Abstract
The breakpoint data format for rainfall simply records the times when the rain rate changes from one steady value to another; the rates are also recorded. Data covering 15 years at Invercargill, New Zealand (46°25′S, 168°20′E), are decomposed into lognormal modes representing rainfall rate and duration variations during convective or frontal precipitation; dry period durations are similarly analyzed. The degree of decomposition and the labeling of the modes are initially arbitrary, but an objective choice is made through discriminant analyses, using parameters derived from the decompositions, and by a comparison with manual weather observations.
The result of the discriminant analysis is that the rain record is divided into “rain” and “shower” events and the dry times that often separate such events. This provides the physical basis for the rainfall model proposed by Sansom and Thomson, which consists of a five-state hidden semi-Markov model with the “rain” and “Shower” states being divided into “wet” and “dry” substates and the fifth state is the dry time between events. Further support for this model is given through a simple analysis of the spatial variation of event parameters using data from 71 stations throughout New Zealand for seven years. The spatial variation follows the climate regionalization, which was established in New Zealand many decades ago.
