STOCHASTIC SPACE-TIME WEATHER MODELS FROM GROUND-BASED DATA

The development of space-time stochastic weather models from data obtained from standard meteorological networks is examined. The approach adopted is first to identify appropriate stochastic point models whose parameters can be calibrated from point data and reliably spatially interpolated. Following standard statistical climatological practice, the space-time covariance structure of the resulting weather anomalies can then be used to develop space-time simulation models which generate correlated anomalies of rainfall and other variables. Two stochastic point daily rainfall models illustrate competing requirements of physical realism and economy of parameterisation. Weather model parameters can be accurately interpolated using statistical methods which incorporate spatially varying dependences on elevation. These methods can also apply objective data smoothing to incorporate serially incomplete data. The truncated power of normal distribution is shown to be a promising candidate for space-time simulation of rainfall, on both monthly and daily time scales, provided problems associated with systematic differences between occurrence-based and intensity-based correlations can be resolved. It is suggested that monthly space-time models, which are reasonably well determined from standard meteorological networks, are sufficient to resolve much ecological and hydrological behaviour, particularly at spatial resolutions of a few kilometres. They also form a useful precursor to daily space-time stochastic models.