chapter  14
Interpolation of Severely NonLinear Spatial Systems with Missing Data: Using Kriging and Neural Networks to Model Precipitation in Upland Areas
ByJoanne Cheesman, James Petch
Pages 14

For strategic planning purposes, water authorities require accurate yield estimates from reservoirs, therefore precipitation gauge interpolation results are critical for providing areal precipitation estimates. However, the interpolation of precipitation amounts in remote, upland areas is one situation in which input data are severely unrepresentative. Precipitation gauge networks are usually of low density and uneven distribution with the majority of gauges located in the lowland regions of catchments. Results of using traditional interpolation techniques are seriously affected both by the complexity of theoretical data surfaces (Lam, 1983) and by the quality of data, especially their density and spatial arrangement. Typically, a standard interpolation technique will fail to model upland precipitation successfully, as the interpolation is likely to be based upon lowland gauges.