Key points are not available for this paper at this time.
Artificial neural nets are used in an empirical down-scaling procedure to derive daily subgrid-scale precipitation from general circulation model (GCM) geopotential height and specific humidity data. The neural net-based transfer functions are developed using a 2°×2·5° gridded data assimilation product from the Goddard Space Flight Center, applied to a 4×4 matrix of grid-cells centred on the Susquehanna river basin. The down-scaled precipitation is a close match to the observed data (temporal correlations at individual grid-points range from 0·6 to 0·84). Doubled CO2 climate change scenarios are produced by applying the same transfer functions to the geopotential height and specific humidity fields from 1×CO2 and 2×CO2 simulations of version II of the GENESIS climate model. The analysis indicates a 32 per cent increase in spring and summer rainfall over the basin, resulting from changes in both moisture availability and the orientation of the storm track over the region. The down-scaled precipitation increases, derived from the change in the GCM's circulation and humidity fields, are considerably larger than the change in the model's actual computed precipitation. © 1998 Royal Meteorological Society.
Crane et al. (1998) studied this question.