Sub-Annual To Annual Dynamics Of Alaskan Ice-Marginal Lakes From Automated Image Classification Using Google Earth Engine

• Main Authors: Hengst, Anthony Matthew, NC DOCKS at Appalachian State University
• Language: English
• Published: 2020
• Subjects: Glacial Lake; Marginal Lake; glacier; glaciers; Alaska
• Online Access: http://libres.uncg.edu/ir/asu/f/Hengst_Anthony_2020_Honors GES Thesis.pdf

Ice-marginal lakes play an important role in glacier dynamics and downstream hydrology. Proglacial lakes may alter glacial mass loss by enabling submarine melt and by providing a body of water into which glaciers may calve, and provide a basin which traps glacial sediment. Ice-dammed lakes play a critical role in the generation of outburst floods and must be monitored for human safety in downstream environments. Observation of ice-marginal lakes from satellite imagery provides valuable insight into these remote systems because in-situ data are difficult to obtain over a large study area. However, even large-scale remote sensing of these lakes is difficult due to their varied spectral appearance and the complex interface between sediment-laden, iceberg filled lakes and their adjacent crevassed and water-covered glaciers. Previous remote sensing studies feature coarse temporal sampling of lake behavior over a multi-decadal timescale. We seek to investigate how ice-marginal lakes evolve over sub-annual to annual timescales. Ice-marginal lakes are intimately connected to glacial systems, which can vary over seasonal cycles and longer-term cycles in the case of some surging glaciers. We develop a robust remote sensing method to provide observations of ice-marginal lakes across Alaska, a region whose ice-marginal lakes have received comparatively little attention.We develop an automated routine implemented in Google Earth Engine to investigate short- term glacial lake area changes across southern Alaska over the Landsat 8 era (2013-present). We create monthly estimates of ice-marginal lake area by applying a supervised Mahalanobis minimum-distance land cover classifier to Landsat 8 imagery. We optimize image processing parameters by running a suite of classifications and selecting the parameters that minimize error against a set of manually-delineated lakes and achieve an F-score from 0.33 in the most challenging test regions to 0.77 at best. In an exploration using Monte Carlo simulations, we interrogate our data to ...