Tropical Pacific subseasonal forecast: the role of mean state biases, model errors, and ocean data assimilation

The tropical Pacific region plays an important role in modulating the global climate system with its prevailing pattern of the warm pool in the western Pacific and cold tongue in the eastern Pacific, the dominant climate modes in different timescales, like ENSO, MJO, and their teleconnections. These modes of variability also are major sources of prediction skill on subseasonal timescales. With most of the predictability on subseasonal timescales in the tropics mainly coming from the MJO and ENSO, which are related to the upper ocean features and air-sea interaction, this study focuses on the subseasonal forecast of the upper ocean and air-sea interaction evolution over Tropical Pacific. In this study, we examine how assimilating in-situ ocean observations would influence the initial ocean state estimate and the subseasonal forecast. We analyze the subseasonal forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF) Integrated Forecast System (IFS) where the ocean states were initialized using different Observation Sensitivity Experiment (OSE) analyses. We quantify the influence of assimilating the ocean in-situ observations on the development of SST biases on subseasonal timescales. We find that the initial ocean state with the in-situ observations assimilated tends to have warmer SST over the equatorial central to eastern Pacific and colder SST in the equatorial western Pacific and the far-eastern Pacific east of the Galapagos Islands. While the initial difference is not statistically significant, the pattern grows in the forecast, which leads to a weaker cold tongue bias later in the forecast with the assimilation of the in-situ observations. While the initial subsurface ocean density structure (i.e., mixed layer depth, MLD) with the assimilation of the in-situ observations has a lesser bias of the MLD, this difference does not sustain in the forecast. Instead, the MLD is dominated by the model errors in the forecast, which have thin MLD biases over the tropical Pacific region, even though the initial states have thick MLD biases. Based on these results, we emphasize how variability and model biases in surface and subsurface ocean features influence the air-sea interaction processes and therefore the subseasonal forecast in the tropical Pacific region.