2017 ESA Annual Meeting (August 6 -- 11)

COS 27-8 - Modeling multi-scale plant species richness in a Piedmont, NC landscape using LiDAR-hyperspectral remote-sensing

Tuesday, August 8, 2017: 10:30 AM
B116, Oregon Convention Center
Christopher Hakkenberg, Curriculum for the Environment and Ecology, University of North Carolina, Chapel Hill, NC, Kai Zhu, Department of Biology, University of Texas, Arlington, Arlington, TX, Robert K. Peet, Department of Biology, University of North Carolina at Chapel Hill, Chapel Hill, NC and Conghe Song, Geography, UNC Chapel Hill
Background/Question/Methods

Recognition of the significance of vascular plant species diversity for habitat conservation and forest ecosystem function has propelled efforts to map and monitor its spatial distribution at landscape scales. This study employs spatially-nested field plots in conjunction with active and passive remotely-sensed data from the Goddard LiDAR, hyperspectral, thermal (G-LiHT) airborne sensor to map multi-scale vascular plant species richness in a compositionally- and structurally-complex Piedmont forest landscape in NC, USA. Nonparametric models employ feature-selected derived remotely-sensed variables, aggregated at multiple spatial resolutions, to predict wall-to-wall vascular plant species richness at 0.01m2, 0.1m2, 1m2, 10m2, 100m2, 400m2, and 900m2 scales. In addition to accuracy assessment of nonparametric models and spatially-explicit uncertainty maps, we explicitly assess the scale-dependence of remotely-sensed predictors in relation to characteristic scales of stand components in the study site, including plant size and density as well as canopy gaps and understory growth patterns.

Results/Conclusions

Based on 10-fold cross-validation, feature-selected models exhibit a general pattern of increasing explanatory power with spatial scale, accounting for 15-70% of variance in plant richness across scales. Post-hoc parametric tests reveal a similar trend among individual predictors, most of which only significantly predict plant richness above a threshold scale sufficient to encompass heterogeneity in the resource environment, capture the ground footprint of the largest individuals, and subsume geolocational errors of remotely-sensed data. Results confirm the predominant role of topography, forest structural complexity, and spectral variability for predicting landscape patterns in vascular plant richness. In addition to applications ranging from landscape management to ecosystem process modeling, results offer insights into the scale-dependence among remotely-sensed predictors of plant richness and multi-scale patterns in plant diversity in the North Carolina Piedmont.