Related Experiment Video
Updated: Sep 21, 2025
Design and Construction of an Urban Runoff Research Facility
Published on: August 8, 2014
Using Random Forest, a machine learning approach to predict nitrogen, phosphorus, and sediment event mean
Mina Shahed Behrouz1, Mohammad Nayeb Yazdi1, David J Sample2
1Department of Biological System Engineering, Virginia Polytechnic Institute and State University, Seitz Hall, 155 Ag-Quad Ln, Blacksburg, VA, 24060, United States; Hampton Roads Agricultural Research and Extension Center, Virginia Polytechnic and State University, 1444 Diamond Springs Rd, Virginia Beach, VA, 23455, United States.
Machine learning models accurately predict urban runoff pollutant loads using watershed characteristics. Land use data proved most influential for estimating Event Mean Concentrations (EMCs), guiding better stormwater management.
Area of Science:
- Environmental Engineering
- Water Quality Management
- Machine Learning Applications
Background:
- Estimating pollutant loads in urban runoff is crucial for reducing nonpoint source pollution and meeting water quality goals.
- Stormwater Control Measures (SCMs) are vital for urban water management, and their design relies on accurate influent pollutant load estimations.
- The Event Mean Concentration (EMC) method is a common approach for estimating pollutant loads in urban runoff.
Purpose of the Study:
- To develop and apply data-driven Random Forest (RF) models for predicting EMCs of Total Nitrogen (TN), Total Phosphorus (TP), Total Suspended Solids (TSS), and Ortho-Phosphorus (Ortho-P) in urban runoff.
- To identify the most influential climatological, land use, and site-specific catchment characteristics for predicting these pollutant EMCs.
Main Methods:
- Utilized Random Forest (RF), a machine learning algorithm, to build predictive models for EMCs.
- Incorporated various parameters: climatological (Antecedent Dry Period, Precipitation Depth, Duration, Intensity), land use (Imperviousness, Saturated Hydraulic Conductivity, Available Water Capacity), and site-specific (Slope, Catchment Size).
- Employed data from the National Stormwater Quality Database (NSQD) for model training and validation.
Main Results:
- Land use characteristics (Imperviousness, Saturated Hydraulic Conductivity, Available Water Capacity) were the most significant predictors for all studied EMCs.
- For TP, TSS, and Ortho-P, site-specific characteristics (Slope, Catchment Size) were more influential than climatological factors.
- Climatological characteristics were more influential than site-specific factors for TN; precipitation characteristics (P, D, I) were more effective than ADP for TN, TP, and TSS.
Conclusions:
- The study successfully identified key parameters influencing urban runoff EMCs, providing valuable insights for stakeholders and SCM designers.
- Improved estimation of nutrient and sediment EMCs can lead to more effective SCM selection and design.
- Optimized SCM performance is essential for effective stormwater treatment, achieving water quality objectives, and protecting downstream aquatic ecosystems.
