Method for constructing dynamic claim threshold of agricultural insurance based on disaster-causing vulnerability coupling

Abstract

The present application relates to the technical field of insurance claim threshold construction, and particularly relates to a kind of agricultural insurance dynamic claim threshold construction method based on disaster-causing-vulnerability coupling;Actual claim loss result is used as the actual observation of crop disaster loss in the present application, and the response relationship between meteorological disaster intensity and crop loss rate is inverted;The present scheme avoids the uncertainty caused by traditional experience assignment or test extrapolation, solves the problem that traditional research pays more attention to statistical correlation and less to machine interpretation;The present application considers the dynamic coupling analysis of disaster-causing-vulnerability of regional and temporal heterogeneity, constructs a dynamic coupling analysis method of disaster-causing-vulnerability containing temporal (monthly / growth period) and spatial (township level) heterogeneity, reveals the synergistic influence mechanism of disaster intensity change and crop system vulnerability evolution on claim loss formation, greatly improves the spatiotemporal adaptability, and deepens the basic understanding of the formation mechanism of agricultural meteorological disaster risk.

Description

Technical Field

[0001] This invention relates to the field of insurance claim threshold construction technology, specifically to a method for constructing dynamic claim thresholds for agricultural insurance based on disaster-vulnerability coupling. Background Technology

[0002] Currently, agricultural insurance meteorological claims generally adopt static thresholds and single-disaster type determination. This approach has the following main shortcomings: First, the static nature of insurance claim thresholds fails to consider the sensitivity of crop growth stages and topographical heterogeneity, making it unsuitable for complex terrains and differences in growth stages, easily leading to situations of "different compensation for the same disaster" and "less compensation for severe disasters." Second, it only establishes a simple statistical relationship between meteorological elements and loss rates, failing to reveal the chain mechanism of meteorological stress-crop physiological response-economic loss. Third, it cannot identify compound disasters; there are no quantitative determination methods for compound disasters such as drought-high temperature and rainstorm-low temperature, making it impossible to identify the superimposed effects. Fourth, it lacks a dynamic update mechanism, making it impossible to iteratively optimize with climate background, crop layout, and claims data. In summary, existing technologies cannot meet the refined, precise, and mechanism-based claims needs of agricultural insurance, meaning that existing agricultural claims solutions suffer from technical problems such as inaccurate loss assessment and poor spatiotemporal adaptability. Summary of the Invention

[0003] The main objective of this invention is to provide a method for constructing dynamic claims thresholds for agricultural insurance based on disaster-vulnerability coupling, aiming to solve the technical problems of inaccurate loss assessment and poor spatiotemporal adaptability in existing agricultural claims schemes.

[0004] The technical solution proposed in this invention is as follows: A method for constructing dynamic claims thresholds for agricultural insurance based on disaster hazard-vulnerability coupling includes: Acquire meteorological monitoring data, agricultural insurance claims data, geographical and topographical data, and crop growth stage data to construct an integrated spatiotemporal database of region, disaster type, growth stage, meteorological intensity, and loss rate. The disaster types include drought, rainstorm, heat damage, and cold damage. Based on the database, principal component analysis and machine learning algorithms are used to extract core disaster-causing factors. Entropy weight-analytic hierarchy process algorithm is used to combine and assign weights to the core disaster-causing factors in order to construct a comprehensive disaster-causing intensity index. Based on the comprehensive disaster intensity index and the claim loss rate as the actual loss observation, a multi-level Bayesian model is constructed, and inversion is performed using a zero-inflation model to obtain the crop vulnerability function. Based on the crop vulnerability function, a disaster-vulnerability coupling degree and coupling coordination degree model is constructed to quantify the synergistic damage effect between the intensity of meteorological disasters and crop vulnerability. Based on the disaster-vulnerability coupling degree and coupling coordination degree model, the initial claim threshold and spatiotemporal heterogeneity correction, a three-dimensional dynamic claim threshold matrix is ​​generated by region, disaster type and reproductive stage, and the claim threshold is determined based on the matrix.

[0005] Preferably, the acquisition of meteorological monitoring data, agricultural insurance claims data, geographical topographic data, and crop growth stage data to construct an integrated spatiotemporal database of region-disaster type-growth stage-meteorological intensity-loss rate includes: The Euclidean distance method was used to match the nearest meteorological monitoring station to the disaster area, and meteorological monitoring data collected by the meteorological monitoring station was obtained. The meteorological monitoring data were subjected to climatological boundary value test, time series consistency test, and spatial interpolation rationality test in sequence. For meteorological monitoring stations with missing values ​​less than the preset percentage, the meteorological monitoring data were supplemented by Kriging interpolation algorithm or neighboring station mean algorithm. Logical consistency checks and on-site cross-validation are performed on agricultural insurance claims data to eliminate erroneous data, and the claims amount in the agricultural insurance claims data is converted into the claims loss rate. The time scale is divided according to the type of disaster: for rainstorms, hourly data is used, with the time scale being the first preset number of days before the claim date; for drought, the time scale is the second preset number of days before the claim date; for cold damage, the time scale is the third preset number of days before the claim date; and for heat damage, the time scale is the fourth preset number of days before the claim date. Using ArcGIS spatial analysis tools, meteorological monitoring data collected from meteorological monitoring stations are interpolated to the township scale through an inverse distance weighted interpolation algorithm, realizing a township-level gridded expression of meteorological disaster intensity. This is then matched with the township spatial attributes of agricultural insurance claims data to form an integrated spatiotemporal database of region, disaster type, reproductive stage, meteorological intensity, and loss rate.

[0006] Preferably, the step of extracting core disaster-causing factors based on the database using principal component analysis and machine learning algorithms includes: Machine learning and principal component analysis algorithms were used to extract core disaster-causing factors for different types of disasters from raw meteorological monitoring data. These core factors were then screened based on out-of-bag error and feature importance to construct a graded and quantitative meteorological disaster intensity index system. The core disaster-causing factors for rainstorms included: The rainfall, maximum hourly rainfall, and number of days of continuous heavy rain are calculated based on the fourth preset number of days prior to the claim date. The core disaster-causing factors of drought include: The daily average relative humidity, average temperature, daily precipitation, soil relative humidity, number of consecutive days without effective precipitation, and sunshine hours are calculated for the first preset number of days prior to the claim date. The core disaster-causing factors of cold damage include: the daily minimum temperature two preset days prior to the claim date, the number of days each year when the daily minimum temperature is lower than the first preset temperature, the number of consecutive days of low temperature, and the accumulated temperature anomaly. The core disaster-causing factors of heat damage include: the highest daily temperature three preset days prior to the claim date, the average relative humidity, the longest consecutive number of days greater than or equal to the second preset temperature, the number of days greater than or equal to the second preset temperature, the effective accumulated high temperature, precipitation, and sunshine hours.

[0007] Preferably, the step of using the entropy-weighted analytic hierarchy process (AHP) algorithm to combine and weight the core disaster-causing factors to construct a comprehensive disaster-causing intensity index includes: The core disaster-causing factors are combined and weighted using an entropy weight algorithm and an analytic hierarchy process (AHP) algorithm to construct a comprehensive disaster-causing intensity index. The consistency ratio of the weight results of the AHP algorithm must be less than or equal to a first preset value. The percentile method was used to classify the intensity of meteorological disasters to determine each intensity level and its corresponding threshold range. The intensity levels of meteorological disasters included four categories: mild, moderate, severe, and extremely severe. The threshold range for the mild level was as follows: The grading threshold range corresponding to the medium level is: The grading threshold range corresponding to the heavy level is: The grading threshold corresponding to the extremely serious level is in the range of [range missing]. .

[0008] Preferably, the step of classifying the intensity of meteorological disasters using the percentile method to determine each intensity level and its corresponding classification threshold range further includes: The interannual variation trend of disaster types is analyzed using trend analysis algorithm, wavelet analysis algorithm, spatial clustering analysis algorithm, and Mann-Kendall trend test algorithm. Spatial variability function is used to analyze the spatial variation characteristics of disaster types in order to determine spatial correlation distances; Meteorological monitoring data from different monitoring areas were obtained, and the disaster-causing characteristics of four types of disasters in each monitoring area were analyzed from three dimensions: time, space, and intensity, in order to determine the occurrence frequency, spatiotemporal distribution pattern, and intensity evolution characteristics of the types of disasters in different monitoring areas. Based on the disaster-causing factor-disaster-loss theory of agricultural disaster science, this study uses literature review combined with structural equation modeling to identify the vulnerability response paths of different crops to different types of disasters. This study analyzes the formation and evolution of crop vulnerability from four dimensions: plant physiological vulnerability, growth and development vulnerability, yield and quality vulnerability, and economic vulnerability. By using structural equation modeling to quantify the direct and indirect effects between core disaster-causing factors, four dimensions of crop vulnerability, and claims loss rate, the mediating role of crop vulnerability in the relationship between disaster-causing factors and claims loss rate is clarified.

[0009] Preferably, the step of quantifying the direct and indirect effects between core disaster-causing factors, four dimensions of crop vulnerability, and claims loss rate using structural equation modeling, clarifying the mediating role of crop vulnerability between disaster-causing factors and claims loss rate, further includes: Using bivariate correlation analysis, nonlinear regression analysis, and quantile regression, we analyzed the nonlinear relationships between the intensity of meteorological disasters and crop vulnerability, and between crop vulnerability and claim loss rate, under different meteorological disaster intensities and different growth stages. The interaction effect analysis method was used to examine the influence of the interaction between crop type, disaster type, and growth stage on the coupling relationship, and the analysis of variance method was used to examine the significance of the interaction effect. Through group comparative analysis, the differential characteristics of crops in the disaster-vulnerability coupling process are revealed, and a framework for analyzing the disaster-vulnerability coupling mechanism is constructed. The rationality of the coupling mechanism framework is verified by structural equation modeling and path coefficients. The intensity of meteorological disaster, the vulnerability of crops in four dimensions, and the loss rate of claims are used as latent variables. The coefficients and significance of each path are quantified to form a quantifiable and verifiable nonlinear coupling mechanism framework of disaster-vulnerability.

[0010] Preferably, the step of constructing a multi-level Bayesian model based on the comprehensive disaster intensity index and the claim loss rate as the actual loss observation, and then combining it with a zero-inflation model for inversion, to obtain the crop vulnerability function, includes: Standardize the intensity of meteorological disasters: Standardize the comprehensive intensity index of each type of disaster from 0 to 1 to eliminate the difference in dimensions and obtain the standardized intensity of meteorological disasters. Standardize the physical loss rate of crops: Standardize the extreme values ​​of the actual crop loss rate at the township level in the disaster-stricken area; Using exploratory data analysis algorithms, by plotting scatter plots and fitting curve trends, we can preliminarily determine the type of response relationship between the intensity of meteorological disasters and the rate of physical loss of crops. Group analysis was conducted for different crops, different types of disasters, and different growth stages to clarify the response relationship characteristics under different scenarios; The distribution characteristics of crop physical loss rate were analyzed using the kernel density estimation algorithm. Then, the Pearson correlation coefficient and Spearman correlation coefficient were used to determine the type and significance of the correlation between the intensity of meteorological disaster and crop physical loss rate.

[0011] Preferably, the step of analyzing the distribution characteristics of crop physical loss rate using a kernel density estimation algorithm, then using Pearson correlation coefficient and Spearman correlation coefficient to determine the type and significance of the correlation between the intensity of meteorological disaster and crop physical loss rate, further includes: Selection of crop vulnerability function type: Nonlinear models are selected as candidate crop vulnerability functions, with standardized meteorological disaster intensity as independent variable and standardized crop physical loss rate as dependent variable. The candidate crop vulnerability functions are fitted and screened to obtain the preferred crop vulnerability function. Mathematical Construction and Probabilistic Expression of Crop Vulnerability Function: This involves calculating the actual claim loss rate of township i during its reproductive stage t. Considered as the intensity of disaster caused by meteorological disasters Potential crop vulnerability status According to Bayes' theorem, the posterior distribution of crop vulnerability, driven by a co-driven stochastic process, can be expressed as: , In the formula, P Represents the probability distribution function; This indicates the actual claim loss rate during the maternity stage in rural areas. Intensity of meteorological disasters Under certain conditions, the potential vulnerability status of crops The posterior probability distribution; This represents the intensity of a given meteorological disaster in township i during the reproductive stage t. Potential crop vulnerability status Under these conditions, the actual claim loss rate was observed. The likelihood function; Indicates the potential vulnerability status of crops The prior probability distribution; These parameters represent the relevant parameters of crop growth stages, and are used to characterize the prior features of crop sensitivity to meteorological disasters at different growth stages. Crop vulnerability is classified into four levels: low, medium, relatively high, and high, based on the calculation results of the optimized crop vulnerability function, so as to realize the quantitative and hierarchical expression of crop vulnerability. A sample segmentation algorithm was used to verify the accuracy of the selected crop vulnerability function.

[0012] Preferably, the step of constructing a disaster-vulnerability coupling degree and coupling coordination degree model based on the crop vulnerability function to quantify the synergistic damaging effect of meteorological disaster intensity and crop vulnerability includes: Characterizing temporal heterogeneity: Using the moving average algorithm and the trend coefficient algorithm, we analyze the changing trends and seasonal fluctuations of the monthly meteorological disaster intensity and crop vulnerability; Characterizing spatial heterogeneity: Using ArcGIS spatial analysis tools, spatial autocorrelation analysis algorithm is used to examine the spatial clustering characteristics of disaster-causing capacity and crop vulnerability at the township level, and geographic weighted regression algorithm is used to quantify the contribution of geographic elements to spatial heterogeneity. Analyze the single-factor and interaction explanatory power of geographical elements on the spatial heterogeneity of disaster-vulnerability; Coupling Degree Model Construction: A coupling coordination degree model is introduced, which treats the intensity of meteorological disasters and the vulnerability of crops as two coupled subsystems. A disaster-vulnerability coupling degree and coupling coordination degree model is constructed to calculate the coupling degree and coupling coordination degree at the township level and on a monthly basis. Analysis of dynamic evolution patterns: Using a centroid migration model, we analyze the migration direction and distance of the centroid of disaster-vulnerability coupling degree and coupling coordination degree over time. A panel regression model was adopted, with the township-level insurance claim loss rate as the dependent variable and the interaction term between the intensity of meteorological disasters and crop vulnerability as the core explanatory variable. Control variables were introduced to construct a fixed-effects panel model to quantify the contribution of the intensity of meteorological disasters, crop vulnerability, and the interaction to insurance losses. The formula for defining the disaster-vulnerability coupling degree and coupling coordination degree model is as follows: , In the formula, C represents the disaster-vulnerability coupling degree and coupling coordination degree model; H represents the disaster-causing intensity of meteorological disasters; and V represents the crop vulnerability index. Synergistic effect quantification: The marginal effect analysis algorithm is used to quantify the synergistic damage effect of meteorological disaster intensity and crop vulnerability at different levels.

[0013] Preferably, the step of generating a three-dimensional dynamic claim threshold matrix based on the disaster-vulnerability coupling degree and coupling coordination degree model, the initial claim threshold, and spatiotemporal heterogeneity correction, and determining the claim threshold based on the matrix, includes: A three-dimensional matrix containing thresholds for single disaster types and joint thresholds for typical compound disaster types is constructed. For compound disasters, the equivalent intensity method is used to convert the compound disaster-causing index into the equivalent value of a single disaster type. The initial claim trigger thresholds for different crops, different types of disasters, and different growth stages are determined by using machine learning algorithms combined with percentile methods. Based on the spatiotemporal heterogeneity analysis results of disaster-vulnerability, a geographically weighted correction method combined with a time series correction method is used to refine the initial claims threshold. The refined correction includes spatial correction and time correction. Using townships as the row dimension and disaster type, crop type, and growth stage as the column dimension, the revised graded claim thresholds are incorporated into a three-dimensional matrix to construct a township-level agricultural insurance claim threshold identification model. Reasonableness verification of the three-dimensional matrix: The historical claims data back-substitution test method is adopted. The threshold matrix is ​​applied to historical claims cases, and the consistency between the claims trigger results calculated by the three-dimensional matrix and the actual claims results is compared.

[0014] The above technical solution can achieve the following beneficial effects: The proposed method for constructing dynamic claims thresholds for agricultural insurance based on disaster-vulnerability coupling addresses the technical problems of inaccurate loss assessment and poor spatiotemporal adaptability in existing agricultural claims schemes; traditional agricultural insurance schemes suffer from inaccurate crop vulnerability assessment results and poor applicability. This application uses actual claims losses as a realistic measurement of crop disaster losses to invert the response relationship between meteorological disaster intensity and crop loss rate. This approach avoids the uncertainties introduced by traditional empirical assignments or experimental extrapolations, and addresses the problem of traditional research emphasizing statistical correlations while neglecting mechanistic explanations. This approach considers the dynamic coupling analysis of disaster-vulnerability due to regional and temporal heterogeneity. Traditional models neglect spatiotemporal differences and suffer from static risk assessments, while also lacking meteorological explanations, resulting in insufficient credibility of claims triggers. This application constructs a dynamic coupling analysis method for disaster-vulnerability that incorporates temporal (monthly / growing season) and spatial (township level) heterogeneity, revealing the synergistic impact mechanism of disaster intensity changes and crop system vulnerability evolution on claims losses, significantly improving spatiotemporal adaptability. It deepens the fundamental understanding of the formation mechanism of agricultural meteorological disaster risks, providing scientific support for insurance claims from phenomenon description to mechanism-driven approaches. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating the first embodiment of a method for constructing dynamic claims thresholds for agricultural insurance based on disaster-vulnerability coupling proposed in this invention. Detailed Implementation

[0017] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0018] This invention proposes a method for constructing dynamic claims thresholds for agricultural insurance based on disaster-vulnerability coupling.

[0019] As attached Figure 1 As shown, in the first embodiment of the method for constructing dynamic claims thresholds for agricultural insurance based on disaster-vulnerability coupling proposed in this invention, this embodiment includes the following steps: Step S110: Obtain meteorological monitoring data, agricultural insurance claims data, geographical topographic data, and crop growth stage data to construct an integrated spatiotemporal database of region, disaster type, growth stage, meteorological intensity, and loss rate. The disaster types include drought, rainstorm, heat damage, and cold damage.

[0020] Step S120: Based on the database, the core disaster-causing factors are extracted using principal component analysis and machine learning algorithms, and the core disaster-causing factors are combined and weighted using the entropy weight-analytic hierarchy process to construct a comprehensive disaster-causing intensity index.

[0021] Specifically, Principal Component Analysis (PCA) is one of the most commonly used unsupervised linear dimensionality reduction algorithms. Its core idea is to project high-dimensional data onto the axes with the largest variances while preserving as much of the original information as possible (i.e., minimizing reconstruction error). PCA achieves dimensionality reduction by finding a new set of orthogonal coordinate systems (principal components), sorting the data according to the variance on each axis, and then selecting the most important axes while discarding the rest. The entropy-weighted analytic hierarchy process (AHP) combines the Analytic Hierarchy Process (AHP) with information from the data itself (entropy weighting) to obtain more comprehensive and reliable weights.

[0022] This step constructs a coupled mechanism framework based on agricultural disaster science theory, revealing the nonlinear relationship between disaster-causing factors, crop vulnerability, and insurance losses.

[0023] Step S130: Based on the comprehensive disaster intensity index and the claim loss rate as the actual loss observation, construct a multi-level Bayesian model and combine it with a zero-inflation model for inversion to obtain the crop vulnerability function.

[0024] Specifically, we will conduct disaster-vulnerability coupling analysis to achieve a quantitative characterization of crop vulnerability and quantify its temporal (monthly) and spatial (township level) heterogeneity and dynamic evolution patterns.

[0025] Step S140: Construct a disaster-vulnerability coupling degree and coupling coordination degree model based on the crop vulnerability function to quantify the synergistic damage effect between the intensity of meteorological disasters and crop vulnerability.

[0026] Specifically, structural equation modeling was used to analyze the transmission path of meteorological factors-physiological stress-loss manifestation.

[0027] Step S150: Based on the disaster-vulnerability coupling degree and coupling coordination degree model, the initial claim threshold and spatiotemporal heterogeneity correction, generate a three-dimensional dynamic claim threshold matrix divided by region, disaster type and reproductive stage, and determine the claim threshold based on the matrix.

[0028] This application takes meteorological disaster intensity and crop vulnerability as the two cores, constructs a coupled model, integrates multi-source meteorological monitoring data, insurance claim actual loss data, and crop growth period data, and forms a complete technical system by Bayesian inversion vulnerability function → identifying nonlinear mutation thresholds → spatiotemporal heterogeneity correction → constructing township-level dynamic threshold matrix → automatic iterative update.

[0029] The proposed method for constructing dynamic claims thresholds for agricultural insurance based on disaster-vulnerability coupling addresses the technical problems of inaccurate loss assessment and poor spatiotemporal adaptability in existing agricultural claims schemes; traditional agricultural insurance schemes suffer from inaccurate crop vulnerability assessment results and poor applicability. This application uses actual claims losses as a realistic measurement of crop disaster losses to invert the response relationship between meteorological disaster intensity and crop loss rate. This approach avoids the uncertainties introduced by traditional empirical assignments or experimental extrapolations, and addresses the problem of traditional research emphasizing statistical correlations while neglecting mechanistic explanations. This approach considers the dynamic coupling analysis of disaster-vulnerability due to regional and temporal heterogeneity. Traditional models neglect spatiotemporal differences and suffer from static risk assessments, while also lacking meteorological explanations, resulting in insufficient credibility of claims triggers. This application constructs a dynamic coupling analysis method for disaster-vulnerability that incorporates temporal (monthly / growing season) and spatial (township level) heterogeneity, revealing the synergistic impact mechanism of disaster intensity changes and crop system vulnerability evolution on claims losses, significantly improving spatiotemporal adaptability. It deepens the fundamental understanding of the formation mechanism of agricultural meteorological disaster risks, providing scientific support for insurance claims from phenomenon description to mechanism-driven approaches.

[0030] Furthermore, this solution constructs a township-level dynamic claims threshold identification model. Traditional solutions lack consideration for regional and growth stage differences in claims thresholds, leading to the application of a single threshold across the entire region (province), resulting in mismatches such as "equal compensation for drought and flood" or "less compensation for severe disasters." This invention comprehensively considers geographical differences and the physiological vulnerability of crops, proposing a novel approach to a township-level dynamic claims threshold identification model. It organically incorporates disaster type, crop category, growth stage, and regional differences into a unified framework, addressing the shortcomings of traditional thresholds being static in time and space and lacking adaptability. This provides a theoretical basis for the scientific setting of agricultural insurance claims thresholds and expands the research and application boundaries of agricultural meteorological disaster risk research.

[0031] In summary, this application has the following specific beneficial effects: 1. Improved damage assessment accuracy: The refined threshold at the township level improves the damage assessment accuracy rate to ≥85%, and the threshold accuracy is improved by ≥20%.

[0032] 2. Reduced basis risk: Dynamic thresholds reduce basis risk by 15% to 20% compared to static thresholds.

[0033] 3. Mechanism is explainable: For the first time, the claims threshold is driven by the coupling of disaster-vulnerability, which is highly explainable.

[0034] 4. Composite disaster identification: Supports the determination of equivalent composite disasters such as drought-high temperature and rainstorm-low temperature.

[0035] 5. Practical and scalable: It can be embedded into insurance claims systems for fully automated operation without manual calculation, and is suitable for agricultural meteorological insurance scenarios nationwide.

[0036] In the second embodiment of the agricultural insurance dynamic claims threshold construction method based on disaster-vulnerability coupling proposed in this invention, based on the first embodiment, step S110 includes the following steps: Step S210: Use the Euclidean distance method to match the meteorological monitoring station closest to the disaster area and obtain the meteorological monitoring data collected by the meteorological monitoring station.

[0037] Specifically, the data obtained includes hourly meteorological monitoring data from national meteorological stations and regional automatic meteorological stations; refined gridded products from the China Meteorological Administration's Land Surface Data Assimilation System CLDAS-V2.0 (temperature, specific humidity, air pressure, and shortwave radiation from 2017 to 2025, with a resolution of 0.0625°×0.0625°); China Regional Multi-Source Fusion Real-Time Analysis ART-1km products (temperature and relative humidity from 2024 to 2025, with a horizontal grid resolution of 0.01°×0.01°); township-level agricultural insurance claims data (including information such as the township where the incident occurred, the time of the incident, crop type, growth stage, affected area, degree of loss, and claim amount); DEM data (digital elevation model data, with a resolution of 30m); soil type; and crop planting zoning and growth period data.

[0038] Step S220: Perform climatological boundary value test, time series consistency test, and spatial interpolation rationality test on the meteorological monitoring data in sequence, and supplement the meteorological monitoring data corresponding to meteorological monitoring stations with missing values ​​less than the preset percentage (5%) using the Kriging interpolation algorithm or the neighboring station mean algorithm.

[0039] Specifically, for hourly data on rainstorm disasters, we will focus on verifying the continuity of time, and for daily data on drought, cold damage, and heat damage, we will conduct trend verification to ensure the spatiotemporal integrity of meteorological data.

[0040] Step S230: Perform logical consistency verification and on-site cross-verification on agricultural insurance claims data to eliminate erroneous data (including false reports and incorrect information entry data), and convert the claims amount in the agricultural insurance claims data into the claims loss rate.

[0041] Specifically, the loss rate is calculated as: actual loss / theoretical yield value. This serves as the core quantitative indicator for crop disaster losses, enabling standardized and comparable claims data across different townships and years. Two to three typical townships are selected from each prefecture, and the sample size for verification must be greater than or equal to 10% of the total sample.

[0042] Step S240: Divide the time scale according to the type of disaster: for rainstorms, use hourly data, with the time scale being the first preset number of days before the claim date (e.g., 7 days, to calculate the daily cumulative precipitation and hourly maximum precipitation); for drought, the time scale is the second preset number of days before the claim date (45 days); for cold damage, the time scale is the third preset number of days before the claim date (20 days); and for heat damage, the time scale is the fourth preset number of days before the claim date (25 days).

[0043] Specifically, the daily matching of meteorological factors, combined with the division of crop growth stages, accurately correlates meteorological disaster processes with specific crop growth stages, determines the time response window from disaster occurrence to loss manifestation, and improves the physiological rationality of the matching.

[0044] Currently, the province has more than 3,800 regional automatic weather stations (with an average station network density of 7.5km×7.5km). Factors such as evaporation and sunshine duration are based on data from 97 national surface weather stations. At the same time, the data is integrated with the China Meteorological Administration's land surface data assimilation system CLDAS-V2.0 refined real-time grid product and the China Regional Multi-Source Fusion Real-Time Analysis (ART-1km) product. The CLDAS data is resampled using the GIS bilinear interpolation method, and the output pixel size is the same as the resolution of the ART format image.

[0045] Step S250: Using ArcGIS spatial analysis tools (the spatial analysis tools included in ArcGIS software), the meteorological monitoring data collected by meteorological monitoring stations is interpolated to the township scale using the inverse distance weighted interpolation algorithm to achieve a township-level gridded expression of meteorological disaster intensity. This is then matched with the township spatial attributes of agricultural insurance claims data to form an integrated spatiotemporal database of region, disaster type, reproductive stage, meteorological intensity, and loss rate.

[0046] Specifically, Inverse Distance Weighting (IDW) is a deterministic spatial interpolation method based on the first law of geography: things that are closer together are more similar.

[0047] Specifically, the integrated spatiotemporal database of region-disaster type-fertility stage-meteorological intensity-loss rate has the following structure: spatial attributes (township code, latitude and longitude, topographic factors) - temporal attributes (year, month, fertility stage, disaster occurrence period) - meteorological attributes (disaster type, comprehensive intensity index, classification) - insurance attributes (crop type, affected area, loss rate, compensation amount). The spatial database is constructed using PostgreSQL+PostGIS and supports spatiotemporal retrieval and model calling.

[0048] Step S210, followed by the following steps: Step S211: Verification and correction of multi-source fused gridded data: Obtain multi-source fused gridded data from the national ground meteorological station closest to the disaster area, and compare the gridded data with the meteorological monitoring data collected by the meteorological monitoring station to determine whether the meteorological monitoring data collected by the meteorological monitoring station is usable.

[0049] Step S212: Compare and analyze meteorological elements, including maximum temperature, minimum temperature, relative humidity, and precipitation. By comparing station data with grid data, analyze the distribution trend of grid data and station measured data, as well as the correlation coefficients of various meteorological events, and verify the quality of fused grid data.

[0050] Step S213: Use three indicators, root mean square error (RMSE), correlation coefficient (R), and bias, to verify the data. The R of the grid data and the station measured data should be ≥0.85, and the RMSE should meet the accuracy requirements of meteorological elements. The verification results are used as the criteria for determining whether the data is usable.

[0051] Specifically, the root mean square error (RMSE) is a commonly used metric to measure the difference between the predicted (or estimated) value and the actual value. It reflects the overall magnitude of the prediction error and is particularly sensitive to larger errors, thus providing a good indication of the stability and accuracy of the prediction model.

[0052] The correlation coefficient is a statistical indicator used to measure the direction and strength of the linear correlation between two variables. The most commonly used is the Pearson correlation coefficient (denoted as R).

[0053] In statistics and data analysis, bias usually refers to systematic error, which is the systematic deviation of the predicted (or estimated) value from the true value. It reflects the accuracy of the method or model itself, rather than the magnitude of random fluctuations.

[0054] Step S214: If available, proceed with S220 and the following steps.

[0055] In the third embodiment of the method for constructing dynamic claims thresholds for agricultural insurance based on disaster-vulnerability coupling proposed in this invention, based on the first embodiment, step S120, which involves extracting core disaster-causing factors using principal component analysis and machine learning algorithms from the database, includes the following steps: Specifically, breaking through the traditional paradigm of simple statistical correlation, a four-step analytical framework is constructed, which includes disaster-causing factor identification, disaster-causing characteristic analysis, vulnerability response path analysis, and nonlinear coupling mechanism analysis. This framework reveals the intrinsic mechanism by which meteorological disasters are transformed into insurance losses through crop vulnerability and the differences in crop characteristics.

[0056] Step S310: Using machine learning algorithms (such as random forest algorithm, XGBoost algorithm, LightGBM algorithm) and principal component analysis algorithm (PCA), core disaster-causing factors of different types of disasters are extracted from the original meteorological monitoring data. The core disaster-causing factors are screened based on out-of-bag error (OOB) and feature importance to construct a graded and quantitative meteorological disaster intensity index system. The feature importance of the core disaster-causing factors must be ≥10%.

[0057] Specifically, Out-of-Bag Error (OOB) is a built-in, efficient model validation technique in the Random Forest algorithm. It uses data that was not used during training (i.e., "out-of-bag" data) for each tree to evaluate model performance without the need for additional validation sets.

[0058] Feature importance is a metric used in machine learning models to measure the contribution of each input feature to the prediction result. It helps us identify which features are most important to the model's decisions, thus aiding in feature selection, model interpretation, and simplification.

[0059] Specifically, The core disaster-causing factors of rainstorms include: The daily rainfall, maximum hourly rainfall, and number of days of continuous rainstorm are calculated by counting back the first preset number of days (7 days) from the day of the claim to construct a comprehensive rainstorm intensity index.

[0060] The core disaster-causing factors of drought include: The daily average relative humidity, average temperature, daily precipitation, soil relative humidity, number of consecutive days without effective precipitation, and sunshine hours are calculated based on the second preset number of days (45 days) prior to the claim date to construct a comprehensive drought intensity index.

[0061] The core causative factors of cold damage include: The daily minimum temperature is calculated by counting back three preset days (20 days) from the date of claim, the number of days each year when the daily minimum temperature is lower than the first preset temperature (e.g., 0℃, -2℃, -4℃, or -5℃), the number of consecutive days of low temperature, and the accumulated temperature anomaly, in order to construct a comprehensive cold damage intensity index.

[0062] Specifically, accumulated temperature anomaly is a commonly used diagnostic indicator in agricultural meteorology and climatology. This indicator can analyze the fluctuations in heat conditions across different years and determine whether the current year's temperature is higher or lower than the average. It is particularly widely used in assessing the impact of low-temperature damage on crops.

[0063] The core disaster-causing factors of heat damage include: The comprehensive heat damage intensity index is constructed by taking the daily maximum temperature, average relative humidity, longest consecutive days greater than or equal to the second preset temperature (35℃), number of days greater than or equal to the second preset temperature (35℃), effective high temperature accumulation, precipitation, and sunshine hours for the fourth preset number of days (25 days) prior to the claim date.

[0064] In the fourth embodiment of the method for constructing dynamic claims thresholds for agricultural insurance based on disaster-vulnerability coupling proposed in this invention, based on the third embodiment, the step S120 of using the entropy weight-analytic hierarchy process to combine and weight core disaster-causing factors to construct a comprehensive disaster-causing intensity index includes the following steps: Step S410: Use the entropy weight algorithm combined with the hierarchical analysis algorithm to assign weights to the core disaster-causing factors in order to construct a comprehensive disaster-causing intensity index. The entropy weight algorithm can reflect the objective laws of the data, while the hierarchical analysis algorithm is based on the experience of agricultural meteorological experts (3-5 provincial chief agricultural meteorological experts are invited to score). The consistency ratio of the weight results of the hierarchical analysis algorithm must be less than or equal to the first preset value (e.g., 0.1).

[0065] Specifically, the consistency ratio (CR) is a key indicator in the Analytic Hierarchy Process (AHP) used to quantify the degree of logical contradiction among experts when constructing a judgment matrix. It is an important basis for verifying whether the matrix has satisfactory consistency, thereby ensuring the credibility of the final weighting results. The formula for calculating the consistency ratio (CR) is: consistency index (CI) divided by random consistency index (RI).

[0066] The consistency index (CI) is used to assess the degree of deviation between the judgment matrix and its theoretically perfectly consistent matrix; the random consistency index (RI) is obtained by averaging the CIs of randomly generated judgment matrices and is used to calibrate the CI values.

[0067] Step S420: The intensity of meteorological disasters is classified using the percentile method to determine each intensity level and its corresponding threshold range. The intensity levels of meteorological disasters include four categories: light, moderate, severe, and extremely severe. The threshold range for the light level is as follows: The grading threshold range corresponding to the medium level is: The grading threshold range corresponding to the heavy level is: The grading threshold corresponding to the extremely serious level is in the range of [range missing]. .

[0068] Based on the Kappa coefficient (kappa≥0.75), the consistency between the classification results of meteorological disaster intensity and the actual disaster results is tested, so as to realize the quantitative and hierarchical expression of disaster intensity.

[0069] The Kappa coefficient (Cohen's Kappa) is a statistical metric used to measure the consistency of classification results between two evaluators (or two methods). It eliminates the influence of random consistency and is therefore more robust than simple accuracy (the proportion of correct classifications).

[0070] In the fifth embodiment of the method for constructing dynamic claims thresholds for agricultural insurance based on disaster-vulnerability coupling proposed in this invention, based on the fourth embodiment, after step S130, the following steps are further included: Step S510: Analyze the interannual variation trend of disaster types using trend analysis algorithm, wavelet analysis algorithm, spatial clustering analysis algorithm (K-means), and Mann-Kendall trend test algorithm to determine whether the interannual variation trend of disaster types has increased or decreased significantly (significance level α=0.05).

[0071] Specifically, spatial clustering analysis algorithms unsupervisedly group objects in geographic space based on "similar years" and "similar attributes" to reveal their inherent distribution patterns and structures; the Mann-Kendall trend test algorithm is a nonparametric statistical method used to assess whether time series data have a monotonic trend (continuous increase or decrease).

[0072] Step S520: Use spatial variation function to analyze the spatial variation characteristics of disaster types in order to determine spatial correlation distances and provide a basis for subsequent spatial interpolation.

[0073] Step S530: Obtain meteorological monitoring data from different monitoring areas, and analyze the disaster-causing characteristics of the four types of disasters in each monitoring area (tea and vegetable growing areas in Hunan Province) from three dimensions: time (interannual, seasonal, and monthly), space (e.g., northern, central, southern, and western Hunan, or hilly and plain types), and intensity (including grade distribution and extreme value characteristics). This will help determine the frequency of occurrence, spatiotemporal distribution patterns, and intensity evolution characteristics of disaster types in different monitoring areas, thereby clarifying the differences in disaster-causing characteristics among different monitoring areas.

[0074] Step S540: Based on the disaster-causing factor-disaster-loss theory of agricultural disaster science, the vulnerability response paths of different crops (tea, vegetables) to different types of disasters are identified by combining literature review with structural equation modeling.

[0075] Structural equation modeling (SEM) is a statistical method commonly used to verify complex theoretical models; the steps for constructing a structural equation model are as follows: A theoretical model was established, indicators were designed, data were fitted, the structural equation model was revised, and goodness-of-fit evaluation indicators were set (specifically: χ² / df=1-3, GFI≥0.9, RMSEA≤0.08, CFI≥0.9). The proportion of the mediating effect of crop vulnerability was quantified through the mediating effect test, and the significance of the mediating role was clarified.

[0076] Step S550: Analyze the formation and evolution of crop vulnerability from four dimensions. The four dimensions of crop vulnerability include plant physiological vulnerability (root vitality, etc.), growth and development vulnerability (delayed growth period, flower and fruit drop, decreased fruit setting rate, etc.), yield and quality vulnerability (reduced yield, deterioration in quality, etc.), and economic vulnerability (economic value corresponding to yield loss and quality loss per unit area).

[0077] Step S560: Quantify the direct and indirect effects between core disaster-causing factors, four dimensions of crop vulnerability, and claims loss rate using structural equation modeling, and clarify the mediating role of crop vulnerability between disaster-causing factors and claims loss rate.

[0078] In the sixth embodiment of the method for constructing dynamic claims thresholds for agricultural insurance based on disaster-vulnerability coupling proposed in this invention, based on the fifth embodiment, step S560 is followed by the following steps: Step S610: Using bivariate correlation analysis algorithm, nonlinear regression analysis algorithm, and quantile regression algorithm, analyze the nonlinear relationship between meteorological disaster intensity and crop vulnerability, and between crop vulnerability and claim loss rate under different meteorological disaster intensity and different growth stages.

[0079] Specifically, bivariate correlation analysis algorithms quantify the direction and strength of the linear or monotonic relationship between two variables. The choice of appropriate algorithm depends on the data distribution, measurement scale, and relationship type. Nonlinear regression analysis algorithms are statistical modeling methods used to capture nonlinear relationships between dependent and independent variables. Nonlinear regression is needed when the scatter plot is clearly not a straight line or the linear model fits poorly. Compared to linear regression, nonlinear regression analysis algorithms can handle complex patterns such as curvature, periodic variations, and threshold effects, and are widely used in environmental science, ecology, economics, and other fields. Quantile regression algorithms are statistical methods that focus on the overall data distribution, helping us understand how the influence relationships between variables change at different levels (e.g., low, middle, and high).

[0080] Specifically, the quantiles for quantile regression were chosen to be 0.25, 0.5, and 0.75, corresponding to low, medium, and high loss rates, respectively. The differences in the coupling relationship between disaster intensity and vulnerability under different loss rates were analyzed. The nonlinear regression analysis algorithm adopted the nonlinear least squares method, and the coefficient of determination R² for the structural equation model fitting needed to be ≥0.6.

[0081] Step S620: Use the interaction effect analysis method to examine the influence of the interaction between crop type, disaster type, and growth stage on the coupling relationship, and use the analysis of variance method to test the significance of the interaction effect (P<0.05) to identify the core driving factors of the difference.

[0082] Specifically, analysis of variance (ANOVA) is a statistical method used to test whether there are significant differences between the means of three or more populations. The core idea is to decompose the total variation into variation caused by different factors and random error, and then determine whether the factors have a significant impact on the observed values ​​by comparing the proportions of the two.

[0083] Step S630: Through grouped comparative analysis (grouped by crop type, disaster type, and growth stage), reveal the differences in the disaster-vulnerability coupling process of crops (tea and vegetables), construct a disaster-vulnerability coupling mechanism analysis framework, and elucidate the inherent laws of nonlinear damage.

[0084] Step S640: The rationality of the coupling mechanism framework is verified by using structural equation modeling and path coefficients. The intensity of meteorological disaster, the vulnerability of crops in four dimensions, and the loss rate of claims are used as latent variables. The coefficients and significance of each path are quantified to form a quantifiable and verifiable nonlinear coupling mechanism framework of disaster-vulnerability.

[0085] In the seventh embodiment of the method for constructing dynamic claims thresholds for agricultural insurance based on disaster-vulnerability coupling proposed in this invention, based on the sixth embodiment, step S130 includes the following steps: Step S710: Standardize the intensity of meteorological disasters: Standardize the comprehensive intensity index of each type of disaster from 0 to 1 to eliminate the difference in dimensions and obtain the standardized intensity of meteorological disasters.

[0086] Specifically, the standardized disaster intensity data is subjected to a normality test (Shapiro-Wilk test). If the normality is not met, the Box-Cox transformation is used to normalize it, ensuring the prerequisite for model fit.

[0087] Specifically, the Shapiro-Wilk test is a statistical method used to test whether sample data comes from a normal distribution, especially suitable for small samples (sample size usually ≤5000), and is one of the most commonly used normality tests. The Box-Cox transformation is a mathematical transformation used to convert non-normal data into an approximately normal distribution, while also stabilizing variance. It is often used in statistical modeling such as linear regression and analysis of variance, which require the assumptions of normality and homogeneity of variance to be satisfied.

[0088] Step S720: Standardize the physical loss rate of crops: Standardize the extreme values ​​of the actual crop loss rate at the township level in the disaster-stricken area to eliminate the influence of extreme outliers, ensure the normal distribution of the data, and lay the foundation for subsequent function fitting.

[0089] Step S730: Using exploratory data analysis (EDA) algorithms, by plotting scatter plots and fitting curve trends, the type of response relationship (linear / nonlinear, such as exponential, logarithmic, S-shaped curves, etc.) between the intensity of meteorological disasters and the rate of physical loss of crops is preliminarily determined.

[0090] Specifically, the core of exploratory data analysis algorithms is a set of techniques and methods whose purpose is not to build models, but to understand the data itself to the greatest extent possible.

[0091] Step S740: Conduct group analysis for different crops, different types of disasters, and different growth stages to clarify the response relationship characteristics under different scenarios.

[0092] Step S750: The kernel density estimation algorithm is used to analyze the distribution characteristics of the physical loss rate of crops. Then, the Pearson correlation coefficient and Spearman correlation coefficient are used to determine the correlation type and significance between the intensity of meteorological disasters and the physical loss rate of crops, providing a data basis for the selection of function type.

[0093] Specifically, kernel density estimation (KDE) is a nonparametric method used to estimate the probability density function of a random variable. Compared to histograms, KDE produces continuous, smooth density curves and is independent of the histogram's interval division method.

[0094] In the eighth embodiment of the method for constructing dynamic claims thresholds for agricultural insurance based on disaster-vulnerability coupling proposed in this invention, based on the seventh embodiment, after step S750, the following steps are further included: Step S810: Selection of Crop Vulnerability Function Type: Select nonlinear models (such as physiological constraint parameters (e.g., accumulated temperature, physiological lower limit of key phenological period, etc.), Logistic curve model, power function model, quadratic polynomial model, etc.) as candidate crop vulnerability functions, use the standardized meteorological disaster intensity as the independent variable and the standardized crop physical loss rate as the dependent variable, perform model fitting and screening on the candidate crop vulnerability functions to obtain the preferred crop vulnerability function.

[0095] Specifically, the comprehensive evaluation system for model selection uses R², SSE (sum of squared errors), and AIC as core indicators, while also considering the physical interpretability of the model to avoid simply pursuing fitting accuracy while ignoring physiological mechanisms. For example, models with a Logistic curve that better fits the "S-shaped" response characteristics of biomass loss are given priority.

[0096] Specifically, SSE stands for Sum of Squares for Error; AIC stands for Akaike Information Criterion (AIC).

[0097] Step S820: Mathematical Construction and Probabilistic Expression of Crop Vulnerability Function: This involves calculating the actual claim loss rate of township i at growth stage t. Considered as the intensity of disaster caused by meteorological disasters Potential crop vulnerability status According to Bayes' theorem, the posterior distribution of crop vulnerability, driven by a co-driven stochastic process, can be expressed as: , In the formula, P Represents the probability distribution function; This indicates the actual claim loss rate during the maternity stage in rural areas. Intensity of meteorological disasters Under certain conditions, the potential vulnerability status of crops The posterior probability distribution; This represents the intensity of a given meteorological disaster in township i during the reproductive stage t. Potential crop vulnerability status Under these conditions, the actual claim loss rate was observed. The likelihood function is used to describe the probabilistic relationship between actual losses and the combined effects of disaster intensity and vulnerability. Indicates the potential vulnerability status of crops The prior probability distribution; These parameters represent the relevant parameters of crop growth stages, used to characterize the prior characteristics of crop sensitivity to meteorological disasters at different growth stages, including the physiological state, stress resistance, and yield formation sensitivity at different growth stages; This indicates the potential crop vulnerability status of township i at the reproductive stage t; This represents the actual claim loss rate of township i during the childbirth stage t; This indicates the intensity of meteorological disasters affecting township i during the reproductive stage t. Likelihood function To characterize the probability distribution of loss occurrence under a given disaster-causing pressure, a Zero-Inflated model is introduced to address the problem of a large number of zero-value (lossless) skewed values ​​in insurance claims data.

[0098] Step S830: Classify crop vulnerability: Based on the calculation results of the optimized crop vulnerability function, crop vulnerability is divided into four levels: low, medium, relatively high, and high, so as to realize the quantitative and graded expression of crop vulnerability.

[0099] Specifically, the loss rate thresholds for crop vulnerability classification are as follows: low vulnerability (loss rate <10%), medium vulnerability (10%≤loss rate <30%), relatively high vulnerability (30%≤loss rate <50%), and high vulnerability (loss rate ≥50%), which are aligned with the agricultural insurance claim classification (deductible, partial compensation, and full compensation).

[0100] Step S840: The sample segmentation algorithm is used to verify the accuracy of the selected crop vulnerability function; the root mean square error (RMSE) and mean absolute error (MAE) of the validation set are calculated to evaluate the model's fitting accuracy and predictive ability. The training set / validation set is divided using stratified random sampling to ensure that the proportion of samples of each disaster type and loss rate level is consistent in the training set and validation set. The RMSE of the validation set is ≤0.15 and the MAE is ≤0.1. The Monte Carlo simulation method is used to analyze the uncertainty of the model parameters, clarify the error range and confidence interval of the vulnerability function, and improve the reliability of the results.

[0101] Specifically, the sample segmentation algorithm is the foundation and prerequisite of the machine learning process. Its core purpose is to divide the original dataset into different subsets and objectively evaluate the model's true generalization ability by simulating the model's performance on "unseen data" to avoid overfitting on the training set. In this embodiment, the ratio of the training set to the validation set of the sample segmentation algorithm is 7:3.

[0102] Specifically, the detailed plan for on-site observation and verification: Typical county selection principles (covering plains, hills, and mountains, major tea / vegetable producing areas, and large sample size of claims): 3-5 observation points are selected in each county. Observation indicators include crop physiological indicators (root activity), growth indicators (plant height, seed setting rate), and yield and quality indicators (yield per unit area, moisture content). The observation frequency is 1-7 days after the disaster. The vulnerability function is verified by the relative error between the measured loss rate and the model predicted loss rate, with a relative error ≤20%.

[0103] In the ninth embodiment of the method for constructing dynamic claims thresholds for agricultural insurance based on disaster-vulnerability coupling proposed in this invention, based on the first embodiment, step S140 includes the following steps: Specifically, the process in this embodiment is as follows: construct a coupling degree-coupling coordination degree model to quantify the synergistic damage effect of disaster intensity and vulnerability; introduce nonlinear catastrophe theory to identify the critical point of loss transition; and use structural equation modeling to analyze the path coefficients of meteorological factors-physiological stress-loss manifestation.

[0104] Step S910: Characterize temporal heterogeneity: Use the moving average algorithm and trend coefficient algorithm to analyze the changing trends and seasonal fluctuations of the monthly meteorological disaster intensity and crop vulnerability.

[0105] Specifically, the temporal variation of disaster-vulnerability in different years and different reproductive stages is quantified by using the coefficient of variation (CV) grading standard (CV<0.1 is weak variation, 0.1≤CV<0.3 is moderate variation, and CV≥0.3 is strong variation), thus clarifying the temporal variation of disaster-vulnerability in different reproductive stages.

[0106] Specifically, the coefficient of variation (CV) is a statistical indicator that measures the relative dispersion of data, defined as the ratio of the standard deviation to the mean. It eliminates the influence of units and the magnitude of the mean, making it easier to compare the degree of fluctuation in datasets with different units or large differences in the mean.

[0107] Step S920: Characterize spatial heterogeneity: Using ArcGIS spatial analysis tools, employ the spatial autocorrelation analysis algorithm (Moran's I index) to examine the spatial clustering characteristics (high-high clustering, low-low clustering, high-low clustering, low-high clustering) of disaster-causing capacity and crop vulnerability at the township level; identify spatial hotspots (high-value clustering) and cold spots (low-value clustering) of disaster-causing vulnerability through hotspot analysis (Getis-Ord Gi index); analyze the contribution of geographical elements through the spatial distribution of regression coefficients, and quantify the contribution of geographical elements such as topography, soil, climate, and planting patterns to spatial heterogeneity using the geographically weighted regression algorithm (GWR), revealing the causes of spatial heterogeneity.

[0108] Specifically, the core of spatial autocorrelation analysis algorithms is that the correlation between geographical features is inversely proportional to their distance. Spatial autocorrelation analysis is mainly divided into two types: global and local, with different algorithms serving different analytical purposes. Spatial autocorrelation analysis algorithms mainly include the following indicators: Moran's I index: used to study whether spatial data within a region exhibits an overall trend of clustering or dispersion; Getis-Ord Gi index: identifies statistically hot and cold areas, i.e., significant clusters of high or low values.

[0109] Geographically Weighted Regression (GWR) is a natural extension following spatial autocorrelation analysis.

[0110] Step S930: Analyze the single-factor explanatory power and interaction explanatory power of geographical elements (topography, soil, climate, planting patterns, etc.) on the spatial heterogeneity of disaster-vulnerability, and identify the core influencing factors (when the q value of a geographical element is ≥0.2, then the geographical element is the core factor).

[0111] Step S940: Coupling Degree Model Construction: Introduce the coupling coordination degree model, taking the intensity of meteorological disasters and the vulnerability of crops as two coupled subsystems, and construct disaster-vulnerability coupling degree and coupling coordination degree models to calculate the coupling degree (reflecting the interaction strength between the two) and coupling coordination degree (reflecting the synergistic development level between the two) at the township level and on a monthly basis.

[0112] Specifically, the calculation formulas for coupling degree and coupling coordination degree are as follows: the value range of coupling degree is [0,1], and the classification standard is low coupling [0,0.3), medium coupling [0.3,0.6], and high coupling [0.6,1]; the classification standard of coupling coordination degree is mismatch (0,0.5) and coordination [0.5,1], which is further subdivided into mild, medium and high coordination.

[0113] Step S950: Dynamic Evolution Law Analysis: Using the center of gravity migration model, analyze the migration direction and distance of the center of gravity of disaster-vulnerability coupling degree and coupling coordination degree in the time series.

[0114] Specifically, it reveals the characteristics of spatial dynamic evolution; by quantifying the temporal evolution trend of coupling characteristics through interannual variability rate, it clarifies the dynamic change patterns under different climatic backgrounds.

[0115] Step S960: Using a panel regression model, with the township-level insurance claim loss rate as the dependent variable and the interaction term between the intensity of meteorological disasters and crop vulnerability as the core explanatory variable, and introducing control variables (topography, planting area, agricultural management level), a fixed-effects panel model is constructed to quantify the intensity of meteorological disasters, crop vulnerability, and the contribution of the interaction to insurance losses.

[0116] Specifically, stationarity tests (LLC test, IPS test) and cointegration tests (Kao test) are performed on panel regression models to avoid spurious regressions; A fixed effects model is a statistical model commonly used in panel data analysis. It is mainly used to control for individual heterogeneity that does not change over time, thereby more accurately estimating the causal effect of independent variables on dependent variables.

[0117] The random effects model is another commonly used method in panel data analysis. It treats individual effects as random variables rather than fixed constants, thus enabling the estimation of the impact of variables that do not change over time, and is more suitable for inference of data randomly sampled from a large population.

[0118] The F-test is a statistical test used to compare whether there are significant differences in the variances of two or more populations, or more commonly, to test the overall significance of a model in regression analysis and analysis of variance. Its core is the F-statistic, which reflects the ratio of the two types of variance (variation).

[0119] The Hausman test is a statistical method used to compare two estimators: one that is more efficient but consistent under the given assumptions, and the other that remains consistent even when the assumptions are violated. In panel data analysis, it is most commonly used to choose between fixed-effects and random-effects models, and can also be used to test for endogeneity, measurement error, etc.

[0120] The variance inflation factor (VIF) is a commonly used indicator to detect the severity of multicollinearity in multiple linear regression. It quantifies the degree of linear correlation between a given independent variable and other independent variables, and the inflationary effect of this correlation on the variance of the regression coefficients.

[0121] The rationality of model selection was determined by the F-test of the fixed effects model and the Hausman test of the random effects model. The variance inflation factor (VIF) of the core explanatory variables was ≤10 to avoid multicollinearity. Through grouped regression (grouped by region, disaster type, and year), the synergistic mechanism of disaster-causing capacity change and vulnerability evolution on the formation of claims losses under different scenarios was revealed, and the key threshold and impact characteristics of the synergistic loss caused by the two were clarified.

[0122] Step S970: Determine the formula for the disaster-vulnerability coupling degree and coupling coordination degree model as follows: , In the formula, C represents the disaster-vulnerability coupling degree and coupling coordination degree model; H represents the disaster-causing intensity of meteorological disasters; and V represents the crop vulnerability index.

[0123] Step S980: Synergistic Effect Quantification: The marginal effect analysis algorithm is used to quantify the synergistic damage effect of meteorological disaster intensity and crop vulnerability at different levels. Specifically, the marginal effect analysis algorithm defines marginal effect as the instantaneous rate of change of the explained variable caused by a one-unit change in an explanatory variable, assuming other factors remain constant. In economics, it can be viewed as a generalization of the slope to more complex models. In simpler models, the marginal effect is simply the regression coefficient; however, in nonlinear models, models with interaction terms, or models with spatial heterogeneity, it is the key to understanding the true impact of variables.

[0124] The marginal effect of the interaction terms is calculated by using the marginal effect analysis algorithm to determine the critical interval of the synergistic effect between disaster hazard capability and vulnerability (e.g., when disaster hazard capability ≥ 0.6 and vulnerability ≥ 0.5, the synergistic effect is significantly enhanced).

[0125] In the tenth embodiment of the method for constructing dynamic claims thresholds for agricultural insurance based on disaster-vulnerability coupling proposed in this invention, based on the first embodiment, step S150 includes the following steps: Step S1010: Construct a three-dimensional matrix containing thresholds for single disaster types and joint thresholds for typical compound disaster types. For compound disasters, use the equivalent intensity method to convert the compound disaster-causing index into the equivalent value of a single disaster type, or add compound disaster type columns such as "drought-high temperature" and "rainstorm-low temperature".

[0126] Step S1020: Using machine learning algorithms combined with percentile methods, determine the initial claim trigger threshold (i.e., meteorological disaster threshold) for different crops, different types of disasters, and different growth stages.

[0127] Specifically, machine learning uses "whether a claim is triggered" as the dependent variable and disaster intensity, crop type, and growth stage as independent variables, with an AUC value ≥ 0.8. The core influencing factors of the threshold are determined through feature importance. The percentile method is combined with the loss distribution method of actuarial science, and the expected loss rate is used as the basis for setting the threshold to ensure that the threshold is consistent with the actuarial logic of insurance compensation. The trigger threshold corresponds to the insurance deductible (e.g., 5%), the partial compensation threshold corresponds to the compensation ratio of 30%, 50%, and 80%, and the full compensation threshold corresponds to the compensation ratio of 100%, which is seamlessly connected with the current agricultural insurance claims rules in Hunan Province.

[0128] Specifically, the AUC (Area Under the Curve) value is a commonly used metric for evaluating the performance of binary classification models. It represents the probability that the model will rank a positive sample before a negative sample when a random pair of positive and negative samples is selected.

[0129] Step S1030: Based on the spatiotemporal heterogeneity analysis results of disaster-vulnerability, the initial claim threshold is refined by using a geographically weighted correction method combined with a time series correction method. The refined correction includes spatial correction and time correction.

[0130] Specifically, spatial correction includes: adjusting the initial thresholds for different townships based on the spatial clustering characteristics and hotspot distribution of township-level disaster-causing capacity and crop vulnerability; appropriately lowering the claim thresholds for townships with high vulnerability and high disaster-causing capacity, and appropriately raising the claim thresholds for townships with low vulnerability and low disaster-causing capacity; the formula for calculating the correction coefficient of the geographically weighted correction method is based on the comprehensive index of township-level disaster-causing capacity and vulnerability, and the formula is: Correction coefficient = 1 - 0.2 × (Comprehensive index / Maximum value of comprehensive index), ensuring the rationality and operability of the correction coefficient; the correction coefficient for townships with high vulnerability / high disaster-causing capacity is ≤0.8, and the correction coefficient for townships with low vulnerability / low disaster-causing capacity is ≥1.2.

[0131] The geographically weighted correction method refers to a series of improvement, expansion and optimization techniques developed around the core geographically weighted regression (GWR) model to address its inherent limitations.

[0132] Specifically, time correction includes: adjusting the initial threshold based on the dynamic coupling characteristics of disaster-vulnerability at different growth stages and in different months; appropriately lowering the claim threshold during critical growth periods of crops (such as tea picking period and vegetable fruiting period); and appropriately raising the claim threshold during non-critical growth periods. The weighting coefficient of the growth period in the time series correction method is set based on the importance of the crop's physiological sensitivity window. The weighting coefficient for critical growth periods is 0.7-0.9, and the weighting coefficient for non-critical growth periods is 1.1-1.3. The weighting coefficients are jointly determined by agricultural meteorological experts and insurance actuaries, and the consistency test CR≤0.1.

[0133] Step S1040: Using townships as the row dimension and disaster type, crop type, and growth stage as the column dimension, incorporate the revised graded claim thresholds (trigger threshold, minor disaster, moderate disaster, severe disaster, and extremely severe disaster threshold) into a three-dimensional matrix to construct a multi-dimensional, refined, and dynamic township-level agricultural insurance claim threshold identification model; each cell in the matrix corresponds to a specific claim threshold for "a certain township - a certain disaster type - a certain crop - a certain growth stage", realizing the spatiotemporal precise expression of claim thresholds.

[0134] Step S1050: Verify the reasonableness of the three-dimensional matrix: Use the historical claims data back-substitution test method to apply the threshold matrix to historical claims cases (2017-2025) and compare the consistency between the claims trigger results calculated by the three-dimensional matrix and the actual claims results.

[0135] Specifically, the applicability of the threshold matrix is ​​evaluated by calculating the compliance rate; at the same time, experts in agricultural meteorology and agricultural insurance are organized to conduct expert reviews to verify the rationality of the threshold matrix from both theoretical and practical perspectives.

[0136] Step S1060: When consistency is achieved, determine the claim threshold based on the three-dimensional matrix.

[0137] Step S1070: Dynamically update the three-dimensional matrix: Based on the verification results of the three-dimensional matrix, and combined with factors such as climate background changes, crop planting pattern adjustments, and agricultural insurance policy optimization, establish a dynamic update mechanism for the three-dimensional matrix, clarify the update cycle and update indicators, and ensure the timeliness and practicality of the threshold matrix.

[0138] This example uses agricultural insurance for tea and vegetables in Hunan Province: 1. Data period: Township-level insurance claims data from 2017 to 2025, data from more than 3,800 regional automatic weather stations across the province, and CLDAS grid data (CMA Land Data Assimilation System, China Meteorological Administration).

[0139] 2. Data source: Hunan Tianqing Meteorological Cloud Platform, PICC Property & Casualty Insurance / Pacific Insurance; 3. Research subjects: Tea (temperature-sensitive crop), vegetables (water-sensitive crop); 4. Output results: Trigger threshold, partial compensation threshold, and full compensation threshold (three levels of claim thresholds); 5. Verification Results: Historical data back-substitution accuracy rate ≥ 85%, on-site observation error ≤ 20%. The final insurance claim threshold results are shown in the table below: Note: T min18 R is the average daily temperature from the day of the claim to the 18 days prior to the claim. H9 The average relative humidity from the date of the claim to the nine days prior to the claim, P 24 P is the cumulative rainfall in the 24 hours prior to the claim. 48 This refers to the cumulative rainfall in the 48 hours prior to the claim.

[0140] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0141] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for constructing dynamic claims thresholds for agricultural insurance based on disaster hazard-vulnerability coupling, characterized in that, include: Acquire meteorological monitoring data, agricultural insurance claims data, geographical and topographical data, and crop growth stage data to construct an integrated spatiotemporal database of region, disaster type, growth stage, meteorological intensity, and loss rate. The disaster types include drought, rainstorm, heat damage, and cold damage. Based on the database, principal component analysis and machine learning algorithms are used to extract core disaster-causing factors. Entropy weight-analytic hierarchy process algorithm is used to combine and assign weights to the core disaster-causing factors in order to construct a comprehensive disaster-causing intensity index. Based on the comprehensive disaster intensity index and the claim loss rate as the actual loss observation, a multi-level Bayesian model is constructed, and inversion is performed using a zero-inflation model to obtain the crop vulnerability function. Based on the crop vulnerability function, a disaster-vulnerability coupling degree and coupling coordination degree model is constructed to quantify the synergistic damage effect between the intensity of meteorological disasters and crop vulnerability. Based on the disaster-vulnerability coupling degree and coupling coordination degree model, the initial claim threshold and spatiotemporal heterogeneity correction, a three-dimensional dynamic claim threshold matrix is ​​generated by region, disaster type and reproductive stage, and the claim threshold is determined based on the matrix.

2. The method for constructing dynamic claims thresholds for agricultural insurance based on disaster-vulnerability coupling as described in claim 1, characterized in that, The acquisition of meteorological monitoring data, agricultural insurance claims data, geographical topographic data, and crop growth stage data to construct an integrated spatiotemporal database of region, disaster type, growth stage, meteorological intensity, and loss rate includes: The Euclidean distance method was used to match the nearest meteorological monitoring station to the disaster area, and meteorological monitoring data collected by the meteorological monitoring station was obtained. The meteorological monitoring data were subjected to climatological boundary value test, time series consistency test, and spatial interpolation rationality test in sequence. For meteorological monitoring stations with missing values ​​less than the preset percentage, the meteorological monitoring data were supplemented by Kriging interpolation algorithm or neighboring station mean algorithm. Logical consistency checks and on-site cross-validation are performed on agricultural insurance claims data to eliminate erroneous data, and the claims amount in the agricultural insurance claims data is converted into the claims loss rate. The time scale is divided according to the type of disaster: for rainstorms, hourly data is used, with the time scale being the first preset number of days before the claim date; for drought, the time scale is the second preset number of days before the claim date; for cold damage, the time scale is the third preset number of days before the claim date; and for heat damage, the time scale is the fourth preset number of days before the claim date. Using ArcGIS spatial analysis tools, meteorological monitoring data collected from meteorological monitoring stations are interpolated to the township scale through an inverse distance weighted interpolation algorithm, realizing a township-level gridded expression of meteorological disaster intensity. This is then matched with the township spatial attributes of agricultural insurance claims data to form an integrated spatiotemporal database of region, disaster type, reproductive stage, meteorological intensity, and loss rate.

3. The method for constructing dynamic claims thresholds for agricultural insurance based on disaster-vulnerability coupling as described in claim 1, characterized in that, The extraction of core disaster-causing factors based on the database using principal component analysis and machine learning algorithms includes: Machine learning and principal component analysis algorithms were used to extract core disaster-causing factors for different types of disasters from raw meteorological monitoring data. These core factors were then screened based on out-of-bag error and feature importance to construct a graded and quantitative meteorological disaster intensity index system. The core disaster-causing factors for rainstorms included: The rainfall, maximum hourly rainfall, and number of days of continuous heavy rain are calculated for the first preset number of days prior to the claim date. The core disaster-causing factors of drought include: The daily average relative humidity, average temperature, daily precipitation, soil relative humidity, number of consecutive days without effective precipitation, and sunshine hours are calculated for the second preset number of days prior to the claim date. The core disaster-causing factors of cold damage include: the daily minimum temperature three preset days prior to the claim date, the number of days each year when the daily minimum temperature is lower than the first preset temperature, the number of consecutive days of low temperature, and the accumulated temperature anomaly. The core disaster-causing factors of heat damage include: the highest daily temperature four preset days prior to the claim date, the average relative humidity, the longest consecutive number of days with a temperature greater than or equal to the second preset temperature, the number of days with a temperature greater than or equal to the second preset temperature, the effective accumulated high temperature, precipitation, and sunshine hours.

4. The method for constructing dynamic claims thresholds for agricultural insurance based on disaster-vulnerability coupling as described in claim 3, characterized in that, The method employs entropy weight-analytic hierarchy process (AHP) to combine and weight core disaster-causing factors to construct a comprehensive disaster intensity index, including: The core disaster-causing factors are combined and weighted using an entropy weight algorithm and an analytic hierarchy process (AHP) algorithm to construct a comprehensive disaster-causing intensity index. The consistency ratio of the weight results of the AHP algorithm must be less than or equal to a first preset value. The percentile method was used to classify the intensity of meteorological disasters to determine each intensity level and its corresponding threshold range. The intensity levels of meteorological disasters included four categories: mild, moderate, severe, and extremely severe. The threshold range for the mild level was as follows: The grading threshold range corresponding to the medium level is: The grading threshold range corresponding to the heavy level is: The grading threshold corresponding to the extremely serious level is in the range of [range missing]. .

5. The method for constructing dynamic claims thresholds for agricultural insurance based on disaster-vulnerability coupling as described in claim 4, characterized in that, The method of classifying the intensity of meteorological disasters using the percentile method to determine each intensity level and its corresponding threshold range also includes: The interannual variation trend of disaster types is analyzed using trend analysis algorithm, wavelet analysis algorithm, spatial clustering analysis algorithm, and Mann-Kendall trend test algorithm. Spatial variability function is used to analyze the spatial variation characteristics of disaster types in order to determine spatial correlation distances; Meteorological monitoring data from different monitoring areas were obtained, and the disaster-causing characteristics of four types of disasters in each monitoring area were analyzed from three dimensions: time, space, and intensity, in order to determine the occurrence frequency, spatiotemporal distribution pattern, and intensity evolution characteristics of the types of disasters in different monitoring areas. Based on the disaster-causing factor-disaster-loss theory of agricultural disaster science, this study uses literature review combined with structural equation modeling to identify the vulnerability response paths of different crops to different types of disasters. This study analyzes the formation and evolution of crop vulnerability from four dimensions: plant physiological vulnerability, growth and development vulnerability, yield and quality vulnerability, and economic vulnerability. By using structural equation modeling to quantify the direct and indirect effects between core disaster-causing factors, four dimensions of crop vulnerability, and claims loss rate, the mediating role of crop vulnerability in the relationship between disaster-causing factors and claims loss rate is clarified.

6. The method for constructing dynamic claims thresholds for agricultural insurance based on disaster-vulnerability coupling as described in claim 5, characterized in that, The text describes the quantification of the direct and indirect effects between core disaster-causing factors, four dimensions of crop vulnerability, and claims loss rates using structural equation modeling. It clarifies the mediating role of crop vulnerability between disaster-causing factors and claims loss rates. The text also includes: Using bivariate correlation analysis, nonlinear regression analysis, and quantile regression, we analyzed the nonlinear relationships between the intensity of meteorological disasters and crop vulnerability, and between crop vulnerability and claim loss rate, under different meteorological disaster intensities and different growth stages. The interaction effect analysis method was used to examine the influence of the interaction between crop type, disaster type, and growth stage on the coupling relationship, and the analysis of variance method was used to examine the significance of the interaction effect. Through group comparative analysis, the differential characteristics of crops in the disaster-vulnerability coupling process are revealed, and a framework for analyzing the disaster-vulnerability coupling mechanism is constructed. The rationality of the coupling mechanism framework is verified by structural equation modeling and path coefficients. The intensity of meteorological disaster, the vulnerability of crops in four dimensions, and the loss rate of claims are used as latent variables. The coefficients and significance of each path are quantified to form a quantifiable and verifiable nonlinear coupling mechanism framework of disaster-vulnerability.

7. The method for constructing dynamic claims thresholds for agricultural insurance based on disaster-vulnerability coupling as described in claim 6, characterized in that, The method involves constructing a multi-level Bayesian model based on the comprehensive disaster intensity index and using the claim loss rate as the actual loss observation, and then combining it with a zero-inflation model for inversion to obtain the crop vulnerability function, including: Standardize the intensity of meteorological disasters: Standardize the comprehensive intensity index of each type of disaster from 0 to 1 to eliminate the difference in dimensions and obtain the standardized intensity of meteorological disasters. Standardize the physical loss rate of crops: Standardize the extreme values ​​of the actual crop loss rate at the township level in the disaster-stricken area; Using exploratory data analysis algorithms, by plotting scatter plots and fitting curve trends, we can preliminarily determine the type of response relationship between the intensity of meteorological disasters and the rate of physical loss of crops. Group analysis was conducted for different crops, different types of disasters, and different growth stages to clarify the response relationship characteristics under different scenarios; The distribution characteristics of crop physical loss rate were analyzed using the kernel density estimation algorithm. Then, the Pearson correlation coefficient and Spearman correlation coefficient were used to determine the type and significance of the correlation between the intensity of meteorological disaster and crop physical loss rate.

8. The method for constructing dynamic claims thresholds for agricultural insurance based on disaster-vulnerability coupling as described in claim 7, characterized in that, The method employs a kernel density estimation algorithm to analyze the distribution characteristics of crop physical loss rates. Then, it uses Pearson and Spearman correlation coefficients to determine the type and significance of the correlation between the intensity of meteorological disasters and crop physical loss rates. This is followed by: Selection of crop vulnerability function type: Nonlinear models are selected as candidate crop vulnerability functions, with standardized meteorological disaster intensity as independent variable and standardized crop physical loss rate as dependent variable. The candidate crop vulnerability functions are fitted and screened to obtain the preferred crop vulnerability function. Mathematical Construction and Probabilistic Expression of Crop Vulnerability Function: This involves calculating the actual claim loss rate of township i during its reproductive stage t. Considered as the intensity of disaster caused by meteorological disasters Potential crop vulnerability status According to Bayes' theorem, the posterior distribution of crop vulnerability, driven by a co-driven stochastic process, can be expressed as: ， In the formula, P Represents the probability distribution function; This indicates the actual claim loss rate during the maternity stage in rural areas. Intensity of meteorological disasters Under certain conditions, the potential vulnerability status of crops The posterior probability distribution; This represents the intensity of a given meteorological disaster in township i during the reproductive stage t. Potential crop vulnerability status Under these conditions, the actual claim loss rate was observed. The likelihood function; Indicates the potential vulnerability status of crops The prior probability distribution; These parameters represent the relevant parameters of crop growth stages, and are used to characterize the prior features of crop sensitivity to meteorological disasters at different growth stages. Crop vulnerability is classified into four levels: low, medium, relatively high, and high, based on the calculation results of the optimized crop vulnerability function, so as to realize the quantitative and hierarchical expression of crop vulnerability. A sample segmentation algorithm was used to verify the accuracy of the selected crop vulnerability function.

9. The method for constructing dynamic claims thresholds for agricultural insurance based on disaster causality-vulnerability coupling as described in claim 1, characterized in that, The model for disaster-vulnerability coupling degree and coupling coordination degree, constructed based on the crop vulnerability function, is used to quantify the synergistic damaging effect of meteorological disaster intensity and crop vulnerability, including: Characterizing temporal heterogeneity: Using the moving average algorithm and the trend coefficient algorithm, we analyze the changing trends and seasonal fluctuations of the monthly meteorological disaster intensity and crop vulnerability; Characterizing spatial heterogeneity: Using ArcGIS spatial analysis tools, spatial autocorrelation analysis algorithm is used to examine the spatial clustering characteristics of disaster-causing capacity and crop vulnerability at the township level, and geographic weighted regression algorithm is used to quantify the contribution of geographic elements to spatial heterogeneity. Analyze the single-factor and interaction explanatory power of geographical elements on the spatial heterogeneity of disaster-vulnerability; Coupling Degree Model Construction: A coupling coordination degree model is introduced, which treats the intensity of meteorological disasters and the vulnerability of crops as two coupled subsystems. A disaster-vulnerability coupling degree and coupling coordination degree model is constructed to calculate the coupling degree and coupling coordination degree at the township level and on a monthly basis. Analysis of dynamic evolution patterns: Using a centroid migration model, we analyze the migration direction and distance of the centroid of disaster-vulnerability coupling degree and coupling coordination degree over time. A panel regression model was adopted, with the township-level insurance claim loss rate as the dependent variable and the interaction term between the intensity of meteorological disasters and crop vulnerability as the core explanatory variable. Control variables were introduced to construct a fixed-effects panel model to quantify the contribution of the intensity of meteorological disasters, crop vulnerability, and the interaction to insurance losses. The formula for defining the disaster-vulnerability coupling degree and coupling coordination degree model is as follows: ， In the formula, C represents the disaster-vulnerability coupling degree and coupling coordination degree model; H represents the disaster-causing intensity of meteorological disasters; and V represents the crop vulnerability index. Synergistic effect quantification: The marginal effect analysis algorithm is used to quantify the synergistic damage effect of meteorological disaster intensity and crop vulnerability at different levels.

10. The method for constructing dynamic claims thresholds for agricultural insurance based on disaster-vulnerability coupling as described in claim 1, characterized in that, Based on the disaster-vulnerability coupling degree and coupling coordination degree model, the initial claim threshold, and spatiotemporal heterogeneity correction, a three-dimensional dynamic claim threshold matrix is ​​generated, categorized by region, disaster type, and reproductive stage. The claim threshold is then determined based on this matrix, including: A three-dimensional matrix containing thresholds for single disaster types and joint thresholds for typical compound disaster types is constructed. For compound disasters, the equivalent intensity method is used to convert the compound disaster-causing index into the equivalent value of a single disaster type. The initial claim trigger thresholds for different crops, different types of disasters, and different growth stages are determined by using machine learning algorithms combined with percentile methods. Based on the spatiotemporal heterogeneity analysis results of disaster-vulnerability, a geographically weighted correction method combined with a time series correction method is used to refine the initial claims threshold. The refined correction includes spatial correction and time correction. Using townships as the row dimension and disaster type, crop type, and growth stage as the column dimension, the revised graded claim thresholds are incorporated into a three-dimensional matrix to construct a township-level agricultural insurance claim threshold identification model. Reasonableness verification of the three-dimensional matrix: The historical claims data back-substitution test method is adopted. The threshold matrix is ​​applied to historical claims cases, and the consistency between the claims trigger results calculated by the three-dimensional matrix and the actual claims results is compared.

Images