Using Vegetation Indices from Hyperspectral Imaging Data to Differentiate Among Invasive Plant Species

The management of invasive plants is a prevalent area of study and is relevant today in many ecosystems. Vegetation indices from remotely sensed hyperspectral data are useful for identifying invasive plants, as these indices are determined by the different physical and chemical features of plants. Drone-based hyperspectral images collected from a field in northwestern Virginia four individual times during 2020 were used to identify certain invasive plant species. From these images, reflectance spectra were sampled from 15 pixels representative of target individuals and transformed into vegetation indices using the R package hsdar. A partial least squares-discriminatory analysis (PLS-DA) was conducted with the vegetation indices to differentiate individual species and determine indices most useful in differentiation. Two invasive shrub species, autumn olive and Dahurian buckthorn, were each compared to all other plants in the field. The greatest variance explained by components 1 and 2 occurred in November for autumn olive, and June for buckthorn. The components explained a combined 49% of variance for autumn olive in November, and a combined 58% of variance for buckthorn in June. Both species separate particularly strongly across component 1 for both of these months. Buckthorn was particularly well-separated by vegetation indices related to chlorophyll and leaf area index, while autumn olive was most well-separated by vegetation indices related to chlorophyll and stress. However, vegetation indices relating to all physiological factors were useful for both species. Furthermore, buckthorn is particularly differentiable in later months, likely due to it losing its leaves very late. The greatest variance explained by the PLS-DA, paired with the degree of separation among the different plant species, leads me to conclude that June and November are the best times of year to identify autumn olive and buckthorn.