Mapping glacial lakes and their changes using cloud processing of optical satellite images

Glacial lakes are an important component of terrestrial meltwater storage and respond to climate change and glacier retreat. Although there is evidence of rapid worldwide growth of glacial lakes, changes in frequency and magnitude of glacier lake outburst floods (GLOFs) under climatic changes are not yet understood. We refine existing methods of water mapping, based on optical remote sensing images, that can be applicable for the mapping of glacial lakes. We propose and discuss methods for regional-scale glacial lake mapping and GLOF detection using large time series of optical satellite images and the cloud processing tool Google Earth Engine in a semi-automatic way. The methods are presented for various temporal scales, from the 2-week Landsat revisit period to annual resolution. The proposed methods show how constructing an annual composite of pixel values such as minimum or maximum values can help to overcome traditional problems associated with water mapping from optical satellite data like clouds, or terrain and cloud shadows. For annual-resolution glacial lake mapping, our method set only involves two different band ratios based on multispectral satellite images. The elevation range, computed from a digital elevation model is used on the postprocessing step to filter out noise associated with image quality and misclassifications. The study demonstrates various examples of how the proposed methods can be applied to detect GLOFs and to produce a complete regional-scale glacial lake inventory, using the Greater Caucasus as an example. We also discuss limitations of the approaches, finding that for applications where reliable detailed maps are required, visual revision of the results is still recommended owing to the often small size of glacial lakes, combined with their large variability in, for example, topographic setting, turbidity, depth, or temporal occurrence.