Evaluating probability distributions for design rainfall estimation and their implications for spillway design at the Kalankuttiya reservoir, Mahaweli system H, Sri Lanka

Abstract

The design of hydraulic structures depends critically on the design flood, which is governed by the choice of probability distribution used to estimate extreme design rainfall. This study evaluates how distribution and estimation method affect design rainfall, design flood, and spillway capacity for the Kalankuttiya Reservoir catchment, Mahaweli System H, Sri Lanka. The Siyambalangamuwa station's annual maximum daily rainfall for 31 years was fitted to six distributions: Gumbel, Generalised Extreme Value, Log-Pearson Type III, Lognormal, Gamma, and Weibull. Anderson-Darling, Kolmogorov-Smirnov, and Chi-square tests were used to evaluate goodness-of-fit. The tail-sensitive Anderson-Darling statistic indicated that Log-Pearson Type III was the best-performing distribution. Design rainfall at the 50-year return period was estimated for each distribution and used to generate design flood hydrographs in a HEC-HMS model. Resulting peak discharge ranged from 230.8 to 293.1 m³/s, corresponding to a required spillway length of 45 to 59 m, a spread of approximately 24 and 27 percent, respectively. Critically, independent of distribution selection, all six distributions show a necessary spillway length that is significantly longer than the current 18 m spillway, indicating a significant capacity deficiency. Log-Pearson Type III is recommended as the primary design basis, with the full distributional range reported as an uncertainty envelope to inform urgent spillway capacity review.