Imaging spectroscopy predicts variable distance decay across contrasting Amazonian tree communities

Abstract

The forests of Amazonia are among the most biodiverse on Earth, yet accurately quantifying how species composition varies through space (i.e., beta‐diversity) remains a significant challenge. Here, we use high‐fidelity airborne imaging spectroscopy from the Carnegie Airborne Observatory to quantify a key component of beta‐diversity, the distance decay in species similarity through space, across three landscapes in Northern Peru. We then compared our derived distance decay relationships to theoretical expectations obtained from a Poisson Cluster Process, known to match well with empirical distance decay relationships at local scales. We used an unsupervised machine learning approach to estimate spatial turnover in species composition from the imaging spectroscopy data. We first validated this approach across two landscapes using an independent dataset of forest composition in 49 forest census plots (0.1–1.5 ha). We then applied our approach to three landscapes, which together represented terra firme clay forest, seasonally flooded forest and white‐sand forest. We finally used our approach to quantify landscape‐scale distance decay relationships and compared these with theoretical distance decay relationships derived from a Poisson Cluster Process. We found a significant correlation of similarity metrics between spectral data and forest plot data, suggesting that beta‐diversity within and among forest types can be accurately estimated from airborne spectroscopic data using our unsupervised approach. We also found that estimated distance decay in species similarity varied among forest types, with seasonally flooded forests showing stronger distance decay than white‐sand and terra firme forests. Finally, we demonstrated that distance decay relationships derived from the theoretical Poisson Cluster Process compare poorly with our empirical relationships. Synthesis. Our results demonstrate the efficacy of using high‐fidelity imaging spectroscopy to estimate beta‐diversity and continuous distance decay in lowland tropical forests. Furthermore, our findings suggest that distance decay relationships vary substantially among forest types, which has important implications for conserving these valuable ecosystems. Finally, we demonstrate that a theoretical Poisson Cluster Process poorly predicts distance decay in species similarity as conspecific aggregation occurs across a range of nested scales within larger landscapes.

Identifying and Quantifying the Abundance of Economically Important Palms in Tropical Moist Forest Using UAV Imagery

“…In addition, the different size of the leaves was reflected in the compactness of the segments, and the fact that the leaves have a different arrangement leads to different values of entropy for the two species. However, differentiating other groups of similar species may be challenging [58]. For example, Cecropia latiloba (a tree with somewhat similar leaf characteristics as M. armata in a UAV image) in some plots tends to be predicted as M. armata.…”

“…A similar case could occur if the palms Socratea exorrhiza and Iriartea deltoidea are found in the same image, as from above, the crown shapes look similar [40]. To overcome this challenge, other types of technologies like hyperspectral imagery or Lidar techniques could be used in the future [58,59].…”

Discovering floristic and geoecological gradients across Amazonia

“…Vegetation mapping has taken advantage of radar images and aerial photographs to identify forest types that differ in structural and terrain characteristics (Duivenvoorden & Lips, 1993;Huber & Alarcón, 1988;Huber, Gharbarran, & Funk, 1995;IBGE, 2004). Several studies have used Landsat data to predict edaphic properties or different aspects of plant communities (species composition, turnover or richness) over landscape extents (Draper et al, 2019;Higgins et al, 2012Higgins et al, , 2011Salovaara, Thessler, Malik, & Tuomisto, 2005;Sirén, Tuomisto, & Navarrete, 2013;Thessler, Ruokolainen, Tuomisto, & Tomppo, 2005;Tuomisto, Poulsen, et al, 2003;Tuomisto, Ruokolainen, Aguilar, et al, 2003). The highest local resolution has been obtained by airborne hyperspectral sensors and LiDAR, which have been used to map forest properties such as canopy height, above-ground carbon stocks and canopy chemistry at the regional extent (Asner et al, 2015(Asner et al, , 2013(Asner et al, , 2014.…”

Reconciling the contribution of environmental and stochastic structuring of tropical forest diversity through the lens of imaging spectroscopy

“…Second, we employed a spectral clustering approach previously utilised by others (Baldeck & Asner, 2013;F eret & Asner 2014;Draper et al, 2018) to reduce the dimensionality of the data and make it suitable for analysis with intuitive and well established techniques. Image pixels were grouped by the similarity of their spectral reflectance into 250 clusters using the minibatch k-means algorithm (Sculley, 2010).…”