This repository contains code for a project trying to detect environmental violations in real-time using satellite imagery. Specifically, we detect instances of winter land application, the practice of farms spreading manure on snow-covered fields.
-
cikm: contains code and artifacts for the paper Detecting Environmental Violations with Satellite Imagery in Near Real Time: Land Application under the Clean Water Act, in Conference of Information and Knowledge Management (CIKM), 2022. This paper and code describes the training, tuning and analysis of the object-detection model. -
elpc_pipeline: contains code and details for a pipeline to run the object detection model and output results of suspected land application events for a given set of locations. This is the pipeline deployed in the winter 2022 field trial with Environmental Law and Policy Center (ELPC) and Wisconsin Department of Natural Resources (WDNR), using the best model from the CIKM paper. It is also a currently functional pipeline that continues to be used by ELPC as they take on an additional piloting and testing of the model in winter 2023. -
field_trial_2022_analysis.ipynb: a Jupyter notebook containing analysis from the winter 2022 field trial with ELPC and WDNR. The data used in this analysis has been published to an open Hugging Face repository here.