Early Detection of Overwintering Fires Using Sentinel-1 InSAR Deformation Time Series

Overview

Overwintering fires (OWFs), in which smoldering peat combustion persists underground through boreal winters and re-ignites the following spring, represent a growing but poorly monitored feedback in the global fire regime. Existing detection methods based on optical and thermal satellite imagery cannot sense subsurface combustion, leaving a critical gap in early warning capability. This study demonstrates that Interferometric Synthetic Aperture Radar (InSAR) time series from Sentinel-1, combined with terrain covariates derived from global digital elevation models, can detect OWFs before surface re-ignition. Three physically interpretable deformation features, measuring the acceleration, velocity change, and cumulative displacement of ground subsidence driven by thermokarst collapse, are consistently elevated over OWF source zones across every season and region tested, as predicted by a first-order mechanistic model of peat combustion and talik expansion. A terrain-aware latent-nucleus detector, formulated as a multiple-instance learning framework that scores the most anomalous compact patch within each previous-burn component while accounting for elevation, component size, surface roughness, slope, and relative elevation, achieves bag-level AUROC of 0.877 under leave-one-season-out cross-validation in Yakutia (Russia) and 0.841 under strict zero-shot transfer to an independent Alaska (USA) site. The terrain covariates suppress physiographic confounds that mimic OWF subsidence signatures, improving the balanced metric by 9.1 % over an InSAR-only baseline (0.771) while requiring no region-specific tuning. With region-specific adaptation, Yakutia AUROC exceeds 0.9. These results establish terrain-augmented InSAR deformation monitoring as a viable, physics-grounded approach for ranking previous-burn areas by OWF risk weeks to months before surface re-ignition.

Competition history

• CSEF 2026 Earth & Environmental Sciences(Senior Division) · Entry S-08-16

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Earth and Environmental Sciences projects
• CSEF Earth and Environmental Sciences projects
• Every CSEF 2026 project