Developed a novel transformer-based neural network capable of multi-label benthic classification, taking processed drone imagery as input and identifying pixels as coral cover, rocks, rubble, sand, algae, etc. This model will be used to inform restoration efforts by providing actionable, specific evidence of where corals should be planted (in areas of higher relative live coral cover).
Through NASA's DEVELOP National Program, we worked with the Langley Research Center (LaRC) and partnered with Washington DC's Department of Energy and Environment (DOEE) to create solar potential maps to inform solar panel installation decisions for the District.
An accepted paper at the NeurIPS CCAI Tackling Climate Change with Machine Learning 2021 Workshop. We modified existing deep learning-based super-resolution models, and applied them to satellite data to increase the resolution of wind speeds and solar irradiance fields for informing short-term, local energy planning. We published machine learning-ready wind and solar datasets.