Impact of upper ocean processes and air-sea fluxes on seasonal SST biases over the tropical Indian Ocean in the NCEP Climate Forecasting System

Abstract

The present study aims to examine the role of air–sea interactions and upper ocean processes in determining the tropical Indian Ocean (TIO) seasonal sea surface temperature (SST) bias in Climate Forecast System version 1 (CFSv1) and version 2 (CFSv2) free runs. CFSv1 displayed dipole like east–west SST bias over the equatorial Indian Ocean from boreal summer to winter and is consistent with errors (bias) in surface winds and upper ocean advection. Large zonal gradients in sea level pressure (SLP) bias and the associated surface wind biases are primarily responsible for the upper ocean current bias. However, over the southern Indian Ocean and parts of Arabian Sea, strong bias in heat flux and mixed layer depth (MLD) have mainly contributed for the SST biases in CFSv1. Equatorial current system is better represented in CFSv2 compared to CFSv1. Improvement in the representation of land-surface processes appears to be contributing towards improving atmospheric circulation and SLP gradients in CFSv2, which may be responsible for the improved ocean circulation. Importantly, east–west dipole like SST bias prevalent in CFSv1 is absent in CFSv2. However, there is a prominent systematic basin-wide TIO cold SST bias in CFSv2. Large biases in surface heat flux (net negative bias) and MLD (deeper) are mainly responsible for SST biases in CFSv2. Negative net heat flux bias in CFSv2 is primarily due to specific humidity bias-induced excess latent heat flux (LHF). Deepening of MLD is mainly due to strong convective mixing, a resultant of anomalous LHF release, which in turn leads to negative SST bias. Models comparison reveals that although representation of SST in CFSv2 is better than in CFSv1, it is essential to improve further the equatorial ocean dynamics and off-equatorial thermodynamics in the form of moist processes and radiative parameterization in order to reduce SST bias in CFSv2.