Influence of topographic and environmental variability on model uncertainty: a case study on snow and ground temperatures in mountain regions

A variety of physically based models to predict and understand the spatio-temporal behaviour of snow and ground temperatures have been developed in recent years. Model evaluation including the analysis of model uncertainty and validation is widely accepted as fundamental in enhancing trust in decisions that are based on model simulations. Due to constraints on resources or lack of distributed validation data, model evaluation is often restricted to one or few locations only, even if the model is applied to make predictions for large spatial areas and time periods. Thus, conclusions about model behaviour entail the tacit assumption that validation at one point can inform decisions about model performance in different environmental conditions. The effect of this assumption on model application and development when modeling phenomena in highly variable terrain or over large distances has rarely been studied. This study is focused on a sensitivity and uncertainty analysis of an energy and mass balance model that simulates snow and ground temperatures. It serves as a case study examining the role of topography and soil on parametric model uncertainty and sensitivity. A sensitivity analysis on individual parameters and a Monte Carlo based uncertainty study are performed at a variety of locations covering the range of topographic and environmental variability typically found in mountain regions. The results indicate that model uncertainties and sensitivities vary strongly under differing environmental conditions. This demonstrates that model evaluation (validation, sensitivity and uncertainty analyses) benefits strongly from the consideration of differing variables and, especially, the environmental variation of their behaviour.