Abstract ID: 3.10005 | Accepted as Talk | Talk | TBA | TBA

From hydrological impact studies and crop monitoring in agronomy to paleoenvironments reconstructions, many studies from various fields require high spatial resolution climate data (< 5 km to subkilometric scales). Although in situ measurements produce reliable estimations of physical parameters locally, limitations regarding their spatial and temporal coverage often leads to using data derived from models. Among these, Global Climate Models (GCM) provide the most comprehensive representation of the climate system and the interactions between its components. Due to high computational costs, the horizontal resolution of such models yet remains restricted to 50-100 km, when multi-decennial or longer simulations are required. In order to refine GCM outputs, two main downscaling approach have been developed over the years. Firstly, dynamical downscaling techniques explicitly resolves atmospheric physics and dynamics at fine scale but generally involves elevated computational costs, limiting domain extent and their usage over long periods of time. Secondly, statistical downscaling does not physically represent the climate, but attempts instead to identify statistical relationships between coarse resolution and local variables. These cost-effective methods are however dependant on the spatial repartition and quality of observations used for calibration, hampering their use in regions with sparse station coverage. Additionally, statistical downscaling relies on a strong hypothesis of stationarity, supposing that the relationship built on the observation period remains valid through time. Although it cannot be tested explicitly, this assumption may not stay valid under very different climate conditions. Less common, an alternative approach focuses specifically on the interactions between regional atmospheric circulation and high-resolution topography which constitutes a major driver of the local climate, producing strong variability over short distances. Since existing topographic downscaling methods did not meet our requirements, we developed a physically based model, adapted to long-term simulations (multi-millennia) at fine spatial scale and taking into account local terrain characteristics derived from Digital Elevation Models and large scale climate signal. The method allows to downscale temperatures and precipitation in mountainous areas and requires limited inputs as well as low computing resources.