Characterizing highway runoff and quantifying the influencing factors in the United States with a machine learning approach
runoff from highways is a significant non-point pollution sources, leading to
the watercourse and aquatic ecosystem impairments. The primary contaminants
include sediments, nutrients, heavy metals, and polycyclic aromatic
hydrocarbons (PAHs) (Trenouth and Gharabaghi, 2016). There is significant
differences in contaminants concentrations among different sample sites. The
large variety of contaminants are attributed to a combination of factors, such
as traffic volume, average daily traffic, vehicles during a storm, antecedent
dry period, rainfall pattern, climatic factors, and the adjacent land use
(Opher and Friedler, 2010; Huber et al., 2016). Although there is a number of
research projects studying the physical, chemical and biological processes
associated with highway runoff events (Crabtree et al., 2006; Alo et al.,
2007), there are still open questions regarding the characteristics of highway
runoff and the complex mutual influences of the many factors.
comprehensive Highway-Runoff database (HRDB) was developed by USGS and
published in Nov. 2018. This database provides planning level information of
highway runoff, by recording 242 highway sites and 6873 storm events data from
1975 to 2017 across the United States. The dataset includes more than 100,000
concentration values for 414 water quality constituents (Granato et al., 2018).
This dataset provides an opportunity to study highway runoff characteristics
and mechanisms across the United States, and thus informing planning approaches
to control highway runoff contaminants. In this study, a data-driven model will
be built based on this database to explore how traffic volume, adjacent land
use and storm characteristics combine to predict highway runoff contaminants.
objective of this research include:
1） To identify several distinguishable
typologies of highway runoff characteristics. Specifically, the 242 highway
sites will be clustered into several groups, with sites in each group sharing
similar characteristics regarding contaminants concentration, rainfall pattern,
traffic volume and land use information.
2） To build a machine learning model, measuring how traffic,
land use, and climatic factors combine to predict highway runoff. The
importance of different factors will be investigated, as well as their mutual
3） To provide planning guidelines
according to the model developed. Stormwater management strategies will be
suggested for highway sites with different characteristics.
past and present studies focus on the model development to estimate highway
runoff. These studies are useful resources as a reference to understand the
complex process of highway runoff. With the application of the new highway
runoff dataset and the machine learning models, a more robust and efficient
model can be developed to estimate highway runoff. The related research
include, but not limited to:
Stochastic Empirical Loading and Dilution Model (SELDM) for runoff-quality
analyses developed by Oregon Water Science Center (Risley and Granoto, 2014).
research project NCHRP 25-53: Approaches for Determining and Complying
with TMDL Requirements Related to Roadway Stormwater Runoff.
research project NCHRP 25-20: Evaluation of Best Management Practices for Highway
This research topic has relevance for researchers, stormwater engineers and planning agencies. It will help improve the understanding of highway runoff processes and mechanisms, as well as find a more effective approach to control pollution from highway runoff.
|Sponsoring Committee:||AKD40, Landscape and Environmental Design
|Research Period:||6 - 12 months|
|RNS Developer:||Runzi Wang and Ming-Han Li|
|Index Terms:||Runoff, Highways, Machine learning, Data files, Contaminants, Data analysis, Traffic volume, Land use, Storms, |
|Cosponsoring Committees:||AKD50, Hydrology, Hydraulics and Stormwater|
Data and Information Technology
Hydraulics and Hydrology