Thu, Aug 18, 2022: 4:15 PM-4:30 PM
516D
Background/Question/MethodsInfestation by invasive plants has threatened plant species listed under the Endangered Species Act in United States. In addition, it is costly to remove invasive plants from forested ecosystems. Chinese tallow and Chinese privet are enlisted as the worst invasive plants in southern Alabama. Hence, the distribution mapping of these species is crucial to fulfilling invasive species management objective. The study’s overall goal is to develop the most accurate and high-resolution distribution maps of Chinese tallow and Chinese privet in the coastal region of Alabama. The study addresses the following questions a) Does the integration of lidar and NAIP imagery improve classification accuracy? b) What is the most accurate approach for developing distribution map of Chinese tallow and Chinese privet? Light detection and ranging (lidar) data and National Agriculture Imagery Program (NAIP) imagery were examined for this purpose. Five approaches to developing spatially explicit maps of Chinese tallow and Chinese privet were investigated: (1) ISODATA clustering, (2) Maximum Likelihood classifier using NAIP alone, (3) Maximum Likelihood classifier using NAIP and lidar, (4) Random Forest (RF) with NAIP alone, and (5) RF using NAIP and lidar. Classification accuracies from using NAIP and NAIP-lidar stacked images were compared for each classification method.
Results/ConclusionsThe integration of lidar and aerial imagery was found to have increased overall classification accuracy from 74.73 % to 83.24% and from 79.58% to 98.62% using Maximum Likelihood classifier and RF, respectively. We observed an increase in users’ classification accuracy from 80% to 94.81% for classifying Chinese tallow after integrating Canopy Height Model derived from lidar to the aerial imagery. Among the classification approaches used, RF classification using NAIP imagery integrated with lidar-derived variables provided the highest overall classification accuracy. Also, the highest producer’s accuracy (88.24%) and user’s accuracy (85.71%) for classifying Chinese privet was observed with this approach. Detection of Chinese privet was more challenging as it was covered by sweetgum, water oak, pine, and other shrubs. Chinese privet was confused with other vegetation and built-up areas due to similar spectral properties. Mapped products from this study contribute to an initial, spatially comprehensive baseline inventory of crucial invasive species within the region. It will help uncover the pattern of invasive plant species distribution across landscapes and inform the development of a framework for broader-scale mapping and monitoring efforts.
Results/ConclusionsThe integration of lidar and aerial imagery was found to have increased overall classification accuracy from 74.73 % to 83.24% and from 79.58% to 98.62% using Maximum Likelihood classifier and RF, respectively. We observed an increase in users’ classification accuracy from 80% to 94.81% for classifying Chinese tallow after integrating Canopy Height Model derived from lidar to the aerial imagery. Among the classification approaches used, RF classification using NAIP imagery integrated with lidar-derived variables provided the highest overall classification accuracy. Also, the highest producer’s accuracy (88.24%) and user’s accuracy (85.71%) for classifying Chinese privet was observed with this approach. Detection of Chinese privet was more challenging as it was covered by sweetgum, water oak, pine, and other shrubs. Chinese privet was confused with other vegetation and built-up areas due to similar spectral properties. Mapped products from this study contribute to an initial, spatially comprehensive baseline inventory of crucial invasive species within the region. It will help uncover the pattern of invasive plant species distribution across landscapes and inform the development of a framework for broader-scale mapping and monitoring efforts.