Abstract
Satellite-derived information on fractional snow cover is essential to resource monitoring, hydrologic modeling, and climate change assessment. Evaluating the accuracy of remotely sensed snow-cover products is important but difficult, largely because point-scale surface observations are spatially sparse and generally nonrepresentative of the remote sensor footprint.
In this study, two remotely sensed snow-cover products [the Interactive Multisensor Snow and Ice Mapping System (IMS) and the Moderate Resolution Imaging Spectroradiometer (MODIS) Climate Modeling Grid (CMG), v.3] are evaluated against ground observations from the Cooperative Observing Network and SNOTEL on a daily basis over the continental United States for calendar year 2000. Ground observations are treated as points in space and time; no physical modeling or statistical interpolation is applied. Hypothesis tests based on discrete and continuous distributions are developed to assess agreement between ground observations and the remotely sensed snow-cover products at 0.25° resolution. (The MODIS CMG product was degraded from 0.05° for this study, thus its potential is not fully evaluated.)
As overall snow extent increases in the course of the season, both MODIS and IMS improve in identifying snow-covered areas (fewer errors of omission), but deteriorate in identifying snow-free areas (more errors of commission). The detection of scattered areas of snow is generally better during ablation than during accumulation. Weaknesses of the statistical methods and assumptions are discussed.
This work will help to identify areas for improvement in snow-cover detection algorithms and provides a framework to assess the accuracy of remotely sensed snow cover used as model input and/or confirmation.
Corresponding author address: Dr. Kaye Brubaker, 1173 Glenn L. Martin Hall, University of Maryland, College Park, College Park, MD 20742. Email: klbrubak@eng.umd.edu
