Author(s): Claudio I. Meier; Geri Pranzini; J. Sebastián Moraga; Peter Molnár
Linked Author(s):
Keywords: Interannual variability; Climate change; Non stationarity; Switzerland; Chile
Abstract: The interannual variability of precipitation is given by the probability distribution (pdf) of annual rainfall, which in temperate zones is typically obtained by fitting a normal (or sometimes a lognormal) distribution to a series of annual rainfall data. There are three problems with this approach: (i) a long record (n> 25~30) is required in order to fit the model; (ii) years with missing data cannot be used for the analysis; and (iii) trends or jumps (non-stationarities) can happen over the rather long period required for adequately fitting a probabilistic model. Therefore, a methodology is needed that would allow for better estimation of the pdf of annual rainfall, without requiring long records. In 1978, Eagleson proposed using derived distributions to obtain the pdf of annual rainfall, by combining the marginal pdfs for storm depth and inter-arrival time. Our aim in this paper is to assess the differences between Eagleson´s approach and the fitting of a normal distribution, looking at two effects: length of the record and level of aggregation of the rainfall data. We quantified the first effect by randomly subsampling the rainfall series, in order to create multiple synthetic records of different lengths, to which we applied both methods. We then re-did our analyses for precipitation data aggregated over 12 and 24-h long periods, the typical information available at most weather stations. We used records for Concepción, Chile (25 yr) ,and for Lugano, Switzerland (32 yr) ,to compare two different humid climates, where a normal distribution would be typically used for fitting annual rainfall data. Annual rainfall is uncorrelated in time, at both sites. Both approaches yield very similar pdfs when using all data available. In Concepción, totalising rainfall every 12 h results in a pdf that is undistinguishable from that obtained for continuous data, but aggregating over 24 h introduces bias. In Lugano, any level of aggregation introduces a bias. The proposed method is a much more consistent way of estimating the pdf of annual rainfall when only a few years of data are available, allowing for a robust estimation even for records as short as 5 years. Thus, derived distributions are a powerful tool for describing interannual variability in rainfall, in the case of short records or when climate might be nonstationary.
Year: 2015