Optimising fire severity mapping using pixel-based image compositing

Fire severity is closely linked to ecosystem responses. As climate change increases the frequency of severe fires, large-scale fire severity monitoring has become increasingly important. The traditional bitemporal approach, which compares single pre- and post-fire images to map fire severity, is effective at local scales but less efficient for larger-scale assessments. Recent advances in compositing multiple images have improved large-scale mapping; however, the effects of compositing parameters on mapping accuracy remain poorly understood. This study evaluates the impact of compositing parameters on composite-based fire severity maps derived from Landsat images across summer fires at multiple scales. At the large scale, we analysed fires in the Iberian Peninsula (2000−2023); at the medium scale, fires in a Mediterranean region of Spain (1985–2015), validated with national statistics; and at the local scale, two fires with field-based severity data. Various compositing parameters were tested, including the period for selecting pre- and post-fire images: lag timing (initial vs. extended assessment), seasonal timing (summer-summer, spring-summer, and summer-autumn), and the criteria for determining pre- and post-fire pixel values (mean-mean, max-min, and mean-min). Accuracy was assessed using the optimality index, correlation with the bitemporal approach, and alignment with field-based severity data, depending on the scale. Both the compositing period and criteria significantly influenced the accuracy of severity maps. In Mediterranean landscapes, the highest accuracy was achieved using post-fire imagery from the same year as the fire, whereas in alpine areas, it was achieved using post-fire imagery from the year after the fire. Seasonally consistent pre- and post-fire images acquired during summer also improved accuracy. The optimal compositing criteria varied depending on the lag timing: mean-min and max-min were most effective for assessing fire severity shortly after the fire, whereas mean-mean performed better for assessments conducted one year later. This study highlights the importance of carefully selecting compositing parameters based on the ecological characteristics of the study area. Additionally, it presents a scalable, automated methodology for large-scale fire severity mapping using Google Earth Engine, which can be applied to larger regions with appropriate adjustments to the compositing parameters. • Image compositing parameters affect fire severity mapping accuracy. • The optimal lag timing (initial vs. extended) varies by ecosystem. • Seasonal timing differences reduce the accuracy of fire severity maps. • The optimal compositing criteria depends on lag timing. • Fire severity maps and GEE code for the Iberian Peninsula are shared.