Dissolved organic carbon (DOC) constitutes one of the largest reduced carbon reservoirs in the ocean and plays a pivotal role in global carbon cycling and climate regulation. How this vast carbon reservoir responds to ongoing and future climate change remains poorly constrained, due to limited long-term observations and inadequate representation of DOC cycling processes in Earth system models. To address this gap, we develop machine learning models trained on global DOC observations, combined with physico-chemical and biological predictors derived from six CMIP6 Earth system models, to project DOC concentrations under both historical and future climate scenarios. This presentation will highlight projected changes in the marine DOC reservoir under future climate scenarios, with a focus on variations in reservoir size, regional patterns of DOC response, and their broader implications for marine carbon cycling and climate feedbacks.