Distinguishing snow and glacier ice melt in High Asia using MODIS
Abstract:In High Mountain Asia, snow and glacier ice both contribute to streamflow, but few in-situ observations exist that can help distinguish between the two melt components. We utilize a suite of satellite based MODIS-derived datasets to distinguish three surface types as they change daily: 1) exposed glacier ice, 2) snow over ice and 3) snow over land. The MODIS products include fractional snow cover from MODSCAG and permanent ice and snow from MODICE, both at 500 m resolution, that are used jointly with albedo or grain size. The method provides a means to systematically analyze the cycle of snow and glacier ice over large regional extents. We compare the time series of these surfaces for sub-basins of the Upper Indus Basin and characterize the variability over the MODIS record. We use the Randolph Glacier Inventory to categorize by glacier size within the sub-basins and analyze small, medium, and large glaciers to characterize their variability and investigate changes to the cryosphere at different scales.
In addition to analyses of the surface conditions, we use the surface classification to understand the source of melt volumes from glacier ice and seasonal snow cover. We model snow and ice melt in the Hunza and Gilgit sub-basins of the Upper Indus basin. We apply two melt models, a temperature index model and an energy balance model. For our temperature index model, we use lapse rates derived from ERA-Interim to downscale temperatures to 500 m and aggregate by elevation bands. Our spatially-distributed energy-balance model requires solar and longwave radiation, temperature, and wind data; we downscale to 500 m from GLDAS NOAH surface simulations. We compare results from the two models with measured streamflow, and evaluate the model computation times, accuracies and ease of diagnosing output errors. We include comparisons of model results using different remote sensing products (MCD43, MOD10A1, MODSCAG, MODDRFS) to partition surface types. Uncertainty is estimated through the use of multiple models and the different remote sensing products used in the surface classification.