GIS-based modelling of spatial pattern of snow cover duration in an alpine area

Snow cover duration patterns of an alpine hillslope (approximately 2 km2) were derived using daily terrestrial photographic remote sensing. We have developed a suite of quantitative models in order to investigate the relative controls of topographic factors, the degree of non-linearity, the effect of seasonal differences and a possible influence of further systematic influences. We have only used data that are relatively easily available to ensure applicability beyond the site. Elevation, slope angle and aspect, and potential irradiation for the winter period can be directly derived from a digital elevation model. The number of days with temperature ≤0°C was included using a regression with elevation. Furthermore, a coarse vegetation classification (forested/not forested) was included. To estimate the necessary degree of non-linearity for such modelling without forming exact assumption about the functional interrelations, results from a linear regression analysis are compared with an artificial neural network (ANN). The results show that a R2 of 71% can be achieved by means of a linear approach, whereas a non-linear approach (ANN) leads to 81%. An indirect estimation demonstrates that a further 6% can be explained without considering data on annually specific weather conditions. The analysis of the residuals shows a clear spatial pattern. This indicates that additional spatial variables may allow a further improvement of the model.