Automating Mapping and Data Outputs for Post-Fire Assessments
A Case Study of the 2019 Sabarah Fire, Lamington National Park
Abstract
Wildfires are a significant threat to ecosystems worldwide, with increasing frequency and severity due to climate change. Accurate mapping and assessment of burnt areas are crucial for understanding fire behaviour, ecological impacts, and guiding post-fire management strategies. Remote sensing techniques have revolutionized our ability to monitor and analyse fire events across large landscapes. The 2019 Sarabah bushfire in Lamington National Park, Queensland, Australia, provides an opportunity to evaluate and enhance these mapping techniques.
This study aims to automate the creation of burnt area and fire severity maps using satellite imagery and advanced spectral indices. We focus on the Burned Area Index for Sentinel-2 (BAIS2), which leverages the satellite's unique combination of visible, red-edge, near-infrared, and shortwave infrared bands. By applying BAIS2 to Sentinel-2 imagery of the Sarabah fire, we demonstrate its effectiveness in capturing burn patterns across the diverse subtropical rainforest ecosystem of Lamington National Park.
Methodology
The study utilized the Digital Earth Australia (DEA) Sandbox environment to process Sentinel-2 satellite imagery covering the Lamington National Park area during the 2019 Sarabah fire event. The methodology involved:
- Acquisition and preprocessing of Sentinel-2 imagery from September to December 2019
- Application of the Burned Area Index for Sentinel-2 (BAIS2) to detect burnt areas
- Development of a delta BAIS2 (dBAIS2) approach to compare pre-fire and post-fire conditions
- Classification of burn severity using standardized Fire Extent and Severity Mapping (FESM) metrics
- Validation against Queensland Parks and Wildlife Service (QPWS) official fire maps
The study site polygon was sourced from QSpatial and used to mask the analysis to the boundaries of Lamington National Park. All data was reprojected to EPSG:9473 for consistent spatial analysis.
Key Findings
- •Strong agreement in burnt area extent: The BAIS2-derived burnt area map (1,499.28 hectares) showed close alignment with QPWS assessments (1,576.55 hectares), with only 5.15% difference.
- •Challenges in severity classification: The automated severity classification showed limited correlation with official assessments, with an overall accuracy of 41.57%.
- •Potential for optimization: Threshold analysis suggested improved classification accuracy could be achieved with optimized parameters, reaching up to 57.79% accuracy.
Burnt Area Comparison
| Method | Area (hectares) |
|---|---|
| BAIS2 (This study) | 1,499.28 |
| QPWS (dNBR) | 1,576.55 |
| Difference | 5.15% |
Severity Classification
| Severity Class | BAIS2 (%) | QPWS (%) |
|---|---|---|
| Low | 48.6% | 23.6% |
| Moderate | 33.6% | 49.7% |
| High | 8.7% | 24.7% |
| Extreme | 9.1% | 2.1% |
Conclusions
This research demonstrates both the potential and limitations of automated fire mapping approaches in subtropical rainforest environments. The successful automation of burnt area mapping through DEA Sandbox offers significant operational advantages for rapid assessment, while the challenges encountered in severity classification emphasize the complexity of fire impacts in rainforest ecosystems.
Key recommendations include:
- Implementation of the automated burnt area mapping workflow as a rapid assessment tool
- Development of a hybrid approach to fire severity mapping that combines automated spectral analysis with expert interpretation and ground validation
- Integration of multiple data sources beyond spectral indices, including LiDAR data, multiple spectral indices, ground-based validation points, and pre-fire vegetation maps
- Incorporation of detailed vegetation classification layers to improve interpretation of fire impacts in heterogeneous landscapes
This study contributes to the evolving field of automated fire mapping by clearly identifying both the capabilities and limitations of current approaches, particularly in complex and ecologically significant landscapes like the Gondwana Rainforests World Heritage Areas.