Comparison of the potential skill of raw and downscaled GCM output for river flow forecasting: a UK case study

The background to this work was to compare the potential skill of raw versus downscaled GCM output for simulating river flow. The results at both tested resolutions highlighted that by using ERA-40 precipitation data as direct input to the PDM rainfall-runoff model, the simulated river flow substantially underestimated the observed river flow in the Dyfi basin. This is likely to be due to the reanalysis data's inability to resolve catchment-scale (or GCM sub-grid scale) atmospheric processes such as orographic enhancement of rainfall over the Welsh Mountains. The direct use of GCM output as input to a rainfall-runoff model is thus not recommended due to its poor ability in reproducing observed river flows. To address the relatively poor performance of the GCM data, a statistical downscaling technique (SDSM) was used. The downscaling process added skill to the river flow simulations for ERA-40, and more particularly for ERA-40 1.5˚ resolution. This improvement is significant for all parts of the river regime (from high flows such as Q5 to low flows such as Q95). This research suggests that much improvement is added to GCM output when combining them with statistical methods. In the future we intend to test the downscaling method used here on seasonal forecasts from GCMs to assess how combining dynamical with statistical techniques could forecast river flow in the UK.