CR 13:61-75 (1999)  -  doi:10.3354/cr013061

ABSTRACT: Based on principal component analysis (PCA) and k-means clustering algorithm, 4 daily circulation patterns, which are associated with daily precipitation in southern Portugal, have been identified from observed daily mean sea level pressure (MSLP) fields over the northeastern Atlantic and western Europe. A weather generator based on daily circulation patterns was calibrated with 3 different schemes to simulate daily precipitation occurrence and a 2-parameter gamma distribution was applied to generate daily precipitation amounts on rain days in southern Portugal. Parameters of the weather generator were estimated for every month in the winter half of the year (October to March) subject to each of the 4 daily circulation patterns. The weather generator was validated by hindcast using independent observational data. The synthetic daily precipitation series generated from the weather generator kept most of the important attributes of the observed series, such as the empirical distribution of daily precipitation amounts, the autocorrelation structure of the sequence of wet and dry days, the distributions of durations of wet and dry spells, and even the distributions of precipitation totals on several consecutive days. Furthermore, the daily MSLP fields over the northeastern Atlantic and western Europe simulated by the Hadley Centre's second generation coupled ocean-atmosphere GCM (HADCM2) control run (HADCM2CON) were validated by comparison with observed daily MSLP fields. It is clear that HADCM2 is able to reproduce very well daily MSLP fields and their seasonal variability over the region. Four daily circulation patterns, associated with daily precipitation in Portugal, identified from the observed daily MSLP fields over the area, were also classified well from the simulated daily MSLP fields. The weather generator was then applied to the sequence of daily circulation patterns in simulated MSLP fields of HADCM2CON, and validated. Essentially, the conclusions of previous validation analysis remained the same in the present case. The results imply that the weather generator can reproduce well important features of the 'present climate', including local precipitation in southern Portugal, and can therefore be applied to obtain future precipitation scenarios from daily MSLP fields simulated by the 2 transient experiments of HADCM2, HADCM2GHG (greenhouse gases) and HADCM2SUL (greenhouse gases plus sulphate aerosols). However, it should be kept in mind that the use of the weather generator in obtaining future precipitation scenarios is dependent on the assumption that the present relationship between local precipitation and large-scale atmospheric circulation will remain valid in the future changing climate.

KEY WORDS: Weather generator · Downscaling · Precipitation · Daily circulation patterns · Southern Portugal