A method for determining a flood-drought disaster threshold for sensitive targets

Abstract

This invention discloses a method for determining the threshold for rapid drought-flood transitions to disasters targeting sensitive objects. It innovatively incorporates society, water resources, and the ecosystem as three core objectives into the threshold determination framework, breaking through the limitations of traditional single-factor fixed thresholds. By treating the disaster threshold as a dynamic multidimensional surface, it can reveal the nonlinear variation law of the critical conditions for disasters under different characteristics of disaster-bearing bodies. This invention achieves quantitative analysis from the description of disaster-loss relationships to the determination of critical conditions through the construction of a nonlinear regression model and its inverse solution. Through comparative analysis of subsystem and comprehensive thresholds, this invention can identify the contribution of each system to the overall threshold, providing a basis for identifying the dominant system and key vulnerable factors. The final threshold surface output by this invention is presented in the form of a three-dimensional graph, which is intuitive and easy to apply, providing scientific support for early warning of rapid drought-flood transition disaster risks, land spatial planning, and disaster prevention and mitigation decision-making.

Description

Technical Field

[0001] This invention belongs to the field of natural disaster risk assessment technology, specifically involving the design of a method for determining the threshold of rapid shift from drought to flood disaster for sensitive targets. Background Technology

[0002] In recent years, influenced by both climate change and human activities, extreme hydrological events have become more frequent. Rapid drought-flood transitions, a special type of complex extreme event—the rapid shift from drought to flooding in the same region within a short period, or vice versa—are characterized by their suddenness and destructiveness, and have become a significant form of new drought and flood disasters in my country, posing a serious threat to agricultural production, socio-economic development, and ecosystems. Accurately identifying the disaster-causing conditions of rapid drought-flood transitions is a key scientific issue for disaster risk early warning and disaster prevention and mitigation decision-making.

[0003] Currently, research on determining the threshold for disasters caused by rapid shifts between drought and flood is still in the exploratory stage. Existing methods mostly employ a fixed threshold strategy, setting a specific value for a rapid shift indicator based on historical disaster statistics or empirical formulas as the disaster threshold; exceeding this value indicates a disaster. However, such methods ignore the spatiotemporal heterogeneity of vulnerability in disaster-bearing entities. That is, the actual losses caused by rapid shifts between drought and flood of the same intensity vary significantly under different population densities, water resource endowments, and ecological conditions. Therefore, fixed thresholds cannot objectively reflect the true disaster-causing conditions. Furthermore, existing research lacks a systematic characterization of disaster-bearing entities, failing to incorporate multi-dimensional vulnerability factors such as social factors, water resources, and ecosystems into a unified analytical framework, resulting in a lack of comprehensiveness in threshold determination results. More critically, although some studies have attempted to use machine learning models to fit the relationship between disaster factors and losses, most stop at correlation analysis or feature importance ranking, failing to further deduce the disaster intensity conditions required to reach the critical loss threshold, thus failing to truly achieve a quantitative determination of the disaster threshold. Summary of the Invention

[0004] The purpose of this invention is to address the problem that existing methods for determining the threshold of sudden drought-flood transitions to disaster have failed to achieve a truly quantitative determination of the disaster threshold. This invention proposes a method for determining the threshold of sudden drought-flood transitions to disaster for sensitive targets, providing a scientific basis for the refined management of disaster risks caused by sudden drought-flood transitions.

[0005] The technical solution of this invention is: a method for determining the threshold of rapid shift from drought to flood disaster for sensitive targets, comprising the following steps: S1. Collect data on the intensity of drought-flood transitions, social system factors, water resource system factors, ecosystem factors, and historical disaster damage in the target area within the time series to be evaluated.

[0006] S2. Process the collected data with a unified spatiotemporal resolution to construct a multi-system coupled dataset that includes the intensity of rapid shifts between drought and flood, factors of each system, and disaster loss variables.

[0007] S3. Using the intensity of rapid shifts between drought and flood and various system factors as input variables, and disaster damage variables as response variables, a nonlinear regression model is constructed and trained to fit the response relationship between disaster damage and multiple factors.

[0008] S4. Set the critical loss value for disaster based on the historical disaster loss distribution or disaster level in the historical disaster loss data.

[0009] S5. Based on the trained nonlinear regression model, the critical conditions for the drought-flood transition intensity required to reach the disaster-causing critical loss value are derived by numerical solution and reverse derivation as a function of various system factors, and the disaster-causing threshold surfaces of the social system, water resource system and ecosystem are obtained respectively.

[0010] S6. Based on all system factors, the critical conditions for the change of drought-flood intensity with key factors are derived in reverse by reducing dimensions or screening key factors to reach the critical loss value of disaster, thus obtaining the disaster threshold surface of the comprehensive system.

[0011] The beneficial effects of this invention are as follows: This invention innovatively incorporates society, water resources, and the ecosystem as three core objectives into the threshold determination framework for rapid drought-flood transitions, breaking through the limitations of traditional single-factor fixed thresholds. It treats the disaster threshold as a dynamic multidimensional surface, revealing the nonlinear variation law of the critical conditions for disaster under different disaster-bearing body characteristics. By constructing a nonlinear regression model and solving it in reverse, it achieves a quantitative analysis from the description of the disaster-loss relationship to the critical conditions. Through comparative analysis of subsystems and the comprehensive threshold, the contribution of each system to the overall threshold can be identified, providing a basis for identifying the dominant system and key vulnerable factors. The final output threshold surface is presented in the form of a three-dimensional graph, which is intuitive and easy to apply, providing scientific support for early warning of rapid drought-flood transition disaster risks, land spatial planning, and disaster prevention and mitigation decision-making.

[0012] Furthermore, the drought-flood transition intensity data in step S1 is either a comprehensive drought-flood transition index constructed based on multi-source data or an existing drought-flood transition index product.

[0013] Social system factor data include the proportion of arable land, the proportion of urban areas, grain output, population, and GDP density.

[0014] Water resource system factors include water production coefficient, water resource vulnerability, soil erosion sensitivity, water consumption per unit area, per capita water consumption, and water consumption per 10,000 yuan of GDP.

[0015] Ecosystem factors include the Landscape Vulnerability Index (LVI), the Normalized Difference Vegetation Index (NDVI), vegetation resilience, and vegetation water use efficiency.

[0016] Historical disaster damage data includes at least one of the following: direct economic losses, affected crop area, and affected population.

[0017] The beneficial effects of the above-mentioned further scheme are: by comprehensively collecting representative factors from the three systems, the disaster-bearing characteristics of the target to the sudden shift from drought to flood can be characterized from multiple dimensions such as exposure, sensitivity and adaptability; disaster damage data, as a direct manifestation of disaster, provides reliable response variables for subsequent modeling; and the fusion of multi-source data lays the foundation for constructing a high-precision, multi-scale threshold surface.

[0018] Furthermore, step S2 includes the following sub-steps: S21. Using a certain reference raster data of the target area as a template, the spatial resolution, projection coordinates and spatial range of the template are used as a unified standard.

[0019] S22. Convert the collected vector format data into raster data using a spatialization method, and register all raster data onto the template grid using a resampling method.

[0020] S23. Imput or remove missing values ​​generated during the registration process to obtain complete multi-system data, so as to construct a multi-system coupled dataset containing the intensity of drought-flood transition, factors of each system, and disaster loss variables.

[0021] The beneficial effects of the above-mentioned further solutions are: by unifying spatiotemporal registration, the problems of inconsistent resolution and coordinate system of multi-source data are solved, ensuring the integrity and comparability of different pixel features in the evaluated time series in subsequent analysis; and the reasonable missing value handling strategy preserves the effective samples to the maximum extent and avoids information loss due to missing data.

[0022] Furthermore, step S3 includes the following sub-steps: S31. Divide the multi-system coupled dataset into a training set and a test set according to the year, with the year of the training set preceding that of the test set.

[0023] S32. Using the intensity of sudden shifts between drought and flood and various system factors in the training set as input variables, and the disaster loss variable as the response variable, train a random forest regression model. The model parameters include the number of decision trees, maximum depth, minimum number of sample splits, and minimum number of samples in the leaf nodes.

[0024] S33. Evaluate model performance using the test set. Evaluation metrics include the coefficient of determination R. 2 And root mean square error (RMSE).

[0025] S34. Output the importance scores of each input variable to obtain the response relationship between disaster damage and multiple factors.

[0026] The beneficial effects of the above-mentioned further solutions are: the random forest regression model can automatically capture the nonlinear relationship and complex interaction between input features and disaster damage, without the need for a pre-defined function form, and has strong robustness and generalization ability; dividing the training set and test set by year can verify the model's time extrapolation ability and ensure that it can be used to predict future scenarios; feature importance scoring provides an objective basis for subsequent steps such as screening key factors and dimensionality reduction visualization.

[0027] Furthermore, the method for setting the critical loss value for disaster based on the historical disaster loss distribution in step S4 is as follows: statistically analyze the cumulative distribution of historical disaster loss data, select a certain quantile as the critical loss value for disaster, and the quantile value ranges from 80% to 95%.

[0028] The method for setting the critical loss value for disaster based on the disaster level in step S4 is as follows: Based on the disaster level standard, the historical disaster loss data is divided into different levels, and the lower limit of the loss corresponding to a certain level is selected as the critical loss value for disaster.

[0029] The beneficial effects of the above-mentioned further solutions are: defining the critical loss value for disasters by using historical data quantiles or existing disaster level standards makes the threshold determination have clear statistical significance or policy basis, avoiding subjective arbitrariness; different critical loss values ​​for disasters can be flexibly selected according to different application scenarios to meet the needs of multi-level risk management.

[0030] Furthermore, step S5 includes the following sub-steps: S51. Select the social system, water resource system, and ecosystem as the systems to be inverted in sequence, and fix the factors of the other two systems as typical values.

[0031] S52. Discretize each factor of the system to be inverted into grid points within its actual value range to form a factor combination space.

[0032] S53. For each combination of factors, the corresponding critical intensity of sudden shift from drought to flood is obtained by using a trained nonlinear regression model. : in This represents the critical loss value for disaster. This represents a trained nonlinear regression model. This represents the factor vector of the system to be inverted. and This represents the factor vectors of the other two systems after they have been fixed.

[0033] S54. Traverse all factor combinations to obtain a series Data points were used to construct disaster threshold surfaces for the social system using interpolation or surface fitting methods. Disaster threshold surface of water resource system and the disaster threshold surface of the ecosystem ,in The surface function representing the disaster threshold of a social system. The factor vector representing the social system, The surface function representing the disaster threshold of the water resource system. Represents the factor vector of the water resource system. The surface function representing the disaster threshold of an ecosystem. A factor vector representing an ecosystem.

[0034] The beneficial effects of the above-mentioned further scheme are as follows: by fixing other system factors, the independent influence of the system to be inverted is isolated, and the quantitative analysis of the subsystem threshold is realized; the combination of grid discretization and numerical solution can handle nonlinear models of any form and has strong applicability; the final threshold surface intuitively shows the critical drought-flood transition intensity required to trigger disaster under different combinations of system factors, and can be directly used for risk zoning and vulnerability assessment.

[0035] Furthermore, the solution method in step S53 adopts the bisection method, specifically as follows: A1. Determine the search interval for the intensity of the sudden shift from drought to flood. And ensure that the predicted loss is monotonic within the search interval and includes the critical loss value for disaster. .

[0036] A2. Calculate the midpoint of the search interval. and the midpoint The combined values ​​of the factors of the system to be inverted, along with the factor values ​​of the other two systems already fixed as typical values, are used as the input feature vector. This input feature vector is then fed into a trained nonlinear regression model, and the corresponding prediction loss is output. ,like Then As a critical value, otherwise according to and Narrowing the range based on size relationship: If Then let ,like Then let ,in This indicates the preset tolerance error.

[0037] A3. Repeat steps A1 to A2 until the convergence condition is met or the maximum number of iterations is reached to obtain the solution.

[0038] The advantages of the above-mentioned further scheme are: the bisection method is simple in principle, converges stably, and does not require the calculation of derivatives, making it suitable for numerical inversion of nonparametric models such as random forests; the preset tolerance error can control the solution accuracy and balance computational efficiency and accuracy.

[0039] Furthermore, step S6 includes the following sub-steps: S61. Based on the importance of the features, select the 2 to 3 key factors with the highest importance from all system factors.

[0040] S62. Fix the remaining non-critical factors to typical values.

[0041] S63. Discretize the key factors within their value range to form a key factor combination space.

[0042] S64. For each combination of key factors, the corresponding critical intensity of sudden shift from drought to flood is obtained by using a trained nonlinear regression model. .

[0043] S65. Iterate through all combinations of key factors to obtain a series of... Data points are used to construct the disaster threshold surface of the integrated system through interpolation or surface fitting methods. ,in The surface function representing the disaster threshold of the integrated system. This represents the key factor vector.

[0044] The beneficial effects of the above-mentioned further scheme are: by screening key factors to achieve dimensionality reduction, the comprehensive threshold surface can be visualized and interpreted; key factors usually represent the dominant system or key vulnerable links, and the threshold surface under their interaction reveals the core law of multi-system coupling effect. Attached Figure Description

[0045] Figure 1 The diagram shown is a flowchart of a method for determining the threshold of sudden shift from drought to flood disaster for sensitive targets, provided by an embodiment of the present invention. Detailed Implementation

[0046] Exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be understood that the embodiments shown and described in the drawings are merely exemplary and are intended to illustrate the principles and spirit of the invention, and are not intended to limit the scope of the invention.

[0047] This invention provides a method for determining the threshold for rapid shift from drought to flood disaster for sensitive targets, such as... Figure 1 As shown, the process includes the following steps S1 to S6: S1. Collect data on the intensity of drought-flood transitions, social system factors, water resource system factors, ecosystem factors, and historical disaster damage in the target area within the time series to be evaluated.

[0048] In this embodiment of the invention, the drought-flood transition intensity data is a comprehensive drought-flood transition index constructed based on multi-source data such as meteorological, hydrological, and soil data, or it can be directly adopted from existing drought-flood transition index products.

[0049] Social system factor data include the proportion of arable land, the proportion of urban areas, grain output, population, and GDP density.

[0050] Water resource system factors include water production coefficient, water resource vulnerability, soil erosion sensitivity, water consumption per unit area, per capita water consumption, and water consumption per 10,000 yuan of GDP.

[0051] Ecosystem factors include the Landscape Vulnerability Index (LVI), the Normalized Difference Vegetation Index (NDVI), vegetation resilience, and vegetation water use efficiency.

[0052] Historical disaster damage data includes at least one of the following: direct economic losses, affected crop area, and affected population.

[0053] In this embodiment of the invention, when historical disaster damage data includes multiple types (such as direct economic losses, affected crop area, and affected population), one of the following methods can be used for modeling: (1) Separate modeling: For each disaster loss variable, a nonlinear regression model is independently constructed to obtain the threshold surface corresponding to different disaster loss types, which is used for special risk assessment.

[0054] (2) Comprehensive loss index: Multiple disaster loss variables are integrated into a comprehensive loss index by normalization and weighted summation or principal component analysis, and then a single model is constructed using this index as the response variable to obtain the comprehensive disaster threshold surface, which is used for overall risk assessment.

[0055] In practical applications, the appropriate method can be selected according to the decision-making needs. If the dimensions of each disaster loss variable are significantly different, it is recommended to prioritize separate modeling in order to provide more refined risk management services for different fields such as agriculture, economy, and population.

[0056] S2. Process the collected data with a unified spatiotemporal resolution to construct a multi-system coupled dataset that includes the intensity of rapid shifts between drought and flood, factors of each system, and disaster loss variables.

[0057] Step S2 includes the following sub-steps S21 to S24: S21. Using a certain reference raster data of the target area as a template, the spatial resolution, projection coordinates and spatial range of the template are used as a unified standard.

[0058] S22. Convert the collected vector format data into raster data using a spatialization method, and register all raster data onto the template grid using a resampling method.

[0059] In this embodiment of the invention, the spatialization method may employ area weight allocation, inverse distance weight interpolation, or kriging interpolation, and the resampling method may employ nearest neighbor, bilinear interpolation, or cubic convolution.

[0060] S23. Imput or remove missing values ​​generated during the registration process to obtain complete multi-system data, so as to construct a multi-system coupled dataset containing the intensity of drought-flood transition, factors of each system, and disaster loss variables.

[0061] In this embodiment of the invention, the interpolation method may be time series linear interpolation, spatial proximity interpolation, or regression interpolation based on machine learning.

[0062] S3. Using the intensity of rapid shifts between drought and flood and various system factors as input variables, and disaster damage variables as response variables, a nonlinear regression model is constructed and trained to fit the response relationship between disaster damage and multiple factors.

[0063] In this embodiment of the invention, the nonlinear regression model adopts the random forest regression model.

[0064] Step S3 includes the following sub-steps S31 to S34: S31. Divide the multi-system coupled dataset into a training set and a test set according to the year, with the year of the training set preceding that of the test set.

[0065] S32. Using the intensity of sudden shifts between drought and flood and various system factors in the training set as input variables (in this embodiment, the input feature matrix X_train), and the disaster loss variable as the response variable (in this embodiment, the output vector Y_train), train a random forest regression model. The model parameters include the number of decision trees, the maximum depth, the minimum number of sample splits, and the minimum number of samples in the leaf nodes.

[0066] S33. Evaluate model performance using the test set. Evaluation metrics include the coefficient of determination R. 2 And root mean square error (RMSE).

[0067] S34. Output the importance scores of each input variable to obtain the response relationship between disaster damage and multiple factors.

[0068] In this embodiment of the invention, in order to enhance the interpretability of the random forest regression model and reveal the marginal impact and interaction effect of each factor on the loss, the Partial Dependence Plot (PDP) or SHAP (SHapley Additive exPlanations) values ​​can be used to interpret and analyze the trained random forest regression model.

[0069] Specifically, the partial dependency plot, by fixing other factors, shows the average trend of predicted loss changes when one or two factors change, which can intuitively identify whether there is a threshold abrupt change point in the marginal effect of the intensity of drought-flood transitions on the loss; the SHAP value quantifies the contribution of each factor in each sample to the prediction result, which helps to understand the dominant vulnerability factors in different regions and years. These interpretive analysis results can serve as evidence for the physical meaning of the threshold surface and can also be used to verify whether the model conforms to domain knowledge.

[0070] S4. Set the critical loss value for disaster based on the historical disaster loss distribution or disaster level in the historical disaster loss data.

[0071] In this embodiment of the invention, the method for setting the critical loss value for disaster based on the historical disaster loss distribution is as follows: statistically analyze the cumulative distribution of historical disaster loss data, select a certain quantile as the critical loss value for disaster, and the quantile value ranges from 80% to 95%.

[0072] The method for setting the critical loss value for disaster based on the disaster level is as follows: Based on the disaster level standard, historical disaster loss data are divided into different levels, and the lower limit of loss corresponding to a certain level is selected as the critical loss value for disaster.

[0073] S5. Based on the trained nonlinear regression model, the critical conditions for the drought-flood transition intensity required to reach the disaster-causing critical loss value are derived by numerical solution and reverse derivation as a function of various system factors, and the disaster-causing threshold surfaces of the social system, water resource system and ecosystem are obtained respectively.

[0074] Step S5 includes the following sub-steps S51 to S54: S51. Select the social system, water resource system, and ecosystem as the systems to be inverted in sequence, and fix the factors of the other two systems as typical values.

[0075] In this embodiment of the invention, typical values ​​are the median, mean, or a specific percentile.

[0076] S52. Discretize each factor of the system to be inverted into grid points within its actual value range to form a factor combination space.

[0077] S53. For each combination of factors, the corresponding critical intensity of sudden shift from drought to flood is obtained by using a trained nonlinear regression model. : in This represents the critical loss value for disaster. This represents a trained nonlinear regression model. This represents the factor vector of the system to be inverted. and This represents the factor vectors of the other two systems after they have been fixed.

[0078] In this embodiment of the invention, the solution method can be the bisection method, Newton's method, or grid search method, with the bisection method being preferred. The specific steps are as follows: A1. Determine the search interval for the intensity of the sudden shift from drought to flood. And ensure that the predicted loss is monotonic within the search interval and includes the critical loss value for disaster. .

[0079] A2. Calculate the midpoint of the search interval. and the midpoint The combined values ​​of the factors of the system to be inverted, along with the factor values ​​of the other two systems already fixed as typical values, are used as the input feature vector. This input feature vector is then fed into a trained nonlinear regression model, and the corresponding prediction loss is output. ,like Then As a critical value, otherwise according to and Narrowing the range based on size relationship: If Then let ,like Then let ,in This indicates the preset tolerance error.

[0080] In this embodiment of the invention, the midpoint of the search interval This refers to the intensity of the sudden shift from drought to flood in the current iteration step, and the predicted loss. This will be directly compared with the disaster-causing critical loss value set in step S4. By comparing the results, we can determine whether the current intensity of the rapid shift from drought to flood is too high, too low, or has reached the critical condition, thereby determining the direction of the contraction of the next search interval.

[0081] A3. Repeat steps A1 to A2 until the convergence condition is met or the maximum number of iterations is reached to obtain the solution result, so as to ensure the stability and efficiency of the solution process.

[0082] S54. Traverse all factor combinations to obtain a series Data points were used to construct disaster threshold surfaces for the social system using interpolation or surface fitting methods. Disaster threshold surface of water resource system and the disaster threshold surface of the ecosystem ,in The surface function representing the disaster threshold of a social system. The factor vector representing the social system, The surface function representing the disaster threshold of the water resource system. Represents the factor vector of the water resource system. The surface function representing the disaster threshold of an ecosystem. A factor vector representing an ecosystem.

[0083] S6. Based on all system factors, the critical conditions for the change of drought-flood intensity with key factors are derived in reverse by reducing dimensions or screening key factors to reach the critical loss value of disaster, thus obtaining the disaster threshold surface of the comprehensive system.

[0084] Step S6 includes the following sub-steps S61 to S65: S61. Based on the importance of the features, select the 2 to 3 key factors with the highest importance from all system factors.

[0085] S62. Fix the remaining non-critical factors to typical values.

[0086] S63. Discretize the key factors within their value range to form a key factor combination space.

[0087] S64. For each combination of key factors, the corresponding critical intensity of sudden shift from drought to flood is obtained by using a trained nonlinear regression model. The solution method is the same as in step S53.

[0088] S65. Iterate through all combinations of key factors to obtain a series of... Data points are used to construct the disaster threshold surface of the integrated system through interpolation or surface fitting methods. ,in The surface function representing the disaster threshold of the integrated system. This represents the key factor vector.

[0089] In this embodiment of the invention, the disaster threshold surfaces of the social system, water resource system, and ecosystem generated in step S5, as well as the disaster threshold surface of the integrated system generated in step S6, are all output in the form of three-dimensional graphs. The output format is intuitive and convenient for different user groups to use according to their needs.

[0090] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A method for determining the threshold for rapid shift from drought to flood disaster for sensitive targets, characterized in that, Includes the following steps: S1. Collect data on the intensity of rapid shifts between drought and flood, social system factors, water resource system factors, ecosystem factors, and historical disaster damage in the target area within the time series to be evaluated. S2. Process the collected data with a unified spatiotemporal resolution to construct a multi-system coupled dataset that includes the intensity of rapid shifts between drought and flood, factors of each system, and disaster loss variables. S3. Using the intensity of rapid shifts between drought and flood and various system factors as input variables, and disaster damage variables as response variables, a nonlinear regression model is constructed and trained to fit the response relationship between disaster damage and multiple factors. S4. Set the critical loss value for disaster based on the historical disaster loss distribution or disaster level in the historical disaster loss data; S5. Based on the trained nonlinear regression model, the critical conditions for the intensity of the rapid shift between drought and flood as required to reach the critical loss value of disaster are derived by numerical solution and reverse derivation, and the disaster threshold surfaces of social system, water resource system and ecosystem are obtained respectively. S6. Based on all system factors, the critical conditions for the change of drought-flood intensity with key factors are derived in reverse by reducing dimensions or screening key factors to reach the critical loss value of disaster, thus obtaining the disaster threshold surface of the comprehensive system.

2. The method for determining the threshold for rapid shift from drought to flood disaster for sensitive targets according to claim 1, characterized in that, The drought-flood transition intensity data in step S1 is either a comprehensive drought-flood transition index constructed based on multi-source data or an existing drought-flood transition index product. The social system factor data includes the proportion of arable land, the proportion of urban areas, grain output, population, and GDP density. The water resources system factors include water production coefficient, water resources vulnerability, soil erosion sensitivity, water consumption per unit area, per capita water consumption, and water consumption per 10,000 yuan of GDP. The ecosystem factors include the Landscape Vulnerability Index (LVI), the Normalized Difference Vegetation Index (NDVI), vegetation resilience, and vegetation water use efficiency. The historical disaster damage data includes at least one of the following: direct economic losses, affected crop area, and affected population.

3. The method for determining the threshold for rapid shift from drought to flood disaster for sensitive targets according to claim 1, characterized in that, Step S2 includes the following sub-steps: S21. Using a certain reference raster data of the target area as a template, the spatial resolution, projection coordinates and spatial range of the template are used as a unified standard. S22. Convert the collected vector format data into raster data using a spatialization method, and register all raster data onto the template grid using a resampling method; S23. Imput or remove missing values ​​generated during the registration process to obtain complete multi-system data, so as to construct a multi-system coupled dataset containing the intensity of drought-flood transition, factors of each system, and disaster loss variables.

4. The method for determining the threshold for rapid shift from drought to flood disaster for sensitive targets according to claim 1, characterized in that, Step S3 includes the following sub-steps: S31. Divide the multi-system coupled dataset into a training set and a test set according to the year, with the years in the training set preceding those in the test set; S32. Using the intensity of sudden shifts between drought and flood and various system factors in the training set as input variables and the disaster loss variable as the response variable, train a random forest regression model. The model parameters include the number of decision trees, maximum depth, minimum number of sample splits, and minimum number of samples in the leaf nodes. S33. Evaluate model performance using the test set. Evaluation metrics include the coefficient of determination R. 2 and root mean square error (RMSE); S34. Output the importance scores of each input variable to obtain the response relationship between disaster damage and multiple factors.

5. The method for determining the threshold for rapid shift from drought to flood disaster for sensitive targets according to claim 1, characterized in that, The method for setting the critical disaster loss value based on the historical disaster loss distribution in step S4 is as follows: statistically analyze the cumulative distribution of historical disaster loss data, select a certain quantile as the critical disaster loss value, and the quantile value ranges from 80% to 95%. The method for setting the critical loss value for disaster based on the disaster level in step S4 is as follows: according to the disaster level standard, the historical disaster loss data is divided into different levels, and the lower limit of the loss corresponding to a certain level is selected as the critical loss value for disaster.

6. The method for determining the threshold for rapid shift from drought to flood disaster for sensitive targets according to claim 1, characterized in that, Step S5 includes the following sub-steps: S51. Select the social system, water resource system and ecosystem as the systems to be inverted in sequence, and fix the factors of the other two systems as typical values; S52. Discretize each factor of the system to be inverted into grid points within its actual value range to form a factor combination space; S53. For each combination of factors, the corresponding critical intensity of sudden shift from drought to flood is obtained by using a trained nonlinear regression model. : in This represents the critical loss value for disaster. This represents a trained nonlinear regression model. This represents the factor vector of the system to be inverted. and This represents the factor vectors of the other two systems after they have been fixed. S54. Traverse all factor combinations to obtain a series Data points were used to construct disaster threshold surfaces for the social system using interpolation or surface fitting methods. Disaster threshold surface of water resource system and the disaster threshold surface of the ecosystem ,in The surface function representing the disaster threshold of a social system. The factor vector representing the social system, The surface function representing the disaster threshold of the water resource system. Represents the factor vector of the water resource system. The surface function representing the disaster threshold of an ecosystem. A factor vector representing an ecosystem.

7. The method for determining the threshold for rapid shift from drought to flood disaster for sensitive targets according to claim 6, characterized in that, The solution method in step S53 adopts the bisection method, specifically: A1. Determine the search interval for the intensity of the sudden shift from drought to flood. And ensure that the predicted loss is monotonic within the search interval and includes the critical loss value for disaster. ; A2. Calculate the midpoint of the search interval. and the midpoint The combined values ​​of the factors of the system to be inverted, along with the factor values ​​of the other two systems already fixed as typical values, are used as the input feature vector. This input feature vector is then fed into a trained nonlinear regression model, and the corresponding prediction loss is output. ,like Then As a critical value, otherwise according to and Narrowing the range based on size relationship: If Then let ,like Then let ,in Indicates the preset tolerance error; A3. Repeat steps A1 to A2 until the convergence condition is met or the maximum number of iterations is reached to obtain the solution.

8. The method for determining the threshold for rapid shift from drought to flood disaster for sensitive targets according to claim 1, characterized in that, Step S6 includes the following sub-steps: S61. Based on the importance of features, select the 2-3 most important key factors from all system factors; S62. Fix the remaining non-critical factors to typical values; S63. Discretize the key factors within their value range to form a key factor combination space; S64. For each combination of key factors, the corresponding critical intensity of sudden shift from drought to flood is obtained by using a trained nonlinear regression model. ; S65. Iterate through all combinations of key factors to obtain a series of... Data points are used to construct the disaster threshold surface of the integrated system through interpolation or surface fitting methods. ,in The surface function representing the disaster threshold of the integrated system. This represents the key factor vector.

Images