Frontiers in Environmental Science (Oct 2022)
Multi-model seasonal prediction of global surface temperature based on partial regression correction method
Abstract
The increased climate change is having a huge impact on the world, with the climatic change sensitive and vulnerable regions at significant risk particularly. Effective understanding and integration of climate information are essential. It helps to reduce the risks associated with adverse weather conditions and to better adapt to the impacts of climate variability and change. Using the hindcast data from Japan Meteorological Agency/Meteorological Research Institute (JMA/MRI) coupled prediction system version 2 (JMA/MRI-CPS2), National Centers for Environmental Prediction (NCEP) Climate Forecast System model version 2 (CFSv2), and Canadian Centre for Climate Modeling and Analysis (CCCma) Coupled Climate Model, versions 3 (CanCM3) seasonal prediction model systems, the performance of seasonal prediction for global surface temperature in boreal summer and winter is comprehensively evaluated and compared for 1982–2011 from the perspective of deterministic and probabilistic forecast skills in this study, and a partial regression correction (PRC) method is introduced to correct seasonal predictions. The results show high prediction skills in the tropics, particularly in the equatorial Pacific, while poor skills on land. In general, JMA/MRI-CPS2 has slightly better prediction performance than CFSv2 and CanCM3 in the tropics. CFSv2 is generally superior to JMA/MRI-CPS2 and CanCM3 in the extratropical northern hemisphere and East Asia, especially for the abnormal low winter temperature prediction in East Asia. CanCM3 shows good deterministic forecast skills in extra-tropics but performs slightly worse in probabilistic forecasting. Based on the respective strengths of each seasonal prediction model, an ensemble forecast correction with observational constraint is implemented by partial regression, and the improvement of skills in ensemble predicting has been analyzed. Compared to the simple multi-model ensemble (MME), the correction improved the global-average temporal correlation coefficient and multi-year mean anomaly correlation coefficient by about 0.1 and 0.13, respectively. The validation tests indicate that the corrected ensemble forecast has higher ranked probability skill scores than that of the MME, which is improved by more than 0.06 in the tropics. Meanwhile, when the training period is sufficiently long, it may have the potential for future seasonal temperature predictions from the perspective of stable zonal partial regression coefficients.
Keywords
