A method for constructing a flood disaster vulnerability curve model

Abstract

The present invention discloses a method for constructing a flood vulnerability curve model, comprising: selecting an optimal daily extreme precipitation threshold and an optimal number of consecutive days, and calculating the annual total extreme precipitation data corresponding to this precipitation parameter combination on a grid-by-grid basis; statistically analyzing land cover change data and slope data using a density calculation method to obtain a corresponding weight matrix; combining the weight matrix, rasterizing the statistically obtained county-level annual total loss data by county and year; and dynamically constructing a flood loss curve using a deep learning multi-objective optimization four-parameter logistic function based on the rasterized county-level annual total loss data and annual total extreme precipitation data. This method can address the technical issues of insufficient data, low accuracy, and limitations on flood events or local cities in constructing flood loss curves, preventing their application to large-scale and large-scale data.

Description

Technical Field

[0001] The present invention relates to the technical field of flood disaster risk warning and loss assessment, and in particular to a method for constructing a flood disaster vulnerability curve model. Background Art

[0002] Scientifically assessing the economic losses and spatiotemporal variations of future flood disasters under global warming and different temperature-rising scenarios is crucial for developing disaster prevention and mitigation strategies, enhancing climate adaptability, and ensuring sustainable development. Establishing vulnerability curves (hazard-loss curves) for scenario analysis has become a key method for estimating flood disaster losses. Flood disaster loss functions primarily describe the relationship between disaster intensity and the extent of losses to affected elements. Related research has primarily focused on urban waterlogging or localized flooding, relying on point surveys or local statistical data. Current methods based on water depth-based damage curves are difficult to apply on a large regional scale due to high data acquisition costs and limited measurement technology. While some studies have attempted to use parameters such as runoff as an alternative to water depth to characterize flood intensity, the accuracy and reliability of these damage curves remain limited due to incomplete data systems. Currently, there is a lack of methods for constructing flood disaster loss curves for large regions. Specifically, there are the following major problems: (1) It is difficult to determine the optimal indicator for reflecting the intensity of flood disasters at a large regional scale; (2) The amount of large-scale, high-quality disaster loss data required to fit the disaster loss curve at a large regional scale is insufficient; (3) The existing functions used to construct flood disaster curves based on local conditions are difficult to reveal the objective law of how disaster losses change with disaster intensity. Summary of the Invention

[0003] In response to the technical problems that most flood loss curve construction methods have, such as insufficient data volume, low accuracy, high data acquisition costs, and being limited to flood events or local cities, and unable to be extended to large-scale and large-scale flood loss curve construction, the present invention aims to provide a flood vulnerability curve model construction method. This method uses extreme precipitation as an indicator to reflect the intensity of flood disasters. It rasterizes the panel data of flood losses under the consideration of disaster intensity and disaster-bearing body characteristics, effectively increasing the amount of data for constructing flood loss curves. It adopts a deep learning method to construct a multi-objective joint loss function to dynamically adjust the parameters of the logistic function to construct the flood loss curve. This effectively reflects the objective change law that the losses caused by floods increase first and then tend to level off as the flood disaster intensifies, and effectively improves the accuracy of loss estimation based on the disaster loss curve. This invention can provide effective support for scientifically assessing the economic losses caused by future flood disasters, formulating effective disaster prevention and mitigation strategies, improving the ability to cope with climate change, and ensuring the sustainable development of the economy and society.

[0004] In order to achieve the above technical objectives, the technical solution adopted by the present invention is:

[0005] A method for constructing a flood disaster vulnerability curve model, the method comprising the following steps:

[0006] S1: Gridded daily precipitation data extracted under different percentile and consecutive day precipitation parameter combinations are statistically converted to annual scale by county to calculate the annual total extreme precipitation. The correlation coefficient between the annual total extreme precipitation and the annual total flood loss at the county scale is calculated. The precipitation parameter combination corresponding to the maximum correlation coefficient is selected as the optimal daily scale extreme precipitation threshold and the optimal consecutive day number. The annual total extreme precipitation data corresponding to this precipitation parameter combination is calculated grid by grid point and represented as a disaster intensity index.

[0007] S2: Land use and land cover change data, as well as slope data calculated using a digital elevation model, are used as indicators reflecting the characteristics of the disaster-bearing body. Density calculation methods are used to statistically analyze the land cover change data and slope data to obtain a corresponding weight matrix. Combined with the weight matrix, the statistically obtained county-level annual total loss data is rasterized by county and year to represent it as a vulnerability indicator for the disaster-bearing body.

[0008] S3, based on the rasterized county-level annual total loss data and annual total extreme precipitation data, a four-parameter logistic function is used to construct the flood disaster loss curve.

[0009] Furthermore, in step S1, the annual total extreme precipitation is calculated using the following formula:

[0010] ;

[0011] Where, express county The annual total extreme precipitation in The county contains grid point, in time Year includes sky; Indicates that the extreme precipitation threshold has been exceeded Grid In the Daily precipitation, , is a constant, which is determined by calculating the percentile and the number of consecutive precipitation days in the time series at each spatial grid point. of The value is unique under the same precipitation parameter combination.

[0012] Furthermore, in step S1, the correlation coefficient between the annual total extreme precipitation and the annual total flood loss at the county level is calculated using the following formula: :

[0013] ;

[0014] in, express county The annual total extreme precipitation in represents the mean of the annual total extreme precipitation data for all counties in all years, express county The total annual flood losses in represents the mean annual total flood losses of all counties in all years, represents the total number of counties, Indicates the annual total.

[0015] Furthermore, in step S2, for the land use and land cover change data, various land types affected by floods, including human activity land, cultivated land, pasture grassland, and economic forest land, are extracted as valid pixels according to the classification system. The allocation weight of each land type is determined by the flood damage department data, and the corresponding LUCC weight matrix is ​​calculated using the following formula:

[0016] ;

[0017] in Represents the weight matrix corresponding to land use and land cover change data, Indicates the The weight of the land type, Indicates the The effective pixel area of ​​land types at the original high spatial resolution, Indicates the area of ​​the pixel at the target low spatial resolution.

[0018] Furthermore, in step S2, a slope of 15 degrees is used as the extraction threshold for valid pixels, and the slope weight matrix is ​​calculated using the following formula:

[0019] ;

[0020] in represents the weight matrix of the slope, Indicates the effective pixel area of ​​the slope data at the original high spatial resolution. Indicates the area of ​​the pixel at the target low spatial resolution.

[0021] Furthermore, combined with the weight matrix, the following formula is used to rasterize the county-level annual total loss data obtained by statistics by county and year:

[0022] ;

[0023] Where, Indicates the county The loss value of the grid point, represents the total loss of the county, Indicates the The annual extreme total precipitation at the grid point, and Respectively represent The weight values ​​of the LUCC weight matrix and slope weight matrix of the grid points.

[0024] Step S3 further comprises:

[0025] The four-parameter Logistic function is used to construct the flood disaster loss curve:

[0026] ;

[0027] Where, is the annual total extreme precipitation at the i-th grid point, is the direct economic loss of the ith grid point after taking the natural logarithm;

[0028] By comparing historical sample points Perform a fit to determine the maximum loss , growth rate , inflection point and minimum loss The initial value of ;

[0029] The maximum loss , growth rate , inflection point and minimum loss As the target variable, the parameter optimization model is constructed based on the neural network structure, with the initial fitting parameters As the starting point, set the maximum number of iterations and function evaluations, iteratively train the parameter optimization model until the loss function changes less than the termination tolerance or the number of iterations reaches the maximum number of iterations, and output the stable optimal solution of the target variable;

[0030] The parameter optimization model includes an input layer, a feature extraction layer, a parameter optimization layer and an output layer; the input layer is used to extract the numerical information, spatial location information and time information contained in the gridded annual total extreme precipitation data as the input feature set of the model; the feature extraction layer is used to adopt a structure of multiple fully connected layers connected to the pooling layer, combined with the skip layer connection method, to automatically extract the high-dimensional expression features that reflect the potential relationship between the input and the target variable from the input feature set; the parameter optimization layer is used to select each sample point Or an input vector containing spatial position information Learning outputs a set of personalized parameter estimates ; Among them, the parameter optimization layer is composed of a multi-layer perceptron, and its input is the standardized observation data The output layer consists of four independent fully connected sub-heads, corresponding to the four parameters of the flood disaster loss curve: maximum loss , growth rate , inflection point and minimum loss .

[0031] Furthermore, the loss function is:

[0032] ;

[0033] ;

[0034] ;

[0035] ;

[0036] in, 、 and is an adjustable hyperparameter; The point-to-point fitting loss is used to minimize the mean square error between the predicted value and the true observed value, so that the model can accurately regress the response value of the individual sample; is the spatial total consistency loss, used for each time step The sum of the predicted values ​​of all spatial units is counted and matched with the corresponding sum of true values ​​to ensure the consistency of the total amount at the macro scale. It is a spatial distribution structure similarity loss. By comparing the cosine similarity between the predicted value and the true value in the spatial dimension at each time step, the model is constrained to maintain the spatial structure characteristics of the predicted value while learning local fitting. represents the total number of counties, represents the annual total, and Respectively county The exact value and valuation of direct economic losses in 2018, and Respectively represent The exact value and estimate of direct economic losses in all regions in 2018, It represents the estimated direct economic loss of the ith grid point after taking the natural logarithm.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] First, the flood disaster vulnerability curve model construction method of the present invention takes into account that it is currently difficult to determine the optimal intensity index reflecting flood disasters on a large scale. Therefore, considering that flood disasters on a large scale have a strong correlation with extreme precipitation, extreme precipitation is selected as a variable reflecting the intensity of flood disasters. The optimal parameter combination of optimal extreme precipitation and optimal number of consecutive rainfall days is determined through an optimization algorithm, which effectively reflects the disaster intensity of floods on a large scale.

[0039] Second, the flood disaster vulnerability curve model construction method of the present invention takes into account that the effectiveness of most flood damage curves is limited by the amount of data, and the amount of large-scale, high-quality disaster loss data required to fit the damage curve on a large regional scale is insufficient. Therefore, a rasterization method is proposed to convert the panel loss data caused by floods into a grid scale, effectively increasing the amount of data for constructing the curve.

[0040] Third, the flood disaster vulnerability curve model construction method of the present invention improves the traditional four-parameter logistic function fitting method based on the evolution characteristics of flood disaster losses. It adopts a deep learning method and dynamically adjusts the parameters by constructing a multi-objective joint loss function, thereby improving the fitting accuracy and stability of the flood disaster loss curve and more accurately reflecting the changing process of losses as the disaster intensifies. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a flow chart of the method for constructing a flood disaster vulnerability curve model of the present invention;

[0042] Figure 2 This is a rasterized flow chart of flood disaster losses in the present invention;

[0043] Figure 3 This is a schematic diagram of the weight matrix calculation of the present invention (taking the urban land type of LUCC as an example);

[0044] Figure 4 Schematic diagram of the four-parameter Logistic parameter optimization model framework constructed by the present invention;

[0045] Figure 5 This is a schematic diagram of the disaster loss curve constructed by the present invention based on rasterized losses. DETAILED DESCRIPTION

[0046] The embodiments of the present invention are described in further detail below with reference to the accompanying drawings.

[0047] The present invention provides a method for constructing a flood disaster vulnerability curve model, the method comprising the following steps:

[0048] S1: Gridded daily precipitation data extracted under different percentile and consecutive day precipitation parameter combinations are statistically converted to annual scale by county to calculate the annual total extreme precipitation. The correlation coefficient between the annual total extreme precipitation and the annual total flood loss at the county scale is calculated. The precipitation parameter combination corresponding to the maximum correlation coefficient is selected as the optimal daily scale extreme precipitation threshold and the optimal consecutive day number. The annual total extreme precipitation data corresponding to this precipitation parameter combination is calculated grid by grid point and represented as a disaster intensity index.

[0049] S2: Land use and land cover change data, as well as slope data calculated using a digital elevation model, are used as indicators reflecting the characteristics of the disaster-bearing body. Density calculation methods are used to statistically analyze the land cover change data and slope data to obtain a corresponding weight matrix. Combined with the weight matrix, the statistically obtained county-level annual total loss data is rasterized by county and year to represent it as a vulnerability indicator for the disaster-bearing body.

[0050] S3, based on the rasterized county-level annual total loss data and annual total extreme precipitation data, a four-parameter logistic function is used to construct the flood disaster loss curve.

[0051] See also Figure 1 , the specific steps of the present invention are as follows:

[0052] 1. Extreme precipitation data preprocessing

[0053] (1) Determination of daily extreme precipitation threshold and consecutive days

[0054] Extreme precipitation is typically defined using percentiles, with commonly used percentiles being 90%, 95%, and 99%. Flooding caused by precipitation may also be related to the duration of precipitation, with the maximum duration typically being five days. To determine the optimal threshold for extreme precipitation and strengthen the correlation between the extracted precipitation and disaster losses, an optimization algorithm is employed to determine the optimal threshold and the optimal number of consecutive days for extreme precipitation. The main approach involves calculating the extracted extreme precipitation using different percentiles (90%, 95%, and 99%) combined with cumulative precipitation over multiple days (1-5 days). This is then aggregated to an annual scale by county to obtain the annual cumulative extreme precipitation. The correlation coefficient between the annual cumulative extreme precipitation and the county-level annual total flood losses is then calculated, thereby determining the optimal parameter combination for the daily-scale extreme precipitation threshold and the number of consecutive days. The specific expression for annual cumulative extreme precipitation is as follows:

[0055] (1);

[0056] Assume that County S contains grid point, in time Year includes Heaven, then represents the total annual extreme precipitation in county s in year t, Indicates that the extreme precipitation threshold has been exceeded The daily precipitation, ,in is a constant, which is determined by calculating the percentile and the number of consecutive precipitation days in the time series at each spatial grid point. of The value is unique under the same parameter combination.

[0057] Based on the precipitation parameter combination corresponding to the maximum correlation coefficient, the optimal extreme precipitation threshold and the optimal number of consecutive days are determined.

[0058] The specific expression of the correlation coefficient is as follows:

[0059] (2);

[0060] in represents the correlation coefficient, represents the total annual extreme precipitation in county s in year t, represents the mean of the annual total extreme precipitation data for all counties in all years, represents the total annual flood loss in county s in year t, represents the mean of the total annual flood losses for all counties in all years.

[0061] (2) Calculation of annual extreme precipitation

[0062] Extract precipitation data based on the parameter combination of the optimal extreme precipitation threshold and the optimal number of consecutive days determined in the previous step, calculate the annual total extreme precipitation by county and year, and obtain gridded annual total extreme precipitation data.

[0063] 2. Rasterization of flood disaster loss panel data

[0064] Based on county-level annual panel economic loss data, gridded extreme precipitation data, land use and land cover change data, and terrain data, we rasterize flood disaster loss panel data while fully considering the intensity of disaster-causing factors and the characteristics of disaster-bearing bodies. This increases the amount of data for constructing disaster loss curves and solves the problem of insufficient large-scale, high-quality disaster loss data required for fitting disaster loss curves on a large regional scale.

[0065] (1) First, the rasterization process proposed in this invention not only takes into account the intensity of flood disasters, but also the characteristics of the disaster-bearing body. This invention uses land use and land cover change (LUCC) data and slope data calculated by digital elevation model (DEM) as indicators reflecting the characteristics of the disaster-bearing body.

[0066] (2) The present invention proposes a density calculation method to retain the effective information in LUCC and slope data while unifying the spatial resolution. In order to unify the spatial resolution of land use and land cover change, slope data and annual total extreme precipitation data, the present invention uses a density calculation method to perform statistics on LUCC and slope data to obtain their corresponding weight matrices. First, LUCC is extracted according to the classification system, and various land types affected by floods, such as human activity land, cultivated land, pastoral grassland, economic forest land, etc., are extracted as effective pixels. Considering the differences in the defense capabilities of each type of disaster-bearing body against flood disasters, the weights allocated to each land type are determined by the flood damage department data. The specific formula of the LUCC weight matrix is ​​as follows:

[0067] (3);

[0068] in represents the weight matrix of LUCC, Indicates the The weight of the land type, Indicates the The effective pixel area of ​​land types at the original high spatial resolution, Indicates the area of ​​the pixel at the target low spatial resolution.

[0069] For slope data, a slope weight matrix is ​​calculated using a method similar to the LUCC weight matrix. A slope of 15 degrees is determined as the extraction threshold for valid pixels, and the slope weight matrix is ​​further calculated. The specific formula is as follows:

[0070] (4);

[0071] in represents the weight matrix of the slope, Indicates the effective pixel area of ​​the slope data at the original high spatial resolution. Indicates the area of ​​the pixel at the target low spatial resolution. Figure 3 This is a schematic diagram of the weight matrix calculation of the present invention (taking the urban land classification of LUCC as an example, the red rectangle in the figure is the 0.25-degree grid point). The classification result extracted according to LUCC is the total area of ​​the urban grid points (1km) divided by the area of ​​the 0.25-degree grid point.

[0072] (3) Secondly, the present invention takes into account the intensity of disasters and the characteristics of disaster-bearing bodies, and rasterizes the county-level annual total loss data obtained by statistics by county and year. The part reflecting the characteristics of the disaster-bearing body used in the rasterization is calculated by formula (3) and formula (4), and the disaster intensity part is represented by the gridded annual total extreme precipitation data obtained in the extreme precipitation data preprocessing step. For each county with 10 pixels, the specific rasterization calculation formula is as follows:

[0073] (5);

[0074] Where, Indicates the county The loss value of pixel (i-th grid point), represents the total loss of the county, represents the annual extreme total precipitation of the i-th pixel, and Respectively represent The weight value of the pixel's LUCC and slope. Figure 2 This is a rasterized flow chart of flood disaster losses according to the present invention.

[0075] 3. Construction and Optimization of Flood Disaster Loss Curve

[0076] Based on the objective law that flood disaster losses change with disaster intensity, this paper considers that as precipitation increases, disaster losses initially grow rapidly, then slow down, and finally steadily increase in smaller increments, with the entire process exhibiting an "S"-shaped evolution characteristic. Therefore, a four-parameter logistic function, whose basic form conforms to this change process, is selected to construct the disaster loss curve. This function can accurately characterize the complex response characteristics of losses changing with precipitation intensity. The four parameters correspond to the minimum loss, growth rate, inflection point, and maximum loss, respectively. They have clear physical meanings and provide a theoretical basis for quantitatively characterizing the evolution of flood disasters driven by extreme precipitation. Its basic form is as follows:

[0077] (6);

[0078] In the formula, is the annual extreme precipitation, is the direct economic loss after taking the natural logarithm; the constant is determined through the fitting process 、 、 and The initial value of This parameter set can be used to compare historical sample points. Achieve basic fitting.

[0079] On the basis of preliminary fitting of the four-parameter logistic function, in order to further improve the generalization ability of the fitting curve in high-dimensional space and ensure that the predicted value meets the actual constraints such as spatial total conservation and time step spatial distribution structure consistency while maintaining individual fitting accuracy, this paper introduces a parameter optimization mechanism based on deep learning to replace the static optimal parameter point and improve the dynamic adaptability of the fitting model. This parameter optimization module is built based on a neural network structure and aims to optimize the performance of each sample point. Or an input vector containing spatial position information Learning to output a set of dynamic parameters , this set of parameters will be used to construct the prediction function:

[0080] (7);

[0081] The neural network model is composed of a multi-layer perceptron (MLP), whose input is the standardized observation data The output layer consists of four independent, fully connected sub-heads, each corresponding to the four parameters of the predicted logistic curve, ensuring that each input point receives a set of personalized parameter estimates. This structure is highly scalable and can combine spatial location information and timestamps as network inputs to achieve regional adaptive modeling.

[0082] Figure 4 Schematic diagram of the four-parameter Logistic parameter optimization model framework constructed in this invention. During the model training process, a multi-objective joint loss function is designed, which specifically includes the following three sub-losses:

[0083] Point-wise fitting loss (prediction accuracy). This ensures that the model can accurately regress the response values ​​of individual samples by minimizing the mean squared error (MSE) between the predicted values ​​and the true observed values:

[0084] (8);

[0085] Spatial total consistency loss (regional sum conservation). At each time step The predicted values ​​of all spatial units are summed and matched with the corresponding true values ​​to ensure the consistency of the total amount at the macro scale:

[0086] (9);

[0087] Spatial distribution structure similarity loss (spatial morphology preservation). By comparing the cosine similarity of the predicted value and the true value in the spatial dimension at each time step, the model is constrained to maintain the spatial structure characteristics of the predicted value while learning local fitting:

[0088] (10);

[0089] The three sub-losses are combined to form the total loss function through weighted combination:

[0090] (11);

[0091] in, , , It is an adjustable hyperparameter that is set according to application requirements to achieve a balance between accuracy and constraints. As a starting point, a sufficiently large maximum number of iterations (10,000) and function evaluations (10,000) are set to ensure full algorithm convergence and avoid falling into local optimality. An early stopping strategy is implemented; when the change in the loss function is less than the termination tolerance, the optimization ends early, indicating that a stable optimal solution has been reached. Through the aforementioned neural network training process, the present invention can dynamically optimize the traditional four-parameter logistic fitting model and significantly improve the model's predictive ability and robustness in complex spatial-temporal scenarios, possessing strong application potential and promotional value. Figure 5 This is a schematic diagram of the disaster loss curve constructed by the present invention based on rasterized losses.

[0092] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application.

[0093] Obviously, those skilled in the art may make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application is intended to include these modifications and variations.

Claims

1. A method for constructing a flood disaster vulnerability curve model, characterized in that: The method comprises the following steps: S1: Gridded daily precipitation data extracted under different percentile and consecutive day precipitation parameter combinations are statistically converted to annual scale by county to calculate the annual total extreme precipitation. The correlation coefficient between the annual total extreme precipitation and the annual total flood loss at the county scale is calculated. The precipitation parameter combination corresponding to the maximum correlation coefficient is selected as the optimal daily scale extreme precipitation threshold and the optimal consecutive day number. The annual total extreme precipitation data corresponding to this precipitation parameter combination is calculated grid by grid point and represented as a disaster intensity index. S2: Land use and land cover change data, as well as slope data calculated using a digital elevation model, are used as indicators reflecting the characteristics of the disaster-bearing body. Density calculation methods are used to statistically analyze the land cover change data and slope data to obtain a corresponding weight matrix. Combined with the weight matrix, the statistically obtained county-level annual total loss data is rasterized by county and year to represent it as a vulnerability indicator for the disaster-bearing body. S3, based on the rasterized county-level annual total loss data and annual total extreme precipitation data, a four-parameter logistic function is used to construct the flood disaster loss curve.

2. The method for constructing a flood disaster vulnerability curve model according to claim 1, characterized in that: In step S1, the annual total extreme precipitation is calculated using the following formula: ； Where, express county The annual total extreme precipitation in The county contains grid point, in time Year includes sky; Indicates that the extreme precipitation threshold has been exceeded Grid In the Daily precipitation, , is a constant, which is determined by calculating the percentile and the number of consecutive precipitation days in the time series at each spatial grid point. of The value is unique under the same precipitation parameter combination.

3. The method for constructing a flood disaster vulnerability curve model according to claim 1, characterized in that: In step S1, the correlation coefficient between the annual total extreme precipitation and the annual total flood loss at the county level is calculated using the following formula: : ； in, express county The annual total extreme precipitation in represents the mean of the annual total extreme precipitation data for all counties in all years, express county The total annual flood losses in represents the mean annual total flood losses of all counties in all years, represents the total number of counties, Indicates the annual total.

4. The method for constructing a flood disaster vulnerability curve model according to claim 1, characterized in that: In step S2, for the land use and land cover change data, various land types affected by floods, including human activity land, cultivated land, pasture grassland, and economic forest land, are extracted as valid pixels according to the classification system. The allocation weight of each land type is determined by the flood damage department data, and the corresponding LUCC weight matrix is ​​calculated using the following formula: ； in Represents the weight matrix corresponding to land use and land cover change data, Indicates the The weight of the land type, Indicates the The effective pixel area of ​​land types at the original high spatial resolution, Indicates the area of ​​the pixel at the target low spatial resolution.

5. The method for constructing a flood disaster vulnerability curve model according to claim 1, characterized in that: In step S2, a slope of 15 degrees is used as the extraction threshold for valid pixels, and the slope weight matrix is ​​calculated using the following formula: ； in represents the weight matrix of the slope, Indicates the effective pixel area of ​​the slope data at the original high spatial resolution. Indicates the area of ​​the pixel at the target low spatial resolution.

6. The method for constructing a flood disaster vulnerability curve model according to claim 1, characterized in that: Combined with the weight matrix, the following formula is used to rasterize the county-level annual total loss data obtained by statistics by county and year: ； Where, Indicates the county The loss value of the grid point, represents the total loss of the county, Indicates the The annual extreme total precipitation at the grid point, and Respectively represent The weight values ​​of the LUCC weight matrix and slope weight matrix of the grid points.

7. The method for constructing a flood disaster vulnerability curve model according to claim 1, characterized in that: Step S3 further comprises: The four-parameter Logistic function is used to construct the flood disaster loss curve: ； Where, is the annual total extreme precipitation at the i-th grid point, is the direct economic loss of the ith grid point after taking the natural logarithm; By comparing historical sample points Perform a fit to determine the maximum loss , growth rate , inflection point and minimum loss The initial value of ; The maximum loss , growth rate , inflection point and minimum loss As the target variable, the parameter optimization model is constructed based on the neural network structure, with the initial fitting parameters As the starting point, set the maximum number of iterations and function evaluations, iteratively train the parameter optimization model until the loss function changes less than the termination tolerance or the number of iterations reaches the maximum number of iterations, and output the stable optimal solution of the target variable; The parameter optimization model includes an input layer, a feature extraction layer, a parameter optimization layer and an output layer; the input layer is used to extract the numerical information, spatial location information and time information contained in the gridded annual total extreme precipitation data as the input feature set of the model; the feature extraction layer is used to adopt a structure of multiple fully connected layers connected to the pooling layer, combined with the skip layer connection method, to automatically extract the high-dimensional expression features that reflect the potential relationship between the input and the target variable from the input feature set; the parameter optimization layer is used to select each sample point Or an input vector containing spatial position information Learning outputs a set of personalized parameter estimates ; Among them, the parameter optimization layer is composed of a multi-layer perceptron, and its input is the standardized observation data The output layer consists of four independent fully connected sub-heads, corresponding to the four parameters of the flood disaster loss curve: maximum loss , growth rate , inflection point and minimum loss .

8. The method for constructing a flood disaster vulnerability curve model according to claim 7, characterized in that: The loss function is: ； ； ； ； in, 、 and is an adjustable hyperparameter; The point-to-point fitting loss is used to minimize the mean square error between the predicted value and the true observed value, so that the model can accurately regress the response value of the individual sample; is the spatial total consistency loss, used for each time step The sum of the predicted values ​​of all spatial units is counted and matched with the corresponding sum of true values ​​to ensure the consistency of the total amount at the macro scale. It is a spatial distribution structure similarity loss. By comparing the cosine similarity between the predicted value and the true value in the spatial dimension at each time step, the model is constrained to maintain the spatial structure characteristics of the predicted value while learning local fitting. represents the total number of counties, represents the annual total, and Respectively county The exact value and valuation of direct economic losses in 2018, and Respectively represent The exact value and estimate of direct economic losses in all regions in 2018, It represents the estimated direct economic loss of the ith grid point after taking the natural logarithm.

Images