EES Seminar: Leveraging Machine Learning for Floodplain Wetland Identification, E. White, Stanford

Details

Coastal freshwater floodplain wetlands (CFFWs) are a critical component of the coastal wetland mosaic and offer numerous ecosystem services (i.e. carbon sequestration, storm surge attenuation, groundwater recharge), however they face an existential threat due to coastal climate change (i.e. sea level rise, storm surge, hurricanes). Previous research documented nearly 14,000 km of CFFWs loss in the North American Coastal Plain from 1996 – 2016 with more than 75% being explained by climate and topographic variables. However, there are critical information gaps regarding the location of and habitat suitability for CFFWs. We leveraged publicly available datasets with advances in machine learning to create the first maps of CFFW extent and climate integrated habitat suitability for the contiguous United States. Both maps used the NOAA Coastal Change Analysis Program 2016 palustrine forested wetland class as the locations for training data with the extent map using Landsat for optical imagery. Our extent map is based on a convolutional neural network with Inception-ResNet-V2 architecture best identifies large features (83% overall accuracy, 0.66 F1- Score, 0.54 Kappa Value) with most of the locations being in river valleys or protected areas. The random forest-based habitat suitability integrates 2050 climate data and projected sea level rise with additional environmental data (e.g. physiography, hydrology, and hydrography) to predict where CFFWs can exist in the near future (86% overall accuracy, 0.86 F1 Score, and0.52 Kappa Value). These new maps are being put directly into action by being used to identify carbon credit opportunities to support small landowners. Additionally, our maps can be updated quickly as new data are made available, which exceeds the current standard that is updated on a 5-year basis. The temporally dynamic nature of our approach allows for rapid assessment of CFFW change for acute events and should help constrain long-term estimates of change.

Elliott White Jr. is an Assistant Professor of Earth System Science in the Stanford Doerr School of Sustainability. He is a coastal ecosystem scientist who leverages his domain expertise in wetland sciences with interdisciplinary training in remote sensing and ecohydrology to investigate climate change related challenges on coastal socio-environmental systems (cSES). Elliott has research on all three US coasts and has expanded internationally to include Bolivia, The Gambia, and Canada. Collaborators in his research include academics, non-profits, community-based organizations, and municipal departments. At Stanford, he is an affiliate of the Center for Comparative Studies on Race and Ethnicity and a Center Fellow, by courtesy, of the Woods Institute for the Environment. Elliott has a PhD in Environmental Engineering Sciences from the University of Florida (2019) and a BS in Biology and Animal Ecology from Iowa State University (2015).