Regional frequency analysis of daily extreme precipitation in Croatia

Short-duration extreme precipitation amounts (on both sub-daily and daily scales) are often required for the design and management of drainage systems at locations where no direct measurements or only limited records exist. Identifying regions with similar spatio-temporal characteristics of extremes can provide a more robust basis for estimating regional probabilities of rare events. Regional design values are expected to yield more reliable results than single-station analyses by reducing the influence of outliers, short records, and potential measurement errors. In this study, we apply a regional frequency analysis based on the L-moments approach to identify representative, homogeneous rainfall regions in Croatia. This task is particularly challenging due to the high spatial variability of precipitation maxima across the country, influenced by complex orography and diverse climatic conditions. Although regions with characteristic rainfall patterns have previously been identified, such classifications largely relied on the mean annual precipitation cycle and subjective criteria. The primary aim of this study is to establish a more objective and robust regionalization based on rainfall extremes. Due to the sparse network of rain gauges recording sub-daily precipitation, the analysis focuses on maximum daily precipitation amounts (Rx1d) derived from a dense network of 180 stations covering the period 1961–2022. A well-established four-step approach is applied, including data screening, identification of homogeneous regions based on geographical characteristics and distance from the sea, and the selection and estimation of the frequency distribution. By analysing different numbers of regions, ranging from three to nine, an optimal division into seven climatological regions was identified, providing the best representation of homogeneous characteristics of annual extreme daily precipitation. The generalized extreme value (GEV) distribution was identified as the most appropriate theoretical model for fitting Rx1d and was used to estimate return levels for individual stations as well as for regions. The characteristics of precipitation in the identified regions were further analysed in relation to prevailing weather types, providing additional insight into the physical drivers of extreme precipitation in these areas.