Exploring the Links Between Biases in Regional Climate Models and their Representation of Synoptic Circulation Types in the European Alps

Abstract:

Climate model simulations can show large departures from observations. These biases are often subtracted and then forgotten, or post-processed for impact modeling using pragmatic methods that typically do not account for the origins of the biases. Yet, enhancing our understanding of the reasons behind these biases is essential for the improvement of climate models and for the design of more robust bias-correction techniques. We pursue these objectives by exploring the links between biases at different spatial scales. We quantify biases at the regional scale in an alpine area (Switzerland) and explore whether they are influenced by the misrepresentation of the frequency, temperature and precipitation of circulation types (CTs) at the synoptic scale. Our analysis relies on simulations from 14 regional climate models (RCMs) forced by general circulation models and produced for the ENSEMBLES project. For each model, we characterized the daily synoptic situation over 1960-2099 by using a CT classification based on a hierarchical cluster analysis of principal components. The classification relies on sea level pressure fields within an area representative of the European Alps. We find significant biases in the CT frequency for 1980-2001 and that some of them lead to biases in temperature and precipitation reported in RCM comparison studies. For instance, in winter, models overestimate both the frequency and precipitation intensity of westerly situations, which carry moist air from the Atlantic Ocean and are responsible for most of the rainfall. We propose this as a main driver for the generalized overestimation of precipitation in winter, which is one of the most concerning biases affecting simulations in our area, as it leads to unrealistic estimates of snow accumulation. A general consequence is that CT-based downscaling methods that do not account for biases in CT frequency will generate biased outputs. Overall, we show that decomposing RCM time series using CTs enables to relate biases to atmospheric processes, flow directions and topographic features and thus to reveal connections normally missed by analyses relying on monthly or seasonal values. We also show that exploring future changes in CTs provides new insights into the robustness of projected changes in precipitation and temperature by the end of the century.