Search

Seasonal precipitation and temperature predictions with global climate models that are developed based on the ocean–atmosphere interactions, contribute to the water resources management and hazard mitigation. To date, multi-model ensemble seasonal climate prediction systems, such as North American multi-model ensemble (NMME), are an effective way to provide useful forecast information a few months ahead especially over regions with strong ocean–atmosphere coupling. Previous studies have evaluated the skill of NMME hindcasts worldwide, however, whether the NMME real-time forecasts perform as well as the hindcasts and how the decadal variability in ocean-atmospheric teleconnections affect the prediction skill remain unclear. In this paper, based on precipitation and temperature datasets from nine models of NMME, the evaluation of forecast skills during hindcast (1982–2010) and real-time forecast (2011–2020) periods is carried out in the Yangtze River basin over China. Results show that although selecting an appropriate time frame for the calculation of climatology can reduce errors of real-time prediction, but the real-time prediction skills are lower than hindcast skills in the Yangtze River basin, with anomaly correlation decreased by 14–51% (38–75%) and error increased by 30–31% (51–55%) for seasonal precipitation (temperature) predictions up to the sixth lead-season, and the skill decrease is larger at longer leads. The failure in representing the decadal variations of ocean-atmospheric teleconnection (especially the association with Indian Ocean surface temperature) during the real-time forecast period can partly explain the decline in the prediction skills. Our findings suggest that a better grasp of decadal variability is needed to improve the real-time climate predictions. [ABSTRACT FROM AUTHOR]