水体遥感指数研究进展

Development of remote sensing water indices:a review

水体遥感指数是当前全球和大区域水体分布和面积估算采用的主要技术.文中对近20多年来提出的水体遥感指数进行总结,分析指数构建原理并比较两种主要指数类型(比值型和差值型)的区别,讨论水体遥感指数应用中的几个主要问题.总的看来,比值型指数对水和非水信息的分离度较强,并能抵消地形的影响,因此可以较好地去除云和阴影的影响.加上比值型指数的变量通常不带系数,因此普适性较强.现有的水体指数中还鲜有被证明可以不用选取阈值就能高精度提取水体的指数.但好的水体指数阈值选取简单,稳定性较强.常用的OTSU法(大律法)获得的阈值已被许多研究证明并不可靠,因此不宜基于OTSU获取的阈值进行水体信息提取及其方法对比.

The remote sensing water index is distinguished by its easy and fast calculation as it generally does not require complicated training.At present,global and large regional water distribution and area estimation are mainly based on the remote sensing water index.This paper summarizes and compares the remote sensing water indices developed in the past 20 years,analyzes the principle of the water index construction,compares the differences between two main water index types(ratio-based and differencing-based),and discusses several commonly concerned issues in the application of remote sensing water indices.In general,the ratio-based water index can enlarge the difference between water and non-water information through a ratio operation,has a strong spectral separation degree,and can offset the influence of terrain.Therefore,it can better remove the influence of cloud and slope shadows.In addition,the ratio-based water indices generally have strong robustness and stable water extraction effect as their variables usually do not have sensor-and/or regional-specific coefficients.Water information cannot be extracted by a water index without a proper threshold value.A good index should have strong stability in threshold values that are easily selected.However,in terms of the existing indices,none of them have been proved by a third party to be able to extract water body with high accuracy without selecting thresholds.The optimal threshold obtained by the commonly-used OTSU method has been shown by many studies to be unreliable.Therefore,water information extraction and the related method comparison should not be performed based on the thresholds obtained by the OTSU method.