Monitor the energy and carbon emissions of process-based models: ProcessC

• ProcessC , a multi-platform energy usage and carbon emission tracking program for process-based models was developed. • Machine-learning models consumed and released less energy (115-fold) and less carbon emissions (29397-fold) compared to the RZ-SHAW in simulating winter drainage. • Different modelling approaches result in drastically varying energy usage and carbon emissions; future studies should consider energy efficiency. • Reducing model training or calibration time and optimizing computation locations are crucial for sustainable modelling. Sustainable modelling to reduce carbon emissions from heavy computations is being adopted in machine learning communities. However, this concept has not been considered in process-based model communities, with carbon emissions rarely monitored and reported. This study developed ProcessC, a multi-platform program to monitor energy usage and carbon emissions in process-based models’ simulations. ProcessC was tested through a case study involving a process-based model, RZ-SHAW, for winter artificial drainage simulation. Results indicated that the RZ-SHAW model consumed 115 times more energy and released 29,397 times more carbon emissions compared to machine learning (ML) models for the winter artificial drainage simulation. The study suggests deploying computing systems in regions with low grid carbon intensity, choosing energy-efficient systems, and reducing simulation time as potential solutions for a more carbon-sustainable modelling. The findings from the current study urge the process-based community to commence considering and reporting carbon emissions in future modelling studies.