Likelihood-based expansion detection for satellite imagery

This repository contains the code for the paper Enhancing Environmental Enforcement with Near Real-Time Monitoring: Likelihood-Based Detection of Structural Expansion of Intensive Livestock Farms in the Journal of Applied Earth Observation and Geoinformation.

In addition to the code to run the proposed maximum-likelihood model, we also provide the code for the Dynamic Detection Model (DDM) of Koltunov et al. (2009, 2020), as it is one of our baselines and could not be found elsewhere.

The repo is organized as follows:

• methods directory

  • likelihood.py code for MLE model
  • ddm.py code for DDM

• notebooks directory

  • inference.ipynb notebook to load and run UNet segmentation model over the imagery.
  • mle.ipynb example notebook for running MLE model
  • ddm_setup.ipynb organizes imagery into weeks, which is required for DDM. DDM is a temporal model; we train one model per week, learning parameters over the previous 10 weeks to predict the subsequent week.
  • ddm_poc.ipynb basic proof-of-concept for DDM. Should help provide some intuition of the model.
  • ddm_train.ipynb code for training and predicting expansions with DDM.

• models directory

  • code to load UNet

The UNet checkpoint, data, and labels may be found on Dropbox.

As per our Planet license, we cannot share the geoinformation of the image files. Therefore, we've stored each image as an array of size (200,200,4). The four bands are red, green, blue, and near-infrared, in that order. You can plot an image array arr in rgb as

import matplotlib.pyplot as plt 
plt.imshow(arr[:,:,:3])

The images for each location are stored as a pickle file, e.g.,loc_0038.p. There should be 1436 pickle files.Each file is a dictionary whose keys are dates (e.g., 20190829 for Aug 29, 2019). The value for each key is the (200,200,4) dimensional image array. If there were multiple images per date, then we add the suffix -2, -3, etc, e.g., 20190829-2.

If using this repo, please cite

@article{Chugg2021enhancing,
title = {Enhancing environmental enforcement with near real-time monitoring: Likelihood-based detection of structural expansion of intensive livestock farms},
journal = {International Journal of Applied Earth Observation and Geoinformation},
volume = {103},
pages = {102463},
year = {2021},
issn = {0303-2434},
doi = {https://doi.org/10.1016/j.jag.2021.102463},
url = {https://www.sciencedirect.com/science/article/pii/S0303243421001707},
author = {Ben Chugg and Brandon Anderson and Seiji Eicher and Sandy Lee and Daniel E. Ho}
}