Mapping the Invisible with AI: Predicting Underground Stormwater Pipelines

Keywords: Stormwater, GeoAI, Transportation Infrastructure, Deep Learning
Abstract Type: Paper Abstract

Abstract

Knowing the locations of underground stormwater pipeline is crucial for transportation infrastructure resilience and effective stormwater management. However, this information is commonly poorly documented or inaccessible. This presentation covers the preliminary development of deep learning method for the DeepPipe project, funded by NCDOT, which aims to detect underground stormwater pipelines leveraging cutting-edge AI solutions. Our approach leverages advanced graphic neural networks, trained on spatial and environmental data, to predict pipeline locations across North Carolina. By applying deep learning techniques to this unique spatial challenge, we aim to bridge gaps in existing data and support future transportation planning efforts. This presentation will showcase our preliminary results, offering insights into the model performance in terms of the capability of detecting stormwater pipelines.