Plastic waste is a significant environmental pollutant that is difficult to monitor. We created a system of neural networks to analyze spectral, spatial, and temporal components of Sentinel-2 satellite data to identify terrestrial aggregations of waste. The system works at wide geographic scale, finding more than 4,000 waste sites in 112 countries across Southeast Asia.

The details of this work are published in the journal PLOS ONE. Code and data is available on GitHub.

Amazon Mining Watch uses machine learning to map the scars of mining activities in the Amazonian countries. By constantly analyzing high-resolution and historical satellite images, this tool aims at identifying the fast-paced growth of open-pit mining in the largest rainforest in the world. This database is here to help journalists, activists, and researchers better understand the causes and impacts of the mining industry.

Code and methodology is available on GitHub.

Old Faithful is a geyser in Yellowstone National Park. After visiting Yellowstone in 2018, I wondered if a neural network might be able to tease out more subtle correlations in the eruption patterns to be able to forecast more accurately. Can we predict its next eruption using a neural network?

This was a simple project to use learned image colorization (DeOldify) images from an electron microscope, which can only produce grayscale outputs. Though the algorithm was heavily biased towards flesh tones for these out-of-domain samples, certain structires can have quite striking results. In the future, I would like to refine the DeOldify model by including some manually colorized SEM images to the training set, or refine the colorization network to incorporate color hints given by the user.