Enhancing Environmental Enforcement with Near Real-Time Monitoring: Likelihood-Based Detection of Structural Expansion of Intensive Livestock Farms

Ben Chugg, Brandon R. Anderson, Seiji Eicher, Sandy Lee, Daniel E. Ho

TL;DR

This paper introduces a novel likelihood-based method for detecting structural expansion of Concentrated Animal Feeding Operations (CAFOs) using readily available, medium-resolution satellite imagery. By combining state-of-the-art building segmentation with a time-series change detection model, this approach offers a more efficient and scalable way to monitor environmental compliance compared to traditional methods.

What Problem Does This Paper Address?

Environmental enforcement in the US often relies on infrequent physical inspections or self-reported data, which are resource-intensive and can miss critical compliance issues. A significant challenge is the unpermitted structural expansion of Concentrated Animal Feeding Operations (CAFOs), which pose substantial health and environmental risks. Existing methods for monitoring such changes often require high-resolution imagery that is not frequently updated, creating a gap in timely enforcement capabilities. This paper aims to bridge this gap by enabling near real-time monitoring using more accessible, albeit lower-resolution, satellite data.

Methods

The authors propose a three-step approach:

  1. Segmentation: They employ a U-Net architecture to segment CAFO structures from satellite imagery, generating pixel-level probabilities of belonging to a CAFO shed. This step is crucial for identifying potential building areas.
  2. Footprint Modeling and Change Detection: A maximum likelihood estimation (MLE) model is fitted to the time-series of segmentation probabilities. This model estimates the initial building footprint ($F^0$), the footprint of the expanded area ($F^+$), and the time of expansion ($t^*$). The core idea is to model the transition of the building footprint over time.
  3. Likelihood-Based Test Statistic: A test statistic is computed by comparing the likelihood of the observed data under a model that allows for expansion versus a model that assumes no expansion. A higher test statistic indicates a greater likelihood of structural expansion.

This method is designed to be lightweight and can work with various segmentation models, making it adaptable.

Datasets

  • Training and Validation: The authors created a new dataset of 145,053 hand-validated segmented satellite images (GeoTiffs) for 1,513 CAFO facilities in Indiana. This dataset covers the period from January 1 to December 31, 2019, using Planet's 3m/pixel resolution imagery. The data was augmented with labels from the Microsoft Building Footprint dataset and the USGS of poultry barns in the DelMarVa peninsula.
  • Ground Truth for Expansion: 22 sites were identified as having constructed at least one new CAFO building during the observation window, while 1,414 showed no evidence of new construction.

Key Insights and Ideas

  • Leveraging Medium-Resolution, High-Cadence Imagery: The paper demonstrates that even with 3m/pixel resolution imagery (which is coarser than previously used for similar tasks), it's possible to achieve robust detection of structural changes by exploiting the high temporal frequency (daily to weekly) of the data.
  • Generalizability of the Approach: The core likelihood-based change detection method is designed to be independent of the specific segmentation model used, making it adaptable to different environments and types of structures beyond CAFOs.
  • Focus on Building Footprint Dynamics: Instead of just detecting anomalies, the method models the evolution of building footprints over time, providing a more nuanced understanding of expansion events.
  • Prioritizing Enforcement Resources: The confidence scores generated by the MLE model allow enforcement agencies to prioritize facilities that are most likely to have expanded, leading to more efficient allocation of limited resources.

Key Results

  • High Performance: The proposed MLE method achieved a balanced accuracy of 78.6% and a macro F1-score of 0.60, significantly outperforming baseline methods like BCP, BFAST, and DDM.
  • Improved Resource Efficiency: By sorting facilities by their confidence score, the MLE approach reduced the number of false positives by over 80% compared to a random search, meaning fewer non-expanded facilities need to be inspected.
  • Correlation with Expansion Size: The MLE method showed a statistically significant positive correlation (0.67) between the confidence of expansion and the actual size of the expansion, suggesting the model is capturing meaningful indicators of growth.
  • Robustness to Segmentation Accuracy: An ablation study showed that even with a 5% degradation in segmentation accuracy, the MLE approach's performance only dropped slightly, indicating its robustness.

Limitations and Future Work

  • Segmentation Model Dependency: While the MLE method is designed to be segmentation-agnostic, its performance is still influenced by the quality of the segmentation model. Developing more accurate CAFO-specific segmentation models is suggested.
  • End-to-End Deep Learning: The authors suggest exploring end-to-end deep learning approaches for changepoint detection with a larger labeled dataset.
  • Generalization to Other Environmental Issues: The paper highlights that the core approach can be adapted to detect structural changes for other environmental challenges, such as zoning violations, habitat modification, and deforestation, by modifying the segmentation model.
  • Incorporating Additional Information: Future work could involve modifying the likelihood model to incorporate other contextual information, such as proximity to existing structures, conformity with building directions, or facility closures.
  • Combining DDM with Segmentation: The authors suggest combining the Dynamic Detection Model (DDM) with segmentation to potentially improve its performance and reduce false positives caused by landscape changes.

Summary updated

Abstract

Much environmental enforcement in the United States has historically relied on either self-reported data or physical, resource-intensive, infrequent inspections. Advances in remote sensing and computer vision, however, have the potential to augment compliance monitoring by detecting early warning signs of noncompliance. We demonstrate a process for rapid identification of significant structural expansion using Planet's 3m/pixel satellite imagery products and focusing on Concentrated Animal Feeding Operations (CAFOs) in the US as a test case. Unpermitted building expansion has been a particular challenge with CAFOs, which pose significant health and environmental risks. Using new hand-labeled dataset of 145,053 images of 1,513 CAFOs, we combine state-of-the-art building segmentation with a likelihood-based change-point detection model to provide a robust signal of building expansion (AUC = 0.86). A major advantage of this approach is that it can work with higher cadence (daily to weekly), but lower resolution (3m/pixel), satellite imagery than previously used in similar environmental settings. It is also highly generalizable and thus provides a near real-time monitoring tool to prioritize enforcement resources in other settings where unpermitted construction poses environmental risk, e.g. zoning, habitat modification, or wetland protection.
BibTeX
@article{chugg2021enhancing,
  author = {Ben Chugg and Brandon R. Anderson and Seiji Eicher and Sandy Lee and Daniel E. Ho},
  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},
  year = {2021},
  doi = {10.1016/j.jag.2021.102463},
}
Figure