Abstract
An investigation has been undertaken to quantify the uncertainty in statistically downscaled catchment precipitation and runoff projections due to different climate models. A significant component of the investigation was the assessment of climate model reproduction of the seasonal (i.e. monthly) cycle of the atmospheric predictors used by a statistical downscaling model. Four climate model simulations forced by the SRES A2 emission scenario were used: the CSIRO Mk3 GCM, the CSIRO Conformal-Cubic Atmospheric Model (CCAM, run at high spatial resolution over Australia with far-field forcing from the Mk3 GCM), the Hadley Centre HadAM3P GCM and the Max Planck Institute ECHAM4 GCM.
The non-homogeneous hidden Markov model (NHMM), a stochastic downscaling model, was used to quantify the impacts of the projected climate change on multi-site, daily precipitation. A catchment water balance model (LUCICAT), calibrated under existing conditions, was driven by the downscaled precipitation to produce runoff projections. Land use and potential evaporation (evapotranspiration potential) were kept fixed for the future climate in order to assess the catchment response to differences in the downscaled precipitation series.
Biases in climate model reproduction of the season cycles of the atmospheric predictors used in downscaling are shown to have significant impacts on simulated precipitation, and hence runoff. The downscaled CCAM and Mk3 results project reductions in annual precipitation ranging from 12 to 14% by mid century (2035–2064), resulting in a decline in runoff ranging from 30 to 44%. Downscaling the HadAM3P output, available at the daily time step only for the period 2070–2099, produces a precipitation decline of 24% and a runoff reduction of 69%. The ECHAM4 downscaled precipitation inadequately reproduced the observed annual cycle and so was not used for runoff projection.
