Change Detection System from High Resolution Satellite Images

An algorithm to extract building footprints utilising machine learning with aerial imagery and LiDAR

The Challenge

The Queensland Government’s Statewide Land cover and Trees Study (SLATS) is a vegetation monitoring initiative using Landsat images of relatively low spatial resolution (25-30m). While the Study provides a comprehensive picture of anthropogenic clearing across the State, the annual data supply meant that compliance responses were often reactive and limited in their effectiveness and timeliness. In recent years, spatial image resolution and frequency of delivery have increased significantly. In line with this, the Queensland Government’s focus has been on developing proactive and targeted compliance approaches that support early engagement with landholders. The challenge was to develop an early detection system, complementary to SLATS, enabling more frequent and expedient detection of recent clearing events across the State.

Partners

The project partners were the QLD Department of Natural Resources and Mines (DNRM), QLD Department of Environment and Heritage Protection (EHP), Queensland University of Technology (QUT) and the University of Queensland Joint Remote Sensing Research Program.

The Solution 

The project aimed to research, develop and validate change detection algorithms for high-spatial resolution, high-frequency, Planet Mosaic image time series for Queensland’s environments (3m spatial resolution, captured daily), including a solution based on the freely available Sentinel-2 imagery (10m spatial resolution, captured every 5 days). The outputs of the project included:

  1. A vegetation change algorithm for the state of Queensland for both Planet and Sentinel data.
  2. An algorithm for detecting change in mining extents.
  3. Processing imagery on a tile basis and ranking the amount of change between tiles.
  4. A change algorithm for ‘rapid change’ using (near) daily Planet and Sentinel data.
  5. Algorithms for dealing with noise (clouds, terrain shadows, land use, ocean).
  6. Evaluation of the geometric and radiometric accuracy of Planet data.
  7. A web interface that allows users to identify suitable change across and within a tile.
  8. A visualisation for change detection.

The online tool can be used to analyse satellite imagery on a regular basis to identify recent vegetation changes across the state. This information is then cross-referenced with data about exemptions, current notifications, and clearing approvals to help identify unexplained clearing of native vegetation. Rapidly detecting recent changes in native vegetation enables a proactive response and early engagement with landholders. Timely response to potentially unlawful clearing events and providing landholders with information around their vegetation management requirements, reduces further clearing activities and impacts to native vegetation.

Impact

Effective management of the state’s woody vegetation is critical for balancing environmental protection and sustainable industry growth. The Vegetation Management Act 1999 identifies regulated vegetation and their conservation value status, and is underpinned by a vegetation management compliance framework. Regulated vegetation represents approximately 80% of the State, relating to over 90% of properties in Queensland. The project has enabled improved operational capabilities for DNRM, increasing responsiveness to legislative requirements around vegetation clearing.

Contact

To learn more, contact FrontierSI at contact@frontiersi.com.au or Project Manager, Alan Woodley, at a.woodley@qut.edu.au.