Journal of Vegetation Science && (2016) Forest structure as a predictor of tree species diversity in the North Carolina Piedmont Christopher R. Hakkenberg, Conghe Song, Robert K. Peet & Peter S. White Keywords Biodiversity; Diversity surrogates; Forest inventory and analysis; Forest structure; Gini Coefficient; North Carolina Piedmont; Support vector regression; Tree species diversity; Weibull function Nomenclature USDA, NRSC (2016) Received 22 January 2016 Accepted 3 June 2016 Co-ordinating Editor: Duccio Rocchini Hakkenberg, C.R. (corresponding author, hakkenberg@unc.edu) 1 , Song, C. (csong@email.unc.edu) 1,2 , Peet, R.K. (peet@unc.edu) 1,3 , White, P.S. (peter.white@unc.edu) 1,3 1 Curriculum for the Environment and Ecology, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599-3135, USA; 2 Department of Geography, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599- 3220, USA; 3 Department of Biology, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599- 3280, USA Abstract Questions: Can forest structure significantly predict tree species diversity in the forests of the North Carolina Piedmont? If so, which structural attributes are most correlated with it, and how effective are they when used in concert in a generalized predictive model of tree species diversity? Location: North Carolina Piedmont, USA. Methods: Using a set of geographically distributed Forest inventory and analy- sis (FIA) plots (n = 972), we analysed Spearman correlations between 15 mea- sures of forest structure and five indices of tree species diversity. We predict tree species diversity based on structural predictors using support vector regression (SVR) models, assessing model fit via ten-fold cross-validation. Results: Results show a consistent and significant relationship between most structural attributes and indices of tree species diversity. Among all structural predictors, maximum height, basal area size inequality (basal area Gini coeffi- cient) and skewness of the basal area distribution (Weibull shape) exhibited the strongest correlations with indices of tree species diversity. Predictive SVR mod- els trained solely with structural attributes explained 44–61% of the variance in tree species diversity in the full Piedmont data set, and 22–71% of the variance in subsets defined by stand origin and forest type. Conclusions: Results confirm that forest structure alone was able to predict a substantial portion of the variance in tree species diversity without accounting for other known predictors of diversity in the North Carolina Piedmont, such as environment, soil conditions and site history. Beyond the theoretical implica- tions of unravelling primary patterns underlying tree species diversity, these findings highlight the empirical basis and potential for utilizing forest structure in predictive models of tree species diversity over large geographic regions. Introduction Concerns over global environmental change and biodi- versity loss have driven efforts to model the spatial distri- bution of taxonomic diversity (Ackerly et al. 2010; Hooper et al. 2012). Despite significant progress in recent decades in the modelling of large-scale patterns in biodi- versity, direct mapping is still greatly limited by techno- logical and cost constraints (Gaston 2000; Rodrigues & Brooks 2007; Asner & Martin 2009). To improve the esti- mation of the spatial distribution of diversity over large continuous areas, scalable proxy variables that link ground measurements with remotely sensed data must be identified and assessed (Kreft & Jetz 2007; Kane et al. 2010; He et al. 2015). Candidate variables should be remotely sensible, ecologically relevant, common in cur- rent vegetation inventory databases, as well as tempo- rally dynamic and scalable to larger landscapes (Turner et al. 2003; Anderson & Ferree 2010). This study investi- gates the utility of employing one such suite of candidate variables, namely those based on forest structure, to pre- dict tree species diversity over large areas of temperate forest. Forest structure reflects abiotic conditions and site his- tory, including competition and stochastic disturbance events, at multiple spatial scales that affect the three- dimensional distribution of biomass in a forest stand. Sev- eral potential mechanisms underlie the relationship 1 Journal of Vegetation Science Doi: 10.1111/jvs.12451 © 2016 International Association for Vegetation Science