AI-driven Evidence-based Decision Making for Sustainable Urban Development

In this talk, Dr Zubair Khalid will provide an overview of the use of technological innovations in data gathering, data analytics, data-driven decision-making, and policy design to address complex problems in three important and intertwined areas of an urban system that are critical for sustainable development: (a) urban sprawl, (b) urban mobility, and (c) urban environment and health. The proliferation of mobile and electronic devices, the availability of publicly accessible sensing modalities, and the increasing digitalization of processes have enabled the generation of unprecedented amounts of data. This, combined with rapid advances in computational power, provides an opportunity to use artificial intelligence and machine learning methods for data-aided analysis, discovery, and decision-making to address the complex issues surrounding urban systems.

Dr. Khalid will present an overview of pilot projects in these areas aimed at establishing a mesh of physical, social, and economic connections in urban systems. In this approach, data collected from a wide variety of modalities (such as remote satellite measurements, surveys, socio-economic indicators, and on-ground sensing networks such as video feeds) are fed into cutting-edge machine learning techniques to model these connections, derive useful insights for informed decision-making, and conceptualize a policy cycle for attaining public value.

Following this overview, Dr Khalid will discuss several recent works in detail. First, STEF-DHNet, a spatiotemporal deep learning model that leverages CNNs and LSTMs to accurately forecast ride-hailing demand while integrating external factors such as weather and time-of-day variations. This model outperforms state-of-the-art baselines and maintains long-term accuracy without frequent retraining. Second, he will showcase an exascale climate emulator, a high-fidelity tool for climate modeling that combines spherical harmonic transforms and mixed-precision GPU computation. Trained on hundreds of billions of data points, it achieves unprecedented accuracy and scalability across leading supercomputers. Finally, Dr. Khalid will present the Variational Mode Graph Convolutional Network (VMGCN) for spatiotemporal traffic prediction—a hybrid framework that decomposes data into interpretable modes using variational mode decomposition before applying graph neural networks, improving forecasting accuracy for both short- and long-term traffic predictions.