Assimilating all available observations in numerical models may lead to deterioration of the analysis. Ensemble Forecast Sensitivity to Observations (EFSO) is a method that helps to identify all such observations which benefit the analyses. EFSO has never been tested in an ocean data assimilation system because of a lack of robust formulation of a squared norm against which beneficiality of observations can be estimated.
Here, we explore the efficacy of EFSO in the ocean data assimilation system that comprises the ocean model, Regional Ocean Modeling System (ROMS), coupled to the assimilation system Local Ensemble Transform Kalman Filter (LETKF), collectively called LETKF- ROMS, in the Bay of Bengal by envisaging a novel squared norm. The Bay of Bengal is known for its higher stratification and shallow mixed layer depth. In view of baroclinicity representing the stratification of the ocean, we use the modulus of the baroclinic vector as the squared norm to evaluate forecast errors in EFSO.
Using this approach, we identify beneficial observations. Assimilating only the beneficial observations greatly improves the ocean state. We also show that the improvements are more pronounced in the head of the Bay of Bengal where stratification is much higher compared to the rest of the basin.
Though this approach doesn’t degrade the ocean state in other regions of the Indian Ocean, a universal squared norm is needed that can be extended beyond the Bay of Bengal basin.