United States Department of Transportation
i
A Machine Learning-Based System for Predicting Peak Flowrates of Nebraska Streams
-
2023-10-01
[PDF-2.09 MB]
Details:
-
Creators:
-
Corporate Creators:
-
Corporate Contributors:
-
Subject/TRT Terms:
-
Publication/ Report Number:
-
Resource Type:
-
Geographical Coverage:
-
Edition:Final Report April 2021 – June 2023
-
Corporate Publisher:
-
Abstract:Accurate, early flood warnings are invaluable for ensuring transportation safety so that flood-prone roads can be closed well before they become hazardous. A flood forecasting system typically uses both hydrologic and hydraulic models. The latter requires peak flow information to calculate flood attributes, such as depth, velocity, and inundated area. For ungauged basins, which lack streamflow observations, accurate calculation of flood attributes is challenging. Regional regression equations are used for this purpose. Such equations provide peak flow estimates based on flow records and basin characteristics of nearby gauged basins. Regression equations used for Nebraska, however, are decades old. The three available sets of equations often produce results that vary by order of magnitude. Therefore, there is a serious need to improve the accuracy of peak flow prediction using recent datasets and advanced methods. In this study, we modeled daily streamflow and peak flow in Nebraska streams using new high-resolution datasets and two machine learning algorithms, Long Short-term Memory Network (LSTM) for daily streamflow and Random Forest (RF) for peak flow. A wide range of predictors were used in the study. Physically based constraints were imposed on the LSTM model for daily streamflow simulations, and the benefits were assessed. We showed that there is a value in adding physics-based constraints even though the constraints do not improve results in all cases. Therefore, constraints can be applied as needed. Additionally, adding newer datasets and advanced machine learning algorithms leads to improved estimates of peak flow.
-
Format:
-
Funding:
-
Collection(s):
-
Main Document Checksum:
-
Download URL:
-
File Type:
