Statistical downscaling of GCM simulations to streamflow

A multi-tiered forecast procedure is employed to simulate real-time operati onal seasonal forecasts of categorized (below-normal, near-normal and above -normal) streamflow at the inlets of twelve dams of the Vaal and upper Tuge la river catchments in South Africa. Forecasts are made for the December to February (DJF) season over an 8-year independent period from 1987/ 1988 to 1994/1995. A physically based model of the atmosphere system, known as a g eneral circulation model (GCM), is used to simulate atmospheric variability over southern Africa, the output of which is statistically downscaled to s treamflow. The GCM used is the COLA T30, and is forced at the boundary with predicted monthly-mean global sea-surface temperatures. The monthly-mean s ea-surface temperature fields are first predicted over lead-times of severa l months using a canonical correlation analysis (CCA) model. GCM simulation s are then obtained for an area including most of southern Africa and adjac ent oceans. The GCM simulations are downscaled to catchment level from coar se resolution gridded climate variables, using a perfect prognosis approach : bias-corrected GCM simulations are substituted into the perfect prognosis equations to provide the downscaled categorized streamflow forecasts. Alth ough surface characteristics of each catchment that affect the variability of streamflow are not considered in the proposed downscaling system, succes sful forecasts of streamflow categories were obtained for some of the years forecast independently. The scheme's operational utility is thus demonstra ted. albeit over short lead-times. (C) 2001 Elsevier Science B.V. All right s reserved.