基于多源时序高分遥感洪水淹没变化检测的水动力模型参数优化率定方法研究

In the 13th Five-Year plan of water conservancy reform and national flood risk thematic map in China, the rapid extraction technology of flood inundation area is an important monitoring requirement. But, the cloud, rain and other factors restrict the water extraction accuracy, and there is a validation problem of hydrodynamic model parameter and simulation inundation area. In this study, four critical issues, which are focus on ground cover information extraction, inundation area change detection during flooding, hydrodynamic parameter optimization and study area application, to formulate an approach for parameter optimization of hydrodynamic model based on multi-source high-resolution remote sensing image flood inundation change detection, are investigated as follow. (1)The proposal formulates an extraction model water area based on the time-spatial-spectrum feature library. (2)The proposal formulates a SAR image segmentation model based on two-dimensional class variance, image active contour features and iteration calculation of boundary pixel gray value difference. (3)The proposal formulates a change detection model based on pixel value different and boundary iteration calculation for multi-source time-series remote sensing image flood inundation area. (4)The proposal formulates a depth neural network model of hydrodynamic parameters optimization based on inundation times, to improve analysis accuracy and trust of real-time inundation area based on hydrodynamic model. From improving the extraction method based on combination multi-source high-resolution remote sensing and hydrodynamic model, the flood inundation area can be rapid extraction and the dynamic inundation process can be simulated by using optimized hydrodynamic model. Based on this, storm flood inundation of Hangjiahu is to be applied and validated in Taihu Lake basin in 2015, to provide data and method support for government real-time flood control and disaster mitigation.

根据我国水利改革"十三五"规划防洪减灾体系建设和洪水风险图编制对淹没范围精准获取的需求，针对洪灾期云、雨影响下淹没范围获取难、水动力模型参数率定精度差和结果验证效率低的难题，本项目以地表信息提取-灾中淹没检测-水动力参数率定-实践验证为主线，研究多源时序高分遥感洪水淹没变化检测来优化率定水动力模型参数，建立地物时空谱特征库水体范围提取、主动轮廓特征和边界像元灰度差异迭代SAR影像水体范围提取模型；建立像元差值和邻域差异迭代多时序遥感洪水淹没变化检测模型；建立基于时序间隔多层循环深度神经网络的水动力模型参数优化率定模型，提高洪水期水动力模型实时淹没分析精度和可信度。本项目将改进多源高分遥感与水动力模型耦合的洪灾期洪水淹没精准监测方法，实现洪水淹没范围精准提取与演进过程淹没范围精准变化分析，并开展2015年太湖流域-杭嘉湖（苏）区暴雨洪水淹没应用与实践，为政府实时防洪减灾提供数据与方法支撑。

`防洪减灾是我国水利改革“十三五”规划的重大需求，高时效、高精度、高可信洪水风险预测评估是防灾减灾的重大任务，也是其“十四五”规划的底线。本基金为了解决洪灾期云、雨影响下淹没范围获取难、水动力模型参数率定精度差和结果验证效率低的难题，研究内容包括：.（1）结合平原区域河网密集且地形复杂洪水淹没特性，利用地物时空谱特性提取地表曼宁糙率系数，建立模拟降雨期水量汇流的预估模型；.（2）收集研究区国产高分辨率资源三号立体像对、激光测高ICESat1/2数据，联合构建区域地表高程模型（DSM），建立区域纳雨能力计算模型，分析近似真实地表情况下降雨量与淹没范围对应关系；.（3）构建主动轮廓特征和边界像元灰度差异迭代SAR影像水体范围提取模型，分析洪灾期洪水边界及淹没范围，评估洪灾风险等级；.（4）结合太湖流域-杭嘉湖（苏）区暴雨洪水淹没应用与实践，验证模型的可行性，并为政府实时防洪减灾提供数据与方法支撑。