Identification of ecological zonation in tidal wetlands from high and medium-resolution satellite imagery

Forecasting the temporal and spatial evolution of coastal wetlands is a complex and challenging endeavour, further complicated by shifting climatic conditions. From a coastal management perspective, however, it remains essential to anticipate local vulnerabilities and the potential compromise of ecological functions. This study aims to assess the skill of subpixel and pixel imagery classification algorithms in sensing the zonation of wetlands through satellite imagery by testing in wetland areas, using high spatial resolution imagery from Unmanned Aerial Vehicles (UAVs) and high and medium spatial resolution satellite imagery and identifying the related challenges. The research was conducted in one of the most important coastal lagoons of Portugal, Ria Formosa Lagoon. Hard and soft Random Forest regression algorithms were employed to estimate the zonation of marsh plants, the former applied to UAV data (centimetric pixel size) and the latter to high and medium resolution satellite imagery (WV-2 & -3 and Sentinel, respectively, metric to decametric pixel size). Employing a gradually increasing pixel size allowed to identify error propagation during the passage from pixel to sub-pixel estimators and lower resolutions. The obtained results provide important insights to the barriers and opportunities related to varying imagery sources, carrying an important message to local managers. When discussed within the context of dominant natural and human drives, the developed maps, along with methodology and monitoring, provide valuable scientific insights into vegetation succession in a mesotidal system. More importantly, they serve as essential tools for local coastal decision-makers in identifying priorities for strategic landscape conservation planning and ensuring the sustainability of ecological function as the carbon sequestration within ecosystems.

Acknowledgements: This study contributes to the projects DEVISE (https://doi.org/10.54499/2022.06615.PTDC) and C-Land (CEXC/4647/2024), both funded by the Fundação para a Ciência e a Tecnologia, and to project TraceLands (ID PP0090200), funded by the European Space Agency.