结合DEM与淹没频率的水库水体动态遥感提取优化方法

由于云、雪及山体阴影造成的无效地表像元影响，利用遥感影像难以获取高精度的水库水体信息动态监测结果.论文在大津法的基础上，提出了结合DEM和淹没频率约束条件对水库遥感提取结果进行优化的方法(DF优化方法).基于Landsat系列卫星影像数据，使用改进的归一化差分水体指数(MNDWI)，通过大津法初步提取水库水体信息.结合多期提取结果综合得到的淹没频率以及当期有效水体像元的淹没频率特征，填补了受云、雪等因素影响的无效像元.在此基础上综合高程及水体淹没频率信息，剔除了山体阴影及其他偶发性因素的影响.遴选全球九个不同类型水库验证了DF优化方法的有效性，结果显示DF优化方法能有效减小水库水体的提取误差.与全球水体数据集JRC MWH的直接修正结果相比，DF优化方法总体分类精度提高了13%，达到了95.67%.同时，结合同步卫星测高数据构建的水位－水面积关系模型，进一步验证了DF水面积提取结果的准确性.相较于JRC MWH的直接修正结果，DF优化方法提取结果构建的水位－水面积关系模型的RMSE降低了30%.此外，该方法可基于GEE平台实现，能快速完成长时序的水体动态提取，为全球大尺度范围内的水库相关研究提供基础.

Reservoir water bodies extraction optimization approach based on DEM and inundation frequency

Due to the invalid surface image elements caused by clouds and mountain shadows， it's difficult to obtain high-precision dynamic monitoring result of reservoir water bodies using remote sensing images. The Otsu method， combined with DEM and inundation frequency as constraints， is introduced as DF optimization approach to solve the problem. The modified normalized difference water index (MNDWI) is calculated from USGS Landsat 5/7/8 Level 2， Collection 2， Tier 1 satellite image datasets. Then the Otsu method is used to extract the preliminary inundation results of the reservoir water bodies. The mountain shadows or other independent patches far away from the reservoirs are removed through DEM threshold and inundation frequency constraints. The globally distributed nine reservoirs are used to verify the optimization approach. The results show that the optimization approach is able to effectively eliminate the reservoir water body misclassification. The overall extraction accuracy， is significantly improved from 82.67% of the results directly derived from the JRC MWH datasets to 95.67% of our DF approach. The average of water area root mean square errors (RMSE) of nine reservoirs is reduced by 30.55% compared with the automatic correction results by using the simulated area from the linear level-area regression model. The DF optimization approach would easily be used in the GEE platform to generate the long-term time reservoir water body dynamics， providing continuous reservoir surface area data as the fundaments to the reservoir-related study on national or global scales.