Empirical quantile mapping is the most balanced way to correct GCM rainfall on six continents

Why it matters

Choosing a bias correction method changes both how accurate and how uncertain corrected rainfall is. The paper’s index lets users weigh these two aims and see where the best choice flips. Picking a method on error metrics alone could miss these shifts.

What they did

The authors corrected daily rainfall from 11 CMIP6 climate models over 1980–2014 using three quantile mapping methods: QDM, EQM and DQM. They trained on 1980–1996 and tested on 1997–2014 against ERA5 data on a 1° grid. They scored each method with ten metrics and ranked them with TOPSIS. They measured uncertainty with Bayesian Model Averaging, then combined both into a Comprehensive Index (CI) under three weightings of performance versus uncertainty.

Key findings

• All three methods beat the raw model rainfall in Taylor diagrams. EQM was closest to the reference data and DQM was furthest.
• EQM scored best on most metrics (RMSE 0.30, MAE 0.18, R2 0.98, KGE 0.87, NSE 0.93, EVS 0.99). It also had the lowest uncertainty (variance 0.0027) across all continents.
• TOPSIS picked EQM in most grid cells on every continent, and DQM was chosen least. In South America QDM had the lowest model uncertainty (variance 0.0025).
• In the CI, EQM was chosen most often under all three weightings. Giving uncertainty more weight raised the number of EQM grids.
• In South America DQM was picked more than QDM when performance was stressed, but QDM was picked more when uncertainty was stressed. Oceania showed the reverse.

Limitations

• Only three quantile mapping methods were tested, because a global run was computationally costly. Adding more methods would make the results stronger.
• The study covers only the historical period and does not use greenhouse gas scenarios, which the authors list as future work.
• The CI results depend on the weights the user picks for performance and uncertainty.

Glossary

• Quantile mapping: A bias correction that adjusts model values so their distribution matches the observed distribution.
• TOPSIS: A ranking method that scores options by how close they are to an ideal best and far from an ideal worst across several criteria.
• Bayesian Model Averaging (BMA): A way to combine several models with weights, which also measures how much the models disagree.
• Comprehensive Index (CI): The authors’ score that combines the TOPSIS performance score with the BMA uncertainty measures.