The Wetland Intrinsic Potential Tool: Identifying Forested Wetlands Through Machine Learning of Lidar Derived Datasets

Wetlands provide a vast array of ecosystem services such as water storage, carbon sequestration, sediment removal, and wildlife habitat. Terrestrial or inland wetlands are the largest reservoir of carbon in North America, with roughly half of the wetland area occurring in forested systems. Wetlands under partial or closed canopy are difficult to directly detect in imagery and therefore have high errors of omissions in national and local wetland inventories. The omission of forested wetlands from wetland inventories may mean we are vastly undervaluing the ecosystem services of forested wetlands at the landscape scale. Additionally, insufficient wetland inventories impedes our ability to understand the impacts of disturbances, such as forest practices, on forested wetlands. For example, without improved maps of location and extent of forested wetlands we cannot accurately estimate their role in the carbon cycle and understand how forest practices impact carbon storage at the landscape scale. The goal of this research project was to assess the accuracy of a random forest model that integrated topographic indices, topographic wetness index, depth to water index, spectral indices, and soils to identify wetlands under heavy canopy in the Pacific Northwest. We developed and tested our model in several watersheds throughout Washington State through collection of training and validation data from image interpretation and field observations. We integrated the results into an ArcGIS toolbox called the Wetland Intrinsic Potential (WIP) Tool.