Remote sensing is a powerful tool for natural resource management and ecology research. However, many of the cutting-edge technologies employed by remote sensing professionals require financial resources and technical expertise that is beyond the reach of many ecologists and land managers. In this study we evaluated whether a remote sensing technique that utilizes free, user-friendly software could be used to map Chinese privet (Ligustrum sinense), an invasive forest shrub common in the southeast USA. Privet can form dense stands in hardwood forests that suppress native biodiversity, and mapping its distribution on a property is an important first step in control operations. We used supervised classification to analyze freely available Sentinel multispectral imagery (10m resolution) using the Semi-Automatic Classification Plug-in in QGIS. Mapping was carried out on a ~2500 ha property in western Alabama that consists of a mix of invaded and uninvaded bottomland hardwood forest, swamps, pine stands, and open fields. An accuracy assessment with 250 random field plots is being used to validate the map. Each field plot represents a single Sentinel pixel and land cover variables including privet presence/absence, privet percent cover, native evergreen percent cover, land cover type, and overstory canopy closure are visually estimated.
Results/Conclusions
Initial data (representing one-third of the plots) showed privet presence/absence was correctly identified 65% of the time. However, if a >10% privet cover threshold is assumed as the minimum necessary for detection, the accuracy increases to 75%. This is similar to the accuracy obtained by other researchers using more sophisticated remote sensing methods. It appears that some of the errors in the accuracy assessment were the result of misalignment among field plots and target pixels. A more accurate GPS unit will be deployed to avoid these errors in future sampling which should be complete by April 2019. Based on our preliminary results, this method appears to be adequate for rapidly and effectively mapping areas with high privet cover but not for detecting low density, incipient invasions. The software is simple to use and should be within the capabilities of anyone with a basic familiarity with GIS. If this approach is proven effective, we plan on publishing a guide for implementing this method that is targeted towards land managers, foresters, wildlife biologists, and ecologists.