Fashoto et al. (2025) Anticipating drought: enhancing prediction models and assessing environmental impact in Eswatini’s Maguga Basin

Identification

• Journal: International Journal of Information Technology
• Year: 2025
• Date: 2025-10-24
• Authors: Stephen Gbenga Fashoto, Petros Mashwama, Mcondisi Ngcebo Nxumalo, Boluwaji Akinnuwesi, Elliot Mbunge, Andile Metfula
• DOI: 10.1007/s41870-025-02750-3

Research Groups

• Department of Computer Science, University of Eswatini, Kwaluseni, Eswatini
• Department of Informatics, Namibia University of Science and Technology, Windhoek, Namibia
• Department of Applied Information System, University of Johannesburg, Johannesburg, South Africa

Short Summary

This study developed and compared drought prediction models for Eswatini's Maguga Basin, finding that a Genetic Algorithm (GA) optimized Long Short-Term Memory (LSTM) model significantly outperformed the Auto-regressive Integrated Moving Average (ARIMA) model in forecasting the Standardized Precipitation Evapotranspiration Index (SPEI) and Maguga Dam water levels. The research provides a robust tool for early drought warning and water resource management in the region.

Objective

• To identify suitable drought indices for quantifying drought events in the Kingdom of Eswatini.
• To compare the accuracy of machine learning models (GA optimized LSTM) against other pertinent forecasting models (ARIMA) for drought prediction.
• To predict SPEI values for the Pigg’s Peak area up to a six-month lead time using machine learning techniques.
• To predict water levels at Maguga Dam for a six-month lead time.

Study Configuration

• Spatial Scale: Pigg’s Peak area, Hhohho region, Kingdom of Eswatini, focusing on the Maguga Dam basin.
• Temporal Scale: Data collected from 2000 to 2019 (20 years) for training and testing, with forecasts generated for a six-month lead time.

Methodology and Data

• Models used:
  • Standardized Precipitation Evapotranspiration Index (SPEI) for drought quantification (3-month timescale).
  • Auto-regressive Integrated Moving Average (ARIMA) model for univariate time series forecasting.
  • Genetic Algorithm (GA) optimized Long Short-Term Memory (LSTM) model for univariate SPEI forecasting and multivariate water level forecasting (including temperature, precipitation, and SPEI as exogenous variables).
• Data sources:
  • Two weather stations in Pigg’s Peak (monthly precipitation, minimum and maximum temperature from 2000 to 2019).
  • KOBWA Company (daily observed water levels for Maguga Dam from 2004 to 2019, converted to monthly).

Main Results

• The Standardized Precipitation Evapotranspiration Index (SPEI) was confirmed as a suitable index for quantifying drought in Eswatini, accurately identifying the severe 2015-2016 drought event.
• The Genetic Algorithm (GA) optimized Long Short-Term Memory (LSTM) model significantly outperformed the Auto-regressive Integrated Moving Average (ARIMA) model in predicting SPEI values. The GA optimized LSTM achieved a Mean Absolute Percentage Error (MAPE) of 0.93 and a Mean Percentage Error (MPE) of -0.74, compared to ARIMA's MAPE of 184.5 and MPE of 71.0.
• The GA optimized LSTM model also demonstrated high accuracy in forecasting Maguga Dam water levels, achieving a MAPE of 0.00217 and an MPE of 0.00113.
• Future forecasts for a six-month lead time indicate near-normal drought conditions and a decline in Maguga Dam water levels, but remaining above 600 meters, aligning with the drought forecast.

Contributions

• This study represents a pioneering effort in Eswatini to develop a drought forecasting model using a computational intelligence perspective, specifically a GA optimized LSTM.
• It provides a highly accurate and reliable drought forecasting model capable of predicting events six months in advance, enabling proactive mitigation strategies.
• The research integrates SPEI forecasting with Maguga Dam water level predictions, offering a comprehensive assessment of drought impact on critical water resources.
• The developed model serves as a valuable decision support tool for government and private organizations in developing economies for effective drought management and water resource optimization.

Funding

This study did not receive any specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Citation

@article{Fashoto2025Anticipating,
  author = {Fashoto, Stephen Gbenga and Mashwama, Petros and Nxumalo, Mcondisi Ngcebo and Akinnuwesi, Boluwaji and Mbunge, Elliot and Metfula, Andile},
  title = {Anticipating drought: enhancing prediction models and assessing environmental impact in Eswatini’s Maguga Basin},
  journal = {International Journal of Information Technology},
  year = {2025},
  doi = {10.1007/s41870-025-02750-3},
  url = {https://doi.org/10.1007/s41870-025-02750-3}
}