Sibiya et al. (2026) Assessing trends and forecasting meteorological drought in South Africa using Savitzky–Golay enhanced hybrid deep learning

Identification

• Journal: Scientific Reports
• Year: 2026
• Date: 2026-04-10
• Authors: Siphamandla Sibiya, Shaun Ramroop, Sileshi Fanta Melesse, Nkanyiso Mbatha
• DOI: 10.1038/s41598-026-46664-x

Research Groups

• School of Mathematics, Statistics, and Computer Science, University of KwaZulu-Natal, Pietermaritzburg Campus, South Africa
• Council for Scientific and Industrial Research, Holistic Climate Change, Smart Places, Pretoria, South Africa
• Pure and Applied Analytics, Faculty of Natural and Agricultural Sciences, North-West University, Potchefstroom, South Africa

Short Summary

This study assessed meteorological drought trends in South Africa's uMkhanyakude District using daily rainfall data from 1980-2023 and developed a novel Savitzky–Golay enhanced hybrid deep learning model (SG–TCN–LSTM) for forecasting, demonstrating superior predictive accuracy compared to other models.

Objective

• To comprehend rainfall variability and enhance meteorological drought prediction in the uMkhanyakude District of KwaZulu-Natal, South Africa, for sustainable water and food security planning.

Study Configuration

• Spatial Scale: uMkhanyakude District, KwaZulu-Natal, South Africa, utilizing data from six meteorological stations.
• Temporal Scale: Daily rainfall records spanning 1980 to 2023.

Methodology and Data

• Models used: Standardized Precipitation Index (SPI) at 6-, 9-, and 12-month time scales, Innovative Trend Analysis (ITA), Savitzky–Golay filter (SG), Temporal Convolutional Network (TCN), Long Short-Term Memory (LSTM), and a novel hybrid SG–TCN–LSTM model. Comparative models included ARIMA, LSTM, TCN, and other hybrid models.
• Data sources: Daily rainfall records from six meteorological stations, obtained from the South African Weather Service (SAWS).

Main Results

• Innovative Trend Analysis (ITA) revealed statistically significant decreasing long-term drought trends at five out of six meteorological stations.
• One station (Riverview) exhibited an increasing drought trend over the study period.
• The novel hybrid SG–TCN–LSTM model consistently achieved superior predictive accuracy and stability for drought forecasting.
• The SG–TCN–LSTM model recorded the lowest Root Mean Square Error (RMSE) values (0.0349–0.1453) and the highest (R^{2}) values (0.95–0.99) across all SPI scales (6-, 9-, and 12-month).
• The integration of signal smoothing with deep learning methodologies significantly enhanced the robustness of the drought forecasts.

Contributions

• Development and validation of a novel Savitzky–Golay enhanced hybrid deep learning model (SG–TCN–LSTM) for meteorological drought forecasting.
• Demonstration of the SG–TCN–LSTM model's superior predictive accuracy and stability compared to traditional and other deep learning models (ARIMA, LSTM, TCN, other hybrids).
• Provision of critical insights for proactive drought risk management through enhanced forecasting robustness.
• Establishment of a framework for reliable early-warning instruments for meteorological drought, which can inform national adaptation strategies.
• Detailed analysis of long-term meteorological drought trends in the uMkhanyakude District, identifying both decreasing and increasing trends at specific stations.

Funding

The authors declare no funding.

Citation

@article{Sibiya2026Assessing,
  author = {Sibiya, Siphamandla and Ramroop, Shaun and Melesse, Sileshi Fanta and Mbatha, Nkanyiso},
  title = {Assessing trends and forecasting meteorological drought in South Africa using Savitzky–Golay enhanced hybrid deep learning},
  journal = {Scientific Reports},
  year = {2026},
  doi = {10.1038/s41598-026-46664-x},
  url = {https://doi.org/10.1038/s41598-026-46664-x}
}