Adrian Albert / Pouya Vahmani Contact: email@example.com
Fast, granular, and accurate modeling and scenario generation of the spatio-temporal distribution of wind speed fields at a local scale is of key importance across a variety of applications in climate, energy, and urban sciences, e.g., estimating the risk to urban infrastructure of extreme wind events to forecasting the potential of wind generation at wind farms situated in heterogeneous environments. The prevalent suite of numerical physics-based models for wind such as the Weather Research and Forecasting (WRF) model are computationally expensive to run at large scale and granular level, rely on empirical parametrizations to resolve more granular scales, are complex and time-consuming to set up, and are not informed by real-time observational data such as from remote-sensing satellites.
The goal of this project is to develop a machine learning-based emulator for a numerical physics-based model such as the WRF. We will investigate the use of conditional generative adversarial networks for learning the complex mappings between affecting factors such as terrain topography, urban building geometry, and meteorological forcing variables measurable via remote-sensing and physics-based model output. Such a hybrid AI emulator will allow much faster inference of wind fields as well as realistic scenario generation to accurately resolve distribution tails for estimating the risk of extreme wind events.
– Comfortable with Python programming (object-oriented programming, numpy, matplotlib, pandas etc.)
– Familiarity with modern machine learning frameworks such as PyTorch or TensorFlow
– Knowledge and familiarity with modern machine learning, including deep learning. At a minimum, class work and projects, but previous research work using/developing deep learning methods preferred (ideally having worked with GANs)
-Interest in scientific applications of AI including one or more of climate, energy, materials