Ocean Circulation and Climate Dynamics

Projections and predictions of Local Precipitation Intensities. Advanced Downscaling using Extreme Value Statistics (PLEIADES)

General information

Precipitation extremes are ma jor natural hazards. Robust knowledge about theirlong term future changes and predictions for the next one or two decades of their expected magnitudes are crucial for societies. Along with an assessment of related uncertainties, this knowledge is required on local scales, where impacts of extreme precipitation are experienced and adaptation measures need to be implemented. Climate change scenarios based on general circulation models (GCMs) currently do not provide reliable information on scales below about 200 km. Hence, to reliably assess hydrological impacts of climate change, downscaling of GCM scenarios is required. Three approaches to downscaling exist: in dynamical downscaling, a high-resolution regional climate model (RCM) is nested over a limited area into the GCM. In perfect prognosis (PP) statistical downscaling, statistical links between observed large-scale and local-scale weather are in-ferred and applied to GCM output. In Model Output Statistics (MOS), a correction function for RCM output is derived based on observational data. The following problems currently limit the downscaling of extreme precipitation (Maraun et al., 2010c):

RCMs(1) in general underestimate the magnitude of precipitation extremes, (2) overestimate the spatial extend of localised summer precipitation extremes; (3) simulate biased localisation of precipitation especially in mountaineous regions; (4) are only available for some regions of the Earth.

PP statistical downscaling(1) in its standard variant underestimates the variability in magnitude as well as spatial variability. It is therefore unsuitable to model localised extreme events; (2) in a modern framework of Generalised Linear Models accounts for variability of magnitudes and spatial variability, but does not account for extreme events by applying extreme value statistics; (3) neglects atmospheric physics at mesoscales and thus an important aspect of the climate system’s complexity.

MOS(1) does not explicitly account for extremes. In its standard variant (2) corrects only the overall distribution of precipitation over the whole calibration period, without accounting for different biases for different weather situations; (3) simply corrects the magnitude of simulated precipitation locally and thus inherits bias in spatial extent and event location from the RCM (local MOS); (4) relies on RCM input, and therefore can only be applied in regions for which RCM simulations exist.

For both statistical approaches the modelling of spatial patterns of extremes poses a problem. The generic models for spatial extremes (max-stable processes, e.g. Smith, 1990) address only events where precipita- tion at all locations is extreme. This makes these models unsuitable to simulate realistic rainfall.