Passive microwave sensors onboard satellites can provide global snow water equivalent (SWE) observations day or night, even under cloudy conditions. However, there are both systematic (bias) and random errors associated with the passive microwave measurements. While these errors are well known, they have thus far not been adequately quantified. In this study, unbiased SWE maps, random error maps and systematic error maps of Eurasia for the 1990-1991 snow season (November-April) have been examined. Dense vegetation, especially in the taiga region, and large snow crystals (>0.3 mm in radius), found in areas where the temperature/vapor gradients are greatest, (in the taiga and tundra regions) are the major source of systematic error. Assumptions about how snow crystals evolve with the progression of the season also contribute to the errors. In general, while random errors for North America and Eurasia are comparable, systematic errors are not as great for Eurasia as those observed for North America. Understanding remote sensing retrieval errors is important for correct interpretation of observations, and successful assimilation of observations into numerical models.
This paper describes an approach to estimate global snow cover using satellite passive microwave data. Snow cover is detected using the high frequency scattering signal from natural microwave radiation, which is observed by passive microwave instruments. Developed for the retrieval of global snow depth and snow water equivalent using Advanced Microwave Scanning Radiometer EOS (AMSR-E), the algorithm uses passive microwave radiation along with a microwave emission model and a snow grain growth model to estimate snow depth. The microwave emission model is based on the Dense Media Radiative Transfer (DMRT) model that uses the quasi-crystalline approach and sticky particle theory to predict the brightness temperature from a single layered snowpack. The grain growth model is a generic single layer model based on an empirical approach to predict snow grain size evolution with time. Gridding to the 25 km EASE-grid projection, a daily record of Special Sensor Microwave Imager (SSM/I) snow depth estimates was generated for December 2000 to March 2001. The estimates are tested using ground measurements from two continental-scale river catchments (Nelson River and the Ob River in Russia). This regional-scale testing of the algorithm shows that for passive microwave estimates, the average daily snow depth retrieval standard error between estimated and measured snow depths ranges from 0 cm to 40 cm of point observations. Bias characteristics are different for each basin. A fraction of the error is related to uncertainties about the grain growth initialization states and uncertainties about grain size changes through the winter season that directly affect the parameterization of the snow depth estimation in the DMRT model. Also, the algorithm does not include a correction for forest cover and this effect is clearly observed in the retrieval. Finally, error is also related to scale differences between in situ ground measurements and area-integrated satellite estimates. With AMSR-E data, improvements to snow depth and water equivalent estimates are expected since AMSR-E will have twice the spatial resolution of the SSM/I and will be able to characterize better the subnivean snow environment from an expanded range of microwave frequencies.