Bias correction methods for decadal sea-surface temperature forecasts

Abstract

Two traditional bias correction techniques: (1) systematic mean correction (SMC) and (2) systematic least-squares correction (SLC) are extended and applied on sea-surface temperature (SST) decadal forecasts in the North Pacific produced by Climate Forecast System version 2 (CFSv2) to reduce large systematic biases. The bias-corrected forecast anomalies exhibit reduced root-mean-square errors and also significantly improve the anomaly correlations with observations. The spatial pattern of the SST anomalies associated with the Pacific area average (PAA) index (spatial average of SST anomalies over 20°–60°N and 120°E–100°W) is improved after employing the bias correction methods, particularly SMC. Reliability diagrams show that the bias-corrected forecasts better reproduce the cold and warm events well beyond the 5-yr lead-times over the 10 forecasted years. The comparison between both correction methods indicates that: (1) prediction skill of SST anomalies associated with the PAA index is improved by SMC with respect to SLC and (2) SMC-derived forecasts have a slightly higher reliability than those corrected by SLC.

• bias correction for decadal forecasts
• CMIP5 decadal forecasts
• CFSv2
• decadal forecast skill
• model drift on decadal scale

5. Acknowledgements

This research was funded by NOAA grant NA09OAR4310137. We gratefully acknowledge National Centers for Environmental Prediction (NCEP) for the CFS v2 model made available to COLA. We also acknowledge S. Saha of EMC/NCEP/NWS/NOAA for making the CFS v2 forecasts available to us. We thank Dr. Tim DelSole of GMU/COLA for helpful discussions. We particularly wish to thank W. Lapenta and L. Uccellini for enabling the collaborative activities.