Anomalous Moisture Signal in Sentinel-2 Imagery Precedes Overwintering Wildfire

Abstract. Not all Canadian wildfires are extinguished by winter snowfall, instead smouldering underground for months before reemerging as flaming wildfires the following spring. Although these overwintering wildfires are difficult to manage and prevent, until recently this has been an uncommon phenomenon. This changed following the record-breaking 2023 wildfire season, when dozens of suspected overwintering wildfires reemerged in the spring of 2024, especially in the western boreal. In this study we identify overwintering wildfires through co-located thermal anomalies from late 2023 and early 2024. We investigate pre-fire multispectral Sentinel-2 imagery within areas burned by suspected overwintering fires, evaluating six spectral indices against their historical distributions to identify anomalies as potential fingerprints of elevated fire susceptibility. We identified 25 suspected overwintering fires, accounting for only 1.3% of the wildfires from 2024 but a disproportionate 22.8% of the total area burned. Relative to the reference period, pre-fire Normalized Difference Moisture Index (NDMI) was −0.0654 and Normalized Difference Vegetation Index (NDVI) was −0.1082. Similar patterns were identified in Green Normalized Difference Vegetation Index (GNDVI), Normalized Difference Water Index (NDWI), and SWIR-Transformed Reflectance (STR). Normalized Difference Snow Index (NDSI) was not substantially different from the reference period. Critically, we did not detect statistically significant differences between overwinter burned areas and comparable unburned areas in the same region, suggesting that conditions conducive to overwintering were widespread. These results suggest that multispectral imagery could be valuable as an early-warning system for the emerging threat of overwintering wildfires.