Evaluating Fixed-Window Landsat Composites for Historical Burned-Area Mapping Using an Object-Based Reference Archive: Pilot Cases from Zakynthos, Greece

Introduction: Historical burned-area mapping from Landsat is constrained by uncertain event dates, irregular image availability, and seasonal surface change. Rather than proposing a new detection algorithm, this study rigorously evaluates two lightweight fixed-window compositing strategies against a detailed object-based historical burned-area archive for Zakynthos, Greece.

Methods: Three purposively selected large-fire years—1992, 2009, and 2011—were analysed using same-sensor Landsat 5 Collection 2 Level 2 imagery. They were chosen because detailed archive polygons were available, the annual fire footprints showed distinct spatial configurations, and image support differed, with PRE/POST scene counts of 3/3, 2/2, and 1/2, respectively. A May median represented pre-fire conditions and was paired with either a September median or a per-pixel September minimum-NBR mosaic. Identical dNBR and NDVI-loss thresholds, water and pre-fire vegetation masks, spatial seed-support filtering, connected-component filtering, and a 10 ha minimum mapping unit were applied island-wide before post hoc comparison with year-dissolved archive polygons.

Results: September-median IoU ranged from 0.129 to 0.176 (mean 0.153), whereas minimum-NBR IoU ranged from 0.114 to 0.151 (mean 0.128). Minimum-NBR reduced omission by only 0.4–5.0 percentage points but increased commission by 2.5–4.0 points and mapped area by 19–41% relative to the median. This trade-off was consistent with the extreme-value composite retaining seasonal drying and unrelated surface changes alongside fire effects.

Conclusions: For these three purposively selected Zakynthos cases, the May-to-September designs produced low archive-relative agreement and substantial overestimation. The cases are diagnostic rather than statistically representative of the full Landsat record or other Mediterranean environments. The study demonstrates the value of a detailed historical archive for stress-testing lightweight compositing strategies before broader application.