Towards rapid assessments of tree species diversity and structure in fragmented tropical forests: A review of perspectives offered by remotely-sensed and field-based data

Abstract

Citation: Ganivet E, Bloomberg M (2019) Towards rapid assessments of tree species diversity and structure in fragmented tropical forests: A review of perspectives offered by remotely-sensed and field-based data. Forest Ecology and Management, 432:40-53.

Regional mapping of spekboom canopy cover using very high resolution aerial imagery

“…Hyperspectral data provide a finer spectral resolution and potentially more informative description of distinguishing spectral features compared to multi-spectral data and have been used quite successfully for species mapping in a number of cases (Ganivet & Bloomberg 2019;Nguyen et al 2019). It is, however, seldom used for landscape or country scale mapping due to associated costs and computational requirements (Adjorlolo et al 2012;Ganivet & Bloomberg 2019). Higher spatial resolutions provide more detailed plant information and can improve classification accuracy.…”

“…Supervised classification is typically used for species mapping (Ganivet & Bloomberg 2019). In this approach, objects (image pixels or regions) with known class labels are used to build ("train") a classifier that can predict labels for novel data.…”

“…the dependence of reflectance on view-illumination geometry) are more pronounced when working with large-area mosaics of imagery, captured at varying times (Fassnacht et al 2016;Gehrke & Beshah 2016;Lelong et al 2008). Habitat, ecotype and phenological variations also increase over spatial and temporal scales and add uncertainty to the problem of classifying species (Fassnacht et al 2016), especially in diverse and heterogeneous habitats (Bradley 2014;Ganivet & Bloomberg 2019). In the vast majority of cases, species mapping studies are applied to small areas (often consisting of a single image) (Boyden et al 2007;Johansen et al 2007;Kollár, Vekerdy & Márkus 2013;Mehner et al 2004;Ouyang et al 2011) and thus do not need to comprehensively address these sources of variation (Fassnacht et al 2016).…”

“…A statistical pattern recognition approach to image classification has been used widely in remote sensing (Fassnacht et al 2016;Ganivet & Bloomberg 2019;Vaz et al 2018). In classification, statistical methods are used to partition the feature space into class regions in a training phase.…”

“…Hyperspectral data is reported to provide superior classification performance due to its finer spectral resolution and ability to capture subtle spectral differences between species (Baldeck et al 2015;Clark & Roberts 2012;Clark, Roberts & Clark 2005). Use of hyperspectral data is, however, limited by its cost and processing requirements (Adjorlolo et al 2012;Ganivet & Bloomberg 2019;Vaz et al 2018). Multi-spectral imagery is increasingly being used due to its availability, coverage, high spatial resolution and relative cost-effectiveness.…”