An Automated Method for Surface Ice/Snow Mapping Based on Objects and Pixels from Landsat Imagery in a Mountainous Region

Abstract

Surface ice/snow is a vital resource and is sensitive to climate change in many parts of the world. The accurate and timely measurement of the spatial distribution of ice/snow is critical for managing water resources. Object-oriented and pixel-oriented methods often have some limitations due to the image segmentation scale, the determination of the optimal threshold and background heterogeneity. Therefore, this study proposes a method for automatically extracting large-scale surface ice/snow from Landsat series images, which takes advantage of the combination of image segmentation, the watershed algorithm and a series of ice/snow indices. We tested our novel method in three different regions in the Karakoram Mountains, and the experimental results show that the produced ice/snow map obtained a user’s accuracy greater than 90%, a producer’s accuracy greater than 97%, an overall accuracy greater than 98% and a kappa coefficient greater than 0.93. Comparing the extraction results under segmentation scales of 10, 15, 20 and 25, the user’s accuracy and producer’s accuracy from the proposed method are very similar, which indicates that the proposed method is more reliable and stable for extracting ice/snow objects than the object-oriented method. Due to the different reflectivity values in the near-infrared band in the snow and water categories, the normalized difference forest snow index (NDFSI) is suitable for Landsat TM and ETM+ images. This study can serve as a reliable, scientific reference for rapidly and accurately extracting ice/snow objects.