Pantropical Modelling Of Canopy Functional Traits Using Sentinel-2 Remote Sensing Data

Tropical forest ecosystems are undergoing rapid transformation as a result of changing environmental conditions and direct human impacts. However, we cannot adequately understand, monitor or simulate tropical ecosystem responses to environmental changes without capturing the high diversity of plant functional characteristics in the species-rich tropics. Failure to do so can oversimplify our understanding of ecosystems responses to environmental disturbances. Innovative methods and data products are needed to track changes in functional trait composition in tropical forest ecosystems through time and space. This study aimed to track key functional traits by coupling Sentinel-2 derived variables with a unique data set of precisely located in-situ measurements of canopy functional traits collected from 2434 individual trees across the tropics using a standardised methodology. The functional traits and vegetation censuses were collected from 47 field plots in the countries of Australia, Brazil, Peru, Gabon, Ghana, and Malaysia, which span the four tropical continents. The spatial positions of individual trees above 10 cm diameter at breast height (DBH) were mapped and their canopy size and shape recorded. Using geo-located tree canopy size and shape data, community-level trait values were estimated at the same spatial resolution as Sentinel-2 imagery (i.e. 10 m pixels). We then used the Geographic Random Forest (GRF) to model and predict functional traits across our plots. We demonstrate that key plant functional traits can be accurately predicted across the tropicsusing the high spatial and spectral resolution of Sentinel-2 imagery in conjunction with climatic and soil information. Image textural parameters were found to be key components of remote sensing information for predicting functional traits across tropical forests and woody savannas. Leaf thickness ((R)2 = 0.52) obtained the highest prediction accuracy among the morphological and structural traits and leaf carbon content (R-2 = 0.70) and maximum rates of photosynthesis (R-2 = 0.67) obtained the highest prediction accuracy for leaf chemistry and photosynthesis related traits, respectively. Overall, the highest prediction accuracy was obtained for leaf chemistry and photosynthetic traits in comparison to morphological and structural traits. Our approach offers new opportunities for mapping, monitoring and understanding biodiversity and ecosystem change in the most species-rich ecosystems on Earth.