Abstract

Information on precipitation values ​​and their frequencies is needed in various projects related to water.In this study, the regional frequency analysis based on L-moments was used to estimate the annual maximum daily precipitation quantiles at any site in an area located in the north east of Algeria. The investigated area was divided into three regions statistically homogeneous in terms of L-moments ratios. Among the different tested distributions, (i) the Generalized Extreme Value (GEV) distribution has been identified as the most appropriate regional distribution for modeling precipitation in the northern region (region 1) characterized by a Mediterranean climate and (ii) the generalized lognormal distribution (LN3) for region 2 of southern part subject to semi-arid climate. Growth curves, derived from the regional distributions, were established for each homogeneous region. The growth curves shapes difference reflects the difference between the three homogeneous regions precipitation patterns. The regional model efficiency was put into evidence via a comparison of the specified return period’s quintiles estimated from the regional and at-site frequency. The root mean square error values (RMSE) were found to be below 6% for a return period of 10 years and 12% for a return period of 100 years.

Key words: Maximal annual daily precipitation, L-moments, regional frequency analysis, GEV, LN3 distributions, set generation.