The automated temporal burn index (ATBI) for accurate and scalable burned area mapping

Wildfires are recognized as essential climate variables (ECVs) due to their widespread impacts on ecosystems, greenhouse gas emissions, air quality, and the global carbon cycle. Accurate and scalable burned area mapping is therefore critical. However, conventional indices such as the differenced normalized burn ratio (dNBR) are prone to commission errors, often misclassifying unburned vegetation, clouds, shadows, or snow as burned. This study introduces the Automated Temporal Burn Index (ATBI), which leverages the co-occurring spectral signature of fire (NIR↓ and SWIR↑) through a multiplicative formulation. The differenced ATBI (dATBI) measures pre–post fire changes, while the multi-temporal variant (dATBI_tm) incorporates a fixed pre-fire image with multiple post-fire scenes to capture fire progression and suppress transient noise. This study evaluated dATBI and dATBI_tm across 50 wildfire events worldwide using Landsat imagery in Google Earth Engine, benchmarking against dNBR and its multi-temporal version (dNBR_tm). Validation with Random Forest reference classifications showed that dATBI achieved substantially higher precision (∼0.92 vs. 0.73), higher F1 scores (∼0.94 vs. 0.84), and slightly better separability (M = 1.38 vs. 1.29) than dNBR, while maintaining similar recall. ROC analysis confirmed superior discrimination, with dATBI consistently reducing false positives at low false positive rates. Threshold analysis revealed that dATBI is far less sensitive to class break adjustments than dNBR, enabling more consistent severity mapping across ecosystems. The multi-temporal dATBI_tm further improved robustness, accurately reconstructing cumulative burn scars across boreal, tropical, Mediterranean, and savanna ecosystems while suppressing cloud, shadow, and snow artifacts. Finally, sensitivity tests demonstrated that SREM-corrected reflectance yielded cleaner dATBI outputs than standard LaSRC products, underscoring the importance of preprocessing. Overall, dATBI and dATBI_tm provide reliable alternatives to dNBR for operational burned area monitoring. Their robustness to commission error, threshold variability, and atmospheric noise makes them suitable for integration into large-scale wildfire assessment and climate resilience strategies.