基于Sentinel-1/2数据的洪水淹没范围提取模型

Research on extraction model of flood inundation range based on Sentinel-1/2 data

遥感是监测洪水淹没范围、掌握洪涝灾情演变的重要手段,而光学影像在洪水发生时往往有较多缺失,全天候的SAR影像在提取水体时精度略低。为快速、精准提取洪水淹没范围,构建了一种综合利用Sentinel-2光学影像和Sentinel-1雷达影像数据的洪水淹没范围提取模型,采用一种自适应阈值分割算法即大津算法(OTSU)分别对两种数据以及该模型进行了水体范围提取试验,并以河北省保定市为例进行了应用分析。结果显示:云量较少的Sentinel-2影像水体提取效果最好,总体精度(OA)达到95.6%;所构建的模型在引入部分可用Sentinel-2数据后,OA达到95%,相比单独使用Sentinel-1数据OA和Kappa系数分别提升1.2%和2.4%。该模型搭载于Google Earth Engine平台,能实现快速、准确、低成本的地表水体空间范围连续输出,不受限于云雾且比单独使用Sentinel-1影像的提取精度更高,在云覆盖严重导致Sentinel-2数据缺少的情况下,该模型可作为洪水淹没范围提取方法的一种选择。

Remote sensing is an important means of monitoring the extent of flood inundation and understanding the evolution of flood disasters.However,optical images often have many deficiencies during floods,and all-weather SAR images have slightly lower accuracy in extracting water bodies.A flood inundation range extraction model based on Sentinel-2 optical images and Sentinel-1 radar image data was constructed to extract the flood inundation range quickly and accurately.An adaptive threshold segmentation algorithm,the Otsu algorithm,was used to extract the water body range of two types of data and the proposed model,and an application analysis was conducted using Baoding City,Hebei Province as an example.The results showed that Sentinel-2 images with less cloud cover had the best water extraction effect,with an overall accuracy(OA)of 95.6%.After introducing some available Sentinel-2 data,the OA of the constructed model reached 95.0%,OA and Kappa coefficient were increased by 1.2%and 2.4%respectively compared to using Sentinel-1 data alone.This model is installed on the Google Earth Engine platform and can achieve fast,accurate,and low-cost continuous output of the spatial range of surface water bodies.Clouds and mist do not limit it and have higher extraction accuracy than Sentinel-1 images alone.In the case of severe cloud coverage leading to a lack of Sentinel-2 data,this model can be used as an alternative method for extracting flood inundation areas.