Integrating public land fire data and satellite imagery improves fire frequency estimates across the landscape

Résumé fourni par la source

Background Effective fire management requires accurate knowledge of fire history, often derived from satellite imagery. However, satellites are not well suited to detecting low-intensity fires. Aims We aimed to improve satellite-derived fire frequency estimates by incorporating mapped fire history data from public land and environmental co-variation. Methods Using a generalisable workflow, we applied boosted regression trees, generalised linear, and generalised additive models to predict fire frequency in an eastern Australia case study. Performance of raw and modelled satellite-derived fire frequencies was tested by correlating them with higher-quality public land fire mapping. Key results Satellite-derived data underestimated fire frequency, especially in infrequently burnt areas (i.e. one to six fires in the past 36 years). Generalised linear and generalised additive models improved the correlations relative to the baseline (Pearson’s r = 0.331, to 0.577 and 0.526 respectively). Conclusions Generalised linear and generalised additive models improved fire frequency estimates and were most useful at low fire frequencies. Generalised linear models also had some utility for mapping higher fire frequencies. Implications Satellite-derived fire mapping is widely used in fire science but is likely to underestimate fire activity. Our approach can improve the accuracy of estimates derived from satellite data for fire management and research.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.