Assessment of enhanced Kohonen self-organizing map, quantile mapping and copula-based bias-correction approaches for constructing basin-scale rainfall forecasts

ABSTRACT

This study evaluates the performance of an enhanced Kohonen self-organizing map (eKSOM)-based bias-correction technique and compares the results with a copula-based bias-correction approach and the traditional quantile mapping (QM) to correct the daily multi-model ensemble short- to medium-range rainfall forecast products of the India Meteorological Department (IMD-MME) in the Hirakud Reservoir catchment in India. The copula and eKSOM outperformed the raw IMD-MME and QM rainfall across all lead times. While both the copula and eKSOM-based bias-corrected forecasts could satisfactorily provide the temporal pattern of the observed rainfall, the corresponding eKSOM-based forecasts presented better skills in capturing seasonality in the observed rainfall. The streamflow at one- to five-day lead times are simulated by forcing a suitable hydrological model, MIKE11 NAM (Nedbør Afstrømnings Model)-HD (Hydrodynamic) with the raw and bias-corrected rainfall inputs. The overall performance evaluation reveals significant improvement in both the rainfall and streamflow forecasts by copula and eKSOM, with the latter performing most accurately at higher lead times.

KEYWORDS:

• bias correction
• copula
• eKSOM
• ensemble
• rainfall forecasting
• quantile mapping

Editor A. Fiori; Associate Editor G. Mascaro

Editor A. Fiori; Associate Editor G. Mascaro

Acknowledgements

This study was organized as part of the sub-project “Impact of Climate Change on Flood Risk” conducted under the Center of Excellence (CoE) in Climate Change studies activity established at IIT Kharagpur and funded by Department of Science and Technology (DST), Government of India. The authors thank the DST for their financial support, and the India Meteorological Department (IMD) and Central Water Commission (CWC), Government of India, for providing the datasets for this research. The authors also thank the anonymous reviewers for their insightful comments to improve the manuscript.

Disclosure statement

No potential conflict of interest was reported by the authors.

Data availability statement

All data, models, and code generated or used during the study appear in the submitted article.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/02626667.2022.2109972