Sincere appreciation to my co-authors: @mchll-ln.bsky.social , Brefo Dwamena Yaw, Jade Boutot, Mary Kang , @drolnick.bsky.social .

@mila-quebec.bsky.social
‪
#MachineLearning #ComputerVision #ClimateAction #RemoteSensing #aiforgood
Look for our work at #ICML2025!

Our work includes evaluating baseline AI algorithms. While results show promise for automated monitoring, they also highlight the task's complexity, inviting further research to enhance scalability & accuracy.

Dataset: zenodo.org/records/1374...
Preprint: arxiv.org/abs/2410.09032

It is aimed to empower environmental action using remote sensing by enabling the scalable identification of such wells, assisting experts to:
(1) Better estimate methane emissions and groundwater contamination.
(2) Prioritize plugging and remediation of wells.

To address this, we developed the Alberta Wells Dataset (AWD): the first large-scale benchmark for this critical challenge. AWD provides over 213,000 well instances (abandoned, suspended, active) in Alberta, Canada, with high-resolution, multi-spectral satellite imagery.

The undocumented nature of countless such wells creates profound environmental uncertainty. In Canada, for example, abandoned wells are the biggest source of uncertainty in methane emissions estimates.