Katambo et al. (2025) Understanding ENSO Teleconnections’ Influence on Drought in Southern Africa: A Machine Learning Approach

Identification

• Journal: Lecture notes in networks and systems
• Year: 2025
• Date: 2025-10-18
• Authors: Jimmy Katambo, Gloria Iyawa, Lars Ribbe, Victor Kongo
• DOI: 10.1007/978-981-96-9709-0_11

Research Groups

• Namibia University of Science and Technology, Windhoek, Namibia
• University of Salford, Salford, UK
• TH Köln, University of Applied Sciences, Cologne, Germany
• Global Water Partnership Tanzania, Dar es Salaam, Tanzania

Short Summary

This study investigates the influence of El Niño-Southern Oscillation (ENSO)-related Sea Surface Temperature (SST) variations on drought patterns across Southern Africa using machine learning. The findings reveal SST's significant and consistent impact across all climate zones, underscoring its value for enhanced drought prediction and adaptation planning.

Objective

• To investigate the teleconnection relationships between El Niño Southern Oscillation (ENSO) and Standardized Precipitation Evapotranspiration Index (SPEI) drought across the different regions of Southern Africa.
• To quantify and rank the influence of Sea Surface Temperature (SST) data on drought conditions across Southern Africa’s climate zones using machine learning.

Study Configuration

• Spatial Scale: Southern Africa, spanning from the equator to 40° South, divided into eight Köppen-Geiger climate zones. Key river basins include the Orange (~1.0 x 10^12 m²), Limpopo (~4.15 x 10^11 m²), Okavango, and Zambezi (~1.37 x 10^12 m²). The Kalahari Desert reaches temperatures over 313.15 K. Temperatures in the Limpopo basin range between 273.15 K and 309.15 K.
• Temporal Scale: Monthly data from 1972 to 2022.

Methodology and Data

• Models used: Transformer architecture (selected after preliminary evaluation against Random Forest and Feedforward Neural Networks), permutation importance function.
• Data sources:
  • Climate station historical data (monthly, 1972-2022) from SASSCAL Information and Data portal (for validation).
  • ERA5 from Copernicus: wind speed (m/s), relative humidity (%), solar radiation (W/m²), precipitation (m), potential evapotranspiration (m), minimum, maximum, and average temperature (K).
  • NASA satellite observed data: soil moisture (1981-2022).
  • Synthetically generated data using Generative Adversarial Networks (GANs) for missing soil moisture (1972-1980) and sunshine duration (seconds).
  • Köppen-Geiger climate classification global maps for zonal divisions.
  • Sea Surface Temperature (SST) data (1972-2022) from National Oceanic and Atmospheric Administration (NOAA).
  • Standardized Precipitation Evapotranspiration Index (SPEI) calculated at 1-, 3-, and 6-month intervals using R software.

Main Results

• The transformer model paired with the 6-month Standardized Precipitation Evapotpiration Index (SPEI6) demonstrated optimal performance among tested configurations.
• Sea Surface Temperature (SST) consistently showed a significant impact on drought variability across all eight Southern African climate zones.
• Feature Importance Analysis (SST ranking among 12 variables):
  • Highest impact (4th place): Cold Semi-Arid Climate (BSk, Zone 7) and Warm Mediterranean Climate (Csa, Zone 8).
  • Moderate impact: Cold Desert Climate (BWk, Zone 6) at 5th, Humid Subtropical Climate (Cwa, Zone 1) at 6th, Humid Subtropical/Subtropical Oceanic Highland (Cwb, Zone 5) at 7th.
  • Lower impact: Tropical Savanna Climate (Aw, Zone 2) at 8th, Warm Semi-Arid (BSh, Zone 3) and Warm Desert (BWh, Zone 4) both at 10th.
• Comparative Model Performance (with vs. without SST data):
  • Inclusion of SST data generally improved model performance metrics across all zones. For example, in Zone 1, the Correlation Coefficient (R) increased from 96.2% to 96.8%, Nash–Sutcliffe Efficiency (NSE) from 84.9% to 87.3%, and accuracy from 96.2% to 96.7%. Mean Bias Error (MBE) decreased from 0.9% to 0.08%, and both Mean Squared Error (MSE) and Mean Absolute Error (MAE) improved from 3.8% to 3.2%.
  • The impact of SST, as measured by variance in performance metrics, was highest in Zone 2 (Tropical Savanna Climate) and lowest in Zone 5 (Humid Subtropical/Subtropical Oceanic Highland).
• Both analytical approaches confirmed the strong influence of SST, and by extension El Niño events, on drought conditions across Southern Africa, despite slight variations in the observed order of effect across zones.

Contributions

• First study to apply machine learning, specifically a transformer architecture, to assess the role of Sea Surface Temperature (SST) in regional drought dynamics across diverse climate zones in Southern Africa.
• Introduced two novel analytical methods: a feature importance analysis to rank SST against twelve other climatic variables and a comparative model approach evaluating performance with and without SST data.
• Provides detailed, zone-specific insights into the primary climate drivers influencing drought, enhancing the accuracy of drought forecasting and supporting effective adaptation planning for stakeholders.
• Expands the geographical scope beyond typical country-level studies by examining a large region divided into distinct climatic zones.

Funding

• Southern African Science Service Centre for Climate Change and Adaptive Land Management (SASSCAL).
• German Government through the Federal Ministry of Education and Research (BMBF), funding No: 01LG2091A.

Citation

@article{Katambo2025Understanding,
  author = {Katambo, Jimmy and Iyawa, Gloria and Ribbe, Lars and Kongo, Victor},
  title = {Understanding ENSO Teleconnections’ Influence on Drought in Southern Africa: A Machine Learning Approach},
  journal = {Lecture notes in networks and systems},
  year = {2025},
  doi = {10.1007/978-981-96-9709-0_11},
  url = {https://doi.org/10.1007/978-981-96-9709-0_11}
}