Hyperspectral remote sensing for mapping species and community composition in a diverse tropical forest

Background/Question/Methods

Understanding the main drivers of community compositional change at different spatial scales is an important goal in ecology, but the limited spatial extent of much ecological field work hampers our ability to explore broad-scale patterns in canopy structure, function, and biodiversity.  However, recent developments in airborne hyperspectral remote sensing (imaging spectroscopy) combined with advances in classification methods have made it possible to remotely identify crowns of individual tree species, even within highly diverse tropical forest canopies.  We used high-fidelity airborne imaging spectroscopy to identify crowns of 23 canopy tree species across the diverse tropical forest of Barro Colorado Island (BCI) in Panama.  We then used these species maps along with detailed elevation, soil, and forest history maps to identify the primary gradients of tree species turnover and the primary environmental drivers of this community turnover across the island. 

Results/Conclusions

We found that our remote species classification model was able to detect the 23 focal species with producer’s accuracies 68–97% at the pixel level.  The mapped crowns allowed us to identify the main axes of community turnover across the island and the primary environmental drivers of this turnover, revealing a primary community compositional gradient corresponding to variation in slope and geological substrate.  Our results demonstrate the ability of airborne remote sensing technology to identify large numbers of tree species within a diverse tropical forest canopy, opening these ecosystems to examination at broad spatial scales not previously possible.