Monday, June 6, 2022

Deep Learning for Unmonitored Water Level Prediction and Risk Assessment

The research uses publicly available data to develop deep learning models to predict river gauge heights at unmonitored locations in Missouri. The geospatial and rainfall data for 20 different catchment areas of Missouri is used in tandem with the clustering and ensemble deep learning approaches to develop a high-performance deep learning model that efficiently captures the interdependencies between the time-series input data values. The models can accurately predict river water level values up to 4 hours ahead in the future with a correlation of greater than 0.82 with most results having a correlation greater than 0.9. The data-based approach applied to develop a deep learning neural networks-based framework can assist the first responders in issuing timely and localized flood warnings for the safety of the general public. This methodology is applied to publicly available datasets obtained from The United States Geological Survey (USGS) and National Weather Service (NWS). The research project is funded by the Missouri Department of Transportation (MoDOT) and Mid-America Transportation Center (MATC).

VIEW FINAL REPORT


Report number: cmr 22-007
Published: June 2022
Project number: TR202111
Authors: Steven M. Corns, PhD; Suzanna K. Long, PhD; Jacob Hale; Bhanu Kanwar; and Samuel Vanfossan
Performing organization: Missouri University of Science and Technology