Remote sensing-based aboveground carbon estimation to assess edge effects in Atlantic forests landscapes

Abstract

Forests and roads are both fundamental to sustainable development, yet their interaction produces complex ecological consequences. Understanding how road infrastructure alters forest structure is critical to mitigating long-term degradation in biodiversity-rich regions. However, the effects of road infrastructure on aboveground carbon (AGC) remain understudied—particularly when considered alongside structural forest attributes such as canopy height or tree density. This study aimed to quantify and characterize edge effects on AGC stocks and forest structure near various types of road infrastructure, employing a combination of remote sensing and machine learning techniques. An ensemble learning model is trained and applied using high-resolution LiDAR and multispectral imagery from satellite and drone platforms to estimate AGC. These predictions are then used to evaluate spatial edge patterns through cumulative diagnostics. The analyses revealed that road-induced edge effects varied by road type and distance, exhibiting non-uniform and non-linear patterns. AGC and canopy structure responded through abrupt, delayed, or irregular changes. The ensemble model outperformed individual learners in predictive accuracy, achieving a high R² of 0.75 during testing. These findings demonstrate that road-induced edge effects are neither uniform nor monotonic, but rather unfold through complex and delayed structural shifts. This nuanced understanding can support more accurate detection of degradation patterns and guide managers in identifying priority edge zones for targeted mitigation and spatial planning.