Mert Katipoğlu, Okan; Achite, Mohammed; Elshaboury, Nehal; Abdelkader, Eslam Mohammed

The Aras Basin in Turkey is highly susceptible to meteorological and hydrological droughts because of the reduced rainfall due to climate change and higher water demand for agriculture. In this context, the main objective of this work was to model the Reconnaissance Drought Index (RDI) over 6- and 12-month periods in two meteorological observation stations in the Aras basin. The prediction performance was evaluated by merging the Artificial Neural Network (ANN) model optimised using the Artificial Bee Colony (ABC) algorithm, namely the ABC-ANN model, with the Variational Mode Decomposition (VMD) and Robust Empirical Mode Decomposition (REMD) methodologies. Some statistical metrics and graphical plots were used to assess the model performance. According to graphical and statistical indicators, the hybrid ABC-ANN (R2 = 0.668-0.967) and VMD-ABC-ANN (R2 = 0.758-0.905) models achieved satisfactory results in estimating RDI values at 6- and 12-month time scales. To evaluate the predictive performance applied models, the Friedman test was employed and revealed statistically significant differences among the models (p < 0.0001). The subsequent Nemenyi post-hoc test identified that ABC-ANN performed significantly better than others. Furthermore, when the temporal variations in forecasts were studied, it was established that the ABC-ANN model was superior to other models in estimating extreme drought values.