Abstract
Restoring and conserving seagrass ecosystems are considered a climate solution due to their exceptional ability to store carbon in their sediments. However, restoration and financing efforts are held back by the inherent challenges of quantifying and monitoring carbon storage in sediments and the highly variable nature of seagrass carbon stocks globally. This research explores the application of machine learning (ML) models, using Earth Observation (EO) derived datasets, to estimate carbon stocks in the seagrass species Zostera marina , across its northern temperate range. A dataset of 176 Z. marina seagrass carbon stocks from 18 eco-regions was collated along with open-source data on 21 associated environmental variables, with the aim of developing a framework for estimating sediment carbon stocks and better understanding the variables that contribute to variability in storage. Ensemble decision trees were the best performing model able to predict nearly 40% of the variability in carbon stocks within a seagrass bed with human modification (e.g. population density and infrastructure), exposure, tidal range and wave height contributing most. Whilst the model performance reflects the complexity and uncertainty inherent in ecological systems, this research demonstrates the potential of ML approaches to estimate seagrass carbon stocks at a multi-regional scale and highlights key areas for future improvement.