The error and bias of supplementing a short, arid climate, rainfall record with regional vs. global frequency analysis

Robust and accurate estimates of rainfall frequencies are difficult to make with short, and arid-climate, rainfall records, however new regional and global methods were used to supplement such a constrained 15-34 yr record in Cyprus. The impact of supplementing rainfall frequency analysis with the regional and global approaches was measured with relative bias and root mean square error (RMSE) values. Analysis considered 42 stations with 8 time intervals (5-360 min) in four regions delineated by proximity to sea and elevation. Regional statistical algorithms found the sites passed discorclancy tests of coefficient of variation, skewness and kurtosis, while heterogeneity tests revealed the regions were homogeneous to mildly heterogeneous. Rainfall depths were simulated in the regional analysis method 500 times, and then goodness of fit tests identified the best candidate distribution as the general extreme value (GEV) Type 11. In the regional analysis, the method of L-moments was used to estimate location, shape, and scale parameters. In the global based analysis, the distribution was a priori prescribed as GEV Type 11, a shape parameter was a priori set to 0.15, and a time interval term was constructed to use one set of parameters for all time intervals. Relative RMSE values were approximately equal at 10% for the regional and global method when regions were compared, but when time intervals were compared the global method RMSE had a parabolic-shaped time interval trend. Relative bias values were also approximately equal for both methods when regions were compared, but again a parabolic-shaped time interval trend was found for the global a short, and climate, rainfall record with regional vs. global frequency analysis 175 method. The global method relative RMSE and bias trended with time interval, which may be caused by fitting a single scale value for all time intervals. (c) 2006 Elsevier B.V. All rights reserved.