Mapping and quantifying early tidal wetland evolution using remotely piloted aircraft system imagery and object-based image analysis

Abstract

This research presents recommendations to effectively monitor the transition from former agricultural land back into salt marsh habitat following the dyke managed realignment at the Converse Marsh. For this analysis, two sets of multispectral RPAS imagery was collected pre- and a post-breach in September 2018 and May 2019 respectively. Both images were classified using Object-based Image Analysis (OBIA) and Random Forest, resulting in overall accuracies of 61% (Kappa 0.55) and 71% (Kappa 0.69). The results were compared with the Inundation Frequency to quantify classes by area and relative positioning. The use of OBIA is a feasible method for intertidal wetland delineation, however, it requires more objectivity in the process. The RGB camera and near-infrared sensor were missing critical sensitivity measures for radiometric calibration. The implementation of Random Forest proved to be a strong algorithm for intertidal habitat mapping. For future analysis, it is critical to have long-term tide recordings available.