Mining global soil carbon datasets: can modern machine learning uncover the missing pieces of process-based models?

Abstract

The future of terrestrial soil carbon stocks plays a crucial role in climate change prediction. Modern machine learning techniques are now widely applied in soil science to predict the spatial distribution of soil properties from observational data. Beyond prediction, the use of machine learning as a data-mining tool offers a promising pathway for improving soil carbon modelling and refining projections of climate–carbon feedbacks. In this paper, we review recent advances in the application of machine learning to global soil carbon modelling as a data-mining tool and highlight its potential to drive an iterative feedback loop that improves the representation of soil carbon dynamics in Earth System Models.