ERTC3 MCTI December 2024 Webinar

Learning with Heterogeneous Datasets - Understanding Environmental Impact of Transportation Systems Using Causal AI, Dr. Xinghui Zhao
Today big data has become the key challenge in virtually every area of human endeavor. AI and machine learning approaches are explored in many disciplines to learn patterns from vast datasets. However, heterogeneous datasets often present challenges, especially if these datasets are collected from different disciplines. In this talk, I will present our recent work in discovering causality relations between the transportation and environmental data. The goal is to enhance the understanding of the environmental impact of transportation systems and decisions. Specifically, we leverage causal AI to performed extensive data analysis on various datasets collected by various agencies, including WSDOT, the National Water Quality Monitoring Council and the National Oceanic and Atmospheric Administration (NOAA). An efficient data analysis workflow is developed to derive the causal effects of various variables. Our work has the potential to mitigate the negative impact on the environment by providing recommendations for policy makers, leveraging the causal relations learned from the historical data.
Investigating the controls of sediment transport in streams and rivers: evidence from sediment hysteresis patterns, Dr. Tyler Mahoney
Excessive sediment transport is a leading non-point source pollutant in streams and rivers across the US, with landscape disturbance being a well-recognized contributor to high sediment loads. An important tool for understanding the processes controlling sediment export is the analysis of sediment hysteresis patterns, which define the non-linear relationship between sediment transport and stream discharge. These patterns can provide valuable insights into potential sediment sources and whether the sediment load is transport-limited or supply-limited. However, there is limited knowledge on how land use and land cover, landscape connectivity, and other hydroclimatic watershed properties alter sediment hysteresis patterns. The objective of this study is to understand the climatic and physiographic controls of sediment hysteresis in watersheds. Herein, we apply machine learning techniques to investigate the controls of sediment hysteresis, including information on hydrologic pathways as informed by tracer data collected in streams in rivers. We utilize unsupervised machine learning techniques to characterize hysteresis patterns broadly across the US. Additionally, we identify parameters required to predict sediment hysteresis using explainable AI techniques. Our results have important implications for better understanding the mechanisms controlling sediment transport in watersheds.