Answers are grounded in this paper and cite their sources.
Rainfall in the Kingdom of Saudi Arabia is highly intermittent and strongly seasonal, which complicates the generation of synthetic rainfall series for water-resources studies. This study develops and evaluates a stochastic generator of monthly rainfall for 28 meteorological stations with record lengths of 14–43 years (1970–2012). Rainfall occurrence is modelled with a first-order, two-state Markov chain whose transition probabilities vary by calendar month, and wet-month rainfall depths are drawn from fitted probability distributions within a 500-realization Monte Carlo framework. Five candidate distributions (Gamma, Lognormal, Exponential, Beta, and Gumbel) were compared using maximum likelihood estimation (MLE) and a mean–maximum matching procedure, and were evaluated with the Akaike information criterion (AIC) and the reproduction of monthly means, variability, and upper quantiles. The Gamma distribution fitted by MLE gave the most balanced performance (R2 = 0.93 for the monthly standard deviation and 0.92 for the 95th percentile), whereas the Gumbel distribution was never preferred by the AIC. In split-sample validation, the generator performed as well as the calibration-period climatology (median R2 of monthly means 0.13 versus 0.14), and 89% of observed annual totals fell within the simulated 5th–95th percentile range. Mean wet- and dry-spell lengths and annual maximum monthly rainfall were reproduced at all stations, but the longest dry spells were underestimated, falling outside the simulated range at 10 of the 28 stations, which reflects the limitation of first-order dependence. Rainfall in the validation period was lower than in the calibration period at 21 of the 28 stations, indicating a non-stationarity that a stationary generator cannot represent.
No takes yet. Share an insight, caveat, or question.
Holik et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: