Subseasonal-to-seasonal monsoon prediction in the Met Office coupled initialized forecasting system GloSea5

Predicting the circulation and precipitation features of the Asian monsoon on time scales of weeks to the season ahead remains a challenge for prediction centres. Current state-of-the-art models retain large biases, particularly dryness over India, which evolve rapidly from initialization and persist into centennial length climate integrations, illustrating the seamless nature of the monsoon problem.

Using the Met Office Unified Model (MetUM), which like other GCMs suffers dry biases and weakened flow compared to observed estimates, we present initial results from our MoES Indian Monsoon Mission collaboration. We use the coupled initialized Global Seasonal Prediction System GloSea5 to explore development of biases in the MetUM and their influence on prediction skill. Using a 14-year hindcast ensemble of integrations in which atmosphere and ocean components are initialized from May start dates, we assess the influence of mean state biases in the monsoon for the season ahead on the seasonal prediction skill of the MetUM in comparison to observations and other initialized coupled models.

In a comparison with hindcasts initialized at other start dates and long-term coupled integrations, our subsequent analysis focuses on short-time scale development of key regional errors, notably cold biases in the Arabian Sea and wet biases in the Western Equatorial Indian Ocean that are known to be detrimental to monsoon precipitation over India.

We also examine seasonal-scale relationships with ENSO: despite an appropriate negative response of the large-scale monsoon circulation to El Niņo warming, regional precipitation does not respond in the correct manner, a problem that we attribute to rapidly-developing coupled SST biases in the equatorial Pacific, incorrect placement of diabatic heating anomalies associated with ENSO and subsequent misplaced subsidence over the Indian Ocean region.

We outline initial case study results from correction and nudging experiments in which the coupled initialized system is pushed back towards the mean state.