The spatial distribution of snowcover in a catchment is determined by complex interactions between meteorological and physiographical factors, integrated over time. The snowcover shows variability over scales ranging from centimeters up to hundreds of kilometers.

An important and necessary decision for modelers is to determine spatial resolution in a distributed model. Since the spatial variability in snowcover may be quite large, even within a few meters, it is difficult to use modeling units small enough so that the snow can be assumed evenly distributed within the unit. A possible method to compensate for this is to use larger units, and describe the snow distribution within each unit by a statistical model (e.g. normal, log-normal, gamma, etc). This technique requires information about spatial statistical properties of snowcover within a unit.

As many of the distributed hydrological models operate on a grid basis, it would be desirable to find a statistical distribution on a sub-grid scale. However, as an initial approach, the study presented here was done on a catchment scale. The catchment scale presented the possibility of incorporating data from several historical snow surveys. These surveys were taken at the time of maximum snow accumulation in various mountainous catchments in Norway. Comparing empirical distribution functions with different theoretical distribution functions, it was shown that a mixed distribution combining two separate log-normal distributions clearly gave the best fit in most of the catchments. This seems to indicate that a mixture of at least two different populations of SWE values exists.