Assessing the Ability to Detect Invasive Plant Species Using Drone-Based Leaf-Scale Visible and Near-Infrared Imaging Spectroscopy

Abstract:

Across Virginia, invasive plants prevent growth of native species, which will drastically alter biodiversity. These widespread changes in species composition also have broader impacts on soil chemistry and forest canopies, with feedbacks on dynamics of carbon, nutrients, water, and energy. Two common invasive shrub species in Virginia, Elaeagnus umbellate (autumn olive) and Rhamnus davurica (Dahurian buckthorn), out-compete and displace native plants, and alter both light and soil nitrogen resources. A critical first step in understanding the ecological impacts of these species is the ability to map their distributions. We utilized a UAV equipped with a high-precision GPS system and Headwall Nano-Hyperspec hyperspectral imager to collect images of canopies of heterogeneous vegetation communities in northwestern Virginia, where these species are common. Images collected included reflectance from 400 to 1000 nm, approximately every 2.2 nm. Spectral signals were extracted from 15 well-lit and representative pixels from individual trees and shrubs of a known identity within images collected at different times throughout the 2019 and 2020 growing seasons. A partial least squares discriminatory analysis of data collected in August 2019 demonstrated that even near peak biomass, species are quite differentiable. Because UAV platforms provide the opportunity to sample multiple times over a growing season, we incorporated phenological characteristics into our analysis to improve the accuracy of classification. We utilized phenological differences in an analysis of imagery collected in April 2020, in which E. umbellate and R. davurica leafed-out prior to neighboring trees and shrubs. We found that including spectral characteristics of red edge shoulder location and reflectance as well as a normalized difference vegetation index (reflectance at 553 and 682 nm) and photochemical reflectance index (reflectance at 531 and 570 nm) allowed for accurate classification. A hyperspectral data analysis methodology to effectively identify and locate targeted invasive plants from aerial images opens doors to better understand temporal and spatial patterns in response to climate and the implications this has on primary productivity, evapotranspiration, and feedbacks on the global carbon, water, and energy cycles.