View Important Policies and System Requirements for this course

Interested in registering 5 or more engineers for a course? Contact us for information and rates.

INSTRUCTORS: 
Xiaofeng Liu, Ph.D, P.E.
Balraj Singh
Vijay Minocha
Caleb Buahin, Ph.D
Corinne Wiesner-Friedman, Ph.D
Charles Holber
Aditya Tyagi

Purpose and Background

These presentations were recorded at the 2024 World Environmental & Water Resources Congress:

Comparative Study of Machine Learning Techniques For Prediction of Scour Depth Around Spur Dikes (13 minutes)

This presentation explores the application of various machine learning techniques to predict the scour depth around spur dikes, structures used to control riverbank erosion. The study compares methods such as Multivariate Adaptive Regression Splines (MARS), M5P3 models, and Group Method of Data Handling (GMDH) based on performance metrics like root mean square error (RMSE) and mean absolute error (MAE). Data was collected from 154 observations, with 80% used for training and 20% for testing the models. The results indicate that MARS outperforms other techniques, offering more accurate predictions of scour depth, which is critical for effective hydraulic design.

Graph Neural Networks for Hydraulic Routing in Collection Systems (13 minutes)

This presentation delves into the use of Graph Neural Networks (GNNs) for hydraulic routing in collection systems, focusing on their ability to model complex networked systems like sewer and stormwater infrastructure. By leveraging the topological structure of these systems, GNNs enhance the accuracy of flow prediction and pressure estimation, crucial for managing water resources. The study demonstrates how GNNs outperform traditional hydraulic models, particularly in handling non-linearities and spatial dependencies within the network. This approach offers a significant advancement in predictive modeling, contributing to more efficient and resilient water management practices.

Differential Modeling for Computational Hydraulics: Bridging Physics-Based and Data-Drive Modeling (17 minutes)

This presentation explores differential modeling as a method to bridge physics-based and data-driven approaches in computational hydraulics. By integrating physical laws with machine learning models, the approach enhances the accuracy of hydraulic simulations while maintaining interpretability. The study highlights the advantages of combining first-principles modeling with data-driven corrections, particularly in scenarios with sparse or noisy data. This hybrid technique promises improved predictive performance in hydraulic engineering, offering a balanced solution between traditional modeling and modern AI-driven methods.

Analyzing Heavily Censored Surface Water Pesticide Concentration Data Using Innovative Statistical Techniques (17 minutes)

This presentation examines the use of innovative statistical techniques to analyze heavily censored surface water pesticide concentration data, often affected by detection limits. The study introduces methods such as survival analysis and multiple imputation to effectively handle non-detects and improve data interpretation. By applying these techniques, the analysis reveals more accurate estimates of pesticide concentrations and their environmental impact. This approach enhances the reliability of water quality assessments, crucial for regulatory compliance and environmental protection efforts.

Transferability of Data-Driven Models to Enhance the Water Level Prediction in Basins with Scarce Data (18 minutes)

This presentation investigates the transferability of data-driven models to predict water levels in basins with limited data availability. It highlights the challenges of applying models trained on data-rich basins to those with scarce data, focusing on methods like transfer learning to adapt these models. The study demonstrates that, with appropriate adjustments, data-driven models can significantly enhance predictive accuracy in under-monitored basins. This approach offers a promising solution for improving water resource management in regions with sparse hydrological data.

Benefits and Learning Outcomes

Upon completion of this course, you will be able to:

• Explain how different machine learning techniques, including MARS and GMDH, can be used to predict scour depth around spur dikes and identify the most effective method based on performance metrics.
• Describe the application of Graph Neural Networks in hydraulic routing and discuss their advantages over traditional models in predicting flow and pressure in collection systems.
• Discuss the integration of physics-based and data-driven approaches in differential modeling and explain how this hybrid method enhances hydraulic simulation accuracy.
• Identify innovative statistical techniques used to analyze heavily censored pesticide concentration data and describe their impact on improving water quality assessments.
• Explain the concept of transferability in data-driven models and discuss how these models can be adapted to improve water level prediction in basins with scarce data.

Assessment of Learning Outcomes

Learning outcomes are assessed and achieved through passing a 10 multiple choice question post-test with at least a 70%.

Who Should Attend?

• Water resource engineers
• Environmental engineers
• Consulting engineers
• Utility engineers
• Public Agency Engineers
• Utility Directors

How to Earn Your CEUs/PDHs and Receive Your Certificate of Completion

This course is worth 0.2 CEUs/2 PDHs. To receive your certificate of completion, you will need to attend the live session and/or watch the recording(s) and complete the post-session survey. When the course is taken On-Demand, there will also be a 10 multiple choice question post-test.

How do I convert CEUs to PDHs?

1.0 CEU = 10 PDHs [Example: 0.1 CEU = 1 PDH]