Subseasonal forecasts of precipitation over maritime continent in boreal summer and the sources of predictability

In this study, subseasonal precipitation forecast skills over Maritime Continent in boreal summer are investigated for the ECMWF and CMA models involved in the S2S Project. Results indicate that the ECMWF model shows generally superior forecast performances than CMA, which is characterized by lower errors and higher correlations compared with the observations. Meanwhile, ECMWF tends to produce wet biases with increasing lead times, while the mean errors of CMA are revealed to be approximately constant throughout lead times of 2-4 weeks over most areas. Besides, the temporal correlations between model outputs and observations obviously decrease with growing lead times, with a high-low distribution presented from north to south. In addition, the roles of large-scale drivers like ENSO and BSISO in modulating subseasonal precipitation forecast skills are also assessed in the models. Both ECMWF and CMA can reasonably capture the ENSO related precipitation anomalies for all lead times, while their capabilities of capturing BSISO related precipitation anomalies decrease with growing lead times, which is more obvious in CMA. The enhanced subseasonal precipitation forecast skills mainly respond to the BSISO associated precipitation variability. For most MC areas such as southern Indochina, western Indonesia, Philippines and the eastern ocean, the forecast skills of both ECMWF and CMA can be improved to a great extent by enhancing the capture of BSISO related precipitation anomalies, with the temporal correlations for both ECMWF and CMA increased by about 0.15 for lead times of 3-4 weeks. It provides an opportunity window for the models to improve precipitation forecasts on the subseasonal timescale.

Automated summary

This study investigates the subseasonal precipitation forecast skills over Maritime Continent in boreal summer for the ECMWF and CMA models. The results show that the ECMWF model generally outperforms the CMA model in terms of lower errors and higher correlations compared with observations. The ECMWF model tends to produce wet biases with increasing lead times, while the mean errors of the CMA model remain relatively constant throughout lead times of 2-4 weeks. The study also assesses the roles of large-scale drivers such as ENSO and BSISO in modulating subseasonal precipitation forecast skills, and finds that both models can reasonably capture the ENSO-related precipitation anomalies for all lead times, while their abilities to capture BSISO-related precipitation anomalies decrease with longer lead times, particularly in the CMA model.