runoff, is determined by the spatial topological relationship analysis method. At the same time, the temporal dependency relationship between each link is determined, a chain disaster chain hazard transmission path map is established, and further, through spatial location fusion and time series registration, a spatiotemporal structure model of disaster chain hazard transmission is constructed.

[0234] Based on the spatiotemporal structure model of disaster chain hazard transmission constructed above, the trigger probability, acceleration probability, and mutation probability of sensitive areas are extracted respectively, and a comprehensive risk index of disaster chain hazard is obtained by weighted fusion calculation. Based on the preset disaster chain hazard level threshold, different levels of disaster chain hazard areas are divided, and the risk range area and the number of affected grids of each area are calculated. After spatial interpolation and rasterization analysis, a real-time updated disaster risk spatial distribution map is generated, realizing real-time identification and refined management of disaster chain hazards during the snow melting and warming process in the study area.

[0235] To more intuitively demonstrate the actual effects of the invention, a detailed comparative analysis was conducted on the disaster chain hazard level identification results and the measured disaster situation in five typical sensitive areas within the study area. The specific data results are shown in Table 1.

[0236] Table 1 Comparison of Predicted and Actual Disaster Levels in Disaster Chain Hazard Situation Table

[0237] Area code Snowmelt trigger probability (%) Deformation acceleration probability (%) Runoff mutation probability (%) Comprehensive Risk Index Forecast Predicted value of disaster hazard level Actual disaster level Area A 92.5 85.6 81.2 87.1 High risk High risk Area B 67.8 54.7 52.9 58.5 Medium risk Medium risk Area C 45.3 32.9 29.8 35.5 Low risk Low risk Region D 80.6 78.2 74.5 77.8 High risk High risk Area E 56.2 48.3 42.1 49.7 Medium risk Medium risk

[0238] As can be seen from the comparison data of the predicted and measured disaster levels of the disaster chain hazard level given in Table 1, the method proposed in this invention exhibits high prediction accuracy in practical applications. Specifically, in region A, the predicted disaster hazard level is high risk, and the actual disaster level is also high risk, indicating that the method of this invention has a good ability to identify high-risk areas. Similarly, in region D, this invention also accurately identified the high-risk level, which is highly consistent with the measured results. In regions B and E, this invention successfully predicted the medium-risk level, and the prediction results are consistent with the actual disaster levels, reflecting the reliability of this invention in predicting medium-risk levels. The low-risk level prediction in region C is consistent with the actual situation, further demonstrating the accuracy of the method of this invention in identifying low risks.

[0239] Based on the above analysis, the improved Bayesian online variable point detection algorithm and the spatiotemporal structure model of disaster chain hazard transmission proposed in this invention effectively enhance the real-time identification capability of snowmelt and warming disaster chain hazards, demonstrating significantly better prediction accuracy and stability than traditional methods, and can effectively support accurate early warning and scientific prevention and control of disaster risks.

[0240] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for identifying potential hazards in the disaster chain of snowmelt and warming process based on big data, characterized in that, include: Long-term remote sensing snow cover data, radar wet snow line observation data, temperature data, flow data from hydrological monitoring stations, and surface deformation data from synthetic aperture radar interferometry were acquired within the study area and preprocessed to obtain the initial input dataset. Based on the temperature data and radar wet snow line observation data in the initial input dataset, an improved Bayesian online change point detection algorithm is used to determine the real-time trigger time point sequence. Based on the real-time trigger time sequence, long-term remote sensing snow cover data and radar wet snow line observation data before and after each trigger time point are extracted and analyzed to obtain a spatial distribution map of the sensitive area of ​​the snow melting and warming process. An improved Bayesian online change point detection algorithm was used to perform online analysis on surface deformation data to obtain a spatial distribution map of sensitive areas for accelerated surface deformation. An improved Bayesian online change point detection algorithm was used to analyze the change point characteristics in flow data and obtain a spatial distribution map of runoff change-sensitive areas. By utilizing spatial distribution maps of sensitive areas during snowmelt warming, accelerated surface deformation, and runoff mutation, a chain-like disaster chain hazard transmission path map is constructed, and a spatiotemporal structure model of disaster chain hazard transmission is obtained. Based on the spatiotemporal structure model of disaster chain hazard transmission, the disaster chain hazard level of each spatial region under snow melting and warming conditions is determined, and the disaster chain hazard during the snow melting and warming process is quantitatively identified.

2. The method for identifying potential hazards in the disaster chain of snow melting and warming process based on big data, as described in claim 1, is characterized in that... The preprocessing includes cloud removal, noise reduction, and interpolation of long-term remote sensing snow cover data; noise removal and smoothing filtering of radar wet snow line observation data; outlier detection and correction of temperature data; missing data interpolation and outlier removal of flow data from hydrological monitoring stations; and phase unwrapping, noise reduction, and smoothing of surface deformation data from synthetic aperture radar interferometry.

3. The method for identifying potential hazards in the disaster chain of snow melting and warming process based on big data, as described in claim 1, is characterized in that... Based on the temperature data and radar wet snow line observation data in the initial input dataset, an improved Bayesian online change point detection algorithm is used to determine the real-time trigger time point sequence, specifically: The temperature data and radar wet snow line observation data of each study area in the initial input dataset are obtained and simultaneously input into the improved Bayesian online change point detection algorithm. In the improved Bayesian online change point detection algorithm, the prior probability at the current time is dynamically adjusted based on the posterior probability of the previous time step, and the Bayesian change point probability is adaptively updated. Calculate the joint change probability of temperature data and wet snow line observation data in each study area, and use the probability threshold to determine whether the joint change probability meets the triggering condition; For the real-time triggering time point determined when the probability of the joint change point first exceeds the probability threshold, a dual dynamic sliding window mechanism is used to verify its effectiveness. The first sliding window is used to verify whether the probability of the change point after the real-time triggering time point continues to exceed the probability threshold for a preset duration. The second sliding window is used to verify whether the changing trends of the temperature data and the wet snow line observation data before and after the real-time triggering time point show statistically stable differences. The cumulative probability increment is calculated for the real-time trigger time points that have passed the verification of the dual dynamic sliding window mechanism. The real-time trigger time points that reach the preset cumulative probability increment are selected to obtain the real-time trigger time point sequence of the start of the snow melting and warming process.

4. The method for identifying potential hazards in the disaster chain of snow melting and warming process based on big data, as described in claim 1, is characterized in that... Based on the real-time trigger time sequence, long-term remote sensing snow cover data and radar wet snowline observation data before and after each trigger time point are extracted and analyzed to obtain a spatial distribution map of the sensitive area of ​​the snowmelt and warming process, specifically: Based on the real-time trigger time sequence, long-term remote sensing snow cover data and radar wet snow line observation data within a preset time period before and after each real-time trigger time are extracted. Calculate the change in snow cover area per unit time in long-term remote sensing snow cover data before and after each real-time triggering time point, and calculate the change in height per unit time in radar wet snow line observation data before and after each real-time triggering time point; The spatial grid division method is adopted to divide the study area into multiple regular grid cells with consistent spatial resolution. The rate of change of snow cover area and the rate of change of wet snow line of each grid cell are calculated based on the change of snow cover area per unit time and the change of height per unit time, respectively. Based on the preset sensitive spatial region identification threshold, spatial grid cells that exceed the preset threshold for the rate of change of snow cover area and the preset threshold for the rate of change of wet snow line per unit time are identified respectively. The identified spatial grid cells are subjected to spatial clustering to obtain their spatial location range; Spatial overlay analysis was performed on the spatial location range to identify the spatial intersection area between the snow cover melting acceleration area and the wet snow line rising rate acceleration area, and to determine the spatial area sensitive to the snow melting and warming process. Based on the sensitive spatial areas of the snowmelt warming process, draw a spatial distribution map of the sensitive areas of the snowmelt warming process.

5. The method for identifying potential hazards in the disaster chain of snow melting and warming process based on big data, as described in claim 1, is characterized in that... The improved Bayesian online change point detection algorithm is used to analyze surface deformation data online to obtain a spatial distribution map of areas sensitive to accelerated surface deformation, specifically: Using the spatial distribution map of sensitive areas during the snowmelt warming process, the spatial location range of sensitive areas during the snowmelt warming process is determined, and surface deformation data at each observation time within the spatial location range is extracted; The improved Bayesian online change point detection algorithm is used to calculate the change point probability of surface deformation data at each observation time, and a dynamic probability threshold judgment is made on the change point probability at each observation time to identify candidate real-time starting nodes with abnormally increased surface deformation rate. For the identified candidate real-time starting nodes, a dynamic sliding time window is set. The sliding time window continuously observes the changes in the surface deformation rate at multiple time points and calculates the cumulative probability increment of the change point at each time point within the window. The validity of the candidate real-time starting node is determined by whether the cumulative change point probability increment at each moment within the sliding time window exceeds the preset statistical threshold, and the candidate real-time starting nodes that meet the statistical conditions are retained to obtain the valid real-time starting nodes. Based on the retained effective real-time starting nodes, spatial grid cells with significantly increased surface deformation rates are identified through spatial grid division methods. Spatial cluster analysis is then performed on the identified spatial grid cells to obtain the spatial location range of the sensitive areas of accelerated surface deformation, and a spatial distribution map of the sensitive areas of accelerated surface deformation is drawn.

6. The method for identifying potential hazards in the disaster chain of snow melting and warming process based on big data, as described in claim 1, is characterized in that... The improved Bayesian online change point detection algorithm is used to analyze the change point characteristics in the flow data and obtain a spatial distribution map of runoff change-sensitive areas, specifically: Using the spatial distribution map of areas sensitive to accelerated surface deformation, the spatial location range of these areas is determined, and flow data from hydrological monitoring stations within this spatial range is extracted. For the extracted traffic data, an improved Bayesian online change point detection algorithm is used to calculate the change point probability of traffic data at each observation time. The probability threshold is dynamically updated to determine whether the change point probability of traffic data exceeds the preset threshold, and to identify candidate sensitive periods when traffic responds rapidly. For the candidate sensitive period, set up a dynamic verification time window that is symmetrical before and after. Calculate the statistical mean and variance of the traffic data in the two time windows before and after the candidate sensitive period, and compare whether the difference in the statistical characteristic values ​​in the two windows meets the preset statistical conditions. Candidate sensitive periods that meet the preset statistical conditions are retained as effective sensitive periods for rapid runoff response. The spatial location range of the river section corresponding to the rapid runoff response is determined by combining the spatial coordinates of each hydrological monitoring station within the spatial location range. The spatial grid division method is adopted to divide the determined spatial location range of the river section into multiple spatial grid units, and the flow change rate in each spatial grid unit is calculated to filter out spatial grid units whose flow change rate exceeds a preset threshold. Spatial clustering analysis is performed on the selected spatial grid cells whose flow change rate exceeds a preset threshold to determine the spatial location range of each spatial cluster region; Based on the spatial location range of the runoff change sensitivity area, draw a spatial distribution map of the runoff change sensitivity area.

7. The method for identifying potential hazards in the disaster chain of snow melting and warming process based on big data, as described in claim 1, is characterized in that... The method utilizes spatial distribution maps of sensitive areas during snowmelt warming, accelerated surface deformation, and runoff mutation to construct a chain-like disaster chain hazard transmission path map, thereby obtaining a spatiotemporal structural model of disaster chain hazard transmission. Specifically: Extract the spatial location range and corresponding start trigger time of each sensitive area from the spatial distribution maps of sensitive areas for snowmelt warming process, sensitive areas for accelerated surface deformation, and sensitive areas for runoff change. Based on the spatial location range of the sensitive areas of snowmelt warming process and the spatial location range of the sensitive areas of accelerated surface deformation, the spatial topological relationship analysis method is used to calculate the spatial intersection area between the two sensitive areas, and to identify the spatial triggering relationship of snowmelt warming process on accelerated surface instability. Based on the spatial location range of the sensitive areas of accelerated surface deformation and the spatial location range of the sensitive areas of sudden runoff change, the spatial topological relationship analysis method is used to calculate the spatial intersection area between the two sensitive areas and identify the spatial triggering relationship of accelerated surface instability to rapid runoff response. The temporal dependency between the sensitive areas is determined by calculating the initial triggering time interval of the sensitive areas for accelerated surface deformation triggered by the snowmelt warming process, and the initial triggering time interval of the sensitive areas for accelerated surface deformation triggered by the sensitive areas for sudden runoff. Based on the established spatial triggering and temporal dependencies, a chain disaster hazard transmission path map is constructed, which involves snowmelt warming triggering, accelerated surface instability, and rapid runoff response. Based on the chain-like disaster chain hazard transmission path map, the spatiotemporal dependencies between each link are determined by spatial fusion and temporal sequence registration of the spatial location range and trigger time information of all sensitive areas; Based on the established spatiotemporal dependencies, a spatiotemporal structural model of the transmission of potential hazards in the disaster chain is constructed.

8. The method for identifying potential hazards in the disaster chain of snow melting and warming process based on big data, as described in claim 7, is characterized in that... Based on the established spatiotemporal dependencies, a spatiotemporal structural model for the transmission of disaster chain hazards is constructed, specifically as follows: Based on the established chain disaster chain hazard transmission path map, the spatial location range and triggering time information of each node in the sensitive areas of snow melting and warming process, the sensitive areas of accelerated surface deformation, and the sensitive areas of runoff change in the disaster chain path map are extracted; The extracted spatial location ranges of each node are uniformly projected onto the same spatial coordinate reference system, and the spatial location range data are standardized to eliminate the differences in spatial scale between different regions. Using the triggering time of the sensitive area of ​​snowmelt and warming process as the starting reference time, the triggering time of the sensitive area of ​​accelerated surface deformation and sensitive area of ​​runoff change are time-registered with the starting reference time, and the standardized relative time difference between each node is calculated and determined. Based on the standardized relative time difference, the time sequence relationship is used as the time dimension basis of the model, and the normalized spatial location range of each node is used as the spatial dimension basis of the model, thus constructing a spatiotemporal three-dimensional structure containing information in both time and space dimensions. The spatial distribution characteristics of each node's spatial location range are represented by the spatial dimension, and the trigger sequence and time interval between nodes are represented by the temporal dimension, thus constructing a spatiotemporal structure model for the transmission of potential hazards in the disaster chain.

9. The method for identifying potential hazards in the disaster chain of snow melting and warming process based on big data, as described in claim 1, is characterized in that... Based on the spatiotemporal structure model of disaster chain hazard transmission, the disaster chain hazard level of each spatial region under snowmelt and warming conditions is determined, and the disaster chain hazard during the snowmelt and warming process is quantitatively identified, specifically as follows: Based on the spatiotemporal structure model of disaster chain hazard transmission, the spatial location range of sensitive areas during snowmelt and warming processes, sensitive areas for accelerated surface deformation, and sensitive areas for sudden runoff changes, along with the corresponding spatiotemporal characteristic parameters of disaster chain hazard transmission, are extracted. For each spatial region, the probability of snowmelt warming, the probability of accelerated surface deformation, and the probability of sudden runoff are calculated separately. Based on the preset weighting factors of each disaster link, the triggering probability, acceleration probability, and sudden runoff probability of each spatial region are weighted and fused for calculation. Based on the weighted fusion calculation results, the comprehensive risk index of disaster chain hazards in each spatial region is determined, and the comprehensive risk index is divided into multiple disaster chain hazard levels according to the preset disaster chain hazard level thresholds; For the identified hazard levels of the disaster chain, a mapping relationship between the hazard level of the disaster chain and the degree of disaster risk is established, and the hazard level of the disaster chain corresponding to each spatial region under the conditions of snow melting and warming is marked; Based on the marked hazard level of the disaster chain, calculate the area of ​​spatial risk range and the number of affected spatial grid units for each spatial region under the hazard level of the disaster chain. By using spatial interpolation and rasterization analysis, the hazard level, risk range area, and number of affected spatial grid cells calculated for each spatial region are spatially fused to generate quantitative identification results.

10. The method for identifying potential hazards in the disaster chain of snow melting and warming process based on big data, as described in claim 1, is characterized in that... The improved Bayesian online change point detection algorithm is as follows: The initial prior distribution and its probability threshold parameters are set, and the Bayesian online change point detection algorithm is initialized by setting the mean and variance parameters of the initial prior distribution. The observation data is acquired time by time, the prior probability of the current time is dynamically adjusted using the posterior probability of the previous time, and the change point posterior probability of the current time is calculated based on Bayes' theorem. Based on the posterior probability of the change point at the current time, the probability threshold at the current time is dynamically updated, and whether the current time is a candidate change point is determined based on whether the posterior probability of the current change point exceeds the probability threshold. When a candidate change point occurs, a dual dynamic sliding window is constructed to verify the validity of the candidate change point. The first sliding window is used to verify whether the posterior probability of the change point at multiple consecutive time points after the candidate change point continuously exceeds the dynamic probability threshold. The second sliding window is used to verify whether there are statistically significant differences in the statistical characteristics of the observed data within the window before and after the candidate change point. After a candidate change point passes the validity verification of a dual dynamic sliding window, the cumulative probability increment of the candidate change point and its subsequent verification window is calculated. It is then determined whether the cumulative probability increment reaches a preset threshold and whether the candidate change point is a true and valid change point.

Images