Factors Affecting the Uncertainty in Remotely Sensed Snow Water Equivalent

J. Dong
NASA’s Goddard Space Flight Center
Hydrological Sciences Branch
Greenbelt, MD 20771 USA
University of Maryland, Baltimore County
Goddard Earth Science and Technology Center (GEST)

J.P. Walker
Department of Civil and Environmental Engineering
University of Melbourne
Parkville, Victoria, 3010 Australia

P.R. Houser
NASA’s Goddard Space Flight Center
Hydrological Sciences Branch
Greenbelt, MD 20771 USA

J.L. Foster
NASA’s Goddard Space Flight Center
Hydrological Sciences Branch
Greenbelt, MD 20771 USA

R.E.J. Kelly
NASA’s Goddard Space Flight Center
Hydrological Sciences Branch
Greenbelt, MD 20771 USA
University of Maryland, Baltimore County
Goddard Earth Science and Technology Center (GEST)

Abstract
Recent passive microwave remote sensing snow water equivalent (SWE) products have considered what they believe to be the most important factors affecting bias and uncertainty. For example, a recent semi-empirical algorithm for SWE estimation has included systematic and random error contributions from environmental factors such as forest cover and snow morphology (crystal size – a function of location and time of year). However, climate and land surface complexities have led to an incomplete consideration of these and unrealized sources of uncertainty, which contribute to significant systematic and random error in the remotely sensed SWE estimates. Joint analysis of independent meteorological records, ground SWE measurements, remotely sensed SWE estimates, and land surface information datasets has provided a unique look at the error structure of these new satellite products. The factors considered were the snow pack mass itself, distance to significant open water bodies, liquid water in the snow pack and/or morphology change due to melt and refreeze, forest cover, snow class, and topographic factors such as large scale root mean square roughness and dominant aspect.

Analysis of the nine-year Scanning Multichannel Microwave Radiometer
(SMMR) SWE data set was undertaken for Canada, due to the intensive number of in-situ measurements available in that region. It was found that the remote sensing product was unbiased with an average root mean square error of approximately 25mm for SWE values less than 100 mm; beyond that the SWE estimate was biased linearly by the snow pack mass and the root mean square error increased to around 150mm. Among the environmental variables assessed, both the distance to open water and average monthly mean air temperature were found to significantly influence the uncertainty. Apart from maritime snow class, which had the greatest snow class affect on root mean square error and bias, all other environmental factors had little affect. Eliminating remote sensing data within 200 km of open water bodies and for monthly mean temperature greater than -2°C, and restricting use to snow packs with less than 100mm SWE makes the remotely sensed product useful for practical applications.

[back]