Suitability of a transitional probability matrix model for catchment scale rainfall projections

Projecting rainfall and water availability in a changing climate is a well researched area. Stochastic and dynamic downscaling techniques are both currently being used to develop rainfall projections from GCM model outputs. While most of these techniques are adequate for reproducing historical rainfall characteristics, a wide range of variability are generally observed in the generated rainfall projections. These lead to a wide range of variability and uncertainty in hydrological and water availability projections. In this paper, we describe how ensembles of daily rainfall data have been generated from 2035 to 2065 using a Transitional Probability Matrix (TPM) model (CRCCH, 2006) for two catchments located in the South Australian Mount Lofty region. These are the Aldgate Creek catchment (7.9 km²) and the Inverbrakie Creek catchment (8.4 km²). It was found that generated daily rainfalls reproduced well the historical rainfall characteristics for these two catchments. The generated replicates of rainfall from 2035 to 2065 were compared with the CSIRO Mk3 (A2 scenario) generated downscaled rainfall replicates. It was found that while the 50th percentile rainfall projections from both models were well matched, the variability of CSIRO Mk3 generated rainfalls was significantly high in comparision to the TPM model generated rainfalls.