Journal
REMOTE SENSING OF ENVIRONMENT
Volume 150, Issue -, Pages 120-131Publisher
ELSEVIER SCIENCE INC
DOI: 10.1016/j.rse.2014.04.026
Keywords
LiDAR; High spatial resolution multispectral imagery; Tree species abundance modeling; Boosted regression trees; Biotic versus abiotic environmental factors
Funding
- Ontario Centre of Excellence for Earth and Environmental Technologies
- Natural Sciences and Engineering Research Council (NSERC)
- Premier's Research Excellence Award (PREA)
- Forest Research Partnership (Tembec, Inc.)
- Forest Research Partnership (Ontario Ministry of Natural Resources)
- Forest Research Partnership (Canadian Forest Service)
- European Research Council [233399]
- European Research Council (ERC) [233399] Funding Source: European Research Council (ERC)
Ask authors/readers for more resources
Tree species display different abundance patterns over the landscape due to a number of hierarchical factors, all of which have implications when modeling their distribution. While climate is often the primary driver for global to regional scale tree species distributions, modeling of presence and abundance patterns at finer scales, and in landscapes with less topographic variation may require predictors that capture biotic processes and local abiotic conditions. Proxies for biotic and disturbance processes may be captured by a combination of multispectral remote sensing and light detection and ranging (WAR) data. LiDAR data have shown great potential for capturing three-dimensional (3D) characteristics of the forest canopy and a number of these characteristics may have strong relationships with drivers of local tree species distributions. The objective of this study was to investigate the importance of remote sensing derived variables related to biotic and disturbance processes in predicting fine-scale abundance patterns of several dominant tree species in a mixed mature forest in the Great Lakes-St. Lawrence Forest Region, Ontario, Canada. Boosted regression trees, an ensemble classification and regression algorithm, was used to compare tree species abundance models that included LiDAR derived topographic variables with models that included spectral and LiDAR derived topographic and vegetation variables. Average model fit (rescaled Nagelkerke R-2) and predictive accuracy (correlation) improved from 0.12 to 0.63 and 0.25 to 0.71, respectively, when spectral and LiDAR derived vegetation variables were included in the tree species abundance models. This indicates that these variables capture some of the variance in local tree species' abundance distributions generated by biotic and disturbance processes in a landscape with limited topographic and climatic variation. Decreased model performance at higher tree species' abundances additionally suggests that our models do not capture all of the local drivers of tree species' abundance. Variables related to historical and current silvicultural practices may be missing. (C) 2014 Elsevier Inc. All rights reserved.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available