Distribution-Based Method-Selection Framework for Reservoir Storage Estimation in a Semi-Arid Snow-Fed Basin

Closed-form critical period methods for reservoir storage rest on different distributional assumptions and are rarely checked against the failure probability they promise. We develop a distribution-based method-selection framework ending in a synthetic-record simulation of achieved reliability, and apply it with five estimators (Gould–Dincer Normal, Log-Normal and Gamma, Alexander, McMahon–Mein) to 42–55-year records at four stations in the Karasu sub-basin, Türkiye. Akaike weights modestly favour skew-aware distributions, without separating Gamma from Log-Normal, so the closed-form capacity is model-averaged (bootstrap interval roughly ±40%). The Normal (Dincer) formula exceeds the Gamma formula by a fixed 36% at 70% regulation and 5% risk, whereas simulation puts the difference between Normal and Gamma flow processes at 51–101%. On 50,000-year records from the fitted Gamma model, Gamma-based capacities deliver 1.5–4.3% annual failure against the 5% target at 70% regulation but up to 7.9% at 80%, and Normal-based 1.0–2.7% and 3.9–5.6%, respectively. Serial correlation of 0.18–0.21 raises non-zero simulated storage requirements by 4–22% on average. Simulation is a benchmark conditional on the fitted two-parameter models: parameter and model-selection uncertainty give it a 95% interval of 0.3–1.6 times its value at 80% regulation. Closed forms remain diagnostics and starting values; all capacities estimated here are inter-annual carry-over storage.