The practical use of climate information in applications is a challenging research problem and requires downscaling of the data from the global climate models. In recent years use of Regional Climate Models nested within the coarser resolution GCM has become increasingly common and affordable way to downscale the climate information to a resolution relevant to regional and local issues. In this presentation we report on the results from an evaluation of a dynamical downscaling approach to seasonal climate forecasting in Australia and in particular in Queensland.

We have evaluated the results from the ensemble of simulations using historical SST data for the period 1965 to 2000 as a forcing for the CSIRO T63/L18 AGCM and NCEP MRF9 T40/L18 AGCM as well as CSIRO DARLAM RCM at 75 km and 15 km horizontal grid spacing and 18 vertical levels. The history files produced by the NCEP AGCM were saved at 6-hourly intervals and used as a lateral boundary conditions for the CSIRO RCM simulations at 75 km spatial resolution over the Australian domain. These simulations were further downscalled using the double nesting approach with RCM at 15 km grid spacing over the area of Queensland. The ensemble of 15 integrations was completed with each model and used to evaluate the simulated rainfall characteristics for various regions of Australia.

Regional rainfall characteristics were compared between observed and simulated quantities for each model in measures such as mean, bias, variance and spatial correlation for the seasonal long-term climatology of 1965 to 2000. We have also evaluated the skill of simulated interannual rainfall variability using anomaly correlation, RSME, and ROC measures. The results show a significant increase in the skill of simulated long-term seasonal rainfall climatology and moderate skill increase for the interannual variability as measured by the anomaly correlation with increasing model resolution. The new generation CSIRO AGCM demonstrated similar skill to the 75 km version of RCM. The simulated rainfall was also callibrated using statistical approach based on SVD and regression methods.

The model simulated rainfall from the hindcast experiments was used to make decisions about stocking rates for the coming summer. This information was then used as an input parameter for the grazing simulation system run at 5 km resolution for Queensland. The model was integrated for the period of 1965 to 2000 and a number of indicators were accumulated, such as animal live weight gain, pasture utilization and runoff. This data was used to evaluate the economic benefits and resource utilization. A number of hindcasts were evaluated in this approach, using data from a perfect knowledge approach (historical rainfall) and comparing it with hindcasts derived from dynamical models and statistical methods such as SOI phase system.