Modelling drought in South Africa: meteorological insights and predictive parameters.

South Africa has grappled with recurring drought scenarios for over two decades, leading to substantial economic losses. Droughts in the Western Cape between 2015 and 2018, especially in Cape Town was declared a national disaster, resulting in the strict water rationing and the "day zero" effect. This study presents a set of simulations for predicting drought over South Africa using Artificial Neural Network (ANN), using Standard Precipitation Index (SPI) as the drought indicator in line with the recommendations of the World Meteorological Organization (WMO). Furthermore, different meteorological variables and an aerosol parameter were used to develop the drought set in four distinct locations in South Africa for a 21-year period. That data used include relative humidity (rh), temperature (tp), soil wetness (sw), evapotranspiration (et), evaporation (ev) sea surface temperature (st), and aerosol optical depth (aa). The obtained R values for SPI3 ranged from 0.49 to 0.84 and from 0.22 to 0.84 for SPI6 at Spring Bok, Umtata 0.83 to 0.95 for SPI3, and 0.61 to 0.87 for SPI6; Cape Town displayed R values from 0.78 to 0.94 for SPI3 and 0.57 to 0.95 for SPI6, while Upington had 0.77-0.95 for SPI3, and 0.78-0.92 for SPI6. These findings underscore the significance of evapotranspiration (et) as a pivotal parameter in drought simulation. Additionally, the predictive accuracy of these parameter combinations varied distinctly across different locations, even for the same set of parameters. This implies that there is no single universal scheme for drought prediction. Hence, the results are important for simulating future drought scenarios at different parts of South Africa. Finally, this study shows that ANN is an effective tool that can be utilized for drought studies and simulations.