A method and system for quantitatively assessing large areas of crop flood damage

Abstract

The present invention discloses a method and system for quantitatively assessing large areas of crop flood damage. The method comprises obtaining flood disaster data and SAR time-series remote sensing data, obtaining the spatial extent of flood inundation from SAR remote sensing images based on a deep learning image semantic segmentation model, estimating the flood inundation depth at the DEM pixel scale using an inundation depth model based on topography and river water level observation data, combining the flood disaster area identified by SAR time-series remote sensing data with ground-observed precipitation and river water level changes, calculating the core assessment indicators of crop flood inundation range, inundation duration, and inundation depth, correlating flood disaster events with crop stages, and constructing a flood damage degree equation using flood stress sensitivity as one of the parameters. The method incorporates the local crop characteristics of the study area, correlating flood disaster events with crop stages, and constructing a flood damage degree equation using flood stress sensitivity as one of the parameters, thereby achieving quantitative calculation of the flood damage degree.

Description

Technical Field

[0001] The present invention relates to the field of assessment, and in particular to a method and system for quantitatively assessing large-scale crop flood-affected areas. Background Art

[0002] Under today's complex and ever-changing climate, crop yields are vulnerable to extreme weather events. In many areas, floods caused by heavy rain are the primary natural disaster responsible for the greatest crop losses. To quantitatively assess losses from large-scale agricultural floods, GIS and remote sensing technologies provide a wealth of data and analytical tools for determining the spatial extent of crop inundation and the spatial distribution of crop damage. GIS can be used for flood risk zoning and flood evolution simulation. In real-time flood monitoring, combined with remote sensing imagery, it can assist in obtaining flood characteristic parameters such as inundation extent and water depth. First, the spatial extent of crop inundation is determined based on remote sensing imagery. Then, the inundation depth and duration are determined at the pixel scale. A crop damage severity function is applied pixel by pixel to achieve quantitative spatial mapping of crop damage caused by floods.

[0003] Existing technologies for extracting flooded areas from remote sensing images mostly rely on vegetation or water body indices derived from optical remote sensing data. However, cloud cover is often severe during periods of heavy rain and flooding, significantly impacting the quality of optical remote sensing images. Polarimetric radar remote sensing data is not restricted by weather conditions and offers near-all-weather observation capabilities, offering significant advantages in flood monitoring. Traditional methods for estimating the duration of flooding rely on differences in water signals from time-series remote sensing images. However, due to the time interval constraints of time-series remote sensing data, this has significant application limitations. Therefore, a method for quantitatively assessing large areas of crop flooding is needed. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for quantitatively evaluating large-scale crop flood damage.

[0005] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0006] The present invention comprises the following steps:

[0007] A. Obtain flood disaster data and SAR time-series remote sensing data, obtain the spatial extent of flood inundation from the SAR remote sensing images based on a deep learning image semantic segmentation model, and estimate the flood inundation depth at the DEM pixel scale using an inundation depth model based on topography and river water level observation data; the deep learning image semantic segmentation model is trained and constructed using the flood disaster data;

[0008] B. Combine the flood disaster areas identified by SAR time-series remote sensing data with the precipitation and river water level changes observed on the ground to comprehensively estimate the flooding duration;

[0009] C calculates the core assessment indicators of crops, associates flood disaster events with crop periods, uses flood stress sensitivity to construct a flood damage degree equation, determines the crop flood disaster degree equation, and completes spatial mapping of the disaster degree; the core assessment indicators include the flood inundation range, inundation duration, and inundation depth.

[0010] Furthermore, the method for estimating the submergence duration includes:

[0011] The prerequisite is that rainfall occurs continuously every day within a period of time and the river water level is higher than the warning level;

[0012] The SAR time series remote sensing data is input into the deep belief network based on the Boltzmann machine, which consists of a hidden layer and a display layer;

[0013] Compute the joint probability density between the hidden and visible layers:

[0014]

[0015]

[0016] The a-th hidden layer is z a , the ath display layer is x a , the bias of the qth hidden layer is v q , the bias of the a-th display layer is w a , No. a The weight between the hidden layer and the qth display layer is l aq , the number of hidden layers is The number of display layers is The parameters are The network function is The normalization factor is E, and the probability joint density between the hidden layer z and the display layer x is p(x, z);

[0017] By learning the probability joint density between the hidden layer and the display layer, the feature extraction of SAR time series remote sensing data is performed to obtain the characteristics of the flooded area;

[0018] Clustering is used to reduce the dimension of flood area characteristics to obtain the reduced dimension features, which are expressed as follows:

[0019]

[0020] The number of flood area characteristics is N, and the characteristic of the u-th flood area is h u , the center value of the c dimension of the y-th cluster is b y (c) The value of the cth dimension of the uth flood area feature is h u (c), the yth cluster is K y , flood area characteristics h uBelongs to cluster d y The distance is d y (h u ), the u-th dimensionality reduction feature of the y-th cluster is D iy ;

[0021] The flood area identification is completed for two consecutive time series remote sensing images. The flood area identification results of the two images are compared and the expression is:

[0022]

[0023] The time variable of the u-th dimension reduction feature is The u-th dimension reduction feature at the s-th moment is h u.s , the u-th dimensionality reduction feature at the s+j-th moment is h u，s+j , the number of dimensionality reduction features that change is N1, and the control coefficient is ζ;

[0024] Obtain overlapping flood areas and non-overlapping flood areas based on time-varying variables;

[0025] For overlapping flood areas, where the flood inundation duration is greater than or equal to the remote sensing time series interval, the time increment Δs based on the SAR time series remote sensing interval t is obtained, and the final inundation duration is t+Δs;

[0026] For non-overlapping areas, the flooding duration estimation method is used.

[0027] Furthermore, the method for obtaining the core evaluation indicators includes:

[0028] Calculate the similarity of flood disaster data:

[0029]

[0030] The rth flood disaster data is h r , the fth flood disaster data is h f , flood disaster data r and flood disaster data f The similarity is Flood disaster data r and flood disaster data f The distance is D(h r , h f ), flood disaster data h r The modulus is h r |, flood disaster data f The modulus is h f |, the tuning parameter is σ, and the kernel width is φ;

[0031] The flood disaster data with the largest number of similarities is taken as the classification cluster center, and the remaining flood disaster data are classified into the classification cluster with the largest similarity to the classification cluster center to obtain the classification cluster;

[0032] Calculate the importance of classification clusters:

[0033]

[0034] The importance of the u-th classification cluster is Z u , the rth flood disaster data is h r , transposed to T, the unit matrix is E, and the transposed of the rth flood disaster data is The cluster center is h o , the number of flood disaster data in the u-th classification cluster is n r , the importance parameter is μ and the error coefficient is ε;

[0035] The flood disaster data of the classification clusters with importance greater than 0.318 are taken as key data, and the flood relevance of the key data is calculated:

[0036]

[0037] The number of key data is n1, and the flood relevance of the rth key data is The flood severity is Flood disaster data r and flood severity The joint probability distribution of Flood disaster data r The marginal probability distribution of is p(h r ), flood severity The marginal probability distribution of

[0038] The key data with flood correlation greater than 0.647 are output as core evaluation indicators.

[0039] Furthermore, a flood inundation depth model is used to determine the potential inundation depth within the river basin. Assuming that floods cause the river water level to rise and flow from high-lying areas to low-lying areas within the basin, it can be expressed as:

[0040] FID ij =ΔH ij -ΔDEM ij

[0041] Where ΔH=H max -H min = maximum water level minus minimum water level, ΔDEM ij is the difference between the DEM at position (i, j) and the DEM at the monitoring point.

[0042] Furthermore, basin analysis using the Hydrological Analyst tool in ArcGIS was performed, combined with DEM data to calculate the extent of each river basin within the study area and the potential inundation depth of each pixel.

[0043] Furthermore, the flood hazard equation is:

[0044] L = f(g, l, d)

[0045] The degree of crop flood damage is L, the growth and development period is g, the flooding duration is l, and the flooding depth is d;

[0046] The flood damage equation outputs a quantitative description of the degree of damage in percentage.

[0047] Furthermore, the flooded areas were treated as new land cover categories for independent model training.

[0048] Furthermore, the ground-observed precipitation data and the river water level observation data are used to construct a flooding duration estimation model.

[0049] The second aspect is a large-scale quantitative assessment system for crop flood damage, including:

[0050] Acquisition module: Used to train and build a deep learning image semantic segmentation model based on a public flood disaster dataset. This model is then applied to SAR remote sensing imagery of target flood disaster events to obtain the spatial extent of flood inundation. Based on topographic and river water level observation data, an inundation depth model is used to estimate the flood inundation depth at the DEM pixel scale.

[0051] Calculation module: This module combines flood disaster areas identified by time-series remote sensing data with ground-observed precipitation and river water level changes to estimate the duration of inundation.

[0052] Output module: used to calculate the three core assessment indicators of crop flood inundation range, inundation duration, and inundation depth, associate flood disaster events with crop stages, construct a flood hazard degree equation based on flood stress sensitivity, determine the crop flood damage degree equation, and complete spatial mapping of the damage degree.

[0053] The beneficial effects of the present invention are:

[0054] The present invention provides a method and system for quantitatively assessing large areas of crop flood damage. Compared with the prior art, the present invention has the following technical effects:

[0055] The proposed near-real-time crop flood damage assessment method has strong universal applicability and can be applied to quantitatively assess losses from large-scale agricultural flood disasters. For different regions or crops, the parameters of the crop damage severity function can be adjusted based on the local meteorological, hydrological, and topographic data required by the model, as well as the crop characteristics. The analysis results can be used by agricultural management departments and agricultural insurance companies to conduct real-time assessments of agricultural losses and assist in the development of response measures.

[0056] In addition, in the absence of historical data on agricultural losses caused by flood disasters, the method of the present invention can be applied to evaluate and analyze flood disaster cases in different regions, so as to obtain the spatiotemporal characteristics of agricultural flood disaster losses in different regions, and then apply agricultural insurance product development tools to realize regional personalized agricultural insurance product development. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is a flowchart of the steps of a method for quantitatively assessing large areas of crop flood damage according to the present invention;

[0058] Figure 2 This is a structural diagram of the U-net model for water body identification in a method for quantitatively assessing large areas of crop flood damage according to the present invention;

[0059] Figure 3 This is a diagram showing the principle of FID calculation in a method for quantitatively assessing large areas of crop flood damage according to the present invention. DETAILED DESCRIPTION

[0060] The present invention will be further described below through specific examples. The illustrative examples and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.

[0061] A method and system for quantitatively assessing large areas of crop flood damage according to the present invention comprises the following steps:

[0062] like Figure 1 As shown, in this embodiment, the following steps are included:

[0063] A. Obtain flood disaster data and SAR time-series remote sensing data, obtain the spatial extent of flood inundation from the SAR remote sensing images based on a deep learning image semantic segmentation model, and estimate the flood inundation depth at the DEM pixel scale using an inundation depth model based on topography and river water level observation data; the deep learning image semantic segmentation model is trained and constructed using the flood disaster data;

[0064] B. Combine the flood disaster areas identified by SAR time-series remote sensing data with the precipitation and river water level changes observed on the ground to comprehensively estimate the flooding duration;

[0065] C calculates the core assessment indicators of crops, associates flood disaster events with crop periods, uses flood stress sensitivity to construct a flood damage degree equation, determines the crop flood disaster degree equation, and completes spatial mapping of the disaster degree; the core assessment indicators include the flood inundation range, inundation duration, and inundation depth.

[0066] In this embodiment, the method for estimating the submergence duration includes:

[0067] The prerequisite is that rainfall occurs continuously every day within a period of time and the river water level is higher than the warning level;

[0068] The SAR time series remote sensing data is input into the deep belief network based on the Boltzmann machine, which consists of a hidden layer and a display layer;

[0069] Compute the joint probability density between the hidden and visible layers:

[0070]

[0071]

[0072] The a-th hidden layer is z a , the ath display layer is x a , the bias of the qth hidden layer is v q , the bias of the a-th display layer is w a , the weight value between the ath hidden layer and the qth display layer is l aq , the number of hidden layers is The number of display layers is The parameters are The network function is The normalization factor is E, and the probability joint density between the hidden layer z and the display layer x is p(x, z);

[0073] By learning the probability joint density between the hidden layer and the display layer, the feature extraction of SAR time series remote sensing data is performed to obtain the characteristics of the flooded area;

[0074] Clustering is used to reduce the dimension of flood area characteristics to obtain the reduced dimension features, which are expressed as follows:

[0075]

[0076] The number of flood area characteristics is N, and the characteristic of the u-th flood area is h u , the center value of the c dimension of the y-th cluster is b y (c) The value of the cth dimension of the uth flood area feature is h u (c), the yth cluster is K y , flood area characteristics h u Belongs to cluster dy The distance is d y (h u ), the u-th dimensionality reduction feature of the y-th cluster is D iy ;

[0077] The flood area identification is completed for two consecutive time series remote sensing images. The flood area identification results of the two images are compared and the expression is:

[0078]

[0079] The time variable of the u-th dimension reduction feature is The u-th dimension reduction feature at the s-th moment is h u.s , the u-th dimensionality reduction feature at the s+j-th moment is h u，s+j , the number of dimensionality reduction features that change is N1, and the control coefficient is ζ;

[0080] Obtain overlapping flood areas and non-overlapping flood areas based on time-varying variables;

[0081] For overlapping flood areas, where the flood inundation duration is greater than or equal to the remote sensing time series interval, the time increment Δs based on the SAR time series remote sensing interval t is obtained, and the final inundation duration is t+Δs;

[0082] For non-overlapping areas, the flooding duration estimation method is used.

[0083] In this embodiment, the method for obtaining the core evaluation indicators includes:

[0084] Calculate the similarity of flood disaster data:

[0085]

[0086] The rth flood disaster data is h r , the fth flood disaster data is h f , flood disaster data r and flood disaster data f The similarity is Flood disaster data r and flood disaster data f The distance is D(h r , h f ), flood disaster data h r The modulus is h r |, flood disaster data f The modulus is h f |, the tuning parameter is σ, and the kernel width is φ;

[0087] The flood disaster data with the largest number of similarities is taken as the classification cluster center, and the remaining flood disaster data are classified into the classification cluster with the largest similarity to the classification cluster center to obtain the classification cluster;

[0088] Calculate the importance of classification clusters:

[0089]

[0090] The importance of the u-th classification cluster is Z u , the rth flood disaster data is h r , transposed to T, the unit matrix is E, and the transposed of the rth flood disaster data is The cluster center is h o , the number of flood disaster data in the u-th classification cluster is n r , the importance parameter is μ and the error coefficient is ε;

[0091] The flood disaster data of the classification clusters with importance greater than 0.318 are taken as key data, and the flood relevance of the key data is calculated:

[0092]

[0093] The number of key data is n1, and the flood relevance of the rth key data is The flood severity is Flood disaster data r and flood severity The joint probability distribution of Flood disaster data r The marginal probability distribution of is p(h r ), flood severity The marginal probability distribution of

[0094] The key data with flood correlation greater than 0.647 are output as core evaluation indicators.

[0095] In this example, a flood inundation depth model is used to determine the potential inundation depth within a river basin. Assuming that floods cause river water levels to rise and flow from high-lying areas to low-lying areas within the basin, it can be expressed as:

[0096] FID ij =ΔH ij -ΔDEM ij

[0097] Where ΔH=H max -H min = maximum water level minus minimum water level, ΔDEM ij is the difference between the DEM at position (i, j) and the DEM at the monitoring point.

[0098] In this embodiment, basin analysis of the Hydrological Analyst hydrological tool of ArcGIS is used, and the DEM data is combined to calculate the range of each river basin within the study area and calculate the potential flooding depth of each pixel.

[0099] In this embodiment, the flood hazard equation is:

[0100] L = f(g, l, d)

[0101] The degree of crop flood damage is L, the growth and development period is g, the flooding duration is l, and the flooding depth is d;

[0102] The flood damage equation outputs a quantitative description of the degree of damage in percentage.

[0103] In this embodiment, the flooded area is treated as a new land cover category and model training is performed independently.

[0104] In this embodiment, the ground-observed precipitation data and the river water level observation data are used to construct a flooding duration estimation model.

[0105] The second aspect is a large-scale quantitative assessment system for crop flood damage, including:

[0106] Acquisition module: Used to train and build a deep learning image semantic segmentation model based on a public flood disaster dataset. This model is then applied to SAR remote sensing imagery of target flood disaster events to obtain the spatial extent of flood inundation. Based on topographic and river water level observation data, an inundation depth model is used to estimate the flood inundation depth at the DEM pixel scale.

[0107] Calculation module: This module combines flood disaster areas identified by time-series remote sensing data with ground-observed precipitation and river water level changes to estimate the duration of inundation.

[0108] Output module: used to calculate the three core assessment indicators of crop flood inundation range, inundation duration, and inundation depth, associate flood disaster events with crop stages, construct a flood hazard degree equation based on flood stress sensitivity, determine the crop flood damage degree equation, and complete spatial mapping of the damage degree.

[0109] Taking rice as an example, multi-source data such as crop distribution, ground hydrological observations, precipitation observations, river network distribution, and SAR time-series remote sensing images during floods were obtained to calculate the three core evaluation indicators of crop flood inundation range, inundation duration, and inundation depth. Finally, an equation for the degree of crop damage was constructed for spatial mapping of the degree of crop damage.

[0110] 1. Submerged range extraction

[0111] This paper trains a deep learning model based on the Cloud to Street flood remote sensing dataset to identify surface water cover. The model uses a U-Net-based CNN neural network to perform semantic segmentation on images, classifying pixels into water bodies and non-water bodies. After model training, the model identifies water cover before and after a disaster. The flood inundation area is then determined by clipping the pre-disaster water cover from the post-disaster water cover.

[0112] The U-Net model was constructed by overlaying SAR remote sensing VH and VV polarimetric imagery with DEM data to create a three-channel image. Data preprocessing included min-max scaling, flipping, and rotation enhancement of each channel's pixel values. Three randomly initialized U-Net convolutional neural networks were trained using the same DICE loss function with different training-test splits. The final prediction was the average of the three model outputs.

[0113] like Figure 2 As shown in the figure, the backbone network block on the left is connected to the decoder on the right via a skip connection. The input is a 512×512 pixel image with three channels: VH, VV, and DEM. The output is the prediction results for the water body and non-water body categories. During model training, three backbone architectures were tested: ResNet34, SeresNet34, and EfficientNetB4. The optimal model was selected for water body recognition in the case study after comparison.

[0114] 2. Calculation of submergence duration

[0115] This invention overcomes the limitations of determining flood inundation duration based solely on time-series remote sensing imagery. Because remote sensing observations occur at fixed intervals, this approach is difficult to apply to flood disasters where the inundation duration is shorter than the time-series interval. This method, based on time-series remote sensing image recognition, adds an inundation duration estimation step based on ground-based meteorological and hydrological data to comprehensively determine inundation duration.

[0116] First, we obtained spatially interpolated time series data of daily precipitation for the study area within the target date range, as well as a watershed demarcation map. The basic inundation duration estimation method is: continuous rainfall occurs daily within a period of time, and the river water level is above the warning level;

[0117] Specifically:

[0118] Step 1: Use two consecutive time-series remote sensing images to identify flooded areas. Compare the flooded area identification results of the two images to obtain overlapping and non-overlapping flooded areas.

[0119] Step 2: For overlapping flooded areas, where the flood inundation duration is greater than or equal to the remote sensing time interval, a time increment Δs based on the time series remote sensing interval t is obtained. This time increment uses the basic inundation duration estimation method described above; the final inundation duration is t + Δs.

[0120] Step 3: For non-overlapping areas, use the basic flooding duration estimation method described above.

[0121] 3. Submergence depth model

[0122] The present invention uses the flood inundation depth (FID) model to determine the potential inundation depth within a river basin. The model assumes that flooding causes the river water level to rise and flow from high-lying areas to low-lying areas within the basin, as follows: Figure 3 As shown;

[0123] The model can be expressed as:

[0124] FID ij =ΔH ij -ΔDEM ij

[0125] Where ΔH=H max -H min = maximum water level minus minimum water level, ΔDEM ij is the difference between the DEM at the (i, j) location and the DEM at the monitoring point;

[0126] like Figure 3 As shown, before using the FID model to calculate potential flooding depth, it is necessary to determine the basin to which the target area belongs. This method uses the basin analysis tool in ArcGIS's Hydrological Analyst, combined with DEM data, to calculate the extent of each river basin within the study area. Based on this, the potential flooding depth of each pixel is calculated.

[0127] 4. Construction of flood hazard equation

[0128] Taking the assessment parameters of crop flood damage in a certain basin as an example, this case uses the relationship between the damage degree and flooding time and depth published by the local government, as well as crop height data at different growth times.

[0129] The degree of crop flood damage L is a function of the growth and development period g, flooding duration l and flooding depth d, that is:

[0130] L = f(g, l, d)

[0131] The present invention determines the parameters of this equation based on a survey of the characteristics of the crop varieties in the study area, and outputs a quantitative description of the degree of damage as a percentage. For example, for rice crops, the relationship between the degree of damage and the duration and depth of flooding can be referred to in the following table:

[0132] Table 1 Relationship between rice damage extent and flooding duration and depth (Philippine Department of Agriculture)

[0133]

[0134]

[0135] Finally, combined with the estimated results of the affected area, that is, the flooded range, the crop disaster loss assessment of the entire study area was completed.

[0136] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for quantitatively assessing large areas of crop flood damage, characterized in that: The following steps are involved: A. Obtain flood disaster data and SAR time-series remote sensing data, obtain the spatial extent of flood inundation from the SAR remote sensing images based on a deep learning image semantic segmentation model, and estimate the flood inundation depth at the DEM pixel scale using an inundation depth model based on topography and river water level observation data; the deep learning image semantic segmentation model is trained and constructed using the flood disaster data; B. Combine the flood disaster areas identified by SAR time-series remote sensing data with the precipitation and river water level changes observed on the ground to comprehensively estimate the flooding duration; C. Calculate the core assessment indicators of crops, associate flood disaster events with crop stages, construct flood damage degree equations using flood stress sensitivity, determine crop flood damage degree equations, and complete spatial mapping of damage degree; The method for estimating the submergence duration includes: The prerequisite is that rainfall occurs continuously every day within a period of time and the river water level is higher than the warning level; The SAR time series remote sensing data is input into the deep belief network based on the Boltzmann machine, which consists of a hidden layer and a display layer; Compute the joint probability density between the hidden and visible layers: The a-th hidden layer is z a , the ath display layer is x a , the bias of the qth hidden layer is v q , the bias of the a-th display layer is w a , the weight value between the a-th hidden layer and the q-th display layer is The number of hidden layers is The number of display layers is The parameter is φ, and the network function is The normalization factor is E, and the probability joint density between the hidden layer z and the display layer x is p(x,z); By learning the probability joint density between the hidden layer and the display layer, the feature extraction of SAR time series remote sensing data is performed to obtain the characteristics of the flooded area; Clustering is used to reduce the dimension of flood area characteristics to obtain the reduced dimension features, which are expressed as follows: The number of flood area characteristics is N, and the characteristic of the u-th flood area is h u , the center value of the c dimension of the y-th cluster is b y (c) The value of the cth dimension of the uth flood area feature is h u (c), the yth cluster is K y , flood area characteristics h u Belongs to cluster d y The distance is d y (h u ), the u-th dimensionality reduction feature of the y-th cluster is D iy ; The flood area identification is completed for two consecutive time series remote sensing images. The flood area identification results of the two images are compared and the expression is: The time variable of the u-th dimension reduction feature is g(h u ), the u-th dimension reduction feature at the s-th moment is h u.s , the u-th dimensionality reduction feature at the s+j-th moment is h u,s+j , the number of dimensionality reduction features that change is N1, and the control coefficient is ζ; Obtain overlapping flood areas and non-overlapping flood areas based on time-varying variables; For overlapping flood areas, where the flood inundation duration is greater than or equal to the remote sensing time series interval, the time increment Δs based on the SAR time series remote sensing interval t is obtained, and the final inundation duration is t+Δs; For non-overlapping areas, the flooding duration estimation method is used; The flood inundation depth model is used to determine the potential inundation depth within the river basin. It is assumed that floods cause the river water level to rise and flow from high-lying areas to low-lying areas within the basin. It can be expressed as: FID ij =ΔH ij -ΔDEM ij Where ΔH=H max -H min = maximum water level minus minimum water level, ΔDEM ij is the difference between the DEM at position (i, j) and the DEM at the monitoring point.

2. A method for quantitatively assessing large areas of crop flood damage according to claim 1, characterized in that: The method for obtaining the core evaluation indicators includes: Calculate the similarity of flood disaster data: The rth flood disaster data is h r , the fth flood disaster data is h f , flood disaster data r and flood disaster data f The similarity is Flood disaster data r and flood disaster data f The distance is D(h r ,h f ), flood disaster data h r The modulus is h r |, flood disaster data f The modulus is h f |, the tuning parameter is σ, and the kernel width is φ; The flood disaster data with the largest number of similarities is taken as the classification cluster center, and the remaining flood disaster data are classified into the classification cluster with the largest similarity to the classification cluster center to obtain the classification cluster; Calculate the importance of classification clusters: The importance of the u-th classification cluster is Z u , the rth flood disaster data is h r , transposed to T, the unit matrix is E, and the transposed of the rth flood disaster data is The cluster center is h o , the number of flood disaster data in the u-th classification cluster is n r , the importance parameter is μ and the error coefficient is ε; The flood disaster data of the classification clusters with importance greater than 0.318 are taken as key data, and the flood relevance of the key data is calculated: The number of key data is n1, and the flood relevance of the rth key data is The flood severity is Flood disaster data r and flood severity The joint probability distribution of Flood disaster data r The marginal probability distribution of is p(h r ), flood severity The marginal probability distribution of The key data with flood correlation greater than 0.647 are output as core evaluation indicators.

3. The method for quantitatively assessing large areas of crop flood damage according to claim 1, characterized in that: Basin analysis using the Hydrological Analyst tool in ArcGIS was performed, combined with DEM data to calculate the extent of each river basin within the study area and the potential inundation depth of each pixel.

4. A method for quantitatively assessing large areas of crop flood damage according to claim 1, characterized in that: The flood hazard equation is: L=f(g,l,d) The degree of crop flood damage is L, the growth and development period is g, the flooding duration is l, and the flooding depth is d; The flood damage equation outputs a quantitative description of the degree of damage in percentage.

5. The method for quantitatively assessing large-scale crop flood damage according to claim 1, characterized in that: The flooded areas were treated as new land cover categories and the model was trained independently.

6. A method for quantitatively assessing large areas of crop flood damage according to claim 1, characterized in that: The ground-based precipitation data and river water level observation data are used to construct a flooding duration estimation model.

7. A large-scale quantitative assessment system for crop flood damage, used to implement the method according to any one of claims 1 to 6, characterized in that: include: Acquisition module: Used to train and build a deep learning image semantic segmentation model based on a public flood disaster dataset. This model is then applied to SAR remote sensing imagery of target flood disaster events to obtain the spatial extent of flood inundation. Based on topographic and river water level observation data, an inundation depth model is used to estimate the flood inundation depth at the DEM pixel scale. Calculation module: This module combines flood disaster areas identified by time-series remote sensing data with ground-observed precipitation and river water level changes to estimate the duration of inundation. Output module: used to calculate the three core assessment indicators of crop flood inundation range, inundation duration, and inundation depth, associate flood disaster events with crop stages, construct a flood hazard degree equation based on flood stress sensitivity, determine the crop flood damage degree equation, and complete spatial mapping of the damage degree.

Images