Forest fire risk assessment method based on compound chain disaster evolution mechanism

Abstract

The present invention discloses a forest fire risk assessment method based on a composite chain disaster evolution mechanism, which relates to the field of forest fire risk assessment. By collecting multi-source heterogeneous data, including remote sensing data, meteorological data, geographic information data and disaster loss data, a chain characteristic field is constructed for the multi-source heterogeneous data to obtain a disaster chain core driving factor tensor. A composite disaster chain coupling analysis is performed based on the disaster chain core driving factor tensor to obtain a disaster chain spatiotemporal cascade trajectory. Dynamic risk assessment and grading are performed based on the disaster chain spatiotemporal cascade trajectory to obtain a five-level risk zoning map. Intelligent early warning analysis is performed based on the five-level risk zoning map to obtain a dynamic forest fire early warning map. The chain driving relationship between drought, high temperature, vegetation dryness and fire risk is integrated, and dynamic thresholds are used for risk grading. This method can capture the dynamic impact of climate variation and vegetation phenology on fire risk, thereby improving the accuracy of forest fire risk assessment.

Description

Technical Field

[0001] The present invention relates to the field of forest fire risk assessment, and in particular to a forest fire risk assessment method based on a composite chain disaster evolution mechanism. Background Art

[0002] Forest fire risk assessment is a core component of forest resource protection and ecological security management. In recent years, with the development of geographic information systems, remote sensing technology, and big data analysis, some studies have attempted to integrate multi-source data to construct comprehensive assessment models. However, most studies remain at the "indicator superposition" level and lack a description of the physical mechanisms and nonlinear dependencies in the evolution of disaster chains. Specifically, they suffer from the following deficiencies:

[0003] 1. Existing methods often focus on the independent assessment of a single risk factor (such as drought or high temperature), ignoring the chain driving relationship between drought, high temperature, vegetation dryness, and fire risk;

[0004] 2. Existing models use fixed thresholds or historical statistical patterns to stratify risk, making it difficult to capture the dynamic impacts of climate variability (such as ENSO events and monsoon anomalies) and vegetation phenology on fire risk.

[0005] 3. Existing technologies for integrating multi-source data such as remote sensing, meteorology, and geographic information remain at the surface level, lacking in-depth coupling based on physical processes. Traditional assessments focus on natural disaster factors (such as combustibles and meteorology) and ignore the spatial distribution and vulnerability analysis of disaster-prone objects (residential areas and infrastructure), resulting in a disconnect between risk zoning and actual disaster losses.

[0006] 4. Existing early warning technologies mostly rely on statistical analysis of historical data, lack the integration of real-time weather forecasts (such as ERA5 data) and deep learning models, and cannot achieve multi-scale (time-space) dynamic early warning.

[0007] In order to solve the above-mentioned defects, a technical solution is now provided. Summary of the Invention

[0008] In order to solve the technical problems raised by the above background technology, the present invention is proposed. The embodiments of the present invention provide a forest fire risk assessment method based on a composite chain disaster evolution mechanism.

[0009] The purpose of the present invention can be achieved by the following technical solution: a forest fire risk assessment method based on a composite chain disaster evolution mechanism comprises the following steps:

[0010] Step S100: multi-source heterogeneous data collection, including remote sensing data, meteorological data, geographic information data, and disaster loss data;

[0011] Step S200: constructing a chain feature field for multi-source heterogeneous data to obtain the core driving factor tensor of the disaster chain;

[0012] Step S300: performing composite disaster chain coupling analysis based on the core driving factor tensor of the disaster chain to obtain the spatiotemporal cascade trajectory of the disaster chain;

[0013] Step S400: Perform dynamic risk assessment and classification based on the spatiotemporal cascade trajectory of the disaster chain to obtain a five-level risk zoning map;

[0014] Step S500: Perform intelligent early warning analysis based on the five-level risk zoning map to obtain a dynamic forest fire early warning map.

[0015] Furthermore, the steps for tensor analysis of the core driving factors of the disaster chain are as follows:

[0016] Based on the multi-source spatiotemporal interpolation results, the smoothed basic spatial field is obtained by inverse Fourier transform and spatial convolution smoothing. The final effective observation data set and the smoothed basic spatial field are seamlessly fused with the preset missing area markers in spatiotemporal order to obtain a spatially continuous and seamless feature field.

[0017] Based on the point fire data, spatial gridding is performed, and the assessment area is divided into grid units. Each fire point is assigned to the corresponding grid through geographic coordinate mapping and the grid fire point count is calculated. The grid count value is binary-encoded and the sign function is used to convert the count value into a basic binary map. The basic binary map is accumulated and summed year by year in the time dimension and the average is taken to obtain the historical fire frequency map. The frequency is then divided into the following categories by a preset threshold: Figure 2 The high-risk fire areas are demarcated and retained to obtain a high-risk fire distribution map. The high-risk fire distribution map is spatially smoothed and the adjacent fire areas are connected through morphological closing operations to obtain a smooth and continuous binary fire distribution map.

[0018] Based on the spatial continuous seamless feature field, nonlinear scale transformation is performed, the smoothed continuous binary fire distribution map is converted into the frequency domain, and a fusion operation is performed to obtain the core driving factor tensor of the disaster chain.

[0019] Furthermore, the steps for analyzing the final valid observation data set are as follows:

[0020] The spatially explicit collaborative fuel load is modulated by the terrain through the TFF function to obtain the terrain fuel mutual feedback field;

[0021] Based on the terrain fuel mutual feed field, tensor product fusion preprocessing is performed to obtain the spatiotemporal extension feature tensor. The spatiotemporal continuous drought and fire risk coupling field is projected into the Fourier frequency domain to obtain the time-frequency domain feature basis set. The spatiotemporal extension feature tensor and the time-frequency domain feature basis set are fused with total variation regularization optimization to obtain the multi-source spatiotemporal interpolation result.

[0022] Quantitative analysis of the degree of baseline deviation of the slope of the assessment area was performed to obtain the terrain weight factor;

[0023] The terrain fuel mutual feedback field and the spatiotemporal continuous drought and fire risk coupling field are spatially fused, and the fused data field with a spatial unified grid is obtained by correcting the terrain weight factor.

[0024] Perform dynamic spatiotemporal threshold determination on the fused data field of the spatial unified grid to obtain a valid identification matrix;

[0025] Based on the slope angle characteristics, the fire risk value is directional adjusted by superimposing the slope calibration items of the spatiotemporal continuous drought and fire risk coupling field to obtain the slope optimization field. Logical multiplication screening is performed based on the slope optimization field and the valid identification matrix to obtain the final valid observation data set.

[0026] Furthermore, the steps of the spatially explicit collaborative combustible load analysis are as follows:

[0027] Based on the cumulative probability, the cumulative probability p is mapped to the standard normal distribution space through logarithmic transformation to obtain the normalized weight;

[0028] The normalized weights are normalized to obtain the standardized precipitation evapotranspiration index, which is then adjusted by vegetation temperature to obtain the plant temperature-adjusted standardized precipitation evapotranspiration index.

[0029] The forest fire danger index is obtained based on meteorological data and geographic information data using the standard algorithm of the Canadian Forest Fire Weather Index System;

[0030] A dynamic drought and fire risk coupling model was established based on the plant temperature-adjusted standardized precipitation evapotranspiration index and the forest fire risk index. The changing dynamics of drought and fire risk were mathematically coupled and the seasonal regulation of vegetation phenology was superimposed to obtain a spatiotemporal continuous drought and fire risk coupling field.

[0031] The morphological coupling and vegetation conversion calculations were performed through leaf area index, vegetation height and normalized difference vegetation index NDVI to obtain the basic amount of vegetation dry matter.

[0032] Based on the slope, aspect and altitude in the geographic information data, the gravity radiation coupling algorithm is used to obtain the terrain adjustment factor;

[0033] The measured values ​​of vegetation moisture content were analyzed using the water stress response equation to obtain the fuel dryness;

[0034] Based on the basic amount of vegetation dry matter, terrain adjustment factors, and combustible dryness, physical coupling multiplication operations are performed to obtain the spatially explicit coordinated combustible load.

[0035] Furthermore, the steps of analyzing the spatiotemporal cascade trajectory of the disaster chain are as follows:

[0036] Based on the high temperature drought forest fire dependent structure parameter set and the core driving factor tensor of the disaster chain, time-lag cross-correlation calculation and disaster trend analysis were performed to obtain the disaster chain intensity field after atmospheric circulation modulation.

[0037] Based on the defined high temperature, drought and wind speed components, the disaster chain state vector is constructed. Based on the disaster chain intensity field modulated by atmospheric circulation, the conditional probability state transition equation of the disaster chain state vector is analyzed through the Bayesian dynamic network to obtain a dynamic disaster triggering probability map.

[0038] Based on the dynamic disaster triggering probability map and terrain combustible mutual feedback field, the cascade dynamics of the disaster chain is simulated through the cascade effect simulation algorithm to obtain the spatiotemporal cascade trajectory of the disaster chain.

[0039] Furthermore, the steps for analyzing the high temperature drought forest fire dependency structure parameter set are as follows:

[0040] Channel separation is performed on the core driving factor tensor of the disaster chain to obtain the disaster chain dynamic derived standardized precipitation evapotranspiration index and the fire risk index change rate. The drought probability v is calculated based on the disaster chain dynamic derived standardized precipitation evapotranspiration index using the Logistic cumulative distribution function (using the drought threshold and steepness parameter). At the same time, the high temperature probability u is calculated based on the fire risk index change rate using the Gaussian cumulative distribution function (combining the historical mean and standard deviation).

[0041] Based on the probability of drought and high temperature, a dependency structure analysis model was established to capture the nonlinear pattern and obtain the dependency structure parameter set of high temperature drought forest fire.

[0042] Furthermore, the analysis steps of the five-level risk zoning map are as follows:

[0043] The rescue path is discretized into several line segments. An anisotropic path cost model is established based on the forest fire risk index and landslide risk rate of each line segment to conduct real-time spatial superposition analysis of fire risk and landslide, and the travel time of the minimum cost path is obtained.

[0044] Based on the travel time of the minimum cost path, the initial resilience against fire, and the risk propagation vector field, a resilience attenuation integral model is established for time-space integration analysis to obtain a dynamic resilience attenuation surface.

[0045] Based on historical forest fire disaster data, the monthly risk benchmark threshold is established through inverse probability transformation, and the benchmark threshold is dynamically adjusted to obtain the monthly dynamic threshold;

[0046] The risk propagation vector field and the dynamic resilience attenuation surface are analyzed by coupling spatial gradient and monthly dynamic threshold through the risk level reconstruction algorithm to obtain a five-level risk zoning map.

[0047] Furthermore, the steps for analyzing the fire initial resilience and risk propagation vector field are as follows:

[0048] Extract the original data of five types of hazard-prone bodies based on geographic information data, obtain the distribution density and spatial location of the hazard-prone bodies, and perform spatial superposition calculation based on the spatiotemporal cascade trajectory of the disaster chain, the distribution density and spatial location of the hazard-prone bodies using the Gaussian attenuation function formula to obtain the exposure heat map.

[0049] Based on the exposure heat map and the spatiotemporal cascade trajectory of the disaster chain, the chain risk stream function is used to couple the terrain boundary vector integral with the terrain wind field disturbance to obtain the risk propagation vector field.

[0050] The fire line intensity was calculated based on the remote sensing vegetation data and local tree species in the assessment area, and the direct loss rate was calculated based on the NDVI index before and after the disaster. A Weibull distribution disaster loss model was established based on the fire line intensity and direct loss rate to fit the historical disaster intensity loss data pairs. The scale parameters and shape parameters were calibrated through maximum likelihood estimation, and the maximum historical fire intensity in the region was taken as input into the Weibull distribution disaster loss model to obtain the initial resilience to fire.

[0051] Furthermore, the forest fire dynamic warning map analysis steps are as follows:

[0052] Based on the five-level risk zoning map and meteorological elements, the temporal convolution long short-term memory coupling architecture is used for analysis to obtain the spatiotemporal warning probability matrix;

[0053] Based on the distribution of vegetation types in the assessment area and the fire frequency in each month, a monthly vegetation fire risk benchmark curve was established to obtain a fire risk benchmark matrix;

[0054] Based on the fire risk benchmark matrix and the spatiotemporal warning probability matrix, the dynamic threshold of the four-color warning is obtained according to the threshold calculation equation;

[0055] Based on the spatiotemporal warning probability matrix, a dynamic forest fire warning map is obtained through dynamic threshold classification and geographic enhancement overlay analysis of four-color warning.

[0056] Furthermore, the cumulative probability analysis steps are as follows:

[0057] Potential evapotranspiration was obtained using the Thornthwaite method based on the monthly mean temperature in the meteorological data. The cumulative water surplus and deficit values ​​for each month were calculated chronologically for a fixed time scale based on the potential evapotranspiration and the monthly precipitation in the meteorological data. The three-parameter Log-logistic probability distribution model was used to fit the cumulative water surplus and deficit value sequence. The three key parameters of the Log-logistic distribution (scale parameter, shape parameter, and location parameter) were calculated using the L-moment method, and the fitted Log-logistic distribution cumulative distribution function was obtained. The cumulative probability was calculated from the fitted Log-logistic distribution cumulative distribution function.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] 1. The present invention collects multi-source heterogeneous data, constructs a chain characteristic field for the multi-source heterogeneous data, obtains the core driving factor tensor of the disaster chain, conducts a composite disaster chain coupling analysis based on the core driving factor tensor of the disaster chain, obtains the spatiotemporal cascade trajectory of the disaster chain, conducts dynamic risk assessment and grade division based on the spatiotemporal cascade trajectory of the disaster chain, and obtains a five-level risk zoning map. Combining the chain driving relationship between drought-high temperature-vegetation dryness-fire risk, the present invention adopts dynamic thresholds for risk grading, captures the dynamic impact of climate variation and vegetation phenology on fire risk, and integrates multi-source data such as remote sensing, meteorology, and geographic information. Based on the deep coupling of physical processes, the spatial distribution and vulnerability analysis of the disaster-prone body are integrated, and the risk zoning is consistent with the actual disaster.

[0060] 2. The present invention conducts intelligent early warning analysis based on the five-level risk zoning map to obtain a dynamic forest fire warning map. The integration of real-time weather forecasts and deep learning models can realize multi-scale dynamic early warning. This method introduces Log-logistic probability distribution, Copula dependency structure, cascade effect simulation algorithm, etc., to achieve breakthroughs from single factor evaluation to chain-driven analysis, from static statistics to dynamic simulation, and from data superposition to physical coupling, thereby improving the mechanistic rationality, spatiotemporal accuracy and decision-making support capabilities of forest fire risk assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. The following drawings are not intentionally scaled to the actual size, and the focus is on illustrating the main purpose of the present invention.

[0062] Figure 1 is a flow chart of the method of the present invention;

[0063] Figure 2 This is a flow chart of step S200 of the present invention;

[0064] Figure 3 This is a flow chart of step S500 of the present invention. DETAILED DESCRIPTION

[0065] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts also fall within the scope of protection of the present invention.

[0066] like Figure 1 As shown in the figure, the forest fire risk assessment method based on the composite chain disaster evolution mechanism includes the following steps:

[0067] Step S100: multi-source heterogeneous data collection, including remote sensing data, meteorological data, geographic information data and disaster loss data.

[0068] Specifically, the analysis of step S100 is as follows:

[0069] Remote sensing data include high temperature points, normalized difference vegetation index (NDVI), combustion ratio, surface temperature, vegetation moisture content, leaf area index (some data are sourced from remote sensing), etc.; meteorological data include temperature and humidity, precipitation, wind speed, monthly average temperature, daily precipitation, maximum temperature, relative humidity, ENSO index, monsoon index, etc.; geographic information data include slope, aspect, altitude, vegetation type, combustible load, leaf area index (some data are sourced from geographic information data), vegetation height, digital elevation, distribution of industrial facilities in residential areas, distribution of power grid facilities, distribution of transportation networks, distribution of agricultural systems, distribution of industrial facilities, etc.; disaster loss data include burned area, historical forest fire event data (including disaster loss intensity index, number of occurrences in each month, etc.), etc.

[0070] Step S200: constructing a chain feature field for multi-source heterogeneous data to obtain the core driving factor tensor of the disaster chain.

[0071] like Figure 2 Specifically, the analysis of step S200 is as follows:

[0072] The potential evapotranspiration is obtained based on the monthly mean temperature in the meteorological data using the Thornthwaite method. , where K is the sunshine hours correction factor, L is the annual heat index, , T represents the monthly average temperature, It represents a nonlinear temperature response parameter, which is calculated based on potential evapotranspiration and monthly precipitation in meteorological data. For a fixed time scale k (such as 3 months, 6 months or 12 months), the cumulative water surplus and deficit value of each month is calculated in chronological order. , where n represents the current month, i represents the cumulative sum index, and P represents monthly precipitation. This results in a cumulative water surplus and deficit sequence, which is then fitted using a three-parameter Log-logistic probability distribution model. This model effectively describes the common skewed distribution characteristics of water surplus and deficit values. Using all values ​​in the cumulative water surplus and deficit sequence, the L-moment method is used to calculate the three key parameters of the Log-logistic distribution: the scale parameter (which controls the degree of data dispersion), the shape parameter (which controls the direction of distribution skewness), and the location parameter (which controls the center position of the distribution). This yields the fitted Log-logistic cumulative distribution function. For each water surplus and deficit value in the cumulative water surplus and deficit sequence, the cumulative probability p is calculated from the fitted Log-logistic cumulative distribution function. This p-value represents the probability of a water surplus or deficit less than or equal to that cumulative water surplus or deficit value occurring under historical climate conditions.

[0073] Based on the cumulative probability, the cumulative probability p is mapped to the standard normal distribution space through logarithmic transformation to obtain the normalized weight. If p≤0.5, the normalized weight , if p>0.5, ;

[0074] The normalized weights are standardized to obtain the standardized precipitation evapotranspiration index SPE. , where C0, C1, C2, d1, d2, and d3 are all standardized constants. The standardized precipitation evapotranspiration index is adjusted by vegetation temperature to obtain the vegetation temperature-adjusted standardized precipitation evapotranspiration index. , where ω is the vegetation adjustment weight, k is the temperature difference sensitivity coefficient, Ts is the surface temperature, which comes from remote sensing data, T is the temperature, which comes from meteorological data, NDVI is the normalized difference vegetation index, which comes from remote sensing data, and is the ratio of the near-infrared band reflectance minus the red band reflectance to the sum of the two values, NDVI max is the maximum value of NDVI in the assessment area.

[0075] The Forest Fire Danger Index (FWI) is obtained based on meteorological data and geographic information data using the standard algorithm of the Canadian Forest Fire Weather Index system. The specific index is based on meteorological observations or reanalysis data (including daily precipitation, maximum temperature, relative humidity and wind speed) combined with vegetation combustible material category parameters, and is obtained through an internationally recognized standardized calculation process. Its methodological specifications were established by scholars such as Van Wagner in 1987 and have been used to this day. The relevant technical details are common industry standards and will not be repeated here.

[0076] A dynamic drought and fire risk coupling model was established based on the plant temperature-adjusted standardized precipitation evapotranspiration index and the forest fire risk index. The dynamic changes of drought and fire risk were mathematically coupled and the seasonal regulation of vegetation phenology was superimposed to obtain a spatiotemporal continuous drought and fire risk coupling field. The dynamic drought and fire risk coupling model includes:

[0077] ,in It is a drought-fire risk coupling field, (x, y) represents the geographic spatial coordinates, and the specific data collection input is spatial raster data, which contains geographic spatial coordinates. is the FWI time rate of change, is the seasonal adjustment coefficient, D0 is the annual cumulative day, the date number in a year.

[0078] The vegetation dry matter basis was obtained by morphological coupling and vegetation conversion calculation using leaf area index, vegetation height and normalized difference vegetation index NDVI. ,in Both represent biomass conversion rate, LAI and Hv represent leaf area index and vegetation height respectively, where leaf area index and vegetation height are geographic information data, and normalized difference vegetation index NDVI is remote sensing data.

[0079] The terrain adjustment factor is obtained by using the gravity radiation coupling algorithm based on the slope, aspect and altitude in the geographic information data. , where θ represents the slope, λ is the slope attenuation coefficient, and ϕ is the south azimuth. is the aspect angle, Aspect gain factor, z is the altitude.

[0080] The measured values ​​of vegetation moisture content were analyzed using the water stress response equation to obtain the fuel dryness , where VWC represents vegetation water content, which comes from remote sensing data, Indicates the saturated water content of vegetation. represents the three-month standardized precipitation evapotranspiration index, and exp represents the exponential function with the natural constant e as the base.

[0081] Based on the basic amount of vegetation dry matter, terrain adjustment factors, and combustible dryness, physical coupling multiplication operations are performed to obtain the spatially explicit coordinated combustible load.

[0082] Specifically, by multiplying the basic amount of vegetation dry matter with the terrain adjustment factor and the dryness of the combustibles, the spatially explicit coordinated combustible load is obtained. The spatial heterogeneity of the combustible load is quantified from three aspects: biomass basis (vegetation dry matter), terrain stress (gravity radiation coupling), and moisture conditions (dryness response), to achieve a refined assessment of the distribution of combustibles and provide a scientific basis for fire risk prediction, ecological management, etc. The current spatially explicit coordinated combustible load is the key link between the original data and terrain modulation. Through physical coupling multiplication operations, multi-source data are integrated to form a combustible load model with both biomass basis and environmental stress characteristics, providing standardized input parameters (such as fuel load and terrain factors) for the TFF function, so that the final output distribution matrix can more realistically reflect the spatial pattern of combustibles driven by terrain, and improve the spatial explicitness and mechanics of the model.

[0083] The terrain-combustible mutual feed field is obtained by modulating the spatial explicit collaborative fuel load through the TFF function. ,in represents the terrain fuel mutual feeding field, It represents the absolute value of the slope of the terrain data, μ represents the slope attenuation coefficient, which controls the impact of the slope on fuel, Asp represents the slope angle, and γ1 represents the slope sensitivity coefficient.

[0084] Based on the terrain fuel mutual feed field, tensor product fusion preprocessing is performed to obtain the spatiotemporal extension feature tensor. The spatiotemporal continuous drought fire risk coupling field is projected in Fourier frequency domain to obtain the time-frequency domain feature basis set. The spatiotemporal extension feature tensor and the time-frequency domain feature basis set are fused with total variation regularization optimization to obtain the multi-source spatiotemporal interpolation result. Specifically, , where GaFil represents the multi-source spatiotemporal interpolation result, F represents Fourier transform, and TV represents total variation regularization. Represents tensor product, feature fusion operation, Represents the L2 norm, Z represents the matrix to be solved, the interpolation target field, argmin represents the minimization solution, and finds the optimal solution. represents the regularization parameter, balancing data fidelity and smoothness, where represents the spatiotemporal extended feature tensor, and F(DS) represents the time-frequency domain feature basis set.

[0085] Quantitative analysis of the degree of baseline deviation of the slope of the assessment area is performed to obtain the terrain weight factor , 45° represents the base slope.

[0086] The terrain fuel mutual feedback field and the spatiotemporal continuous drought and fire risk coupling field are spatially fused, and the fusion data field of the spatial unified grid is obtained by correcting the terrain weight factor. , where max(TFF) is the maximum value of the spatiotemporal continuous drought-fire risk coupling field, and × represents the scalar pixel-by-pixel multiplication operation.

[0087] Perform dynamic spatiotemporal threshold determination on the fusion data field of the spatial unified grid to obtain the effective identification matrix , where σt represents the dynamic time threshold, σt=0.5-0.2×SPE, σs represents the dynamic space threshold, ,in Represents the slope modulus of terrain data.

[0088] Based on the slope angle characteristics, the slope calibration items of the spatiotemporal continuous drought and fire risk coupling field are superimposed to adjust the fire risk value in a directional manner to obtain the slope optimization field. Based on the slope optimization field and the effective identification matrix, logical multiplication screening is performed to obtain the final effective observation data set. , where Aspect represents the orientation angle of the slope, and sin(Aspect) represents the sinusoidal function transformation of the aspect angle.

[0089] Based on the multi-source spatiotemporal interpolation results, the smoothed basic spatial field is obtained by inverse Fourier transform and spatial convolution smoothing. , represents the inverse Fourier transform, which converts the frequency domain data back to the spatial domain. Gσ is the Gaussian smoothing kernel function, which controls the smoothing intensity. * represents the spatial convolution operation, which seamlessly integrates the final effective observation data set and the smoothed basic spatial field with the preset missing area marker in time and space to obtain a spatially continuous and seamless feature field. , Mask represents the preset missing area identifier, 0 represents the final valid observation dataset, 1 represents the multi-source spatiotemporal interpolation result area, the binary matrix defines the data coverage status, Mask=0, the area directly outputs the final valid observation dataset, Mask=1, the area outputs the source spatiotemporal interpolation result.

[0090] Based on the point fire data, spatial gridding is performed, and the assessment area is divided into grid units (size 1km×1km). Each fire point is assigned to the corresponding grid through geographic coordinate mapping and the grid fire point count is calculated. The grid count value is binary-encoded and converted into a basic binary map of 0 (no fire point) or 1 (fire point) using the sign function. The basic binary map is accumulated and summed year by year in the time dimension and the average is taken to generate a historical fire frequency map with a value range of [0, 1]. The frequency is then converted into a historical fire frequency map with a value range of [0, 1] using a preset threshold (the specific threshold is 0.25, indicating that fires have occurred in more than 25% of the years). Figure 2The high-risk fire areas are delineated and retained to obtain a high-risk fire distribution map. Finally, the high-risk fire distribution map is spatially smoothed, and small holes (small gaps inside the fire area) are filled through morphological closing operations (dilation followed by erosion), the boundaries are smoothed, and adjacent fire areas are connected. Finally, a binary fire distribution map with noise eliminated and spatial continuity is obtained, which is marked as a smoothed and continuous binary fire distribution map.

[0091] Based on the spatial continuous seamless feature field, nonlinear scale transformation is performed, the smooth continuous binary fire distribution map is transformed into the frequency domain, and a fusion operation is performed to obtain the core driving factor tensor of the disaster chain. ,in represents the 95% quantile of TFF when historical fires occurred, represents the fast Fourier transform, Represents a smooth continuous binary fire distribution map.

[0092] Specifically, potential evapotranspiration is calculated using the Thornthwaite method, combined with monthly precipitation to generate a cumulative water deficit sequence. The distribution parameters are then fitted using a log-logistic probability distribution model. Remote sensing and geographic information data are integrated to generate a plant-temperature-adjusted standardized precipitation evapotranspiration index and a forest fire risk index. A physical coupling algorithm is used to calculate spatially explicit coordinated fuel loads and implement terrain modulation. This step quantifies fundamental risk factors from three perspectives: meteorology, topography, and vegetation. This allows for the construction of a spatiotemporally continuous drought-fire risk coupling field and fuel load model, providing a physical mechanism for the evolution of complex disasters. The output data serves as the core input for subsequent analysis of disaster chain drivers and calculation of risk for hazard-bearing bodies.

[0093] Step S300: Perform composite disaster chain coupling analysis based on the core driving factor tensor of the disaster chain to obtain the spatiotemporal cascade trajectory of the disaster chain.

[0094] Specifically, the analysis of step S300 is as follows:

[0095] The core driving factor tensor of the disaster chain is subjected to channel separation operation to obtain the disaster chain dynamic derived standardized precipitation evapotranspiration index and the fire risk index change rate. Based on the disaster chain dynamic derived standardized precipitation evapotranspiration index, the drought probability v is calculated by the Logistic cumulative distribution function (using the drought threshold and steepness parameter). At the same time, based on the fire risk index change rate, the high temperature probability u is calculated by the Gaussian cumulative distribution function (combining the historical mean and standard deviation). ,in represents the disaster chain dynamics-derived standardized precipitation evapotranspiration index, Represents the rate of change of the fire risk index, Re represents the real part of the complex number, Represents the complex domain division operation, separates the physical channels of the CDFR tensor, and the Logistic cumulative distribution function is , k represents the curve steepness coefficient, which controls the probability change rate (>0), SPE0 represents the drought threshold, the center point of the Logistic distribution, and the Gaussian cumulative distribution function is , where u0 represents the historical average rate of change, the mean of the Gaussian distribution, σ represents the standard deviation of the rate of change, controls the width of the probability distribution, and erf represents the error function, the Gaussian cumulative distribution function.

[0096] Based on the drought probability and high temperature probability, a dependency structure analysis model is established to capture the nonlinear pattern and obtain the high temperature drought forest fire dependency structure parameter set. The dependency structure analysis model includes:

[0097] ,in Indicates the disaster chain triggering intensity, Represents the Copula function, parameter θ determines the high temperature-drought joint distribution, θ represents the Copula parameter, represents the extreme value dependence operator, which amplifies the extreme event dependence. , disaster chain triggering intensity , high temperature and drought joint distribution parameters θ and The extreme value dependency parameters constitute the high temperature drought forest fire dependency structure parameter set, where τ can trigger a forest fire alarm, and θ determines the dependency structure of drought and high temperature. Determine the interval where extreme events occur.

[0098] Based on the high temperature drought forest fire dependent structure parameter set and the core driving factor tensor of the disaster chain, the time-lag cross-correlation calculation and disaster trend analysis are carried out to obtain the disaster chain intensity field after atmospheric circulation modulation. , where H is the high temperature drought forest fire dependent structural parameter set, ENSO is the El Niño-Southern Oscillation Index, which is obtained by calculating the temperature anomaly in key equatorial Pacific sea areas (such as the Niño 3.4 area). The data is released in real time by NOAA and other agencies. The monsoon index is a measure of the monsoon circulation intensity and location. It is calculated by reanalyzing wind field data (such as the 850hPa zonal wind of ECMWFERA5) or pressure field gradient (such as the India-South China Sea surface pressure difference) and quantifying the monsoon circulation intensity and location through regional average calculation. The specific method for obtaining it is well known in the art and will not be described here. represents the time-delay cross-correlation calculation, represents the Laplace operator.

[0099] Construct disaster chain state vector based on defined high temperature, drought and wind speed components , of which the high temperature component , T is the current temperature, is the historical mean temperature and the standard deviation of the temperature during the same period, and the drought component , where SPI is the Standardized Precipitation Index, reflecting precipitation shortage, and VCI is the Vegetation Condition Index, reflecting vegetation stress. are weight coefficients, wind speed components , where v is the current speed, Indicates the critical wind speed for forest fire spread. Based on the disaster chain intensity field modulated by atmospheric circulation, the Bayesian dynamic network is used to analyze the conditional probability state transfer equation for the disaster chain state vector to obtain the dynamic disaster triggering probability map. ,in It represents the benchmark of high fire incidence, the mean value of X on historical fire days, Indicates the state fluctuation range, the X standard deviation of historical fire days, represents the prior probability of the kth climate state, represents the state distribution of the k-th climate state, a Gaussian probability density function.

[0100] Based on the dynamic disaster triggering probability map and terrain combustible mutual feedback field, the cascade dynamics simulation of the disaster chain is carried out through the cascade effect simulation algorithm to obtain the spatiotemporal cascade trajectory of the disaster chain. The cascade effect simulation algorithm is ,in represents the risk change rate, the risk evolves over time, Represents the cascade gain, controls the spreading speed, Represents terrain damping, which controls the terrain blocking effect. represents the gradient of the dynamic disaster triggering probability graph, represents the spatial gradient of the dynamic disaster triggering probability graph, ,in Represents the convolution kernel of the Sobel operator in the x direction, represents the discrete spatial convolution operator, and the spatiotemporal cascade trajectory of the disaster chain is obtained by time-integrating the risk change rate. .

[0101] Specifically, the core driving factor tensor of the disaster chain was separated, and the probabilities of drought and high temperature were calculated using logistic and Gaussian cumulative distribution functions. A high temperature-drought dependency structure model was established based on the Copula function, and the triggering intensity of the disaster chain was analyzed in combination with atmospheric circulation factors. The state vector of the disaster chain was constructed, and a dynamic disaster triggering probability map was generated through a Bayesian dynamic network. The spatiotemporal cascade trajectory of the disaster chain was simulated using a cascade effect simulation algorithm. This step captures the nonlinear coupling relationship between drought and high temperature, quantifies the dependency intensity of extreme events, and achieves an upgrade in risk analysis from a single factor to a chain-like evolution. The output provides a spatiotemporal risk foundation for subsequent risk propagation analysis and early warning models.

[0102] Step S400: Perform dynamic risk assessment and level classification based on the spatiotemporal cascade trajectory of the disaster chain to obtain a five-level risk zoning map.

[0103] Specifically, the analysis of step S400 is as follows:

[0104] Extract the original data of five types of hazard-prone bodies based on geographic information data. The five types of hazard-prone bodies include residential industrial facilities (population clusters), power grid facilities (substations / transmission towers), transportation networks (highways / railways), agricultural systems (farmland / irrigation facilities), and industrial facilities (factories / oil storage tanks). Obtain the distribution density of hazard-prone bodies. i (the number of disaster-prone bodies per unit area), the spatial location of the disaster-prone body Loc i (Coordinates are extracted from GIS vector data). Based on the spatiotemporal cascade trajectory of the disaster chain, the distribution density of the hazard-bearing body, and the spatial position of the hazard-bearing body, the Gaussian attenuation function formula is used for spatial superposition calculation to obtain the exposure heat map. ,in represents the weight of each disaster-prone body, Indicates the risk attenuation radius and controls the risk impact range. Represents the Euclidean distance, the distance from the risk source to the disaster-stricken body.

[0105] Based on the exposure heat map and the spatiotemporal cascade trajectory of the disaster chain, the terrain boundary vector integral and the terrain wind field disturbance coupling calculation are performed through the chain risk stream function to obtain the risk propagation vector field. Specifically, , where CRF represents the risk propagation vector field, represents the integral along the watershed boundary and the terrain unit, represents the boundary direction vector, Represents the wind speed vector, which comes from meteorological data, div represents the divergence operator, a mathematical operation, Represents the digital elevation of geographic information data, Represents the terrain-wind field coefficient, which controls the terrain modulation of the wind field. Represents a tensor product.

[0106] The fire line intensity (IB) was calculated based on remote sensing vegetation data and local tree species in the assessment area. IB=0.007×FL×H, where the fuel load FL was obtained by inverting vegetation biomass using the Landsat-8 infrared vegetation index (NDVI), and the effective combustion heat H was determined in the laboratory based on the calorific value database of local tree species samples.

[0107] The direct loss rate λfire was calculated based on the NDVI index before and after the disaster, where the direct loss rate = (1-NDVI index after the disaster / NDVI index before the disaster).

[0108] Based on the fire line intensity IB and direct loss rate, a Weibull distribution disaster loss model is established to fit the historical disaster intensity loss data. The model includes: , where η and α1 are scale parameters and shape parameters respectively. The scale parameters and shape parameters are calibrated by maximum likelihood estimation. The maximum historical fire intensity in the region is input into the Weibull distribution disaster loss model to obtain the initial resilience to fire , IBmax represents the maximum historical fire intensity in the region.

[0109] The rescue path is discretized into several line segments. An anisotropic path cost model is established based on the forest fire risk index (FWI) and landslide risk rate of each line segment to conduct real-time spatial superposition analysis of fire risk and landslides, and the travel time of the minimum cost path is obtained. The anisotropic path cost model includes:

[0110] , where s is the path integral variable, vmax is the maximum speed of the rescue vehicle, F0 is the fire hazard threshold, LSR(s) is the landslide risk rate, LSR = slope × soil moisture / rock mass strength, α and β are the fire risk weight coefficient and landslide weight coefficient respectively, path is the continuous trajectory of the rescue path to be evaluated in space, cost is the time spent, and the minimum value of each path is obtained to obtain the travel time of the minimum cost path.

[0111] Based on the travel time of the minimum cost path, the initial resilience against fire and the risk propagation vector field, a resilience attenuation integral model is established to conduct time-space integral analysis and obtain a dynamic resilience attenuation surface. , where RDS(t) represents the dynamic toughness decay surface, Indicates time integral, cumulative effect calculation, represents the response efficiency coefficient, represents the travel time of the minimum cost path, represents the efficiency function and the emergency response saturation curve;

[0112] Based on historical forest fire disaster data, the risk benchmark threshold for each month is established through inverse probability transformation , where m∈{1, 2, ..., 12} represents the month, N m is the total number of forest fire events that occurred in month m in the historical data, is the disaster loss intensity index of the kth event, , It is the inverse function of the standard normal distribution cumulative distribution function. It integrates real-time meteorological anomaly data and vegetation drought index, dynamically adjusts the benchmark threshold, and obtains the monthly dynamic threshold. , where Tano is the monthly temperature anomaly, Pano is the monthly precipitation anomaly, σT and σP are the standard deviations of temperature and precipitation in the same period of history, κ2 and λ2 are climate sensitivity coefficients, VDI is the vegetation drought index, and γ3 is the vegetation sensitivity coefficient.

[0113] The risk propagation vector field and the dynamic resilience attenuation surface are analyzed by coupling the spatial gradient and the monthly dynamic threshold through the risk level reconstruction algorithm to obtain a five-level risk zoning map. ,in It represents a five-level risk zoning map. The final risk classification is 1-5 levels. Sigmoid represents the normalization function. Specifically, [0.0, 0.2), [0.2, 0.4), [0.4, 0.6), [0.6, 0.8), and [0.8, 1.0] correspond to levels 1 to 5 respectively. ∇ represents the gradient, and max represents the maximum value.

[0114] Specifically, data from five types of hazard-prone objects were extracted and combined with a Gaussian attenuation function to calculate exposure heat maps. Disaster losses were assessed based on a risk propagation vector field and a fire line intensity model. A Weibull distribution disaster loss model was established, and a resilience decay integral model was constructed using rescue path cost analysis. Historical and real-time data were integrated to dynamically adjust monthly risk thresholds and generate a five-level risk zoning map. This step couples physical risk with the vulnerability of hazard-prone objects, quantifies the impact of disasters on social-ecological systems, and considers emergency response resilience, achieving a comprehensive assessment from risk source to actual losses. The resulting risk zoning map and loss model provide a spatial benchmark and historical reference for early warning models.

[0115] Step S500: Perform intelligent early warning analysis based on the five-level risk zoning map to obtain a dynamic forest fire early warning map.

[0116] like Figure 1 Specifically, the analysis of step S500 is as follows:

[0117] Based on the five-level risk zoning map and meteorological elements, the temporal convolution long short-term memory coupling architecture is used for analysis, and the spatiotemporal warning probability matrix WarnMat=RiskLevel⊕ConvLSTM(ERA5 fcst ), where ⊕ represents a tensor concatenation operation (aligning the risk field and the weather forecast field in a grid and merging them into a high-dimensional tensor), ConvLSTM represents a weather forecast deep learning model, and outputs a probability value [0, 1], ERA5 fcst Represents ERA5 weather forecast data (including temperature, humidity, wind speed, etc.);

[0118] Based on the distribution of vegetation types in the assessment area and the fire frequency in each month, a monthly vegetation fire risk benchmark curve is established to obtain a fire risk benchmark matrix. The monthly vegetation fire risk benchmark curve includes:

[0119] ,in Represents the fire risk benchmark matrix, v represents vegetation type, coniferous forest, shrub, grassland, m represents month, January to December, y represents year, the historical data of the past 10 years, represents the number of historical fires, α(v) is the vegetation flammability coefficient, coniferous forest = 0.85, grassland = 0.65, shrub = 0.75.

[0120] Based on the fire risk benchmark matrix and the spatiotemporal warning probability matrix, the dynamic threshold of the four-color warning is obtained according to the threshold calculation equation. ,in represents the dynamic threshold of the four-color warning, u represents the annual average risk value, ,in , where N represents the total number of spatial points in the spatiotemporal warning probability matrix, k represents the point, Indicates spatial point In time The risk probability, σ1 represents the risk volatility coefficient, .

[0121] Based on the spatiotemporal warning probability matrix, a dynamic forest fire warning map is obtained through dynamic threshold classification and geographic enhancement superposition analysis of the four-color warning. ,in It represents the dynamic warning map of forest fire, and E[WarnMat] represents the mathematical expectation of the “space-time warning probability matrix WarnMat”.

[0122] Specifically, a ConvLSTM model was used to couple a five-level risk zoning map with meteorological forecast data to generate a spatiotemporal warning probability matrix. A monthly fire risk baseline curve was established based on vegetation type and historical fire frequency, dynamically adjusting the four-color warning threshold. Finally, a four-color warning map was generated through geographic enhancement and overlay analysis. This step integrates historical patterns with real-time forecasts to achieve multi-scale dynamic warnings, improving the timeliness and spatial accuracy of risk prediction. The resulting dynamic forest fire warning map can be directly used for emergency decision-making.

[0123] The above is an illustration of the present invention and should not be considered as limiting thereof. Although several exemplary embodiments of the present invention have been described, it will be readily understood by those skilled in the art that many modifications may be made to the exemplary embodiments without departing from the novel teachings and advantages of the present invention. Therefore, all such modifications are intended to be included within the scope of the present invention as defined by the claims. It should be understood that the above is an illustration of the present invention and should not be considered as being limited to the specific embodiments disclosed, and modifications to the disclosed embodiments and other embodiments are intended to be included within the scope of the appended claims. The present invention is defined by the claims and their equivalents.

Claims

1. A forest fire risk assessment method based on a composite chain disaster evolution mechanism is characterized by: The following steps are involved: Multi-source heterogeneous data collection, including remote sensing data, meteorological data, geographic information data, and disaster loss data; Constructing a chain feature field for the multi-source heterogeneous data to obtain a core driving factor tensor of the disaster chain; The steps for tensor analysis of the core driving factors of the disaster chain are as follows: Based on the multi-source spatiotemporal interpolation results, the smoothed basic spatial field is obtained by inverse Fourier transform and spatial convolution smoothing. The final effective observation data set and the smoothed basic spatial field are seamlessly fused with the preset missing area markers in spatiotemporal order to obtain a spatially continuous and seamless feature field. Based on the point fire data, spatial gridding is performed, dividing the assessment area into grid cells. Each fire point is assigned to the corresponding grid through geographic coordinate mapping and the grid fire point count is calculated. The grid count value is binary-encoded and converted into a basic binary map using the sign function. The basic binary map is accumulated and summed year by year in the time dimension and the average is taken to obtain a historical fire frequency map. The frequency map is binarized using a preset threshold, and high-risk fire areas are delineated and retained to obtain a high-risk fire distribution map. The high-risk fire distribution map is spatially smoothed and adjacent fire areas are connected through morphological closing operations to obtain a smooth and continuous binary fire distribution map. Based on the spatial continuous seamless feature field, nonlinear scaling transformation is performed, the smoothed continuous binary fire distribution map is transformed into the frequency domain, and a fusion operation is performed to obtain the core driving factor tensor of the disaster chain; Based on the core driving factor tensor of the disaster chain, a composite disaster chain coupling analysis is performed to obtain the spatiotemporal cascade trajectory of the disaster chain; The steps for analyzing the spatiotemporal cascade trajectory of the disaster chain are as follows: Based on the high temperature drought forest fire dependent structure parameter set and the core driving factor tensor of the disaster chain, time-lag cross-correlation calculation and disaster trend analysis were performed to obtain the disaster chain intensity field after atmospheric circulation modulation. Based on the defined high temperature, drought and wind speed components, the disaster chain state vector is constructed. Based on the disaster chain intensity field modulated by atmospheric circulation, the conditional probability state transition equation of the disaster chain state vector is analyzed through the Bayesian dynamic network to obtain a dynamic disaster triggering probability map. Based on the dynamic disaster triggering probability map and terrain fuel mutual feedback field, the cascade dynamics simulation of the disaster chain is carried out through the cascade effect simulation algorithm to obtain the spatiotemporal cascade trajectory of the disaster chain; Based on the spatiotemporal cascade trajectory of the disaster chain, dynamic risk assessment and level classification are carried out to obtain a five-level risk zoning map; Based on the five-level risk zoning map, intelligent early warning analysis is performed to obtain a dynamic forest fire early warning map.

2. The forest fire risk assessment method based on the composite chain disaster evolution mechanism according to claim 1 is characterized in that: The steps for analyzing the final valid observation data set are as follows: The spatially explicit collaborative fuel load is modulated by the terrain through the TFF function to obtain the terrain fuel mutual feedback field; Based on the terrain fuel mutual feed field, tensor product fusion preprocessing is performed to obtain the spatiotemporal extension feature tensor. The spatiotemporal continuous drought and fire risk coupling field is projected into the Fourier frequency domain to obtain the time-frequency domain feature basis set. The spatiotemporal extension feature tensor and the time-frequency domain feature basis set are fused with total variation regularization optimization to obtain the multi-source spatiotemporal interpolation result. Quantitative analysis of the degree of baseline deviation of the slope of the assessment area was performed to obtain the terrain weight factor; The terrain fuel mutual feedback field and the spatiotemporal continuous drought and fire risk coupling field are spatially fused, and the fused data field with a spatial unified grid is obtained by correcting the terrain weight factor. Perform dynamic spatiotemporal threshold determination on the fused data field of the spatial unified grid to obtain a valid identification matrix; Based on the slope angle characteristics, the fire risk value is directional adjusted by superimposing the slope calibration items of the spatiotemporal continuous drought and fire risk coupling field to obtain the slope optimization field. Logical multiplication screening is performed based on the slope optimization field and the valid identification matrix to obtain the final valid observation data set.

3. The forest fire risk assessment method based on the composite chain disaster evolution mechanism according to claim 2 is characterized in that: The steps of the spatially explicit collaborative fuel load analysis are as follows: Based on the cumulative probability, the cumulative probability is mapped to the standard normal distribution space through logarithmic transformation to obtain the normalized weight; The normalized weights are normalized to obtain the standardized precipitation evapotranspiration index, which is then adjusted by vegetation temperature to obtain the plant temperature-adjusted standardized precipitation evapotranspiration index. The forest fire danger index is obtained based on meteorological data and geographic information data using the standard algorithm of the Canadian Forest Fire Weather Index System; A dynamic drought and fire risk coupling model was established based on the plant temperature-adjusted standardized precipitation evapotranspiration index and the forest fire risk index. The changing dynamics of drought and fire risk were mathematically coupled and the seasonal regulation of vegetation phenology was superimposed to obtain a spatiotemporal continuous drought and fire risk coupling field. The morphological coupling and vegetation conversion calculations were performed through leaf area index, vegetation height and normalized difference vegetation index to obtain the basic amount of vegetation dry matter. Based on the slope, aspect and altitude in the geographic information data, the gravity radiation coupling algorithm is used to obtain the terrain adjustment factor; The measured values ​​of vegetation moisture content were analyzed using the water stress response equation to obtain the fuel dryness; Based on the basic amount of vegetation dry matter, terrain adjustment factors, and combustible dryness, physical coupling multiplication operations are performed to obtain the spatially explicit coordinated combustible load.

4. The forest fire risk assessment method based on the composite chain disaster evolution mechanism according to claim 1 is characterized in that: The steps for analyzing the high temperature drought forest fire dependent structure parameter set are as follows: Channel separation is performed on the core driving factor tensor of the disaster chain to obtain the disaster chain dynamic derived standardized precipitation evapotranspiration index and the fire risk index change rate. The drought probability is calculated based on the disaster chain dynamic derived standardized precipitation evapotranspiration index through the Logistic cumulative distribution function, while the high temperature probability is calculated based on the fire risk index change rate through the Gaussian cumulative distribution function. Based on the probability of drought and high temperature, a dependency structure analysis model was established to capture the nonlinear pattern and obtain the dependency structure parameter set of high temperature drought forest fire.

5. The forest fire risk assessment method based on the composite chain disaster evolution mechanism according to claim 1 is characterized in that: The analysis steps of the five-level risk zoning map are as follows: The rescue path is discretized into several line segments. An anisotropic path cost model is established based on the forest fire risk index and landslide risk rate of each line segment to conduct real-time spatial superposition analysis of fire risk and landslide, and the travel time of the minimum cost path is obtained. Based on the travel time of the minimum cost path, the initial resilience against fire, and the risk propagation vector field, a resilience attenuation integral model is established for time-space integration analysis to obtain a dynamic resilience attenuation surface. Based on historical forest fire disaster data, the monthly risk benchmark threshold is established through inverse probability transformation, and the benchmark threshold is dynamically adjusted to obtain the monthly dynamic threshold; The risk propagation vector field and the dynamic resilience attenuation surface are analyzed by coupling spatial gradient and monthly dynamic threshold through the risk level reconstruction algorithm to obtain a five-level risk zoning map.

6. The forest fire risk assessment method based on the composite chain disaster evolution mechanism according to claim 5 is characterized in that: The steps for analyzing the initial resilience and risk propagation vector field against fire are as follows: Extract the original data of five types of hazard-prone bodies based on geographic information data, obtain the distribution density and spatial location of the hazard-prone bodies, and perform spatial superposition calculation based on the spatiotemporal cascade trajectory of the disaster chain, the distribution density and spatial location of the hazard-prone bodies using the Gaussian attenuation function formula to obtain the exposure heat map. Based on the exposure heat map and the spatiotemporal cascade trajectory of the disaster chain, the chain risk stream function is used to couple the terrain boundary vector integral with the terrain wind field disturbance to obtain the risk propagation vector field. The fire line intensity was calculated based on the remote sensing vegetation data and local tree species in the assessment area, and the direct loss rate was calculated based on the NDVI index before and after the disaster. A Weibull distribution disaster loss model was established based on the fire line intensity and direct loss rate to fit the historical disaster intensity loss data pairs. The scale parameters and shape parameters were calibrated through maximum likelihood estimation, and the maximum historical fire intensity in the region was taken as input into the Weibull distribution disaster loss model to obtain the initial resilience to fire.

7. The forest fire risk assessment method based on the composite chain disaster evolution mechanism according to claim 1 is characterized in that: The steps for analyzing the forest fire dynamic early warning map are as follows: Based on the five-level risk zoning map and meteorological elements, the temporal convolution long short-term memory coupling architecture is used for analysis to obtain the spatiotemporal warning probability matrix; Based on the distribution of vegetation types in the assessment area and the fire frequency in each month, a monthly vegetation fire risk benchmark curve was established to obtain a fire risk benchmark matrix; Based on the fire risk benchmark matrix and the spatiotemporal warning probability matrix, the dynamic threshold of the four-color warning is obtained according to the threshold calculation equation; Based on the spatiotemporal warning probability matrix, a dynamic forest fire warning map is obtained through dynamic threshold classification and geographic enhancement overlay analysis of four-color warning.

8. The forest fire risk assessment method based on the composite chain disaster evolution mechanism according to claim 3 is characterized in that: The cumulative probability analysis steps are as follows: Potential evapotranspiration was obtained using the Thornthwaite method based on the monthly mean temperature in the meteorological data. The cumulative water surplus and deficit values ​​for each month were calculated chronologically for a fixed time scale based on the potential evapotranspiration and the monthly precipitation in the meteorological data. The three-parameter Log-logistic probability distribution model was used to fit the cumulative water surplus and deficit value sequence. The three key parameters of the Log-logistic distribution, including the scale parameter, shape parameter, and location parameter, were calculated using the L-moment method to obtain the fitted Log-logistic distribution cumulative distribution function. The cumulative probability was calculated from the fitted Log-logistic distribution cumulative distribution function.

Images