Icy lakes extraction and water-ice classification using Landsat 8 OLI multispectral data
Icy lakes extraction and water-ice classification using Landsat 8 OLI multispectral data
复制标题
DOI：
10.1080/01431161.2018.1447165
复制
发表时间：
2018-01-01
影响因子：
3.4
通讯作者：
Merminod, Bertrand
中科院分区：
文献类型：
--
作者：
Barbieux, Kevin;Charitsi, Antigoni;Merminod, Bertrand
我们提出了一种算法，使用 Landsat 8 的多光谱数据来区分冰湖上的冰和开放水域。首先使用新的辐射指数 ILI（冰湖指数）从图像中提取湖泊，该指数利用了水和冰在短波红外波段反射率的稳定性。与修正归一化水指数 (MNDWI) 或水比指数 (WRI) 等最先进指数的定量比较表明，ILI 比当前指数更好地将混合冰/水体与陆地分离，在手动标记的参考数据下，卡帕系数 (.) 始终达到 0.93 以上。此外，这些结果表明，与其他指标相比，ILI 具有非常高的最佳阈值稳定性。在划定的湖泊区域中，我们使用辐射特性以及纹理特性（例如每个波段的局部标准偏差和平均梯度）通过决策树进行监督分类。这种分类的一个关键特征是水-冰分类指数（WICI），也是基于纹理的，它可以有效地区分浅水和冰。我们通过将分类算法的结果与手动数字化参考数据以及拉多加湖的并发 Sentinel-1 合成孔径雷达 (SAR) 数据进行比较，证明了分类算法的稳健性。在这两种情况下，比较都会导致。范围从 0.84 到 0.97。
We present an algorithm discriminating ice from open water on icy lakes using multispectral data from Landsat 8. Lakes are first extracted from the images using a new radiometric index coined ILI (Icy Lakes Index) which uses the stability of the reflectance of water and ice in the shortwave infrared bands. Quantitative comparisons to state-of-the-art indexes such as the Modified Normalised Difference Water Index (MNDWI) or the Water Ratio Index (WRI) show the ILI separates mixed ice/ water bodies from land better than the current indexes, consistently achieving kappa coefficients (.) above 0.93 with manually labelled reference data. Additionally, these results suggest that the ILI has a very high optimal threshold stability, compared to other indexes. In the delineated lake area, we perform a supervised classification with a decision tree using radiometric properties, but also texture properties such as the local standard deviations and average gradients in each band. A key feature of this classification is the Water-Ice Classification Index (WICI), also texture-based, which discriminates shallow water from ice efficiently. We prove the robustness of our classification algorithm by comparing its results to manually digitised reference data, but also to concurrent Sentinel-1 Synthetic Aperture Radar (SAR) data in the case of Lake Ladoga. In both cases, the comparison leads to. ranging from 0.84 to 0.97.