Vegetation change analysis from 2010-2018 using aerial photography and RTK-GNSS to assist Lake Mattamuskeet Restoration Efforts in North Carolina, USA

Abstract

Mapping vegetation species and documenting their changes is critical to achieving the Lake Mattamuskeet National Wildlife Refuge restoration and habitat management efforts of controlling the spread of invasive species. This study maps the dominant vegetation species and inventories their changes in the Lake Mattamuskeet waterfowl impoundment 4 using time series aerial images collected 2010 - 2018. RTK-GNSS surveys were conducted in the field to record the locations of dominant species used as reference data. The reference data were matched to their respective vegetation patch delineated by object-based image analysis. The Random Forest machine learning classification algorithm was used to accurately predict the unknown locations of the dominant species. The algorithm had an overall accuracy [less-than-or-equal-to]76 and Kappa statistic [less-than-or-equal-to]66. Phragmites australis (phragmites) expanded in 2016 but was constrained in 2018 by Echinochloa walteri (Walter's millet). The Refuge's goal of achieving 50% good waterfowl species was not met during the time series investigated, as the largest cover of good waterfowl habitat was only 25% in 2018. The results of this research provide insights about the effectiveness of current vegetation management techniques implemented at the Refuge for waterfowl impoundment 4 and have implications for global wetland mapping and change analysis.