Key points are not available for this paper at this time.
Abstract This study proposes a modeling framework based on piecewise constant regression with a Periodic Autoregressive (PAR) remainder. Model complexity is controlled using Rissanen’s Minimum Description Length (MDL) criterion, while Fourier-based compression reduces dimensionality without compromising likelihood, improving computational efficiency. A multi-island Genetic Algorithm (GA) jointly estimates changepoint (CP) configurations and seasonal order. Monte Carlo (MC) experiments under Dobrogea conditions show reliable detection of all three change points when shifts exceed 1.5 innovation standard deviations. Applied to Medgidia precipitation (1965–2019), the method identifies an early shift while retaining a parsimonious AR(1) structure. By separating true climatic shifts from noise, the framework provides a robust basis for water resource assessment and infrastructure planning.
Saliba et al. (Wed,) studied this question.