A Comparison Study of the Bay of Bengal Summer Monsoon Onset Simulated in Atmosphere-Only and Air-Sea Coupled Experiments With FGOALS-f3-L Model

The onset timing of the Bay of Bengal summer monsoon (BOBSM) has critical influences on the seasonal march of the Asian summer monsoon, but the skills of its simulation and prediction are limited for climate models. This study examined the ability of the finite-volume version 3 of the Flexible Global Ocean–Atmosphere-Land system model (FGOALS-f3-L) to simulate BOBSM onset dates and investigated the potential causes of the largely delayed onset in air-sea coupled simulations by comparing the atmosphere-only experiment datasets derived from the CMIP6 atmospheric model inter-comparison project (AMIP) simulations and the air-sea coupled experiment datasets derived from the CMIP6 Historical simulations. The AMIP simulations can almost accurately reproduce the climatological onset date, but it delays by about four pentads in the Historical simulations. However, the AMIP and Historical simulations can reasonably capture the abrupt transformations of precipitation and atmospheric circulations during the BOBSM onsets. In the Historical simulations, the overestimated surface winds in the preceding winter increase surface latent heat loss over the Arabian Sea and BOB, leading to cold sea surface temperature (SST) biases that persist into spring. Meanwhile, the overestimated land surface sensible heat flux over the Indochina Peninsula in April introduces strong northwestlies over the Eastern BOB, producing a strong anticyclone over the BOB and the resultant SST cooling because of induced upwelling by offshore ocean currents. The atmospheric anticyclone and cold SST bias are unfavourable for the convection initiation over the BOB, resulting in delayed BOBSM onset. For the interannual variability, both AMIP and Historical simulations can capture the significant positive correlation between the onset dates and ENSO, indicating that the FGOALS-f3-L has the potential ability for BOBSM onset prediction.