Non‐Stationary Dry‐Spell Hazard Probabilities for Spain

ABSTRACT This study provides a comprehensive assessment of long‐term changes in dry‐spell dynamics across Spain for the period 1961–2024, using both classical non‐parametric methods and a novel non‐stationary probabilistic framework applied to exceedance series of dry‐spell durations. Using daily precipitation data from a dense network of quality‐controlled stations, dry‐spells were identified based on four thresholds (0.1, 1, 5, and 10 mm) to characterize their duration statistics, and generalized Pareto distributions (GPDs) were fitted under both stationary and non‐stationary specifications. A systematic percentile‐based evaluation was performed to determine optimal exceedance thresholds, and a bootstrap‐based significance test was applied to compare stationary and non‐stationary return levels. Complementary analyses examined trends in annual frequency, mean duration, and maximum dry‐spell length. Results consistently reveal that dry‐spell behaviour across Spain is dominantly stationary. Only a very small fraction of stations show significant non‐stationary effects in the GPD location parameter, whereas non‐stationary scale, shape, or full‐parameter models provide no robust improvement and introduce substantial uncertainty. Traditional trend analyses corroborate this finding: most stations (> 70%–85%, depending on threshold) exhibit no statistically significant changes in dry‐spell frequency, mean duration, or maximum duration. Spatial patterns of change are weak, and return levels associated with long recurrence periods (e.g., 50 years) remain highly stable throughout 1961–2024. Overall, these results support long‐term stationarity in the probability of extreme dry‐spells across Spain.