UC Merced

Invasive plants are non-native species that have detrimental economic, ecological, environmental or environmental effects on their surroundings and can spread rapidly. In aquatic ecosystems, they are particularly harmful because they can affect water quality and availability by disrupting flow patterns. In areas such as the Sacramento-San Joaquin River Delta in California, USA, management programs are in place which use imaging spectroscopy to map and track annual changes in invasive species patterns to inform treatment. These maps are constructed from imagery collected by an airborne imaging spectrometer. Advances in unmanned aircraft systems have enabled small imaging spectrometers such as the Headwall Photonics Nano-Hyperspec (Nano) to provide even higher spatial resolution imagery than manned flights. In this case a 5.1 cm pixel spatial resolution imagery in the VIS-NIR (400-1000 nm) was gathered by the Nano, while HyMap captured 1.7 m VIS-SWIR (400-2500 nm) imagery. For species mapping applications, the higher spatial resolution provided by the Nano allows detection of invasive weeds before they spread over larger areas. To compare the mapping capabilities and utility of unmanned aircraft and manned aircraft for mapping invasive species, aerial imagery was collected concurrently at a wetland study site using a Nano on board a unmanned aircraft (DJI M600P) and the HyMap sensor mounted on board a piloted aircraft (1975 Rockwell International 500-S). Maps were then created from similar remote sensing products using a random forest model. The Nano-Hyperspec was capable of identifying smaller patches of invasive plants, which did not appear in maps generated from the HyMap sensor. Three experiments were used to determine input criteria for the best model for map creation, which included training and test data proportions, flight direction, and acquisition. Results showed that increasing training data proportion resulted in higher median overall accuracies but the decrease in test data caused an increase in distribution width and interquartile range of model accuracy. Acquisition date also impacted model performance, but there were several confounding factors making it difficult to ascertain which, if a single variable was responsible. Flight direction relative to solar position was also significant to model performance; the Nano-generated map performed best when trained on labelled data collected in both flight directions then applied or tested only on data acquired while flying away from the solar plane. Map comparisons between those made from the Nano and those from the HyMap sensor show that the Nano performs as well as the HyMap as a source for spectral data to generate classification maps, with a higher overall accuracy for 2019, and comparable overall accuracies for other years. In addition, the higher resolution of the Nano imagery allowed detection of patches of water hyacinth present in the study site that the HyMap maps could not. However, it would not be feasible to operate the Nano as a replacement to the HyMap despite its improved detection capabilities to due to area coverage limitations. But the Nano could be used to supplement an existing invasive species management program to build an improved map of targeted areas.