Atmospheric Downscaling for Glaciated mountain environments (DoG)

Understanding and quantifying the response of mountain glaciers to changing climate and weather conditions has been a pressing task for climate scientists and glaciologists all over the world. To date, models that relate glacier mass and dynamics to atmospheric variations are well established. It has remained as a major challenge, however, to adequately describe the atmospheric variability that affects the glaciers - for the past, the present, and the future. The research hypothesis investigated by the proposed project (DoG) is the following: The uncertainty in global and regional glacier simulations can be reduced significantly with the use of the appropriate atmospheric input data sets, and downscaling algorithms. To test this hypothesis, we will systematically apply and rigorously validate a modern downscaling framework for glaciated mountain environments situated in different climatic and geo-environmental settings. To this aim, elements of state-of-art statistical and so-called hybrid downscaling techniques will be combined and extended to all atmospheric target variables required in process-based glacier mass balance models. The variables will be targeted at high spatial and temporal resolutions, under consideration of site- and predictand variable interdependencies. All available global reanalysis data sets with relevance for glacier mass modelling will be tested as atmospheric predictors, and a regionally differentiated assessment of their skill will be performed. The transfer of the downscaling framework to glaciated mountain sites in a diversity of topographic and climate settings will be cross-validated based on available in situ observations from stations near to- and on the glaciers. DoG will present an immediate step towards quantifying and reducing the uncertainty introduced by reanalysis data as input in process-based glacier mass balance models. Beyond glaciology, the scale mismatch between atmospheric models and the local-scale weather affects a broad range of scientific fields (in particular, weather prediction, alpine ecology, renewable energy from wind and hydro power, hydrology, water resources management, and geomorphology). Overall, the project will provide the basis for an integral assessment of the uncertainty in general circulation models (and related products) for glaciated mountain regions, which still represent one of the major difficulties in the field of atmospheric modeling to date.

Accurate information about weather and climate is crucial for many aspects of life, including water availability, energy production and consumption, health care, hazard protection, farming, forestry, traffic, tourism, ecology, amongst others. The output by global climate models is generally impratical for immediate applications, because global climate models are limited in their spatial resolution and therefore lack important regional features. Downscaling methods aim at producing regional- to local-scale weather and climate data by combining output from global climate models with detailed information about local-scale conditions. For mountain regions, downscaling methods are of a particular importance; firstly, because global climate models have general difficulties with representing the atmosphere above mountains, and secondly, because atmospheric measurements - needed for the training and evaluation of the global climate models - are very limited in mountain regions. The DoG project investigated two different types of downscaling: statistical downscaling and intermediate complexity downscaling, for different glacierized mountain regions around the globe. Statistical downscaling empirically corrects global climate model output towards local-scale observations, whereas intermediate complexity downscaling applies simplified versions of the atmospheric equations of motion and of mass and energy conservation at high resolutions for a limited domain of interest within the coarser global climate model grid. Intermediate complexity downscaling has the advantage to be computationally less expensive than regional climate modeling, another, common type of downscaling, which applies the full (not simplified) set of the atmospheric equations. The DoG project team refined the statistical downscaling code sDoG and the Intermediate Complexity Atmospheric Research Model ICAR for applications in glacierized mountain regions, following the principle: "as simple as possible and as complex as necessary". The investigation revealed common pitfalls for obtaining correct results, but for the wrong reasons (i.e., seemingly realistic results but due to model artifacts or overfit), and strategies were elaborated how to best circumvent these pitfalls. DoG project output includes both concrete adjustments and extensions of the sDoG and ICAR model codes to allow for appropriate applications in data-scarce environments such as mountain regions, as well as the development of new and profound evaluation strategies for atmospheric models in general. The outcomes demonstrated the importance of an active interplay between the different types of downscaling for analytical model understanding and thus the development of robust information about atmospheric processes on a local scale. Due to the given complexity of atmospheric models and their actual relevance, revisiting the current practices of model validation from the perspective of computational epistemology represents an essential avenue for future research.