Creating US Atlantic Coastal Plain wetlands prediction models using hyperspatial UAS LiDAR and multispectral data and machine learning approaches

Keywords: hyperspatial LiDAR, multispectral, UAS, wetlands
Session Type: Virtual Paper Abstract
Day: Sunday
Session Start / End Time: 2/27/2022 03:40 PM (Eastern Time (US & Canada)) - 2/27/2022 05:00 PM (Eastern Time (US & Canada))
Room: Virtual 20

Abstract

Wetlands provide critical ecosystem services across a range of environmental gradients and are at heightened risk of degradation from natural and anthropogenic drivers and continued development, especially in coastal regions. There is a growing need for high resolution, spatially and temporally, accurate habitat identification, and precise delineation of wetlands across a variety of stakeholder groups, including wetlands loss mitigation programs. Traditional wetland delineations are costly, time-intensive and can physically damage the systems that are being surveyed, while aerial surveys are relatively fast and relatively unobtrusive. We test the comparative accuracies of a wetland prediction model fitted with UAS and airborne LiDAR terrain and topographic derivatives along with UAS-collected multispectral data at nine coastal sites using machine learning approaches. When mounted on low-flying UASs, LiDAR sensors can measure elevation data even underneath dense canopy cover characteristic of much of the Atlantic Plains forested wetlands. We show that the UAS hyperspatial LiDAR derivatives outperform the airborne LiDAR data in deriving high resolution and high classification accuracy rates, especially when trained with field-collected habitat data. Hyperspatial resolution of LiDAR-derived topography models can fill wetlands mapping needs and increase accuracy and efficiency of detection and prediction of sensitive wetland ecosystems.