A large-scale distribution function model was used to investigate the eect of diering parameter mapping schemes on the quality of hydrological predictions.
Precipitation was mapped over a large forested catchment area (163 km 2 ) using both one-dimensional linear and three-dimensional non-linear interpolation schemes. Lumped stream ¯ow predictions were found to be particularly sensitive to the dierent precipitation maps, with the three-dimensional map predicting 12% higher mean annual precipitation, resulting in 36% higher modelled stream ¯ow over a three-year period. However, spatial predictions of stream ¯ow appeared worse when derived from the three-dimensional map, which is considered the better of the two precipitation maps. This implies uncertainty in either the model's response to precipitation or the precipitation mapping process (the 12% precipitation dierence was strongly determined by a single, short term gauge). Leaf area index (LAI) was mapped using both remote sensing and species based methods. The two LAI maps had similar lumped mean values but exhibited signi®cant spatial dierences. The resulting lumped predictions of stream ¯ow did not vary. This suggests a linear response of water balance to LAI in the non-water-limited conditions of the study area, and de-emphasizes the importance of quantifying relative spatial variations in LAI. Topographic maps were created for a small experimental subcatchment (15 ha) using both air photographic interpretation and ground survey. The two maps diered markedly and lead to signi®cantly dierent spatial predictions of runo generation, but nearly identical predicted hydrographs. Thus, at scales of small basins, accurate topographic mapping is suggested to be of little importance in distribution function modelling because models are unable to make use of complex spatial data.
Predictions of water yield can be very sensitive (in the case of precipitation) or insensitive (in the case of smallscale topography) to changes in spatial parameterization. In either case, increased complexity in spatial parameterization does not necessarily result in better, or more certain prediction of hydrological response.