final_program_abstracts[1]

11 IMSC Session Program
Water resources in South-west Western Australia: model
uncertainty in climate change adaption
Thursday - Poster Session 2
Richard E. Chandler 1 , Stephen P. Charles 2 and Bryson C. Bates 3
1
Department of Statistical Science, University College London, UK
2 CSIRO Land and Water, Wembley, Australia
3 CSIRO Climate Adaptation Flagship, Wembley, Australia
Southwest Western Australia (SWWA) is currently experiencing an extended period
of drought and, according to the IPCC Fourth Assessment Report, winter precipitation
in the region is “very likely” to decrease over the coming century. Since 70-80% of
the region’s rainfall falls in winter, this has serious water resource implications. Water
resource managers therefore face the challenge of putting in place the mechanisms
and infrastructure to cope with reduced water availability in the future. The decisions
they make will be informed by projections of future climate based on deterministic
atmosphere-ocean general circulation models (GCMs). However, these projections
can vary widely between GCMs, with differences becoming more pronounced at the
relatively fine spatial and temporal scales relevant for hydrological applications. It is
therefore natural to ask how best to combine information from different GCMs.
Here we present a transparent, logically coherent and interpretable framework that
formalises the issues involved and provides the opportunity to frame the relevant
questions in an unambiguous manner. The framework uses a hierarchical statistical
model to represent features that account for inter-GCM differences, shared biases and
outlying GCMs; it is in principle sufficiently general that it could be applied in any
climate impacts study. To develop projections for use in SWWA water resource
planning, the methodology is applied in conjunction with a statistical downscaling
tool to develop projections of winter (May-October) rainfall for a network 30 sites in
the region. The results suggest that naïvely combining the projections from different
GCMs and treating the resulting ensemble as a probability distribution may
substantially underestimate the uncertainty involved.
Abstracts 276