A Floodplain Map of the Conterminous United States Developed Using Random Forest Classification
Citation:
Baynes, J., S. Woznicki, S. Panlasigui, M. Mehaffey, AND A. Neale. A Floodplain Map of the Conterminous United States Developed Using Random Forest Classification. 2019 Esri User Conference, San Diego, CA, July 08 - 12, 2018.
Impact/Purpose:
Presentation at the ESRI Convention to demonstrate a method to quickly identify unmapped flood prone areas (i.e., areas with a 1% annual probability of flood.
Description:
Floodplains perform several important ecosystem services, including storing water during precipitation events and reducing peak flows, thereby reducing flooding of adjacent communities. Understanding the relationship between flood inundation and floodplains is critical for ecosystem and community health. FEMA’s Flood Insurance Rate Map (FIRM) zones cover approximately 60% of the CONUS; however, detailed flood modeling is expensive and time-consuming. We sought a method to quickly identify unmapped flood prone areas (i.e., areas with a 1% annual probability of flood). For each level-4 hydrologic unit code (HUC-4), random samples of the FIRM dataset were used as the response variable along with ten nationally available, biophysical datasets as explanatory variables to train and apply a random forest classification at 30-meter resolution. Additional random samples of the FIRM dataset were used to access the accuracy of our model. Hit Rate, the percentage of FIRM area the random forest model correctly classified, was 0.79 for the CONUS but varied geographically.
