A LiDAR Topography-based Random Forest Model Using Advanced Processing Techniques and Geomorphic Edge Detection to Identify Environmental Planning-Scale Wetlands
Abstract
Wetlands are important ecosystems that support biodiversity, improve water quality, and provide climate change resilience. Anthropogenic pressures threaten wetlands, prompting increased efforts to protect those that remain. Accurate digital wetland inventories can support these efforts; however, creating these inventories can be challenging and require costly input data. The growing collection of remote sensing data has created new opportunities for wetland identification, and Light Detection and Ranging (LiDAR) data specifically has been widely embraced by the wetland science community. LiDAR digital elevation models (DEMs) are increasingly available and provide surface hydrologic information at the level of detail needed to identify small-scale wetlands. We propose an open source, automated model that performs a Random Forest (RF) classification of LiDAR DEM metrics (topographic wetness index, curvature, and cartographic depth-to-water index) to identify likely wetland areas based on near-surface soil saturation estimates. We use sophisticated smoothing and conditioning techniques that improve the performance of our LiDAR DEM metrics. Additionally, we reconfigure the classification process by thinning RF predictions based on proximity to similarly classified pixels and dilating the skeleton to distinct landform edges, resulting in wetlands represented as geomorphic features rather than isolated pixels. Preliminary results show that when applied to four study sites situated in varying ecoregions of Virginia, USA, and tested against manually surveyed wetlands, the wetland identification model detects significantly more wetlands than the commonly used National Wetland Inventory (NWI) (26-63% increase in wetland recall). The wetland model has the potential to support wetland conservation and management efforts by targeting small-scale wetlands and using widely-available data and open source processes.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2018
- Bibcode:
- 2018AGUFM.H31K2067O
- Keywords:
-
- 1819 Geographic Information Systems (GIS);
- HYDROLOGYDE: 1855 Remote sensing;
- HYDROLOGYDE: 1856 River channels;
- HYDROLOGYDE: 1857 Reservoirs (surface);
- HYDROLOGY