Flood inundation mapping method and system integrating social media data and SAR images

Abstract

The present invention provides a flood inundation mapping method and system integrating social media data and SAR imagery, relating to the field of emergency management technology. The method comprises: obtaining text data from social media related to rainstorm and flood events and containing POI address information; generating a rough flood inundation range map and a flood inundation probability map for the affected area based on DEM data, POI inundation points, and corresponding water depth information; determining flood inundation probability threshold intervals based on searched on-site image and video data, and obtaining a flood inundation range map for urban built-up areas by taking the intersection; extracting water body data from different periods based on multi-view SAR imagery before, during, and after the disaster to generate a river flood inundation range map; and merging the urban built-up area flood inundation range map and the river flood inundation range map to obtain an inundation range map for the affected area. The present invention fully utilizes the advantages of social media data and time-series SAR imagery in urban flood detection and river flood detection, respectively, to achieve high-precision mapping of large-scale flood events.

Description

Technical Field

[0001] The present invention relates to the technical field of flood disaster emergency management, and in particular to a flood inundation mapping method and system integrating social media data and SAR images. Background Art

[0002] Floods, as natural disasters, cause severe damage to human life, agricultural production, and the ecological environment. In recent years, with global climate change and an increase in extreme weather events, the frequency and destructiveness of floods have continued to intensify. Therefore, timely and accurate monitoring and mapping of flooded areas is crucial for flood prevention and mitigation, emergency response, and post-disaster reconstruction.

[0003] With the prevalence of social media, people often share relevant information about natural disasters such as floods through platforms like Weibo, WeChat, and TikTok. This social media data, characterized by real-time availability, widespread availability, and diversity, provides a rich source of information for flood monitoring. Researchers have already achieved considerable success in social media data mining, flood event identification, and flood range prediction. Valuable information is extracted from large amounts of social media text, images, and videos. By mining social media data, we can obtain information on the spatiotemporal distribution, impact area, and severity of flood events, providing data support for flood monitoring. Researchers have also processed social media data by constructing text classification and image recognition models, enabling the automatic identification of flood events.

[0004] Synthetic Aperture Radar (SAR) is an active remote sensing technology with all-weather, all-day, and high-resolution capabilities. It can penetrate clouds and fog to obtain surface information. Time-series SAR imagery plays a vital role in flood monitoring, being used to identify flooded areas and assess flood impacts. By analyzing SAR images taken at different time points, time-series SAR imagery can be used to identify flooded areas. Researchers have used methods such as threshold segmentation, edge detection, and classification to process time-series SAR images, achieving automatic extraction of flooded areas.

[0005] Although social media data and time-series SAR imagery have achieved remarkable results in flood monitoring, existing methods do not sufficiently integrate social media data with SAR imagery, resulting in limitations in flood monitoring and inundation range prediction, making it difficult to meet the needs of real-time and accurate flood monitoring. Summary of the Invention

[0006] To address the above issues, the present invention provides a flood inundation mapping method that integrates social media data and SAR imagery. By fusing social media data and time-series SAR imagery, this method enables rapid extraction and range mapping of flood inundation areas, providing strong support for flood prevention, disaster reduction, and disaster emergency response.

[0007] To achieve the above objectives, the present invention provides the following technical solutions.

[0008] The flood inundation mapping method integrating social media data and SAR images includes the following steps:

[0009] Obtain text data related to rainstorm and flood events from social media with POI address information, detect water depth information in the text data, and obtain a point set of multiple POI flooded points with water depth information;

[0010] Obtain DEM data for urban built-up areas; based on the DEM data, POI inundation points, and corresponding water depth information, obtain the water level elevation values ​​for each POI inundation point; and use the Kriging interpolation algorithm based on the DEM data and the water level elevation values ​​for each POI inundation point to generate a rough flood inundation range map for the affected area.

[0011] Each grid center point in the DEM data of the urban built-up area is traversed one by one to determine the nearest POI inundation point for each grid center point. Based on the water level elevation values ​​of the grid center point and the nearest POI inundation point, the flood inundation probability at each grid center point is obtained and a flood inundation probability map is constructed.

[0012] Based on the on-site image and video data obtained through the search, typical flood inundation locations are determined; these typical flood inundation locations are superimposed on the flood inundation probability map, and the flood inundation probability threshold interval is determined. The flood inundation range map of the urban built-up area is obtained by intersecting the flood inundation probability range within the threshold interval with the rough flood inundation range map;

[0013] Based on multi-view SAR images before, during, and after the disaster, the random forest algorithm was used to extract water body data from different periods and generate river flood inundation range maps.

[0014] The flood inundation range map of urban built-up areas and the river flood inundation range map are merged to obtain the inundation range map of the affected areas of flood events.

[0015] Preferably, the obtaining of text data related to rainstorm and flood events and containing POI address information from social media, and detecting water depth information in the text data, comprises the following steps:

[0016] Using keywords to obtain text data related to a specific rainstorm and flood event from social media; the social media is a Weibo platform;

[0017] The ChatGLM3-6B model is used to extract text data with POI address information;

[0018] A water depth lookup table is constructed to detect water depth information from text data containing POI address information. The water depth lookup table detects water depth information based on four factors: direct description of depth, reference to human height, reference to building height, and reference to the height of other objects.

[0019] Preferably, the construction of the rough flood inundation range map comprises the following steps:

[0020] Obtain the DEM data of the urban built-up area, and based on the DEM data and the inundation depth of the POI inundation point, back-calculate the water level elevation value at the POI inundation point;

[0021] Using these POI inundation points with water level elevation values, the ordinary Kriging interpolation algorithm is used to interpolate the flood water level surface of the affected area;

[0022] The interpolated flood water level surface of the affected area is subtracted from the DEM data, and the locations where the water level elevation is higher than the DEM data are used as the global inundation range map;

[0023] The urban built-up area boundary data publicly released on the Internet was used to mask the global flood range map to obtain a rough flood inundation range map of the urban built-up area.

[0024] Preferably, the construction of the flood inundation probability map comprises the following steps:

[0025] Mark the grid row and column numbers of all POI submerged points with water depth information on the DEM data;

[0026] All grid center points on the DEM data are used as the flood inundation probability locations to be calculated. The KD tree is used to index the POI inundation points, and the POI inundation point closest to each grid center point is determined and marked.

[0027] The flood inundation probability at each grid center point is calculated based on the nearest POI inundation point.

[0028] Preferably, the calculation of the flood inundation probability at each grid center point comprises the following steps:

[0029] The current grid center point to be calculated is j, and the nearest known POI flooding point is i. The higher the elevation of point j, the less likely point j is to be flooded. The flooding probability is negatively correlated with the distance between i and j.

[0030] Define the height difference between POI submerged point i and grid center point j as:

[0031]

[0032] Among them, H i and H j are the heights of POI submerged point i and grid center point j, DH ij is the height difference between POI submerged point i and grid center point j. If the position of point j is higher than point i, point j will not be submerged. i <H j When, DH ij The value of H is 0; i ≥H j When, DH ij The value of is the height difference between the two points;

[0033] Then the probability P of the grid center point j being flooded is j for:

[0034]

[0035] Among them, d ij represents the Euclidean distance between POI submerged point i and grid center point j;

[0036] P j The maximum and minimum normalization is performed on the area that is not 0 or 1. The calculation formula is:

[0037]

[0038] in, is the normalized value of the flooding probability of the grid center point j, max(P) represents the maximum value of the flooding probability of all locations, and min(P) represents the minimum value of the flooding probability of all locations.

[0039] Preferably, the generation of the river flood inundation range map comprises the following steps:

[0040] Preprocessing of SAR images, including radiation correction, speckle noise suppression, geometric terrain correction and backscatter coefficient conversion;

[0041] Select VV, VH, and SDWI bands, and construct training samples and test samples in a ratio of 7:3 from the preprocessed images for random forest model training;

[0042] The trained random forest model was used to extract water body data before, during and after the disaster. The water bodies before, during and after the disaster were superimposed and merged to obtain the river flood inundation range extracted from SAR images.

[0043] Preferably, the generation of the flood range map of the disaster-stricken area includes the following steps:

[0044] The urban built-up area flood inundation range map and the river flood inundation range are raster-merged to generate a new binary raster map; the binary raster map is the flood range map of the affected area.

[0045] A flood inundation mapping system integrating social media data and SAR images, the system comprising:

[0046] processor;

[0047] a memory having stored thereon a computer program executable on the processor;

[0048] Wherein, when the computer program is executed by the processor, the steps of the flood inundation mapping method integrating social media data and SAR images are implemented.

[0049] A computer-readable storage medium stores a data processing program, which, when executed by a processor, implements the steps of the flood inundation mapping method integrating social media data and SAR images.

[0050] Beneficial effects of the present invention:

[0051] This paper proposes a flood inundation mapping method and system that integrates social media data and SAR imagery. By integrating social media data and time-series SAR imagery, this method enables rapid extraction, probabilistic assessment, and range mapping of flooded areas. This generates a flood range map of the affected area that combines the flood range of urban built-up areas with the flood range of river channels. This accurately captures the inundation range of the affected area, meeting the needs of real-time and accurate flood monitoring. Leveraging the advantages of social media data and time-series SAR imagery for urban and river flood detection, respectively, this method achieves high-precision mapping of large-scale flood events, facilitating scientific analysis of flood event processes and providing technical support for government disaster management decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is a flow chart of a flood inundation mapping method integrating social media data and SAR images according to an embodiment of the present invention;

[0053] Figure 2 This is an example of Weibo data downloaded from the Weibo platform using flood event keywords;

[0054] Figure 3 is the flooding point extracted from the Weibo text of a flood event in a certain city; Figure 3 (a) is the flood inundation point, Figure 3 (b) is the submerged point with water depth information;

[0055] Figure 4It is a rough flood inundation range map obtained by spatial interpolation of DEM data and inundation points with water depth information;

[0056] Figure 5 It is a schematic diagram of the method for calculating the probability of flooding;

[0057] Figure 6 It is a flood inundation probability map generated by combining DEM data and inundation points with water depth information;

[0058] Figure 7 is based on the flooding probability map, Figure 7 (a)- Figure 7 (c) The possible flooding range maps generated using different thresholds;

[0059] Figure 8 It is a schematic diagram to verify the possible flooding range map based on the photos and video information in the news reports;

[0060] Figure 9 It is a flood inundation range map of the urban built-up area obtained by taking the intersection of the confirmed final possible inundation range map and the rough flood inundation range map;

[0061] Figure 10 It is the SAR image data of a flood event in a certain city before, during and after the disaster. Figure 10 (a) is the SAR image data before the disaster. Figure 10 (b) is the SAR image data during the disaster. Figure 10 (c) is the post-disaster SAR image data;

[0062] Figure 11 is the water body range generated by the three-view SAR images before, during and after the disaster, among which, Figure 11 (a) is the water body range generated by SAR image before the disaster. Figure 11 (b) The water body range generated by SAR images during the disaster. Figure 11 (c) Water body extent generated for post-disaster SAR images;

[0063] Figure 12 It is the flood inundation range of a certain flood event in a certain city generated based on time-series SAR images;

[0064] Figure 13 It is the final flood inundation range map obtained by combining the flood inundation range generated by time-series SAR images and the flood inundation range generated by social media data. DETAILED DESCRIPTION

[0065] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0066] Example 1

[0067] Social media data and time-series SAR images have achieved remarkable results in flood monitoring, but existing methods still have the following shortcomings: First, the quality of social media data is uneven, and there is a large amount of noise information, which affects the accuracy of flood monitoring. Second, the time-series SAR image processing method is complex and requires high computing resources and professional knowledge, which limits its promotion in practical applications. Third, the degree of integration between social media data and SAR images is insufficient, and there is a lack of effective fusion methods, which affects the efficiency of flood monitoring. Fourth, existing methods have limitations in predicting flood inundation ranges, and it is difficult to meet the needs of real-time and accurate flood monitoring. To this end, the present invention provides a flood inundation mapping method that integrates social media data and SAR images. The specific process is as follows. Figure 1 As shown, the following steps are included:

[0068] S1: Obtain text data related to rainstorm and flood events in social media with POI address information, detect water depth information in the text data, and obtain a point set of multiple POI flooded points with water depth information.

[0069] S2: Obtain DEM data for the urban built-up area; based on the DEM data, POI inundation points, and corresponding water depth information, obtain the water level elevation value of each POI inundation point; based on the DEM data and the water level elevation values ​​of each POI inundation point, use the Kriging interpolation algorithm to generate a rough flood inundation range map of the affected area.

[0070] S3: Traverse each grid center point of the DEM data of the urban built-up area one by one to determine the nearest POI inundation point for each grid center point; obtain the flood inundation probability at each grid center point based on the water level elevation values ​​of the grid center point and the nearest POI inundation point, and construct a flood inundation probability map.

[0071] S4: Based on the on-site pictures and video data obtained through the search, determine the typical flood inundation locations; superimpose the typical flood inundation locations on the flood inundation probability map, confirm the flood inundation probability threshold interval, and intersect the flood inundation probability range within the threshold interval with the rough flood inundation range map to obtain the flood inundation range map of the urban built-up area.

[0072] S5: Based on the multi-view SAR images before, during and after the disaster, the water body data of different periods are extracted based on the random forest algorithm to generate the river flood inundation range map.

[0073] S6: Combine the urban built-up area flood inundation range map and the river flood inundation range map to obtain the flood disaster area inundation range map.

[0074] In the prior art, Chinese patent CN202210633324.3, a method for estimating flood inundation range based on multi-source data, proposes a method for generating a flood inundation probability map. However, this patent independently generates a flood inundation probability map for each POI inundation location. For example, if there are 100 flood inundation points in the disaster-stricken area, 100 flood inundation probability maps will be generated. In contrast, the present invention generates a complete flood inundation probability map for all POI inundation points with water depth information. At the same time, the present invention normalizes the flood inundation probability so that the inundation probability in the disaster-stricken area is comparable. Furthermore, the present invention uses multi-source data to determine weights based on multiple flood inundation probability maps, and the final result is still a flood inundation probability map. However, the present invention uses a single flood inundation probability map in combination with photos and videos of actual inundation locations published in online news reports to determine the appropriate flood inundation probability interval as the actual inundation range. This is a fundamental difference in technology.

[0075] In this embodiment, the social media platform is the Weibo platform. In S1, keywords are specifically used to download Weibo text related to a specific rainstorm and flood event from the Weibo platform.

[0076] The data collection method for the Weibo texts related to the rainstorm and flood events here is mainly to use the API interface provided by Sina Weibo, combine specific keywords such as "XX rainstorm" or "XX flood" characterized by a certain place name XX, and download a large amount of Weibo text data posted by Weibo users from the Sina Weibo platform according to the time of the flood. Figure 2 As shown in FIG, there are multiple examples of microblog data downloaded from the microblog platform. The downloaded data includes information such as user name, microblog publishing time, and microblog content.

[0077] The ChatGLM3-6B large language model was used to extract flood event microblog texts with POI address information.

[0078] ChatGLM3-6B is an open-source model in the ChatGLM3 series. While retaining many of the excellent features of the previous two generations, such as smooth conversation and low deployment requirements, it also provides a more diverse training dataset, more comprehensive training steps, and a more rational training strategy. It extracts POI addresses from Weibo text by constructing appropriate prompt words and providing specific examples to help the model better understand the expected output content.

[0079] The task instruction for the present invention to interact with ChatGLM3-6B is "Please identify and extract all specific place name information from the following Weibo text. The information I need is the specific place name, not the name of the administrative division such as province, city and county. Place categories include tourist attractions, life service facilities, hotel accommodation, shopping consumption, catering, medical care, companies and enterprises, etc."

[0080] For example, if the text of the Weibo post is "[Rescue of disabled people in the rain] At 20:00 on July 18, at the west gate of Yuandingyuan Community, Pingyuan Road, the road was flooded due to heavy rain. A disabled woman in her 50s got her electric tricycle stuck in the water and couldn't push it out. Patrol police officer XXX and auxiliary police officer XXX of the No. 2 patrol car of the High-tech Branch helped the disabled woman push the electric tricycle out of the water, and then sent the disabled woman home", then the place name extracted by ChatGLM3-6B is 'Pingyuan Road, Yuandingyuan Community'. The extracted place names are output in JSON format. The format example is: {"place 1":"name","place 2":"name",...}. Do not output any other explanations. For example Figure 3 (a) shows a Weibo text about a flood incident in a certain city. The ChatGLM3-6B large language model is used to parse and obtain all the flooded locations.

[0081] Furthermore, a water depth lookup table is established to detect water depth information from flood event microblog texts with POI address information.

[0082] In the microblog texts related to rainstorms and floods, in addition to frequently mentioning the location of the flood, the water depth range or maximum water depth is often included. According to the content, they are divided into four categories: (1) directly describing the depth, such as "the water depth is now 1.5 meters"; (2) using a person's height as a reference, such as "How heavy is this rain? It is well known that the intersection of Daoqing Road and Xinzhong Road is flooded every time it rains in Xinxiang. The water level here is already waist-high." The water depth at the intersection of Daoqing Road and Xinzhong Road in Xinxiang City is set to 1.2 meters; (3) using the height of buildings as a reference; (4) using the height of other specific objects as a reference. The program filters the microblog texts containing POI locations and finally obtains the flooded POI locations containing water depth information. The water depth search expression example is as follows:

[0083] Table 1 Water depth lookup table

[0084]

[0085]

[0086] like Figure 3 (b) shows a microblog text about a flood event in a certain city. By using the water depth lookup table, the water depth information is extracted from the microblog text with the flooded location. Figure 3 (b) shows the flooded locations with water depth information.

[0087] Furthermore, the construction of the rough flood inundation range map in S2 includes the following steps:

[0088] S2.1: Based on high-resolution DEM data (spatial resolution of 5 meters or less), combined with the inundation depth at the POI inundation point, reversely calculate the water level elevation value at the POI inundation point.

[0089] S2.2: Use these POIs with associated water level values ​​and apply the ordinary Kriging algorithm to interpolate flood water levels over the affected area. Kriging is a spatial interpolation method based on statistics and probability theory that can provide more accurate and reliable predictions than other interpolation methods.

[0090] S2.3: Subtract the interpolated flood water level surface of the affected area from the high-resolution DEM data, and use the locations where the water level elevation is higher than the DEM elevation as the global inundation range map.

[0091] S2.4: Use the urban built-up area boundary data publicly available on the Internet and mask it with the global inundation extent map to obtain a rough flood inundation extent map of the urban built-up areas in the affected area.

[0092] Figure 4 It is a rough inundation range map of a flood event in a certain city generated by combining high-resolution DEM data and flood location information with water depth information using the Kriging interpolation algorithm.

[0093] Some areas in the flooded range calculated by S2 are far away from the flooded points with known water levels and are scattered, so the accuracy of water level inference is low. Therefore, it is necessary to calculate the flooding probability of each location to more accurately infer whether the area is flooded. Figure 6 It is a flood inundation probability map generated by combining DEM data and inundation points with water depth information. S3 specifically includes the following steps:

[0094] S3.1: Mark the grid row and column numbers of all POI flooded points with water depth information on the DEM of the disaster area.

[0095] S3.2: Using all grid center points on the DEM of the affected area as the locations to be calculated for flood inundation probability, identify the closest POI to each grid center point. To speed up processing and avoid the slow calculations caused by exhaustive enumeration, this method uses a KD tree to index POI inundation points, allowing for faster calculation of the closest POI to each grid center point.

[0096] S3.3: Traverse each grid center point one by one, find the POI flood point closest to the grid center point, and calculate the flood inundation probability at the grid center point. Figure 5 This is a schematic diagram of the method for calculating the probability of flooding. The calculation principle is as follows:

[0097] Assuming that the current grid center point to be calculated is j, and the nearest known POI flooded point to this point is i, then the higher the elevation of point j, the less likely point j is to be flooded, and the flooding probability is negatively correlated with the distance between i and j;

[0098] Define the height difference between POI submerged point i and grid center point j as:

[0099]

[0100] Among them, H i and H j are the heights of POI submerged point i and grid center point j, DH ij is the height difference between POI submerged point i and grid center point j. If the position of point j is higher than point i, point j will not be submerged. i <H j When, DH ij The value of H is 0; i ≥H j When, DH ij The value is the height difference between the two points.

[0101] Then the probability P of the grid center point j being flooded is j for:

[0102]

[0103] Among them, d ij represents the Euclidean distance between POI submerged point i and grid center point j;

[0104] P j The maximum and minimum normalization is performed on the area that is not 0 or 1. The calculation formula is:

[0105]

[0106] in, The normalized flooding probability at grid center point j is represented by max(P), which represents the maximum flooding probability across all locations, and min(P), which represents the minimum flooding probability across all locations. After normalization, all probability values ​​are scaled to the range of 0 to 1, ensuring comparability between flooding probabilities at different locations and eliminating dimensional differences introduced during the calculation process.

[0107] Furthermore, S4 mainly searches for photos and videos in real-time news reports about flood events through search engines, compares the flood site with flood inundation probability maps under different thresholds, and finally selects the flood inundation probability map that best matches the on-site scenario. Figure 7 Maps of possible flooding range generated using different thresholds. Figure 8 This is a schematic diagram of verifying the possible flood range map based on the photos and video information in the news reports. By investigating the flooding scenarios at different locations and comparing them with the flood inundation range maps under different probability thresholds, the most appropriate flood inundation range map under the flood probability threshold is finally determined.

[0108] like Figure 9 shown.

[0109] Specifically, S5 performs the following steps:

[0110] Using multiple images before, during, and after a disaster, the random forest algorithm was used to extract water body data from different periods and generate the flood inundation range of the river channel.

[0111] This step is mainly implemented through the Sentinel-1GRD dataset provided by the GEE platform. The SAR image is preprocessed directly on the GEE platform, including radiation correction, coherent speckle noise suppression, geometric terrain correction, and backscatter coefficient conversion.

[0112] Select the VV, VH, and SDWI bands. VV polarization is sensitive to surface roughness and can effectively distinguish between water and non-water bodies. VH polarization is sensitive to vegetation and can improve classification accuracy in areas with vegetation cover. SDWI is a water index based on a combination of VV and VH polarizations. Taking advantage of the fact that water has a lower backscatter coefficient than soil and vegetation in SAR imagery, it multiplies dual-polarization data to enhance the contrast between water and non-water bodies, improving water identification.

[0113] SDWI=ln(10×VV×VH)

[0114] After preprocessing, training and testing samples were constructed using a 7:3 ratio for random forest model training. Finally, the pre-disaster, mid-disaster, and post-disaster water bodies were superimposed and merged to obtain the river flood inundation extent extracted from the SAR imagery. Figure 10 It is the SAR image data of a flood event in a certain city before, during and after the disaster. Figure 11 The water body range is generated based on the SAR images before, during and after the disaster using the above water body extraction index; Figure 12 It is the flood inundation range of a flood event in a certain city generated based on time-series SAR images. This range is obtained by taking the union of the water body ranges generated by the above three images.

[0115] In S6, the urban area flood inundation range and the river flood inundation range are merged to obtain the inundation range map of the final flood event.

[0116] Since the urban area flood inundation range map and the river flood inundation range are both binary raster data, this step mainly involves directly merging the two sets of data into a new binary raster map. Figure 13 It is the final flood inundation range map obtained by combining the flood inundation range generated by time-series SAR images and the flood inundation range generated by social media data.

[0117] The above is a method for flood inundation mapping that integrates social media data and SAR images, provided in one embodiment of this embodiment. Based on the same idea, this embodiment also provides a corresponding flood inundation mapping system that integrates social media data and SAR images. For the specific definition of the flood inundation mapping system that integrates social media data and SAR images, please refer to the definition of the flood inundation mapping method that integrates social media data and SAR images above, which will not be repeated here. Each module in the above-mentioned flood inundation mapping system that integrates social media data and SAR images can be fully or partially implemented by software, hardware, and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above-mentioned modules.

[0118] This embodiment also provides a computer-readable storage medium, which stores a computer program that can be used to execute the above Figure 1 A flood inundation mapping method integrating social media data and SAR imagery is provided.

[0119] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), tape, floppy disk, flash memory or optical memory, etc. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0120] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements 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 flood inundation mapping method integrating social media data and SAR images, characterized in that: The following steps are involved: Obtain text data related to rainstorm and flood events from social media with POI address information, detect water depth information in the text data, and obtain a point set of multiple POI flooded points with water depth information; Obtain DEM data of urban built-up areas; Based on DEM data, POI inundation points, and corresponding water depth information, the water level elevation values ​​of each POI inundation point are obtained. Based on the DEM data and the water level elevation values ​​of each POI inundation point, a rough flood inundation range map of the affected area is generated using the Kriging interpolation algorithm. Each grid center point in the DEM data of the urban built-up area is traversed one by one to determine the nearest POI inundation point for each grid center point. Based on the water level elevation values ​​of the grid center point and the nearest POI inundation point, the flood inundation probability at each grid center point is obtained and a flood inundation probability map is constructed. Based on the on-site image and video data obtained through the search, typical flood inundation locations are determined; these typical flood inundation locations are superimposed on the flood inundation probability map, and the flood inundation probability threshold interval is determined. The flood inundation range map of the urban built-up area is obtained by intersecting the flood inundation probability range within the threshold interval with the rough flood inundation range map; Based on multi-view SAR images before, during, and after the disaster, the random forest algorithm was used to extract water body data from different periods and generate river flood inundation range maps. Merge the urban built-up area flood inundation range map and the river flood inundation range map to obtain the flood disaster area inundation range map; The method of obtaining text data related to a rainstorm and flood event and containing POI address information from social media and detecting water depth information in the text data includes the following steps: Using keywords to obtain text data related to a specific rainstorm and flood event from social media; the social media is a Weibo platform; The ChatGLM3-6B model is used to extract text data with POI address information; Constructing a water depth lookup table to detect water depth information from text data containing POI address information; the water depth lookup table detects water depth information based on four factors: direct description of depth, reference to human height, reference to building height, and reference to the height of other objects; The construction of the flood inundation probability map includes the following steps: Mark the grid row and column numbers of all POI submerged points with water depth information on the DEM data; All grid center points on the DEM data are used as the flood inundation probability locations to be calculated. The KD tree is used to index the POI inundation points, and the POI inundation point closest to each grid center point is determined and marked. Calculate the flood inundation probability at each grid center point based on the nearest POI inundation point; The generation of the river flood inundation range map includes the following steps: Preprocessing of SAR images, including radiation correction, speckle noise suppression, geometric terrain correction and backscatter coefficient conversion; Select VV, VH, and SDWI bands, and construct training samples and test samples in a ratio of 7:3 from the preprocessed images for random forest model training; The trained random forest model was used to extract water body data before, during and after the disaster. The water bodies before, during and after the disaster were superimposed and merged to obtain the river flood inundation range extracted from SAR images.

2. The flood inundation mapping method integrating social media data and SAR images according to claim 1, characterized in that: The construction of the rough flood inundation range map includes the following steps: Obtain the DEM data of the urban built-up area, and based on the DEM data and the inundation depth of the POI inundation point, back-calculate the water level elevation value at the POI inundation point; Using these POI inundation points with water level elevation values, the ordinary Kriging interpolation algorithm is used to interpolate the flood water level surface of the affected area; The interpolated flood water level surface of the affected area is subtracted from the DEM data, and the locations where the water level elevation is higher than the DEM data are used as the global inundation range map; The urban built-up area boundary data publicly released on the Internet was used to mask the global flood range map to obtain a rough flood inundation range map of the urban built-up area.

3. The flood inundation mapping method integrating social media data and SAR images according to claim 1, characterized in that: The calculation of the flood inundation probability at each grid center point includes the following steps: The current grid center point to be calculated is j, and the nearest known POI flooding point is i. The higher the elevation of point j, the less likely point j is to be flooded. The flooding probability is negatively correlated with the distance between i and j. Define the height difference between POI submerged point i and grid center point j as: Among them, H i and H j are the heights of POI submerged point i and grid center point j, DH ij is the height difference between POI submerged point i and grid center point j. If the position of point j is higher than point i, point j will not be submerged. i <H j When, DH ij The value of H is 0; i ≥H j When, DH ij The value of is the height difference between the two points; Then the probability P of the grid center point j being flooded is j for: Among them, d ij represents the Euclidean distance between POI submerged point i and grid center point j; P j The maximum and minimum normalization is performed on the area that is not 0 or 1. The calculation formula is: in, is the normalized value of the flooding probability of the grid center point j, max(P) represents the maximum value of the flooding probability of all locations, and min(P) represents the minimum value of the flooding probability of all locations.

4. The flood inundation mapping method integrating social media data and SAR images according to claim 1, characterized in that: The generation of the flood range map of the disaster-affected area includes the following steps: The urban built-up area flood inundation range map and the river flood inundation range are raster-merged to generate a new binary raster map; the binary raster map is the flood range map of the affected area.

5. A flood inundation mapping system integrating social media data and SAR imagery, characterized in that: The system comprises: processor; a memory having stored thereon a computer program executable on the processor; When the computer program is executed by the processor, the steps of the flood inundation mapping method integrating social media data and SAR images are implemented as described in any one of claims 1 to 4.

6. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a data processing program, which, when executed by a processor, implements the steps of the flood inundation mapping method integrating social media data and SAR images according to any one of claims 1 to 4.

Images