Blake et al. (2025) The impact of convection-permitting model rainfall on the dryland water balance

Identification

• Journal: Hydrology and earth system sciences
• Year: 2025
• Date: 2025-12-12
• Authors: George Blake, Katerina Michaelides, Elizabeth Kendon, Mark Cuthbert, Michael Bliss Singer
• DOI: 10.5194/hess-29-7093-2025

Research Groups

• School of Geographical Sciences, University of Bristol, Bristol, United Kingdom
• Met Office Hadley Centre, Exeter, United Kingdom
• School of Earth and Environmental Sciences, Cardiff University, Cardiff, United Kingdom
• Earth Research Institute, University of California, Santa Barbara, USA

Short Summary

This study quantifies the impact of convection-permitting climate model rainfall on the dryland water balance in the Horn of Africa, demonstrating that explicitly resolving convection improves rainfall characteristics, leading to significantly higher soil moisture, transpiration, surface runoff, and potential groundwater recharge compared to models that parameterize convection.

Objective

• To determine if explicitly resolving convection in climate models improves the representation of dryland rainfall characteristics and hydrologically relevant rainfall metrics in the Horn of Africa.
• To assess if convective representation affects simulated potential evapotranspiration (PET) dynamics in the Horn of Africa.
• To investigate if the impact of convective representation on rainfall characteristics and PET influences how water is partitioned between different stores in drylands.

Study Configuration

• Spatial Scale: Horn of Africa (HOA), focusing on four sites along an aridity gradient: a humid site in the Ethiopian Highlands, a semi-arid site in southern Kenya, an arid site in eastern Ethiopia, and a hyper-arid site in northern Somalia. Climate models cover a pan-African domain. Hydrological simulations are one-dimensional (1-D) vadose zone models with a 3 meter soil profile.
• Temporal Scale: "Historical" climate model simulations for the period 1997–2007. Satellite rainfall data (IMERG) from June 2000 to February 2007. Hydrological simulations run for 10 years, using hourly resolution for rainfall and potential evapotranspiration (PET) inputs.

Methodology and Data

• Models used:
  • Climate Models: Convection-Permitting Model for Africa (CP4A) at 4.5 km horizontal resolution, and a parameterised 25 km regional model (P25), both configurations of the Met Office Unified Model.
  • Hydrological Model: Hydrus 1-D v4.17 for simulating vertical water redistribution in the soil subsurface.
• Data sources:
  • Satellite: Integrated Multi-satellitE Retrievals for GPM (IMERG) for rainfall (0.1° spatial, 30 minute temporal resolution). Hourly Potential Evapotranspiration (hPET) dataset (0.1° spatial, hourly temporal resolution).
  • Reanalysis: ERA5-Land variables (used to derive hPET).
  • Other: CGIAR-CSI for aridity index, iSDAsoil database for soil texture, National Centers for Environmental Information AVHRR LAI dataset for Leaf Area Index.

Main Results

• Rainfall Characteristics:
  • CP4A significantly reduces the "drizzle" bias (overestimation of low-intensity rainfall events) observed in P25, particularly in drylands. P25 simulates 51.5% of annual dryland rainfall as drizzle (≤ 1 mm/h), compared to 14.1% for CP4A and 13.0% for IMERG.
  • CP4A's rainfall intensity distribution shows a markedly better match to IMERG (Kolmogorov–Smirnov statistic: 0.03 for CP4A vs. 0.24 for P25).
  • P25 underestimates maximum dry spell length (median dryland consecutive dry days: ~18 days for P25 vs. ~38 days for IMERG), while CP4A overestimates in some areas (~100 days).
  • CP4A improves the representation of extreme rainfall magnitude (99th percentile of wet hours), with median 99th percentile values over 110% higher in CP4A compared to P25 in drylands.
  • A greater proportion of annual rainfall is delivered via "heavy" events in CP4A dryland simulations (interquartile range 13.5%–25.4%) compared to P25 (3.8%–11.0%).
• Potential Evapotranspiration (PET):
  • Both CP4A and P25 capture the seasonal and diurnal cycles of PET, and spatial patterns, with comparable annual magnitudes.
  • Differences in PET between the models are less pronounced than for rainfall, primarily driven by variations in temperature, dew point temperature, and meridional wind speed in drylands.
• Water Partitioning (Hydrus 1-D simulations):
  • Despite simulating lower total rainfall, CP4A forcing results in higher depth-integrated soil moisture in drylands. For example, at the hyper-arid site, median soil moisture is 23.1% higher at 1.2 meters below ground level with CP4A.
  • CP4A leads to deeper wetting fronts in drylands, with soil moisture remaining above the wilting point for longer periods (e.g., 83% of the simulation at the arid site with CP4A vs. 44% with P25).
  • Evaporative losses are lower with CP4A in drylands (67%–72% of infiltration returned to the atmosphere vs. 79%–83% with P25).
  • Transpiration is substantially higher at semi-arid and arid sites when forced with CP4A.
  • Surface runoff is significantly greater with CP4A in drylands (6%–10% of rainfall vs. 0.3%–2% with P25).
  • Bottom drainage (indicative of potential groundwater recharge) is substantially higher with CP4A in drylands (e.g., 286 mm vs. 23 mm at the semi-arid site; 52 mm vs. 2 mm at the hyper-arid site over 10 years).
  • In humid regions, water partitioning is less sensitive to rainfall characteristics and hydrological fluxes more closely follow annual rainfall totals.
  • While PET can influence vadose-zone hydrological outcomes, dryland hydrology is more sensitive to the impact of convective representation on rainfall characteristics.

Contributions

• First study to assess how climate model representation of convection affects externally calculated potential evapotranspiration (PET) from model atmospheric variables.
• Quantifies the impact of convection-permitting model rainfall on the 1-D water balance across an aridity gradient in the Horn of Africa.
• Demonstrates the high sensitivity of dryland vadose zone hydrology to rainfall characteristics (frequency, intensity, extremes) rather than solely to total rainfall or PET.
• Highlights the critical risk of misrepresenting societally relevant hydrological fluxes (soil moisture, groundwater availability, surface runoff) in future water resource projections if using climate models that parameterize convection in drylands.
• Provides a direct comparison between a convection-permitting model (CP4A) and a traditional parameterised regional climate model (P25) with the same underlying model physics and driving global climate model, effectively isolating the impact of convective representation.

Funding

• UKRI Natural Environment Research Council (reference NE/S007504/1)
• Horizon 2020 DOWN2EARTH project (grant no. 869550)
• UK Met Office
• University of Bristol Research Fellowship (Katerina Michaelides)
• Leverhulme Research Fellowship (grant no. RF-2023-591/4) (Katerina Michaelides)

Citation

@article{Blake2025impact,
  author = {Blake, George and Michaelides, Katerina and Kendon, Elizabeth and Cuthbert, Mark and Singer, Michael Bliss},
  title = {The impact of convection-permitting model rainfall on the dryland water balance},
  journal = {Hydrology and earth system sciences},
  year = {2025},
  doi = {10.5194/hess-29-7093-2025},
  url = {https://doi.org/10.5194/hess-29-7093-2025}
}