Remote Sensing Mapping of the Invasive Plant Species, Kudzu

With the rapid advances in globalization, invasive plants have raised critical pressure on both the local ecosystem and human society. As an invasive plant species with wide ecological tolerance and fast infestation, kudzu has posed persistent challenges to the ecosystems and economy in the southeastern United States. However, current research on kudzu invasion is notably insufficient without a thorough understanding of the characteristics of kudzu growth and its spatiotemporal dynamics. The development of remote sensing technology, coupled with diverse data sources and enhanced processing capabilities, enables efficient detection of kudzu invasion across various time and spatial scales. This research focuses on mapping kudzu invasion on multi-source remote sensing images and investigating the dynamics and influential factors of kudzu growth across continuous time and space. The research was conducted in Knox County, Tennessee, which is one of the major regions affected by the kudzu invasion, resulting in significant adverse impacts on both ecology and the economy. First, a new kudzu mapping method was developed using multispectral Sentinel-2 images at a moderate resolution, integrating spectral unmixing analysis and phenological characteristics. The method consists of three steps, which are linear unmixing, phenology-based masking, and nonlinear unmixing, and finally reaches relatively high accuracy, with user’s accuracy (UA), producer’s accuracy (PA), Jaccard, and Kappa index values of 0.858, 0.907, 0.789, and 0.725, respectively. The method is then adapted to a multi-temporal mapping strategy to monitor kudzu distribution based on repeated Sentinel-2 observations. Our results indicate a steady increase in kudzu invasion areas from 2020 to 2022, with a phenology of the growth season from May to September (peak in August) and a rapid defoliation in the fall. We found precipitation is not a limiting factor for kudzu growth in Knox County, while temperature has strong correlations with kudzu growth, exhibiting seasonal variations. Human activities also play an important role in kudzu distribution. Finally, a high-resolution kudzu mapping method was developed based on the Segment Anything Model (SAM) using a high-resolution NAIP image. The box prompts mode provides the most reasonable and accurate kudzu map with inputs of precise bounding boxes of kudzu patches, which also outperforms Support Vector Machine (SVM)-based kudzu mapping. The developed kudzu mapping methods are important references for future research in mapping the distribution of invasive plants with remote sensing technology at different spatial and temporal scales. The observed unique kudzu growth pattern and its interactions with various environmental factors are also crucial to the prediction, prevention, and management of kudzu and other invasive plants in the southeastern United States.