Geospatial patterns of carbon storage in relation to protection status and road infrastructure in an insular forest landscape

Forests play a crucial role in climate change mitigation through carbon storage. Nevertheless, these ecosystems face increasing threats from human activities, such as infrastructure development and Land Use/Land Cover (LULC) changes. To date, limited research has focused on understanding how roads impact carbon stocks in forests, and how this relation is influenced by protection regimes, especially on islands. This study on the island of Cyprus aims to assess Machine Learning (ML) techniques for estimating key forest variables such as Canopy Cover (CC) and to analyze the spatial dynamics of carbon stocks around roads in relation to LULCs and protection regimes. Remote Sensing (RS) data, including Landsat imagery and orthophotos, are combined with ML to create an ensemble model for detailed LULC classifications. The Integrated Valuation of Ecosystem Services and Trade-offs (InVEST) tool is utilized to estimate carbon stocks for each LULC and statistical analysis is used to evaluate interactions between forests, roads, and protection regimes. The analysis revealed that protected sites store significantly 17 % more carbon than unprotected areas whilst proximity to roads exhibits complex effects on carbon stocks, with varying patterns depending on the protection status. The ensemble model outperforms individual models, achieving 92 % accuracy and a kappa of 0.91, showing the advantages of combining algorithms for more robust predictions. The research highlights the impact of integrating ML with ecosystem service models to improve understanding of interactions between roads, LULC, and forests. It also emphasizes the importance of conservation and roadside vegetation management for ecosystem resilience and sustainable carbon storage.