The tenth IMSC, Beijing, China, 2007 - International Meetings on ...

Steve.Charles@csiro.au
Bryson C. Bates
CSIRO Marine and Atmospheric Research
Guobin Fu
CSIRO Land and Water
A stochastic downscaling framework based on a nonhomogeneous hidden Markov
model (NHMM) identifies how regional precipitation changes are related to large-scale
atmospheric forcing. <strong>The</strong> NHMM stochastically simulates multi-site, daily precipitation
conditional on large-scale atmospheric forcing. When driven by observed atmospheric
predictors, from NCEP/NCAR Reanalysis, the NHMM is able to reproduce the statistical
properties of daily, intra-seasonal, inter-annual and inter-decadal multi-site precipitation for
several regions across temperate, mid-latitude southern Australia.
Although originally proposed as a tool for climate change projection, we highlight the
benefits of using such stochastic downscaling models for ‘forensic climatology’ – to understand
the drivers of significant decreases in both variability and amount of precipitation experienced
across southern Australia in recent decades. Links to detection and attribution research, and
other recent investigations into how these changes relate to hemispherical circulation, will also
be discussed.
Improving the simulation of extreme events by stochastic weather generators
Speaker: Richard W. Katz
Richard W. Katz
National Center for Atmospheric Research
rwk@ucar.edu
Eva M. Furrer
National Center for Atmospheric Research
Stochastic weather generators are used for a number of purposes, including
disaggregation of climate forecasts and statistical downscaling of the output of climate change
experiments. Although entire sequences of daily weather are usually required, in many
applications it is extreme weather events (such as high precipitation amounts) that are
particularly important. Although much is known about the characteristics of extreme weather
events (e.g., the apparent heavy upper tail of daily precipitation amount), such information is
seldom explicitly taken into account in conventional stochastic weather generators.
One approach to stochastic weather generation is resampling. This approach is
nonparametric, with the advantage that the shape of the distribution of a weather variable is
not assumed a priori. However, it has the disadvantage that no value outside the span of the
86