Bivariate spatial modeling of snow depth and snow water equivalent extremes in Austria

In many applications like hydrological studies focusing on potential runoff, or climate studies investigating water availability, a spatial representation of snow depth (HS) and snow water equivalent (SWE) and their extremes is of great interest. In Austria, although many locations with long-term HS measurements exist, SWE observations are very rare and mostly short. To provide a spatial extreme value model for SWE in Austria, firstly SWE is modeled stationwise from long-term HS observations, employing a newly developed snow layer model, which derives SWE solely from daily HS, without any other meteorological input. Secondly, with a model selection procedure suitable covariates are selected, for modeling the margins of the GEV distribution. Among mean SWE and topographical parameters, a hidden property of the snowpack, namely the time difference between the occurrence dates of the maximum SWE and HS within a winter season is used. Then, different bivariate max-stable processes (H ¨usler-Reiss, Extremal-Gaussian, Extremal-t) are fitted to Austrian HS and modeled SWE data. As expected, the bivariate Extremal-t max-stable process remains as the most suitable model, allowing for the estimation of conditional return levels and the use in risk analysis due to its spatial extremal dependence structure. First validation results consolidate the bivariate approach against a smooth model and a univariate Extremal-t max-stable model.

Recommended citation: Schellander, H., & Hell, T. (2018). "Bivariate spatial modeling of snow depth and snow water equivalent extremes in Austria." International Snow Science Workshop, 2018, Innsbruck, Austria..
Download Paper