A method for estimating flood inundation range based on multi-source data

Abstract

The present invention discloses a method for estimating flood inundation range based on multi-source data. This method performs inverse distance weighting according to in-disaster microblog flood inundation point data and DEM elevation data to generate an inundation probability distribution of the study area based on a single microblog point. Based on optical remote sensing images and remote sensing rainfall products, the soil moisture index characterized by the MNDWI index of the remote sensing image and the rainfall index characterized by the remote sensing rainfall product within a certain area around the microblog point are calculated to obtain the confidence weight of each microblog point. Finally, the flood inundation probability distribution of all microblog points is weighted and integrated through the confidence weight to obtain the estimation result of the flood inundation probability distribution considering multi-source data fusion. The present invention can estimate the flood inundation probability distribution without relying on in-disaster remote sensing images, and to a certain extent solve the problem that large-scale inundation information cannot be obtained due to the lack of in-disaster remote sensing data, which is beneficial to flood disaster assessment and emergency response.

Description

Technical Field

[0001] The present invention relates to the technical field of flood disaster assessment, and in particular to a method for estimating flood inundation range during a disaster based on multi-source data. Background Art

[0002] During the flood season every year, extreme rainfall and flood disasters occur frequently, causing serious damage to people's safety, economic construction and social development. Flood disaster information acquisition and emergency management are major needs for national development and an important guarantee for the sustainable development of social economy and ecological environment.

[0003] Disaster information acquisition and disaster emergency response are inseparable from the support of data and information. Traditional remote sensing satellite data and social media data, which have been widely used in recent years, are commonly used data for flood disaster research. Different data sources have their own advantages and disadvantages in flood disaster application scenarios. For example, remote sensing data has a wider coverage and its accuracy is guaranteed as measured data, but the return cycle and meteorological conditions limit the application of remote sensing data in certain scenarios.

[0004] Social media data has the advantages of being massive, timely, and rich in public opinion information. However, as non-authoritative data, its authenticity and accuracy need to be verified.

[0005] Therefore, based on the data availability when a real flood occurs, comprehensive use of multi-source data can break through the performance bottleneck of a single data source and dig out more accurate and comprehensive disaster information, which has important practical significance for disaster emergency response and disaster prevention and mitigation. Summary of the invention

[0006] The present invention provides a flood inundation range estimation method based on multi-source data fusion. By fusing short-term remote sensing images after a disaster, social media data during a disaster, remote sensing rainfall products and terrain data of a study area, the probability distribution of flood inundation during a disaster is estimated. This can, to a certain extent, solve the technical problem of being unable to obtain large-scale inundation information due to the lack of remote sensing data during a disaster.

[0007] In order to solve the above problems, the present invention provides a flood inundation range estimation method based on multi-source data fusion, and the estimation method includes the following contents:

[0008] Step S 100 : Through the acquisition and preprocessing of multi-source data, a multi-source flood database of specific flood events is obtained;

[0009] The multi-source flood database includes social media data during the disaster, CHIRPS remote sensing rainfall products, short-term optical remote sensing images after the disaster, and DEM elevation data of the study area;

[0010] Step S 200: Multi-source data fusion for estimating flood inundation probability distribution, and the multi-source data fusion for estimating flood inundation probability distribution includes the following sub-steps:

[0011] Step S 201 : Create an inundation probability index distribution layer based on each Weibo flood inundation point;

[0012] For each Weibo flood inundation point, combining the point inundation water depth data and the DEM elevation data of the study area, introducing inverse distance weighting, and taking this inundation point as the center, create an inundation probability index distribution layer based on this point;

[0013] Step S 202 : Perform comprehensive weighting on all the generated inundation probability index distribution layers, and introduce a weighting method based on Gaussian surface;

[0014] Perform synthesis on all the generated inundation probability index distribution layers. The weight of each layer is obtained by comprehensively weighting the MNDWI index calculated from the optical remote sensing image pixels within a certain range around this inundation point and the rainfall value around the corresponding point of the CHIRPS remote sensing rainfall product, and introduce a smoother weighting method based on Gaussian surface;

[0015] Step S 203 : Weight and synthesize the n inundation probability distribution layers based on n Weibo points according to the weight values to obtain a comprehensive flood inundation probability distribution map of the study area based on all Weibo inundation points;

[0016] Step S 300 : Combine the actual inundation situation to verify the estimation result of the flood inundation probability distribution.

[0017] Further, in step S 100 : The acquisition of the multi-source data includes in-disaster social media data and remote sensing data, where:

[0018] The in-disaster social media data includes Weibo data, and the acquisition method of the Weibo data includes: obtaining the Weibo data through a combination of web crawler and API. Among them, the fields of the Weibo data include release time, text content, pictures, video content, and geographical location;

[0019] The remote sensing data includes optical remote sensing images covering the flood disaster area in the short term after the flood disaster.

[0020] Further, in step S 100 : The preprocessing of the in-disaster social media data includes data cleaning, duplicate removal, Chinese word segmentation, and stop word removal.

[0021] Further, in step S 100Among them, the screening conditions for the social media data during the disaster include:

[0022] Condition 1: The social media data during the disaster carries accurate geographical location information; and / or

[0023] Condition 2: The text of the social media data during the disaster contains detailed and determined inundation data; and / or

[0024] Condition 3: The social media data during the disaster carries pictures or videos that can clearly reflect the real inundation situation for verifying the authenticity and accuracy of the data; and / or

[0025] Condition 4: Verify the authenticity and accuracy of the inundation water depth according to the official inundation point data, Baidu Map confirmed location data, text content, pictures and video information; and / or

[0026] Condition 5: For the situation where the research area is relatively large, the duplicate data of the same township or street needs to be aggregated into one point, and the duplicate point data is deleted.

[0027] Further, in step S 100 Among them, the preprocessing of the short-term optical remote sensing images after the disaster includes radiometric correction, atmospheric correction and resampling.

[0028] Further, in step S 100 Among them, the selected DEM elevation data is the DEM data with a resolution of 30m officially released by NASA.

[0029] Further, in step S 100 Among them, the selected CHIRPS remote sensing rainfall product is the CHIRPS data with a spatial resolution of 0.05° and a daily scale time resolution.

[0030] Further, in step S 201 Among them, according to each Weibo inundation point, calculate the inundation probability distribution of the entire research area to form an inundation probability distribution map, and the calculation method is based on the following criteria:

[0031] Criterion 1: The verified flood inundation points released on the Weibo platform reflect the flood inundation facts in a certain range of areas. Moreover, the closer the area is to the inundation point, the higher the possibility of flood inundation; conversely, the farther the area is from the inundation point, the lower the possibility of inundation.

[0032] Criterion 2: Within a certain range around the Weibo inundation point, the lower the terrain, the greater the probability of being inundated; on the contrary, the higher the terrain, the smaller the probability of being inundated. The terrain height is mainly combined with the DEM elevation data and the Weibo inundation water depth data.

[0033] Further, in step S 201Among them, the calculation of the inundation probability distribution based on a single Weibo point is as follows:

[0034] In the three-dimensional space of the study area, let the coordinates of the Weibo inundation point be i(x i , y i ), and any other point in the study area is represented as j(x j , y j ). In the DEM distribution, the elevation of point i is H i , the elevation of point j is H j , and the inundation water depth of the inundation point i is H w . Then the inundation probability of point j

[0035]

[0036] where H ij represents the inundation water depth of point j, and the calculation is as follows:

[0037]

[0038] where D ij represents the Euclidean distance between point i and point j

[0039]

[0040] where the exponential parameters α and β are weight parameters used to adjust the influence intensity of H ij and D ij on the result.

[0041] Furthermore, in step S 202 , the research criteria for the weight calculation based on the MNDWI soil moisture index and CHIRPS rainfall include:

[0042] Criterion 3: The soil moisture content within a certain range around the Weibo inundation data point can, to a certain extent, reflect the flood inundation probability of the area. The higher the MNDWI water content, the higher the possibility that the area has been covered by floods; the higher the overall soil moisture around the Weibo point, the higher the credibility of the flood inundation probability distribution calculated from this point;

[0043] Criterion 4: The in-disaster rainfall data in the area around the Weibo inundation data point can, to a certain extent, reflect the flood inundation situation in these areas. When the rainfall exceeds a certain range, the greater the rainfall, the greater the probability that these areas will be flooded;

[0044] Criterion 5: Around the Weibo inundation data point, the MNDWI value and rainfall value of the pixels closer to point i have a higher impact on the overall inundation probability of point i; conversely, the farther away from the point, the lower the impact on the overall inundation probability, which conforms to the first law of geography.

[0045] Furthermore, in step S 203 In the calculation process of the comprehensive flood inundation probability distribution map based on all micro-blog inundation points in the study area, the following is the calculation process:

[0046] Assume that the joint probability density of two-dimensional continuous random variables (x, y) is:

[0047]

[0048] Among them, μ1, μ2 represent the means of variables x and y respectively, σ1, σ2 represent the standard deviations of variables x and y respectively, μ1, μ2, σ1, σ2, ρ are all constants, and σ1>0, σ2>0, |ρ|<1,

[0049] Then (x, y) is said to obey a two-dimensional normal distribution with parameters μ1, μ2, σ1, σ2, ρ.

[0050] The density function of the two-dimensional normal distribution presents an inverted bell-shaped distribution in three-dimensional space and satisfies:

[0051]

[0052] The probability density function of two-dimensional Gaussian normal distribution within a certain range from microblog point i is used to define the influence of MNDWI values and rainfall values of pixels at different distances on point i.

[0053] Taking microblog point i as the origin, in the three-dimensional geographic space, the two-dimensional Gaussian surface is symmetrically distributed along the x and y axes with point i as the center, with mean μ1=μ2=0, σ1=σ2, and taking the constant ρ as ρ=0, the joint probability density function is generally simplified to the following formula:

[0054]

[0055] Where λ = σ1 2 =σ2 2 is a probability density function parameter, which is used to control the range of the region of interest around point i that is included in the calculation.

[0056] Compared with the prior art, the present invention has significant advantages and beneficial effects, which are specifically embodied in the following aspects:

[0057] By integrating short-term post-disaster remote sensing optical images, in-disaster social media data, remote sensing rainfall products, and topographic data of the study area, the in-disaster flood inundation probability distribution is estimated. In the study area, first, inverse distance weighting is performed based on the in-disaster Weibo flood inundation point data and DEM elevation data to generate the inundation probability distribution of the study area based on a single Weibo point; based on the optical remote sensing image and the remote sensing rainfall product, the soil moisture index characterized by the MNDWI index of the remote sensing image and the rainfall index characterized by the remote sensing rainfall product within a certain area around the Weibo point are calculated to obtain the confidence weight of each Weibo point; finally, the flood inundation probability distribution of all Weibo points is weighted and integrated through the confidence weight to obtain the estimation result of the flood inundation probability distribution considering multi-source data fusion. The present invention can estimate the flood inundation probability distribution without relying on in-disaster remote sensing images, and to a certain extent solves the problem of being unable to obtain large-scale inundation information caused by the lack of in-disaster remote sensing data, which is beneficial to flood disaster assessment and emergency response. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a flowchart of the method for estimating the flood inundation probability range by multi-source data fusion in the embodiment of the present invention;

[0059] Figure 2 It is a schematic diagram for calculating the inundation water depth based on a single Weibo inundation point i and DEM elevation in the embodiment of the present invention;

[0060] Figure 3 It is the situation of the study area of the Henan Province rainstorm flood disaster event in the embodiment of the present invention;

[0061] Figure 4 It is a schematic diagram of the multi-source data coverage of the Henan Province rainstorm flood disaster event in the embodiment of the present invention;

[0062] Figure 5 It is a schematic diagram of the inundation probability distribution generated from four Weibo inundation water depth points numbered 02, 12, 20, and 26 along the Wei River selected in the embodiment of the present invention;

[0063] Figure 6 It is a relationship diagram between the weighted area range of the Weibo point and the weighted result in the embodiment of the present invention;

[0064] Figure 7 It is a diagram of the estimated result of the flood inundation probability distribution around Xinxiang and Hebi cities after weighted integration in the embodiment of the present invention;

[0065] Figure 8 It is a schematic diagram for verifying and comparing the result of the flood inundation probability distribution estimation method with the officially released flood inundation range in the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0066] To make the above objects, features, and advantages of the present invention more apparent and understandable, the following provides a detailed description of specific embodiments of the present invention in conjunction with the accompanying drawings.

[0067] Please refer to Figure 1 As shown, an embodiment of the present invention provides a method for estimating flood inundation extent based on multi-source data fusion. The estimation method includes the following steps:

[0068] Step S 100 : Through the acquisition and preprocessing of multi-source data, a multi-source flood database for a specific flood event is obtained;

[0069] The multi-source flood database includes in-disaster social media data, CHIRPS remote sensing rainfall products, short-term optical remote sensing images after the disaster, and DEM elevation data of the study area;

[0070] S 101 : Acquisition of multi-source data

[0071] Weibo data:

[0072] In terms of social media data, through a combination of web crawlers and APIs, Sina Weibo data related to floods during the occurrence period of a specific flood event is obtained.

[0073] Among them: The Weibo data fields include post time, text content, pictures, video content, and post geographical location, etc.

[0074] In the embodiment of the present invention, the reason for selecting Weibo data is that social media data has advantages such as a large amount, strong timeliness, and rich public opinion theme information.

[0075] Remote sensing data:

[0076] In terms of remote sensing data, prepare optical remote sensing images covering the flood disaster area in the short term after the flood occurs.

[0077] In the embodiment of the present invention, the reason for selecting remote sensing data is that remote sensing data has a wider coverage, can obtain flood inundation information in a larger area, and the accuracy is guaranteed as measured data.

[0078] S 102 : Preprocessing of multi-source data

[0079] Weibo data:

[0080] Through preprocessing work such as data cleaning, duplicate removal, Chinese word segmentation, and stop word removal, a Weibo flood corpus is formed;

[0081] Through screening and verification under certain conditions, Weibo flood point data that can be used for multi-source data fusion experiments is obtained. The specific data screening conditions and verification criteria are as follows:

[0082] Condition 1: Weibo data must contain accurate geographic location information;

[0083] Condition 2: The microblog text contains detailed and confirmed flooding data, such as "×× street, village, flooding depth is close to ×× meters";

[0084] Condition 3: The microblog contains pictures or videos that can clearly reflect the actual flooding situation, which can be used to verify the authenticity and accuracy of the data;

[0085] Condition 4: Verify the authenticity and accuracy of the flooding depth based on official flooding point data, Baidu Map confirmed location data, text content, pictures and video information;

[0086] Condition 5: Due to the huge amount of data related to floods on Weibo, for a wide study area, it is necessary to aggregate the duplicate data of the same township or street into one point and delete the duplicate point data.

[0087] Remote sensing data:

[0088] According to the specific remote sensing data, radiation correction, atmospheric correction, resampling and other preprocessing are performed. DEM data and CHIRPS (Climate Hazards Group Infrared Precipitation, CHIRPS) remote sensing rainfall products are used as auxiliary data for the fusion experiment. The 30m resolution DEM data officially released by NASA and the CHIRPS data with 0.05° spatial resolution and daily time resolution are selected.

[0089] Therefore, in the embodiment of the present invention, the multi-source data involved is used for fusion experiment, so it is necessary to perform data preprocessing first.

[0090] In the embodiment of the present invention, based on the data availability when a real flood occurs, multi-source data is comprehensively utilized to break through the performance bottleneck of a single data source, and more accurate and comprehensive disaster information can be mined, which has important practical significance for disaster emergency response and disaster prevention and mitigation.

[0091] Step S 200 :Estimating flood inundation probability distribution by multi-source data fusion

[0092] Please refer to the attached Figure 1 As shown, in an embodiment of the present invention, a multi-source data fusion estimation method is proposed, and the estimation method includes the following steps:

[0093] Step S 201 : Create an exponential distribution layer of flood probability based on each Weibo flood inundation point;

[0094] For each Weibo flood inundation point, combining the inundation depth data of the point and the DEM elevation data of the study area, inverse distance weighting is introduced, and with this inundation point as the center, an inundation probability index distribution layer based on this point is created;

[0095] It should be further noted that for each individual inundation point in this layer, if there are n Weibo points, then n inundation probability index distribution layers are generated.

[0096] Generate the inundation probability distribution based on a single Weibo point

[0097] Calculate the inundation probability distribution of the entire study area according to each Weibo inundation point to form an inundation probability distribution map. The calculation method is based on the following criteria:

[0098] Criterion 1: The verified flood inundation points published on the Weibo platform reflect the flood inundation facts in a certain range of areas. Moreover, the closer the area is to the inundation point, the higher the possibility of flood inundation; conversely, the farther the area is from the inundation point, the lower the possibility of inundation;

[0099] Criterion 2: Within a certain range around the Weibo inundation point, the lower the terrain, the greater the probability of being inundated; on the contrary, the higher the terrain, the smaller the probability of being inundated. The terrain height is mainly combined with the DEM elevation data and the Weibo inundation depth data.

[0100] Please refer to the attached Figure 2 As shown, in the three-dimensional space of the study area, let the coordinate of the Weibo inundation point be i(x i ,y i ), and any other point in the study area be represented as j(x j ,y j ). In the DEM distribution, the elevation of point i is H i , the elevation of point j is H j , and the inundation depth of the inundation point i is H w .

[0101] Then the inundation probability of point j is

[0102]

[0103] where H ij represents the inundation depth of point j, and the calculation is as follows:

[0104]

[0105] where D ij represents the Euclidean distance between point i and point j

[0106]

[0107] Among them, the exponents α and β are weight parameters used to adjust the influence intensity of H ij and D ij on the results.

[0108] Step S 202 : Perform comprehensive weighting on all generated flood probability exponential distribution layers, and introduce a weighting method based on a Gaussian surface;

[0109] Perform synthesis on all generated flood probability exponential distribution layers. The weight of each layer is obtained by comprehensively weighting the MNDWI index calculated from the optical remote sensing image pixels within a certain range around the flood point position and the rainfall value around the corresponding position of the CHIRPS remote sensing rainfall product. A smoother weighting method based on a Gaussian surface is introduced.

[0110] Weight calculation based on the MNDWI soil moisture index and CHIRPS rainfall

[0111] Calculate the weight of each individual layer based on the MNDWI soil moisture index and CHIRPS rainfall, and perform weighted synthesis to obtain the final flood probability distribution.

[0112] The normalized moisture index MNDWI (Modified Normalized Difference Water Index, MNDWI) is used to represent the surface water body or soil moisture content in the study area, and the rainfall in the area around the Weibo point during the heavy rainfall of the flood characterized by the CHIRPS rainfall product. These variables can reflect the inundation situation of the area around the point to a certain extent. The research is based on the following basic criteria:

[0113] Criterion 3: The soil moisture content within a certain range around the Weibo inundation data point can reflect the flood probability of the area to a certain extent. The higher the MNDWI water content, the higher the possibility that the area has been covered by flood; the higher the overall soil moisture around the Weibo point, the higher the credibility of the flood probability distribution calculated from this point;

[0114] Criterion 4: The in-disaster rainfall data in the area around the Weibo inundation data point can reflect the inundation situation of these areas to a certain extent. When the rainfall exceeds a certain range, the greater the rainfall, the greater the probability that these areas will be inundated by flood;

[0115] Criterion 5: Around the Weibo inundation data point, the MNDWI value and rainfall value of the pixels closer to point i have a higher impact on the overall inundation probability of point i; on the contrary, the farther away from the point, the lower the impact on the overall inundation probability, which conforms to the first law of geography.

[0116] How to quantitatively describe the influence of geographical location distance? In the embodiments of this study, the probability density function of two-dimensional Gaussian normal distribution within a certain range from the Weibo point i is used to define the influence weights of the MNDWI values and rainfall values of pixels at different distances on point i.

[0117] Let the joint probability density of the two-dimensional continuous random variable (x, y) be:

[0118]

[0119] where μ1 and μ2 represent the means of variables x and y respectively, σ1 and σ2 represent the standard deviations of variables x and y respectively, and μ1, μ2, σ1, σ2, ρ are all constants, and σ1 > 0, σ2 > 0, |ρ| < 1.

[0120] Then it is said that (x, y) follows a two-dimensional normal distribution with parameters μ1, μ2, σ1, σ2, ρ.

[0121] The density function of the two-dimensional normal distribution presents an inverted bell-shaped distribution in three-dimensional space and satisfies:

[0122]

[0123] Taking the Weibo point i as the origin, in the three-dimensional geographical space, the two-dimensional Gaussian surface is symmetrically distributed around point i along the x and y axes, with the means μ1 = μ2 = 0, σ1 = σ2, and taking the constant ρ = 0, then the joint probability density function is simplified to the following formula:

[0124]

[0125] where λ = σ1 2 = σ2 2 is the parameter of the probability density function, which is used to control the range of the region of interest included in the calculation around point i. The smaller λ is, the steeper the inverted bell-shaped two-dimensional Gaussian surface is, and the greater the contribution of the MNDWI value of the pixel closer to point i; conversely, the larger λ is, the more pixels participate in the calculation, and the relatively smaller the contribution of the pixel closer to i.

[0126] For the CHIRPS rainfall product, since the distribution of ground rainfall monitoring stations is very sparse, we use the remote sensing rainfall product with a spatial resolution of 0.05°. Since the spatial resolution of the remote sensing rainfall product is much lower than that of Sentinel-2 images (10m) and DEM data (30m), the pixel value at the pixel where the Weibo water depth point is located on the remote sensing rainfall product, that is, the rainfall value during the disaster within the range of 0.05°×0.05° around it, is weighted, and the data needs to be normalized.

[0127] Step S 203: Weight the n flood inundation probability distribution maps based on n Weibo points according to the weight values, and obtain a comprehensive flood inundation probability distribution map of the study area based on all Weibo inundation points in the study area;

[0128] Step S 300 : Combine the actual inundation situation to verify the estimation results of the flood inundation probability distribution.

[0129] Example 1

[0130] Taking the Henan rainstorm event in 2021 as an example, from July 19th to 23rd, 2021, affected by the large amount of water vapor transported by Typhoon In-Fa and the combined effect of local topography, Henan Province encountered extreme heavy rainfall, and cities such as Zhengzhou and Xinxiang encountered extremely heavy rainstorms rarely seen in history.

[0131] Please refer to the appendix Figure 3 As shown, during the flood disaster occurrence period, there are no suitable optical remote sensing images and SAR images in some areas such as within the scope of Hebi City. Therefore, it is difficult to obtain large-scale flood inundation information for this area with existing data. The method of multi-source data fusion in the embodiment of the present invention is used to calculate the flood inundation range during the disaster in this area. The study area of multi-source data fusion is located around Xinxiang City in the northern part of Henan Province, which is a severely affected area.

[0132] S 100 : Acquisition and preprocessing of multi-source data;

[0133] Please refer to the appendix Figure 4 As shown, the appendix Figure 4 As shown is the distribution of specific data. The data used here mainly includes: Sentinel-2 optical images on July 26th shortly after the flood disaster occurred, Weibo data with geographical location information related to the flood disaster in Weibo during the flood disaster, daily average data of CHIRPS remote sensing rainfall products from July 19th to 21st during the heavy rainfall period, and DEM digital elevation data of the study area.

[0134] The first step: Obtain the Weibo flood disaster dataset of the Henan rainstorm flood disaster event in 2021

[0135] In terms of social media data, through the combination of web crawler and API, obtain the Sina Weibo data related to the flood disaster during the occurrence period of the Henan rainstorm flood disaster event in 2021:

[0136] Among them: The Weibo data fields include posting time, Weibo text content, posting location, etc. Through data preprocessing work such as data cleaning, duplicate removal, Chinese word segmentation, and stop word removal, obtain the Weibo flood disaster dataset based on the Henan rainstorm flood disaster event in 2021.

[0137] The second step: Obtain the data of Weibo inundation water depth points

[0138] According to the above-mentioned Weibo water depth point data screening and verification criteria, screening was carried out from the formed Henan rainstorm Weibo flood corpus in 2021, and finally 27 Weibo inundation water depth point data around Xinxiang City and Hebi City were obtained.

[0139] Step 3: Obtain remote sensing data

[0140] Regarding remote sensing data, prepare the Sentinel-2 optical images opened by the European Space Agency (ESA) covering the flood disaster area in the short term after the flood occurred. The imaging time is July 26 in the short term after the Henan flood. The original images need to be preprocessed by radiometric correction, atmospheric correction, and resampling to prepare for calculating the MNDWI index in the subsequent steps.

[0141] Step 4: Obtain DEM data and CHIRPS remote sensing rainfall products

[0142] The DEM data and CHIRPS (Climate Hazards Group Infrared Precipitation) remote sensing rainfall products are used as auxiliary data for the fusion experiment. Select the DEM data with a resolution of 30m released by NASA official, and select the CHIRPS data with a spatial resolution of 0.05° and a daily scale time resolution.

[0143] Step 5: Estimate the flood inundation probability distribution during the disaster based on multi-source data

[0144] First, generate the inundation probability distribution based on the Weibo points. Taking each Weibo inundation point as the center, according to the Weibo water depth data and the DEM elevation data of the study area, the flood inundation probability distribution in the entire study area can be estimated, forming an inundation probability map based on this Weibo point.

[0145] In the embodiment of the present invention, a total of 27 Weibo points in the study area are selected, and finally flood inundation probability distribution maps with the same study scope based on these 27 points are generated.

[0146] As shown in the appendix Figure 5 The following is the inundation probability result map of 4 selected Weibo points. The change in raster color represents the change in flood inundation probability from low to high. It can be seen that:

[0147] 1. Generally speaking, for the inundation probability maps generated based on different Weibo inundation points, the inundation probability mainly changes from high to low with this point as the center, which is related to the inverse distance weight;

[0148] 2. The inundation probability distribution is closely related to the DEM elevation distribution and change. Around the Weibo point, the higher the elevation area, the lower the probability of being inundated. For example, the northwest of the study area is the Taihang Mountains with a higher elevation, so the possibility of this area being inundated is lower;

[0149] Secondly, for the flood probability distribution maps of the entire study area generated based on different Weibo inundation points, further processing is required, that is, image synthesis is performed to form a flood probability distribution that combines all Weibo points.

[0150] To explore the influence of the size of the region of interest on calculating the weight of point i and thus find the most suitable region size, this study calculated the weighted MNDWI values of point i corresponding to the radii of the regions of interest ranging from 1 grid to 80 grids (800 m) with a step size of 1 grid.

[0151] As shown in the appendix Figure 6 As shown, in the embodiment of the present invention, as the region of interest increases, the calculated weighted MNDWI value at point i gradually increases and reaches a saturation value at around 350 m (corresponding to 35 grid numbers). Therefore, in this embodiment, the radius of the region of interest is taken as 350 m, and the weighted MNDWI value of point i is calculated accordingly.

[0152] For the remote sensing rainfall product, since the spatial resolution of the remote sensing rainfall product is much lower than that of the Sentinel-2 image (10 m) and the DEM data (30 m), we take the pixel value at the pixel where the Weibo water depth point is located on the remote sensing rainfall product, that is, the average rainfall value during the disaster within the range of 0.05°×0.05° around it, to calculate the weight, and the data needs to be normalized.

[0153] Based on the MNDWI values of the pixels within a certain range around Weibo point i, weighted calculation is performed, and the rainfall data values of the corresponding pixels of the remote sensing rainfall product are integrated to obtain the weight of the flood probability distribution map corresponding to point i. Then, weighted calculation is performed on all probability distributions.

[0154]

[0155] Thus, a flood probability distribution prediction map that integrates Weibo flood disaster points, Sentinel-2 optical images, and DEM elevation data within the study area is obtained. As shown in the appendix Figure 7 As shown, from Figure 7 it can be concluded that:

[0156] The flood probability distribution has a roughly the same distribution trend as the points of the Weibo flood disaster data. In addition, it is closely related to the DEM terrain trend of the study area;

[0157] Taking the Weibo flood disaster data points as the center, the closer the distance to the point and the lower the elevation, the higher the flood probability. Conversely, the farther the distance to the point and the higher the elevation, the lower the flood probability;

[0158] Sixth step: Verification of the estimation results

[0159] For the method for estimating the flood inundation probability distribution involving multi-source data fusion in the embodiments of the present invention, in combination with the Henan rainstorm flood event in 2021, the estimation results are verified through the flood inundation conditions in the surrounding areas of Xinxiang and Weihui officially released during the flood disaster.

[0160] Please refer to Figure 8 As shown, in the embodiments of the present invention, the attached Figure 8 is a comparison chart of experimental results for the areas surrounding Xinxiang and Weihui.

[0161] Among them: the attached Figure 8 (a) is the flood inundation probability distribution of some areas surrounding Xinxiang City and Weihui City obtained according to the flood inundation probability estimation method in the embodiments of the present invention;

[0162] The attached Figure 8 (b) is the flood inundation range officially released by the National Remote Sensing Center and relevant departments during the Henan rainstorm. The flood inundation range of some areas surrounding Xinxiang and Weihui is obtained from the ICEYE constellation images during the flood disaster period and combined with the water body range before the disaster.

[0163] It can be seen from the comparison results in the figure that the inundated areas surrounding Xinxiang City and Weihui City are mainly the areas along the Wei River.

[0164] In the flood inundation probability distribution estimated by the estimation method in the embodiments of the present invention, the parts with higher inundation probabilities have a roughly the same distribution trend as the officially released flood inundation range.

[0165] From this, it can be seen that the method involved in the embodiments of the present invention can relatively accurately estimate the probability distribution range of flood inundation during the disaster. Therefore, in the areas surrounding Hebi City where remote sensing data is missing during the disaster, we believe that the estimated flood inundation probability distribution in this study has a high degree of credibility.

[0166] Step 7: Conclusion

[0167] Through the verification of the experimental results, it can be considered that by fusing the microblog flood disaster information and remote sensing rainfall products during the disaster, the optical remote sensing images in the short term after the disaster, and the DEM elevation data of the study area, the flood inundation probability distribution that conforms to the actual situation can be deduced. This indicates that by using real and verified social media data and making full use of the advantages of the large volume and timeliness of social media data, the deficiency of the lack of measured data such as remote sensing under extreme conditions during the disaster can be made up for.

[0168] Although the present disclosure is disclosed as above, the protection scope of the present disclosure is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present disclosure, and these changes and modifications will all fall within the protection scope of the present invention.

Claims

1. A method for estimating flood inundation extent based on multi-source data, characterized in that, The estimation method includes: Step S 100 : Obtain a multi-source flood database for a specific flood event through the acquisition and preprocessing of multi-source data; The multi-source flood database includes in-disaster social media data, CHIRPS remote sensing rainfall products, short-term post-disaster optical remote sensing images, and DEM elevation data of the study area; Step S 200 : Estimating the flood inundation probability distribution through multi-source data fusion, and the multi-source data fusion for estimating the flood inundation probability distribution includes the following sub-steps: Step S 201 : Create a flooded probability exponential distribution layer based on each Weibo flood point; For each Weibo flood inundation point, combining the inundation water depth data of the point and the DEM elevation data of the study area, introducing inverse distance weighting, and creating an inundation probability index distribution layer centered on the inundation point; Step S 202 : Comprehensively weight all generated inundation probability exponential distribution layers, and introduce a weighting method based on a Gaussian surface; Integrate all the generated inundation probability index distribution layers. The weight of each layer is obtained by comprehensively weighting the MNDWI index calculated from the pixels of the optical remote sensing image within a certain range around the inundation point and the rainfall value around the corresponding point of the CHIRPS remote sensing rainfall product. A more smoothed Gaussian surface-based weighting method is introduced; Step S 203 : Weightedly synthesize n flood inundation probability distribution maps based on n Weibo points according to the weight values to obtain a comprehensive flood inundation probability distribution map of the study area based on all Weibo inundation points; Step S 300 : Verify the estimation results of the flood inundation probability distribution in combination with the actual inundation situation.

2. The method for estimating flood inundation range based on multi-source data according to claim 1, wherein In step S 100 the acquisition of the multi-source data includes in-disaster social media data and remote sensing data, where: The in-disaster social media data includes Weibo data, and the acquisition method of the Weibo data includes: Obtain the Weibo data through a combination of web crawler and API. Among them, the fields of the Weibo data include release time, text content, pictures, video content, and geographical location; The preprocessing of the in-disaster social media data includes data cleaning, duplicate removal, Chinese word segmentation, and stop word removal; The remote sensing data includes optical remote sensing images covering the flood disaster area in the short term after the flood occurs.

3. The method for estimating flood inundation range based on multi-source data according to claim 1, characterized in that, In step S 100 , the screening conditions for the social media data during the disaster include: Condition 1: The in-disaster social media data has accurate geographical location information; and / or Condition 2: The text of the in-disaster social media data contains detailed and determined inundation data; and / or Condition 3: The in-disaster social media data has pictures or videos that can clearly reflect the real inundation situation for data authenticity and accuracy verification; and / or Condition 4: Verify the authenticity and accuracy of the inundation water depth according to the official inundation point data, Baidu Map confirmed location data, text content, pictures, and video information; and / or Condition 5: For the situation where the study area is relatively large, it is necessary to summarize the duplicate data of the same township or street into one point and delete the duplicate point data.

4. The method for estimating flood inundation range based on multi-source data according to claim 1, characterized in that, In step S 100 , the preprocessing of the short-term optical remote sensing images after the disaster includes radiometric correction, atmospheric correction, and resampling.

5. The method for estimating flood inundation range based on multi-source data according to claim 1, wherein In step S 100 , the selected DEM elevation data is the DEM data with a resolution of 30m officially released by NASA.

6. The method for estimating flood inundation range based on multi-source data according to claim 1, wherein In step S 100 , the selected CHIRPS remote sensing rainfall product is CHIRPS data with a spatial resolution of 0.05° and a temporal resolution of daily scale.

7. The method for estimating flood inundation range based on multi-source data according to claim 1, characterized in that In step S 201 , the inundation probability distribution of the entire study area is calculated based on each Weibo inundation point to form an inundation probability distribution map, and the calculation method is based on the following criteria: Criterion 1: The verified flood inundation points published on the Weibo platform reflect the flood inundation facts in a certain range of areas. The closer the area is to the inundation point, the higher the possibility of flood inundation; on the contrary, the farther the area is from the inundation point, the lower the possibility of inundation; Criterion 2: Within a certain range around the Weibo inundation point, the lower the terrain, the greater the probability of being inundated; on the contrary, the higher the terrain, the smaller the probability of being inundated. The terrain height is mainly combined with the DEM elevation data and the Weibo inundation water depth data.

8. The method for estimating flood inundation range based on multi-source data according to claim 7, characterized in that, In step S 201 The calculation of the inundation probability distribution based on a single Weibo point is as follows: In the three-dimensional space within the study area, let the coordinates of the Weibo inundation point be i(x i , y i ), and any other point within the study area be represented as j(x j , y j ). In the DEM distribution, the elevation of point i is H i , the elevation of point j is H j , and the inundation depth of the inundation point i is H w . Then the inundation probability of point j where H ij represents the submergence depth at point j and is calculated as follows: Among them, D ij represents the Euclidean distance between point i and point j Among them, the exponents α and β are weight parameters used to adjust the influence intensity of H ij and D ij on the result.

9. The method for estimating flood inundation range based on multi-source data according to claim 1, wherein In step S 202 , the research criteria for calculating the weights of the MNDWI soil moisture index and CHIRPS rainfall include: Criterion 3: The soil moisture content within a certain range around the Weibo inundation data point can, to a certain extent, reflect the flood inundation probability of this area; the higher the MNDWI water content, the higher the possibility that this area has been covered by floods; the higher the overall soil moisture around the Weibo point, the higher the credibility of the flood inundation probability distribution calculated from this point; Criterion 4: The in-disaster rainfall data in the area around the Weibo inundation data point can, to a certain extent, reflect the situation of these areas being flooded; when the rainfall exceeds a certain range, the greater the rainfall, the greater the probability of these areas being flooded; Criterion 5: Around the microblog flooding data point, the closer the MNDWI value and rainfall value of the pixel to point i, the greater the impact on the overall flooding probability of point i; on the contrary, the farther away from the point, the lower the impact on the overall flooding probability, which is in line with the first law of geography.

10. The method for estimating flood inundation range based on multi-source data according to claim 9, wherein, In step S 203 The calculation process of the comprehensive flood inundation probability distribution map of the study area based on all Weibo inundation points is as follows: Assume that the joint probability density of two-dimensional continuous random variables (x, y) is: Among them, μ1, μ2 represent the means of variables x and y respectively, σ1, σ2 represent the standard deviations of variables x and y respectively, μ1, μ2, σ1, σ2, ρ are all constants, and σ1>0, σ2>0, |ρ|<1, Then (x, y) is said to obey a two-dimensional normal distribution with parameters μ1, μ2, σ1, σ2, ρ. The density function of the two-dimensional normal distribution presents an inverted bell-shaped distribution in three-dimensional space and satisfies: The probability density function of two-dimensional Gaussian normal distribution within a certain range from microblog point i is used to define the influence of MNDWI values and rainfall values of pixels at different distances on point i. Taking microblog point i as the origin, in the three-dimensional geographic space, the two-dimensional Gaussian surface is symmetrically distributed along the x and y axes with point i as the center, with mean μ1=μ2=0, σ1=σ2, and taking the constant ρ as ρ=0, the joint probability density function is generally simplified to the following formula: where λ = σ1 2 = σ2 2 is the parameter of the probability density function, which is used to control the range of the region of interest included in the calculation around point i.

Images