Estimating species abundance in a northern temperate forest using spectral mixture analysis

Effective, reliable methods for characterizing the spatial distribution of tree species through remote sensing would represent an important step toward better understanding changes in biodiversity, habitat quality, climate, and nutrient cycling. Towards this end, we explore the feasibility of using spectral mixture analysis to discriminate the distribution and abundance of two important forest species at the Bartlett Experimental Forest, New Hampshire. Using hyper-spectral image data and simulated broadband sensor data, we used spectral unmixing to quantify the abundance of sugar maple and American beech, as opposed to the more conventional approach of detecting presence or absence of discrete species classes. Stronger linear relationships were demonstrated between predicted and measured abundance for hyperspectral than broadband sensor data: R-2 = 0.49 (RMSE = 0.09) versus R-2 = 0. 16 (RMSE = 0.19) for sugar maple; R-2 = 0.36 (RMSE = 0.18) versus R-2 = 0.24 (RMSE 0.33) for beech. These results suggest that spectrally unmixing hyperspectral data to estimate species abundances holds promise for a variety of ecological studies.