A Multi-Method Approach to Analyzing Precipitation Series and Their Change Points in Semi-Arid Climates: The Case of Dobrogea
Youssef Saliba, Alina Bărbulescu
Abstract
The Dobrogea region, located in southeastern Romania, experiences a semi-arid climate. This study provides a deep analysis of monthly precipitation series from 46 meteorological stations spanning 1965–2005, exploring mean and variance characteristics and detecting structural changes in precipitation patterns. The series normality was assessed using the Lilliefors test, and transformation, such as the Yeo–Johnson method, was used to address skewness. Analyses of mean and variance included parametric (t-tests, ANOVA) and non-parametric (Mann–Whitney U, Fligner–Killeen) tests to address the homogeneity/inhomogeneity of the data series in mean and variance. Change points were detected using a Minimum Description Length (MDL) framework, modeling the series as piecewise linear regressions with seasonal effects and autocorrelated errors. Pairwise comparisons indicate the low similarity of the series means, and variances, so spatial and temporal variability in precipitation is notable. Validation of the proposed MDL approach on synthetic datasets demonstrated high accuracy, and application to real data identified significant shifts in precipitation regimes. Applied to the monthly series collected at the ten main hydro-meteorological stations, a MDL framework provided at least two change points for each.