Rainfall Classification Using Breakpoint Pluviograph Data

Abstract

Breakpoint data derived from the manual digitization of pluviographs from Invercargill, New Zealand, (46°25?S, 168°20?E) is considered. The breakpoints, recording changes of intensity from one steady value to another, are digitized and processed into a steam of data pairs: the rainfall rate, which includes zero, and the duration of that rate. Viewed this way, rainfall appears to be nonrandom and composed of two types of events that can be interpreted, respectively, as rain and showers. Each type has its own lognormal distribution of intensifies and durations, both wet and dry. For rain there is a ?0.9 correlation between rate and duration, with rates ranging from 0.1 mm h?1 to 6 mm h?1 and durations from 2 min to 1.5 h.; dry times within a rain event range from 6 min to 4.5 h. For showers there is a ?0.44 correlation between the rate and duration, with rates ranging from 0.3 mm h?1 to 16 mm h?1 and durations from 1.5 min to 1 h., dry times within a shower event range from 15 min to 1.2 days. Also, there is a third type of dry period, which corresponds to the intervals between rain and shower events, whose durations range from 5 h to 1 week. Thus, without recourse to other meteorological parameters or its time sequence, the data can be directly classified in a simple and natural way. The classes could be used in a Markovian model of precipitation in which the system states would be ?rain event,? ?shower event,? and ?dry time between events,? and the first two states would have ?wet? and ?dry? substates. This model could then he applied to problems such as the better prediction of rainfall amounts, intensifies, and durations on daily time scales or the better parametedution of rainfall in climate modeling.