final_program_abstracts[1]

11 IMSC Session Program
Downscaling of GCM-simulated precipitation using Model
Output Statistics
Tuesday - Parallel Session 5
Martin Widmann and Jonathan M. Eden
School of Geography, Earth and Environmental Sciences, University of Birmingham,
UK
Regional or local-scale precipitation changes can not be directly inferred from
precipitation simulated by General Circulation Models (GCMs) because of the limited
spatial resolution of GCMs, but also because of systematic errors even on the resolved
scales. One possibility to overcome the problem is to estimate regional precipitation
through statistical downscaling. For climate change studies the statistical links
between large and small spatial scales are usually derived from real-world
observations and then applied to the output of GCM simulations for the future
climate. This approach requires the large-scale predictors from the GCM to be
realistically simulated and is therefore known as ‘Perfect-Prog(nosis)’ (PP)
downscaling.
An alternative approach, which is known as and which is routinely used in numerical
weather prediction, is to derive empirical corrections for simulated variables, for
instance by formulating statistical models that correct simulated precipitation. MOS
usually combines a correction and a downscaling step. The use of MOS for climate
change simulation is hampered by the fact that standard GCM simulations for historic
periods do not represent the temporal evolution of random variability. If MOS
corrections were fitted based on such simulations, there would be a risk that
differences in simulated and observed variables, such as biases, scaling factors or
more general differences in distribution parameters, would be falsely attributed to
model errors and thus would be falsely modified by the MOS approach, when they are
actually caused by random differences in the simulated and observed distribution of
large-scale weather states. Moreover, in such a setting the type of statistical models
that can be formulated to link simulated and observed variables is strongly restricted.
As a consequence the MOS approach has not yet been used for estimating
precipitation changes directly from GCM climate change simulations.
In order to derive MOS corrections for simulated GCM precipitation we have
conducted a simulation for the period 1958-2001 with the ECHAM5 GCM in which
key circulation and temperature variables are nudged towards the ERA-40 reanalysis.
This simulation thus is consistent with reality with respect to the large-scale weather
variability, and MOS corrections that link simulated precipitation with regional
observed precipitation can be derived from it. For this approach it is crucial that the
simulated precipitation is not nudged towards observations and is calculated purely by
the precipitation parameterisations in the GCM.
We have used simple local scaling as well as different regression-based MOS
downscaling methods that use non-local predictors (Maximum Covariance Analysis,
PC multiple linear regression) to estimate regional monthly precipitation (1958-2001)
from the nudged ECHAM5 simulations. The observations to fit and validate the
methods have been taken from the global GPCC gridded dataset, which has a spatial
resolution of 0.5˚ x 0.5˚. Cross-validation shows that ECHAM5 precipitation is in
many areas a very good predictor for the real precipitation given realistic synoptic-
Abstracts 136