Pham-Thanh et al. (2026) Seasonal precipitation prediction over Vietnam: evaluation of RegCM dynamical downscaling and statistical bias correction of NCEP CFS forecasts

Identification

• Journal: Theoretical and Applied Climatology
• Year: 2026
• Date: 2026-03-13
• Authors: Ha Pham-Thanh, Tan Phan-Van, Thanh Nguyen-Xuan, Dzung Nguyen-Le
• DOI: 10.1007/s00704-026-06117-x

Research Groups

• Department of Meteorology and Climate Change, University of Science, Vietnam National University (VNU), Hanoi, Vietnam
• Department of Space and Earth Sciences, University of Science and Technology of Hanoi (USTH), Vietnam Academy of Science and Technology (VAST), Hanoi, Vietnam

Short Summary

This study evaluates the performance of dynamically downscaled seasonal precipitation forecasts over Vietnam using RegCM-NH driven by NCEP CFS, and the improvements obtained through statistical bias correction. It finds that multiple linear regression (MLR) significantly enhances forecast accuracy, reducing systematic biases and improving interannual variability representation across Vietnam's climatic sub-regions.

Objective

• To evaluate the performance of seasonal precipitation prediction over Vietnam using RegCM-NH version 4.9.5 dynamically downscaling NCEP CFSv2 at a finer 20 km horizontal resolution.
• To quantify the extent to which statistical bias correction methods (climatological adjustment and multiple linear regression) can further improve forecast skill relative to the raw RegCM_CFS output.

Study Configuration

• Spatial Scale: Vietnam, divided into seven climatic sub-regions (Northwest (R1), Northeast (R2), Red River Delta (R3), North Central (R4), Central South (R5), Central Highlands (R6), and Mekong River Delta (R7)). The RegCM model domain covers 2°N to 29°N and 93°E to 122°E, with a horizontal grid spacing of 20 km and 18 vertical sigma levels up to 50 hPa.
• Temporal Scale: The study period is from January 1982 to December 2020 (39 years), evaluating monthly precipitation forecasts with lead times from 1 to 6 months.

Methodology and Data

• Models used:
  • Dynamical Downscaling: Non-hydrostatic Regional Climate Model version 4.9.5 (RegCM-NH) by ICTP, using Holtslag planetary boundary layer, SUBEX large-scale precipitation, Grell cumulus convection, Zeng ocean surface flux, BATS land surface model, and standard RegCM radiation scheme.
  • Driving Global Model: National Centers for Environmental Prediction (NCEP) Climate Forecast System version 2 (CFSv2) reforecasts (1° × 1° resolution).
  • Bias Correction:
    • Climatological Adjustment (CLIM): Multiplicative scaling based on observed and model climatological means.
    • Multiple Linear Regression (MLR): Predicts monthly precipitation anomalies using outgoing longwave radiation (OLR), 2-meter specific humidity (q2), 2-meter relative humidity (RH2), vertical velocity at 700 hPa (W700), meridional wind shear between 200 hPa and 850 hPa (U200–U850), and model-predicted precipitation anomaly.
• Data sources:
  • Observations: Monthly gridded rainfall data at 0.25° × 0.25° latitude–longitude resolution for 1981–2020, derived from an updated version of the Vietnam Gridded Precipitation (VnGP) dataset, based on 481 quality-controlled stations from the Vietnam Hydro-Meteorological Service (VHMS).
  • Reanalysis/Forecasts: NCEP CFSv2 reforecasts (1° × 1° resolution) for initial and lateral boundary conditions for RegCM.

Main Results

• Raw RegCM_CFS forecasts show significant regional contrasts in skill, with regional climate characteristics exerting a stronger influence than forecast lead time (1–6 months).
• Northern sub-regions (R1–R4) are better represented in terms of mean seasonal precipitation structure, while southern sub-regions (R5–R7) exhibit higher skill in reproducing interannual variability during their climatologically dominant rainy seasons.
• Raw RegCM_CFS forecasts display persistent wet biases across most sub-regions, with relative mean absolute error (RMAE) values ranging from approximately 0.44 to over 7.7.
• Both climatological adjustment (CLIM) and multiple linear regression (MLR) bias correction methods substantially improve forecast accuracy.
• MLR provides the most robust and consistent enhancement, effectively reducing systematic biases, increasing agreement with observed interannual variability, and improving spatial coherence across sub-regions.
• Bias correction, particularly MLR, removes the spurious bimodal seasonal structure in southern Vietnam's raw forecasts, restoring the observed unimodal rainfall cycle.
• The added value (AV) of bias-corrected forecasts relative to raw RegCM_CFS is consistently positive, with MLR showing peak AV values up to 0.92, indicating substantial improvement.
• These improvements persist across lead times, demonstrating the benefit of combining dynamical downscaling with statistical post-processing for long-lead seasonal precipitation forecasting.

Contributions

• Provides an updated evaluation of RegCM (v4.9.5) dynamical downscaling of NCEP CFSv2 for seasonal precipitation prediction over Vietnam at a finer 20 km resolution, building upon previous coarser-resolution studies.
• Quantifies the significant additional benefits of applying statistical bias correction (CLIM and MLR) to dynamically downscaled seasonal precipitation forecasts, demonstrating that MLR offers robust and consistent improvements.
• Highlights that statistical post-processing is crucial for achieving operationally relevant forecast accuracy, complementing the physical realism provided by dynamical downscaling, especially in complex monsoon-dominated regions.
• Offers a practical and computationally efficient integrated framework for improving seasonal precipitation prediction over Vietnam, with direct relevance for agriculture, water resource management, and disaster preparedness.

Funding

• Vietnam Academy of Science and Technology (VAST) under Grant THTETN.01/25–26.

Citation

@article{PhamThanh2026Seasonal,
  author = {Pham-Thanh, Ha and Phan-Van, Tan and Nguyen-Xuan, Thanh and Nguyen-Le, Dzung},
  title = {Seasonal precipitation prediction over Vietnam: evaluation of RegCM dynamical downscaling and statistical bias correction of NCEP CFS forecasts},
  journal = {Theoretical and Applied Climatology},
  year = {2026},
  doi = {10.1007/s00704-026-06117-x},
  url = {https://doi.org/10.1007/s00704-026-06117-x}
}