Mapping freshwater marsh species in the wetlands of Lake Okeechobee using very high-resolution aerial photography and lidar data

Abstract

ABSTRACTAccurate marsh species maps are needed to study their abundance, distribution, habitat change, and to support the ongoing and future restoration activities in Lake Okeechobee, Florida, U.S.A. In this study, we integrated very high-resolution aerial photography with a light detection and ranging (LiDAR)-derived digital elevation model (DEM) for the discrimination of six freshwater marsh species (Salix caroliniana, Spartina bakeri, Polygonum spp., Typha spp., Phragmites australis, and Cladium jamaicense). Four techniques were combined in the mapping procedure, including object-based image analysis, machine learning classifiers, texture analysis, and ensemble analysis. The results showed that both texture and topography features were invaluable for marsh species mapping. A synergy of spatial, spectral, and topographical features achieved an overall accuracy (OA) of 85.3% and kappa coefficient of 0.83 using the Random Forest (RF) classifier. Ensemble analysis of the outputs from Support Vector Machine (SVM), RF, and Artificial Neural Network (ANN) did not increase the classification accuracy, but produced an uncertainty map to identify regions with a robust classification and areas difficult to map. Ensemble analysis was beneficial in getting further insight into marsh species mapping. The developed digital procedure is a promising alternative to traditional field survey and manual interpretation methods for generating marsh maps to support the lake restoration.