Presenter: Travis Sterne (U.S. Geological Survey)

Description:
Coastal resources are increasingly impacted by erosion, extreme weather events, sea-level rise, nuisance flooding, and other potential hazards related to climate change. These hazards have distinct impacts on coastal landforms due to the numerous spatial, geologic, sociologic, oceanographic, and ecological factors that exist at a given location. The U.S. Geological Survey’s (USGS) coastal change likelihood (CCL) project synthesizes existing datasets across agencies and disciplines in a framework that predicts the likelihood of coastal change along the U.S coastline within the coming decade. This pilot study conducted in the Northeastern U.S. (Maine to Virginia) is comprised of a decision tree-based coastal landscape dataset (a.k.a. fabric dataset) that includes landcover, elevation, slope, long-term (>150 years) shoreline change trends, dune height, and marsh stability data. Additionally, a coastal hazard database is defined and divided into event hazards (e.g., flooding, wave power, and probability of storm overwash) and persistent hazards (e.g., relative sea-level rise rate, short-term (decadal) shoreline erosion rate, and storm recurrence interval). The fabric dataset is then merged with the coastal hazards databases to create a training dataset made up of hundreds of polygons to/for Support Vector Machine Learning classification. Results from this pilot study are location-specific at 10-meter resolution and are made up of four raster datasets that include (1) quantitative and qualitative information used to determine the landscape’s resistance to change, (2 & 3) potential coastal hazards, and (4) machine learning output based on the cumulative effects of both fabric and hazards.  Final outcomes are intended to be used as a first order planning tool to determine which areas of the coast may be more likely to change in response to future potential coastal hazards, as well as to examine elements of the coast that make this change in a given location more likely.