Seasonal predictions initialised by assimilating sea surface temperature observations with the EnKF

Abstract

This study demonstrates that assimilating SST with an advanced data assimilation method yields prediction skill level with the best state-of-the-art systems. We employ the Norwegian Climate Prediction Model (NorCPM)-a fully-coupled forecasting system-to assimilate SST observations with the ensemble Kalman filter. Predictions of NorCPM are compared to predictions from the North American Multimodel Ensemble (NMME) project. The global prediction skill of NorCPM at 6-and 12-month lead times is higher than the averaged skill of the NMME. A new metric is introduced for ranking model skill. According to the metric, NorCPM is one of the most skilful systems among the NMME in predicting SST in most regions. Confronting the skill to a large historical ensemble without assimilation, shows that the skill is largely derived from the initialisation rather than from the external forcing. NorCPM achieves good skill in predicting El Niño-Southern Oscillation (ENSO) up to 12 months ahead and achieves skill over land via teleconnections. However, NorCPM has a more pronounced reduction in skill in May than the NMME systems. An analysis of ENSO dynamics indicates that the skill reduction is mainly caused by model deficiencies in representing the thermocline feedback in February and March. We also show that NorCPM has skill in predicting sea ice extent at the Arctic entrance adjacent to the north Atlantic; this skill is highly related to the initialisation of upper ocean heat content.

Impact of Ocean and Sea Ice Initialisation On Seasonal Prediction Skill in the Arctic

“…(2) Increasing the error of hydrographic observations underneath the sea ice. In the North Atlantic, the prediction skill of our ocean only DA is worse than assimilating only the nearest SST observation in the same system (Wang et al, ). We suspect that the skill reduction in our system is due to a large localization radius for hydrographic data here.…”

“…In the GIN Seas, which are determined by the export through Fram Strait and the atmosphere, we could not derive skillful retrospective predictions with DA of ocean data alone. Note that Wang et al () obtained better prediction skill in SIE in the Labrador, GIN, and Barents Seas despite worse prediction skill in SST. The authors used the same system (NorCPM) but assimilated only SST data.…”

“…In retrospective prediction experiments we assess the impact of our assimilation approaches on the seasonal prediction skill in the Arctic. Retrospective predictions assume perfect historical external forcing, as standard when assessing the impact of assimilation approaches on the seasonal prediction skill (Bushuk et al, ; Wang et al, ). The difference between historical simulations and retrospective predictions provide an estimate of the skill that can be gained with the same system in real predictions.…”

Relating model bias and prediction skill in the equatorial Atlantic

“…NorCPM demonstrated good skill in controlling the upper ocean heat content in the equatorial and north Pacific, the north Atlantic subpolar gyre region and the Nordic Seas seas by assimilating surface temperature anomalies (SSTAs, Counillon et al 2016). In Wang et al (2019) NorCPM with assimilation of SST reaches skill comparable to the top-performing prediction systems of the North American Multi Model ensemble (NMME, Kirtman et al 2014) in most regions, but performance in the tropical Atlanticwere the model has large biases-were found to be poor. One can thus expect anomaly coupling to be particularly beneficial there, by correcting the model bias, improving the representation of the dynamics and enhancing the prediction skill.…”

“…The Norwegian Climate prediction model (Counillon et al 2014(Counillon et al , 2016Wang et al 2019) assimilates observations with the EnKF. Here, we assimilate SST and hydrographic profiles as described in Counillon et al (2014Counillon et al ( , 2016 and Wang et al (2017).…”