Enhancing the Detectability of Clouds and Their Shadows in Multitemporal Dryland Landsat Imagery: Extending Fmask
Enhancing the Detectability of Clouds and Their Shadows in Multitemporal Dryland Landsat Imagery: Extending Fmask
复制标题
DOI：
10.1109/lgrs.2015.2390673
复制
发表时间：
2015-06-01
影响因子：
4.8
通讯作者：
Schmidt, Michael
中科院分区：
文献类型：
--
作者：
Frantz, David;Roeder, Achim;Schmidt, Michael
我们开发了一种新的两步方法，用于自动遮蔽陆地卫星图像中的云及其阴影。第一步是利用Fmask算法在每幅Landsat图像中独立地检测云和云阴影。我们修改了原始Fmask的两个功能：我们放弃了阴影匹配的终止标准，我们附加了一个黑暗过滤器，以抵消在bifidly结构的旱地地区的误报。第二步利用第一步的逐场景检测以及云和云阴影概率的附加时间序列。一个像素的所有晴空观测值被用来估计概率的中值和标准差。任何偏离中位数的标准差倍数以上的观测值都被视为离群值，因此是剩余的云或云阴影。这一方法是专门为水有限的旱地地区设计的，在这些地区，基于事件的降水占主导地位。作为一种效应，绿色植被峰值在时间、幅度和频率上高度可变，对常用的基于傅立叶的离群值检测方法具有不利影响。该方法的设计是强大的，即使时间密集的数据覆盖不可用。
We developed a new two-step approach for automated masking of clouds and their shadows in Landsat imagery. The first step consists of detecting clouds and cloud shadows in every Landsat image independently by using the Fmask algorithm. We modified two features of the original Fmask: we dropped the termination criterion for shadow matching, and we appended a darkness filter to counteract false positives in bifidly structured dryland areas. The second step utilizes the scene-by-scene detections of the first step and additional time series of cloud and cloud shadow probabilities. All clear-sky observations of a pixel are used to estimate the probabilities' median and standard deviation. Any observation that deviates more than a multiple of the standard deviation from the median is considered an outlier and thus a remaining cloud or cloud shadow. The method was specifically designed for use in water-limited dryland areas, where event-based precipitation is predominant. As an effect, green vegetation peaks are highly variable, in timing, magnitude, and frequency, with adverse effects on commonly used Fourier-based outlier detection methods. The method is designed to be robust even if temporally dense data coverage is not available.