Integrating Hybrid and Probabilistic Climatic Models for Short-Term Drought Forecasting in Northwestern Morocco

Short-term drought forecasting remains challenging in Mediterranean regions because of strong interannual climate variability. This study develops and verifies an integrated framework for one- to three-month-ahead forecasting of meteorological drought in the Tangerois watershed (northwestern Morocco) using monthly TerraClimate precipitation and temperature data for 1981–2024 extracted from 17 meteorological stations that share four distinct precipitation series. Drought is quantified with the one-month Standardized Precipitation Temperature Index (SPTI), computed per calendar month with parameters of the 1981–2009 training period. A hybrid ARIMA–ANN forecast feeds three ESP-type ensembles of increasing complexity: an equal-weight ensemble (EEDP), a Gaussian-kernel weighted ensemble driven by an integrated climate predictor with an ARIMA-forecasted bandwidth (MWEDP), and a copula-based conditional ensemble that couples precipitation and temperature (CEDP). Forecasts for June–August, issued at the end of May with predictors available at that date, were verified at four representative sites over 2010–2024 against climatology and persistence, and model differences were tested with a block bootstrap over test years. EEDP and MWEDP were statistically indistinguishable from climatology (mean squared error skill score, MSESS = −0.005 and +0.002) and significantly better than persistence (MSESS = +0.24 and +0.25), whereas CEDP was less accurate than climatology (MSESS = −0.17). Differences between the three ensembles were not significant, consistent with a predictor screening that lacked field significance. A wet-season experiment (October–December) gave the same conclusions. All forecasts inherited a wet bias from the drier 2010–2024 period. At these lead times, the framework adds no forecast skill beyond climatology, which should serve as the benchmark for operational drought outlooks in this region.