On the use of the ESRI image service for mapping Little Ice Age glacier extents

Glacier extents are mainly mapped by a semi-automated classification of multispectral satellite images (e.g. Landsat, Sentinel-2) with manual corrections of unmapped regions (e.g. ice in cast shadow or under debris cover). The quality of such corrections improve towards higher spatial resolution sensors, but such data were so far only seldom available for direct digitizing in a GIS. With the increasing availability of web map services (wms) such as the ESRI image service or national services the situation has strongly changed and first studies already analysed the potential of such services in geoscience.

The ESRI wms can be embedded into the professional mapping environment of ArcMap or QGIS. It provides mostly cloud and snow free mosaics of very high-resolution (0.31 - 0.5 m) GeoEye and Worldview images up to a scale of 1:5000. The images can be shown in the background as an information layer, but not further processed. The user has no control over the images provided (e.g. their acquisition date) or how they are mosaiced and orthorectified, locally resulting in snow covered or shifted images. The acquisition date and sensor used for each image part can be extracted using the information tool. Due to its recent availability, the ESRI wms has not yet been widely used and its huge potential especially for geomorphological and paleoglaciological mapping has still to be explored.

In this study, which is performed in the framework of the EU Horizon 2020 project PROTECT (protect-slr.eu) we present (1) a workflow for mapping Little Ice Age (LIA) glacier extents using the ESRI wms, (2) a detailed uncertainty analysis and (3) first results of glacier area changes since the LIA for selected regions in Alaska, Baffin Island, Novaya Zemlya and the tropics. Additionally to the ESRI wms, we used Sentinel-2 images, the ArcticDEM and modern glacier outlines from the Randolph Glacier Inventory (RGI). Geomorphological indicators (trim lines, moraines, vegetation free zones) and glaciological considerations were considered to guide the digitizing. Geolocation uncertainties were determined against independent data sources and the interpretation and reproduction uncertainties were quantified by multiple digitising experiments. The possible timing of the former LIA maximum extents was obtained to the extent possible from the literature, but here large uncertainties remain.

In total, outlines for 371 LIA glaciers were created and compared to today relative area changes of -20%, -15%, -26% and -58% were found for Alaska, Baffin Island, Novaya Zemlya and the tropics, respectively. Reproduction uncertainties were calculated for a sample of 18 glaciers to be on average 1.4 ±1.3%, interpretation uncertainties for a sample of 17 glaciers 1.9 ±10%. The digitization of LIA glacier extents with 10 m Sentinel-2 images is only rarely possible due to the difficulties identifying small scale moraines and resulted in much higher . We conclude that wms such as the ESRI World imagery layer provide, despite their shortcomings, an excellent opportunity to precisely map LIA maximum extents of glaciers around the world.