Mouassom et al. (2025) Hydrodynamics of rainfall peaks in homogeneous regions clustered using the K-means algorithm in Central Africa

Identification

• Journal: Climate Dynamics
• Year: 2025
• Date: 2025-11-13
• Authors: Fernand L. Mouassom, Alain T. Tamoffo, Elsa Dos Santos Cardoso‐Bihlo
• DOI: 10.1007/s00382-025-07932-0

Research Groups

• Department of Mathematics and Statistics, Memorial University of Newfoundland, St. John’s, NF, Canada
• Climate Service Center Germany (GERICS), Helmholtz-Zentrum Hereon, Hamburg, Germany

Short Summary

This study identifies three homogeneous rainfall subregions in Central Africa using K-means clustering on 1984–2023 daily reanalysis data, revealing distinct rainfall peak patterns and their underlying hydrodynamic and thermodynamic mechanisms.

Objective

• To accurately delineate homogeneous rainfall subregions in Central Africa using machine learning clustering algorithms and to investigate the hydrodynamic and thermodynamic mechanisms driving rainfall peaks within each identified subregion.

Study Configuration

• Spatial Scale: Central Africa (CA), spanning from 15°S to 15°N latitude and 5°E to 35°E longitude, at a grid-point scale.
• Temporal Scale: Daily data covering the period 1984–2023.

Methodology and Data

• Models used:
  • K-means clustering algorithm (primary method)
  • MiniBatchKMeans and hierarchical Ward’s clustering (for robustness evaluation)
  • Optimal cluster number determination: Elbow approach, Silhouette score, Hierarchical graph, Gap statistic
  • Mechanism analysis: Vertically integrated moisture flux convergence, Congo Basin cell (mass-weighted stream-function), Shallow Meridional Overturning Circulation (SMOC), Moist Static Energy (MSE), Equivalent Potential Temperature (θe), Vertical velocity (ω), Latent static energy (LSE).
• Data sources:
  • Reanalysis: ERA5 (fifth-generation European Centre for Medium-Range Weather Forecasts reanalysis), 0.25° resolution (approximately 31 km), hourly data averaged to daily.
  • Observational:
    • CHIRPS2 (Climate Hazards Group InfraRed Precipitation with Station data version 2), 0.05° resolution, daily.
    • TAMSAT 3.1 (Tropical Applications of Meteorology using Satellite data and ground-based observations version 3.1), 0.25° resolution, daily.

Main Results

• Central Africa is optimally divided into three homogeneous rainfall subregions:
  • Equatorial Central Africa (ECA): Centered on the equator, characterized by a long rainfall period (October–May) with two distinct peaks in April and November.
  • Northern Central Africa (NCA): Located north of the equator, exhibiting a unimodal rainfall pattern peaking from May to October (maximum in August).
  • Southern Central Africa (SCA): Situated south of the equator, displaying a unimodal rainfall pattern peaking from November to March (maximum in December/February).
• Physical processes driving rainfall peaks are subregion-dependent:
  • NCA and ECA: Rainfall peaks are primarily associated with moisture convergence originating from low-level westerlies, strengthened by the Congo Basin cell. The shallow meridional overturning circulation redistributes moisture, leading to both shallow and deep convection, with rainfall maxima predominantly arising from deep convection. The contribution of the northern African Easterly Jet (AEJ-N) is comparatively weaker.
  • SCA: The AEJ-N dominates moisture convergence, and in conjunction with the deep Hadley cell, facilitates mid-tropospheric deep convection.
• The ERA5 reanalysis adequately reproduces the spatial distribution and seasonality of these subregions and their rainfall peaks, with differences generally below 0.0014 m/day compared to CHIRPS2 and TAMSAT.
• Rainfall peaks across all subregions are consistently associated with strong vertically integrated moisture flux convergence.
• Atmospheric instability (indicated by Moist Static Energy, equivalent potential temperature, and latent static energy) and upward vertical velocity profiles confirm convective activity. NCA and ECA show instability from lower atmospheric layers (peaking at 700 hPa), while SCA exhibits deeper convection originating from mid-level instability (peaking at 600 hPa).

Contributions

• Provides an accurate and objective grid-point-scale regionalization of Central Africa into homogeneous rainfall subregions using machine learning (K-means clustering).
• Offers a detailed, subregion-specific analysis of the hydrodynamic and thermodynamic mechanisms driving rainfall peaks, moving beyond the traditional Intertropical Convergence Zone (ITCZ) paradigm.
• Identifies key circulation features, including the Congo Basin cell, Shallow Meridional Overturning Circulation (SMOC), African Easterly Jet-North (AEJ-N), and the deep Hadley cell, and quantifies their varying contributions to rainfall peaks across different subregions.
• Challenges the conventional assumption of a homogeneous bimodal rainfall regime across the entire Central Africa, demonstrating unimodal patterns in the Northern and Southern Central Africa subregions.
• The findings are expected to improve the reliability of numerical weather forecasting, enhance climate hazard mitigation strategies, and refine regional climate projections across Central Africa.

Funding

• Canada Research Chairs program
• NSERC Discovery Grant program
• Humboldt-Stiftung (Humboldt Research Fellowship for Alain T. Tamoffo)

Citation

@article{Mouassom2025Hydrodynamics,
  author = {Mouassom, Fernand L. and Tamoffo, Alain T. and Cardoso‐Bihlo, Elsa Dos Santos},
  title = {Hydrodynamics of rainfall peaks in homogeneous regions clustered using the K-means algorithm in Central Africa},
  journal = {Climate Dynamics},
  year = {2025},
  doi = {10.1007/s00382-025-07932-0},
  url = {https://doi.org/10.1007/s00382-025-07932-0}
}