Rethinking design rainfall estimation under non-stationary extremes and imperfect reanalysis data

Abstract

Design rainfall return levels are critical inputs for engineering applications such as urban drainage design, flood mitigation, resilient transport infrastructure and climate-sensitive planning. However, their estimation is significanlty challenged by changing rainfall regimes, which can weaken the stationarity assumptions traditionally used in design rainfall analysis. A second challenge is data availability: design estimates rely on long gauge records that are often spatially sparse, while reanalysis products such as ERA5-Land provide spatial continuity but may poorly represent point-scale extremes, especially in convectively active and topographically complex settings. This study develops a reproducible GEE workflow to jointly examine these challenges by assessing temporal changes in extreme precipitation return levels and evaluating the reliability of ERA5-Land for design-oriented rainfall screening. Annual maxima from ERA5-Land daily precipitation are modelled using Gumbel and GEV distributions, with the workflow demonstrated globally and evaluated locally in Thessaloniki, Greece, against a long in-situ gauge record for return periods of 5, 10, 50 and 100 years. Results show that ERA5-Land systematically underestimates gauge-based design rainfall, with discrepancies increasing toward rarer events; consequently, gauge-to-ERA5-Land correction factors range from 1.17 to 1.29 under Gumbel and from 1.13 to 1.46 under GEV. A reference-to-recent period comparison further indicates substantial gauge-inferred intensification, including an approximately 33% increase in the 100-year Gumbel return level, while ERA5-Land shows little corresponding change. The proposed framework is reproducible and transferable for screening applications, but correction factors remain site-specific and require local gauge validation before engineering use.