Naik et al. (2026) Effect of bias correction on CORDEX simulations of hydrological droughts in the Western Cape region of South Africa

Identification

• Journal: Theoretical and Applied Climatology
• Year: 2026
• Date: 2026-01-01
• Authors: Myra Naik, Babatunde J. Abiodun
• DOI: 10.1007/s00704-025-05951-9

Research Groups

• Climate System Analysis Group, Department of Environmental and Geographical Science, University of Cape Town, Cape Town, South Africa
• Nansen-Tutu Centre for Marine Environmental Research, Department of Oceanography, University of Cape Town, Cape Town, South Africa

Short Summary

This study investigates how bias correction of CORDEX climate data improves simulations of hydroclimatic variables and hydrological droughts in South Africa's Western Cape. It finds that bias correction significantly enhances hydrological variable simulations, with multivariate methods performing best, but offers minimal improvement for meteorological drought characteristics due to the inherent standardization in both the bias correction and drought index calculation processes.

Objective

• To investigate the extent to which bias correction of the COordinated Regional Downscaling Experiment (CORDEX) dataset can improve the quality of hydroclimatic variables and drought simulations over South Africa’s Western Cape region.

Study Configuration

• Spatial Scale: Western Cape region, South Africa (30° S − 35° S; 14° E − 25° E), covering approximately 129,462 km². The study focuses on four major river basins: Breede (12,348 km²), Berg (7,715 km²), Gouritz (45,715 km²), and Olifants (46,220 km²). CORDEX data with a native spatial resolution of approximately 0.44° (about 50 km) was bilinearly interpolated to a 0.5° grid and then downscaled to hydrological response units (HRUs) for the hydrological model.
• Temporal Scale:
  • Global Meteorological Forcing Dataset (GMFD) spans 1901–2012.
  • SWAT+ model calibration period: 1980–1989 (10 years, following a 5-year spin-up).
  • SWAT+ model validation period: 1990–2006 (16 years).
  • Main simulation experiments (GMFD, original CORDEX, and bias-corrected CORDEX): 1971–2000 (30 years, following a 5-year spin-up).
  • Drought indices were calculated with monthly data at a 12-month scale.

Methodology and Data

• Models used:
  • Hydrological Model: Soil Water Assessment Tool Plus (SWAT+).
  • Bias-Correction Methods:
    • Quantile Delta Mapping (QDM) - univariate.
    • Multivariate Bias Correction with Spearman rank correlation dependence structure (MBCr) - multivariate.
    • Multivariate Bias Correction using the N-dimensional probability density function transform (MBCn) - multivariate.
  • Calibration Tool: Integrated Parameter Estimation and Uncertainty Analysis Tool Plus (IPEAT+).
• Data sources:
  • Climate Input:
    • COordinated Regional Downscaling Experiment (CORDEX-Africa) regional climate model (RCM) simulations.
    • Global Meteorological Forcing Dataset (GMFD; version 3.0, Princeton University) at 0.5° resolution, used as a reference for bias correction and for forcing SWAT+ during calibration and validation.
  • Hydrological Observations: Daily streamflow data from The Department of Water Affairs at four stations (E2H003, G1H013, H7H006, J1H019) in the Western Cape.
  • Geographic Information Systems (GIS) Data for SWAT+ setup:
    • Digital Elevation Model (DEM): Shuttle Radar Topography Mission (SRTM).
    • Digital soil information: Food Agricultural Organization (FAO) global soil databases (version 3.6) via Waterbase.Org.
    • Land Use and Land Cover (LULC) maps: USGS Global Land Cover Characterization database via Waterbase.Org.

Main Results

• The SWAT+ model, forced with GMFD data, credibly simulates streamflow in the Western Cape river basins, capturing observed annual cycles and inter-annual variability, with best performance showing Nash-Sutcliffe Efficiency (NSE) ≥ 0.5, coefficient of determination (R²) ≥ 0.7, and percentage bias (PBIAS) ≤ 31%.
• Original CORDEX (ORG) simulations reproduce the annual cycle and spatial distribution of hydroclimate variables but exhibit systematic biases, including a cold bias (up to 3.5 °C), wet bias in February–July (up to 5 mm month⁻¹), dry bias in October–December (up to 10 mm month⁻¹), underestimation of potential evapotranspiration (up to 80 mm month⁻¹), and overestimation of climate water balance.
• All three bias-correction methods effectively reduce systematic biases in CORDEX climate variables, improving agreement with GMFD. For instance, they reduce the root mean square error (RMSE) in annual precipitation (from > 9.7 to about 1.07 mm month⁻¹), potential evapotranspiration (from > 16.7 to < 3.2 mm month⁻¹), and climate water balance (from > 20.9 to < 3 mm month⁻¹), and improve spatial correlation to 1.0 for some variables.
• Bias correction enhances the quality of hydrological simulations, improving runoff patterns (correlation from < 0.4 to > 0.93) and decreasing RMSE (from > 2.59 to < 1.18). However, the improvement for hydrological variables is less pronounced than for climate variables, with PBIAS reaching up to 80% for runoff, indicating error magnification.
• The multivariate bias-correction method MBCn generally performs best in improving the quality of hydrological simulations over the Western Cape river basins.
• None of the three bias-correction methods consistently improve the quality of meteorological drought indices (Standardized Precipitation Index (SPI) and Standardized Precipitation Evapotranspiration Index (SPEI)) from ORG simulations. Improvements in drought intensity and frequency are minimal, and spatial correlation remains poor (r ≤ 0.3). This is attributed to the inherent standardization in both drought index calculation and bias-correction methods.
• Conversely, all three bias-correction methods substantially improve the quality of hydrological drought indices (Soil Water Index (SWI), Run-off Index (RFI), Percolation Index (PERCI), Surface Flow Index (SFI), and Water Yield Index (WLDYI)), increasing spatial correlation (e.g., SWI from ≤ 0.27 to > 0.75, RFI from ≤ 0.01 to > 0.73, WYLDI from ≤ 0.12 to > 0.8) and reducing RMSE.

Contributions

• This study provides a comprehensive assessment of the impact of univariate (QDM) and multivariate (MBCr, MBCn) bias-correction techniques on CORDEX regional climate model simulations for hydrological variables and drought characteristics in the Western Cape, South Africa.
• It quantifies the differential effectiveness of bias correction, demonstrating significant improvements for hydrological variables but minimal impact on standardized meteorological drought indices, offering a crucial insight into the limitations of bias correction for certain drought metrics.
• The research highlights the superior performance of multivariate bias-correction methods (specifically MBCn) in improving hydrological simulations, guiding future climate impact studies.
• The findings are directly applicable to improving the reliability of hydrological variable and drought simulations in the Western Cape, supporting water resource management and adaptation strategies in a drought-prone region.

Funding

• Water Research Commission of South Africa
• National Research Foundation of South Africa (Reference Number: RPOAM231218201502)
• Pan-African and Transdisciplinary Lens in the Margins: Tackling the Risks of Extreme Events (PALM-TREEs) project
• Centre for High-Performance Computing (CHPC, South Africa) (computing facility)

Citation

@article{Naik2026Effect,
  author = {Naik, Myra and Abiodun, Babatunde J.},
  title = {Effect of bias correction on CORDEX simulations of hydrological droughts in the Western Cape region of South Africa},
  journal = {Theoretical and Applied Climatology},
  year = {2026},
  doi = {10.1007/s00704-025-05951-9},
  url = {https://doi.org/10.1007/s00704-025-05951-9}
}