A Data-Driven Framework for Estimating Extreme Rainfall in the Thessaly River Basin, Greece

Background: Accurate estimation of extreme rainfall is essential for hydrological design and flood risk management. In Greece, this is commonly performed using Intensity–Duration–Frequency (IDF) curves, although these may not fully capture the spatial and statistical variability of rainfall. This study proposes an alternative data-driven approach based on observed rainfall data from the Thessaly River Basin.

Methods: Daily rainfall data from 53 hydrometeorological stations were analysed. An Exploratory Data Analysis (EDA) was first conducted to assess data quality, statistical behaviour, and temporal consistency. Trend and change-point tests were applied to evaluate stationarity and guide the Rainfall Frequency Analysis (RFA). Five probability distributions (GEV, GLO, Gumbel, Lognormal, and Pearson Type III) were fitted using Maximum Likelihood Estimation (MLE) and L-moments. Model selection was based on Kolmogorov–Smirnov and Anderson–Darling tests, supported by Q–Q and return level plots. Spatial interpolation of rainfall extremes was performed using Inverse Distance Weighting (IDW), with 43 stations used for mapping and 10 for validation.

Results: L-moments slightly outperformed MLE, being selected as the best method in 52.8% of stations. The GLO distribution was the most frequently preferred model. IDW reproduced the general spatial pattern of rainfall well but tended to smooth extreme values, particularly for longer return periods. Overall, IDW-based rainfall maps showed similar spatial patterns to IDF-based products but generally lower magnitude values.

Conclusions: The proposed framework provides a simple, data-based alternative to traditional IDF methods. The results were directly based on the available observed rainfall data, which generally led to lower extreme rainfall values. However, the spatial patterns produced by the method were consistent with those obtained from IDF curves. The differences in values are mainly related to the characteristics and limitations of the available dataset. Overall, the method can be used as a complementary tool for hydrological design, flood risk assessment, and water resources management, especially when detailed observed data are available.