Session: 16-01-01: Government Agency Student Poster Competition

Paper Number: 150542

150542 - Data-Driven Approach to Assess Green Stormwater Infrastructure (Gsi) Performance 

Assessing the performance and effectiveness of green stormwater infrastructure (GSI) is crucial for sustainable urban stormwater management. However, as these systems become increasingly prolific in the urban landscape this task becomes increasing onerous, requiring ample costs and manpower. Quantifying GSI performance ensures the intended function and longevity of the system, as well as enables guidance for future planning. Prediction of GSI performance, often through ponding time or overflow, is typically done through physics derived hydrological modelling, requiring either simplifications of the highly dynamic nature hydrology of these systems (limiting accuracy) or complex data inputs (limiting the spatial and temporal resolution). However, recent advancement linking in observational data and machine learning (ML) have created an opportunity to improve prediction of GSI performance overtime under changing conditions. Machine learning (ML) has achieved tremendous success on a wide range of applications and has become appealing and powerful tools in water resource engineering. This study applies different ML models, including Random Forest (RF), Extreme Gradient Boosting (XGB), Support Vector Machine (SVM), to predict ponding time and maximum ponding depth of storm water in GSI. Villanova Center for Resilient Water Systems (VCRWS) maintains large database of quality-controlled stormwater observations data for different type of GSIs. This prediction was made using derived features from weather as well as observed data at a bioretention facility. A total of 294 storm event were identified from 10 years of monitoring data at the GSI site. Data of five predictors calculated for each storm i.e., precipitation, average air temperature, storm duration, average intensity and maximum intensity were considered for training/testing ML models. Precipitation and storm duration are found as the two most important features in decision making process for all the models. RF, XGB, and SVM models had R² scores of 0.70, 0.74, and 0.75, and RMSE scores of 14, 14, and 13, respectively, using only weather data to predict ponding time. RF, XGB, and SVM models had R² scores of 0.74, 0.79, and 0.88, and RMSE scores of 0.23, 0.21, and 0.16, respectively, when ponding depth was used as the target variable. Among all the models, SVM shows best result both for training and testing evaluations. This investigation seeks to identify important features for assessing GSI performances and expand this modelling exercise for other types of GSI. This performance prediction, in terms of recession of ponded water and maximum ponding depth to know the overflow occurrence, is an important stride towards sustainable management of GSI at the city-scale.

Presenting Author: Musfiqur Rahman Villanova University

Presenting Author Biography: Musfiqur Rahman is a Doctoral Researcher at the Department of Civil and Environmental Engineering at Villanova University in Pennsylvania, USA. He has a master’s degree in Hydro-informatics and Water Management, a joint master's degree under Erasmus Mundus Scholarship Program from University of Nice Sophia Antipolis (France), Brandenburg University of Technology Cottbus - Senftenberg (Germany) and Newcastle University (England). Prior to joining the Villanova Center for Resilient Water Systems research group, Musfiqur had worked as an Urban Water Specialist for the past ten (10) years on several pre-feasibility, feasibility and research projects in water resources engineering sector at Institute of Water Modelling in Bangladesh. Musfiqur’s current research work focuses on Storm Management Practices (SMP), and its influence on urban watersheds. He is particularly interested in the application of machine learning model to help improve on the sustainable function of SMP to ensure resilience and longevity through proper designs and future planning of the Green Stormwater Infrastructure.

Authors:

Musfiqur Rahman Villanova University
Bridget Wadzuk Villanova University
Peleg Kremer Villanova University
Xun Jiao Villanova University
Achira Amur Villanova University
Madeline Scolio Villanova University
Ruixuan Wang Villanova University
Virginia Smith Villanova University