Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Harnessing Deep Learning and Satellite Imagery for Post-Buyout Land Cover Mapping

Hakan T. Otal, Elyse Zavar, Sherri B. Binder, Alex Greer, M. Abdullah Canbaz

TL;DR

This work addresses the challenge of understanding land-use outcomes on properties acquired through hazard mitigation buyouts by combining public datasets (FEMA HMGP addresses) with high-resolution satellite imagery and CNN-based land-cover classification. The authors compare five CNN architectures, select DenseNet-201 as the best performer (achieving a ROC-AUC of 98.86% in multi-class land-cover classification), and provide parcel- and image-level insights into post-buyout landscapes. Their results reveal a mix of vegetation and built, impervious features on buyout parcels, highlighting both ecological opportunities and ongoing infrastructure, which informs resilience planning and policy. The approach offers a scalable pathway to monitor and evaluate post-buyout land use, with potential extensions including parcel-level segmentation, temporal analysis, and closer integration with urban planning and disaster policy.

Abstract

Environmental disasters such as floods, hurricanes, and wildfires have increasingly threatened communities worldwide, prompting various mitigation strategies. Among these, property buyouts have emerged as a prominent approach to reducing vulnerability to future disasters. This strategy involves governments purchasing at-risk properties from willing sellers and converting the land into open space, ostensibly reducing future disaster risk and impact. However, the aftermath of these buyouts, particularly concerning land-use patterns and community impacts, remains under-explored. This research aims to fill this gap by employing innovative techniques like satellite imagery analysis and deep learning to study these patterns. To achieve this goal, we employed FEMA's Hazard Mitigation Grant Program (HMGP) buyout dataset, encompassing over 41,004 addresses of these buyout properties from 1989 to 2017. Leveraging Google's Maps Static API, we gathered 40,053 satellite images corresponding to these buyout lands. Subsequently, we implemented five cutting-edge machine learning models to evaluate their performance in classifying land cover types. Notably, this task involved multi-class classification, and our model achieved an outstanding ROC-AUC score of 98.86%

Harnessing Deep Learning and Satellite Imagery for Post-Buyout Land Cover Mapping

TL;DR

This work addresses the challenge of understanding land-use outcomes on properties acquired through hazard mitigation buyouts by combining public datasets (FEMA HMGP addresses) with high-resolution satellite imagery and CNN-based land-cover classification. The authors compare five CNN architectures, select DenseNet-201 as the best performer (achieving a ROC-AUC of 98.86% in multi-class land-cover classification), and provide parcel- and image-level insights into post-buyout landscapes. Their results reveal a mix of vegetation and built, impervious features on buyout parcels, highlighting both ecological opportunities and ongoing infrastructure, which informs resilience planning and policy. The approach offers a scalable pathway to monitor and evaluate post-buyout land use, with potential extensions including parcel-level segmentation, temporal analysis, and closer integration with urban planning and disaster policy.

Abstract

Environmental disasters such as floods, hurricanes, and wildfires have increasingly threatened communities worldwide, prompting various mitigation strategies. Among these, property buyouts have emerged as a prominent approach to reducing vulnerability to future disasters. This strategy involves governments purchasing at-risk properties from willing sellers and converting the land into open space, ostensibly reducing future disaster risk and impact. However, the aftermath of these buyouts, particularly concerning land-use patterns and community impacts, remains under-explored. This research aims to fill this gap by employing innovative techniques like satellite imagery analysis and deep learning to study these patterns. To achieve this goal, we employed FEMA's Hazard Mitigation Grant Program (HMGP) buyout dataset, encompassing over 41,004 addresses of these buyout properties from 1989 to 2017. Leveraging Google's Maps Static API, we gathered 40,053 satellite images corresponding to these buyout lands. Subsequently, we implemented five cutting-edge machine learning models to evaluate their performance in classifying land cover types. Notably, this task involved multi-class classification, and our model achieved an outstanding ROC-AUC score of 98.86%
Paper Structure (12 sections, 7 figures, 1 table)

This paper contains 12 sections, 7 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methodology
  3. Datasets
  4. Preprocessing
  5. Image Classification Models
  6. Best Model Selection
  7. Best Confidence Threshold
  8. Hardware Requirements
  9. Experimental Results
  10. Interpreting the Results
  11. Conclusion
  12. Acknowledgements

Figures (7)

  • Figure 1: Land use class distribution of UC Merced Dataset (training data)
  • Figure 2: Training losses of different computer vision models over epochs
  • Figure 3: Validation losses of different computer vision models over epochs
  • Figure 4: DenseNet201's F1-scores over various confidence thresholds
  • Figure 5: The sample of model's land use class predictions of collected satellite images
  • ...and 2 more figures