Prediction of Dimethylsulfide in Large Rivers Using Explainable Machine Learning and Online Monitoring

Dimethylsulfide (DMS) is a key climate-active gas and infochemical vital to global biogeochemistry. However, the driving factors and spatiotemporal variations of DMS in rivers remain unclear. This study collected and analyzed 3630 field DMS and water quality data from the Pearl River, South China. Results showed that measured DMS ranged from 0.01–2.88 nM, with an average of 0.47 nM. Unlike lakes and oceans, the relationship between DMS and chlorophyll (Chl-a) in rivers was linear models (mean R2 = 0.3) but not unimodal models. Path analysis, ExtraTrees model (test sets R2 = 0.74–0.82), and TreeSHAP identified water temperature and Chl-a as the primary direct drivers of DMS, with total nitrogen and total phosphorus indirectly affecting DMS through Chl-a. Prediction DMS based on online monitoring showed that DMS in the Pearl River increased and then decreased from January to December, exhibiting annual cyclic variations. Some prediction sites showed clear sinusoidal periodicity, but some data deviated from the peak values, exceeding the Pearl River average 3-fold. This study demonstrates the use of interpretable machine learning and online monitoring for riverine DMS prediction. It offers new insights into global sulfur cycling and provides a starting point and framework for freshwater DMS research.