Harnessing AI to Model Urban Boundary Layer Flows and Energy Systems for Sustainable Cities

This study explores applications of Artificial Intelligence (AI) and Machine Learning (ML) methods for complex weather processes and their cascading effects, focusing on two critical urban challenges. The first case addresses urban extreme precipitation, which poses significant forecasting challenges in complex urban centers like New York City. Using a fully urbanized Weather Research and Forecasting (uWRF) model as a baseline, we developed an AI-based approach to correct systematic biases in spatial placement and timing of rainfall. By applying an Attention U-Net architecture utilizing hourly uWRF precipitation fields as inputs and Multi-Radar/Multi-Sensor System (MRMS) data as the target the model effectively corrected magnitude and distribution errors. Additionally, urban-scale downscaling using the FourCastNet deep learning model demonstrated that data-driven methods can substantially enhance prediction accuracy, supporting improved flood risk mitigation. The second application focuses on predicting power transmission tower failures during extreme hydrometeorological events in Puerto Rico. This framework integrates high-resolution weather forecasts, tower fragility curves, and real-time mechanical displacement data from tower-mounted sensors into a Random Forest Classifier (RFC). The model calculates failure probabilities based on structural materials and WRF-simulated wind speeds. Validated against data from Hurricane María (2017), the ML model achieved a precision of 0.59 and an F1-score of 0.57 for the positive failure class. These results reflect the potential of AI/ML to improve the predictability of extreme weather and its cascading implications for sustainable urban infrastructure.