Modelling drought in South Africa: meteorological insights and predictive parameters

Area of Science:

• Environmental Science
• Meteorology
• Artificial Intelligence

Background:

• South Africa faces persistent drought challenges, causing significant economic impact.
• The Western Cape experienced a severe drought from 2015-2018, leading to national disaster status and strict water rationing in Cape Town.
• Recurring droughts necessitate improved prediction models for water resource management.

Purpose of the Study:

• To develop and evaluate Artificial Neural Network (ANN) models for drought prediction in South Africa.
• To assess the efficacy of various meteorological variables and aerosol parameters in drought simulation.
• To identify key parameters and location-specific factors influencing drought prediction accuracy.

Main Methods:

• Utilized Artificial Neural Network (ANN) for drought prediction simulations.
• Employed the Standard Precipitation Index (SPI) as the primary drought indicator, following WMO guidelines.
• Incorporated meteorological data (relative humidity, temperature, soil wetness, evapotranspiration, evaporation, sea surface temperature) and aerosol optical depth across four South African locations over 21 years.

Main Results:

• ANN models demonstrated significant predictive capabilities, with R² values varying by location and SPI timescale (SPI3 and SPI6).
• Evapotranspiration (et) was identified as a crucial parameter for accurate drought simulation.
• Predictive accuracy varied considerably across different locations, indicating no universal drought prediction scheme.

Conclusions:

• Artificial Neural Networks (ANN) are effective tools for drought studies and simulations in South Africa.
• Location-specific factors significantly influence drought prediction models.
• The study provides valuable insights for simulating future drought scenarios across South Africa, aiding water resource planning.