Application of a One-Class Classifier and a Linear Spectral Unmixing Method for Detecting Invasive Species in Central Chile

Chile has a large number of endemic species due to its isolated location, which are increasingly endangered by a variety of invasive species. This development is further intensified by high disturbance rates. In this study, we evaluate a Sentinel-2 time-series of one phenological cycle in 2016 and additional geo-information from TanDEM-X as well as infrastructure and soil data to detect the spread of Ulex europaeus onto agricultural areas with a one-class classifier (MaxEnt) and a non-negative least square (nnls) unmixing algorithm. The results were validated with UAS-based image mosaics acauired in March 2016. First results led to overall accuracies of up to $\mathrm{R}^{2}=0.7$ for Ulex europaeus for the linear unmixing and overall accuracies between 79 % and 89 % for the one-class classifier, depending on the included data-set. The mapped pattern reflects the observed on-field spreading pattern on the Isla de Chiloe (Chiloe Island) quite well, showing the most affected areas in the more intense agriculturally used north-east of the island. Furthermore, the results were notably influenced more by the included data-sets than by the detection method. Generally, an increased number of Sentinel-2 images as well as additional geo-information improved the performance.