How much cryosphere model complexity is just right? Exploration using the conceptual cryosphere hydrology framework
Abstract. Making meaningful projections of the impacts that possible future climates would have on water resources in mountain regions requires understanding how cryosphere hydrology model performance changes under altered climate conditions and when the model is applied to ungaged catchments. Further, if we are to develop better models, we must understand which specific process representations limit model performance. This article presents a modeling tool, named the Conceptual Cryosphere Hydrology Framework (CCHF), that enables implementing and evaluating a wide range of cryosphere modeling hypotheses. The CCHF represents cryosphere hydrology systems using a set of coupled process modules that allows easily interchanging individual module representations and includes analysis tools to evaluate model outputs. CCHF version 1 (Mosier, 2016) implements model formulations that require only precipitation and temperature as climate inputs – for example variations on simple degree-index (SDI) or enhanced temperature index (ETI) formulations – because these model structures are often applied in data-sparse mountain regions, and perform relatively well over short periods, but their calibration is known to change based on climate and geography. Using CCHF, we implement seven existing and novel models, including one existing SDI model, two existing ETI models, and four novel models that utilize a combination of existing and novel module representations. The novel module representations include a heat transfer formulation with net longwave radiation and a snowpack internal energy formulation that uses an approximation of the cold content. We assess the models for the Gulkana and Wolverine glaciated watersheds in Alaska, which have markedly different climates and contain long-term US Geological Survey benchmark glaciers. Overall we find that the best performing models are those that are more physically consistent and representative, but no single model performs best for all of our model evaluation criteria.