Trends and spatial variation of oceanic dimethyl sulfide under a warming climate revealed by an artificial neural network model

Abstract. Marine dimethyl sulfide (DMS), a climatically active gas generated through microbial degradation of dimethyl sulfoniopropionate (DMSP), plays a key role in the Earth’s climate system by modifying its radiation budget. However, the sea-to-air flux and future variations under climate change are still uncertain. Simulations from Earth System Models (ESMs) provide divergent trends. Here, we developed an artificial neural network (ANN) model trained using DMS observations and eight observational environmental parameters, along with model parameters extracted from the simulations of CESM2-WACCM to predict variations of DMS concentrations and sea-to-air flux for both historical (1850–2014) and SSP5-8.5 scenario (2015–2100). Our simulation indicates that DMS concentrations will generally decline by the end of this century. Specifically, from 2015 to 2050, the DMS concentrations are projected to decrease at a rate of 0.40±0.13 % per decade. From 2050 to 2100, the rate of decrease is expected to accelerate to 0.89±0.08 % per decade. The sea-to-air flux of DMS exhibits a non-monotonic trend. It is projected to increase at a rate of 0.51±0.16 % per decade from 2015 to 2050. However, from 2050 to 2100, the flux is expected to decrease at a rate of 0.37±0.11 % per decade. We further explore the attribution of DMS changes by running a series of sensitivity tests. We find that elevated sea surface temperature (SST) and photosynthetically active radiation (PAR), along with nutrient depletion, are projected to lead to the decline in DMS concentrations by the end of this century. Furthermore, our geospatial analysis indicates that mixed layer depth (MLD) emerges as the predominant driver in the Southern Ocean, and nutrient-dependent effects strongly correlate with DMS in the open seas (trades and westerlies). Our findings suggest that site-specific modeling schemes are needed to accurately model DMS dynamics.