资源环境科技发展态势分析平台

Forest managers require reliable tools to evaluate post-fire recovery across different geographic/climatic contexts and define management actions at the landscape scale, which might be highly resource-consuming in terms of data collection. In this sense, remote sensing techniques allow for gathering environmental data over large areas with low collection effort. We aim to assess the applicability of remote sensing tools in post-fire management within and across three mega-fires that occurred in pine fire-prone ecosystems located along an Atlantic-Transition-Mediterranean climatic gradient. Four years after the wildfires, we established 120 2x2m plots in each mega-fire site, where we evaluated: (1) density of pine seedlings, (2) percentage of woody species cover and (3) percentage of dead plant material cover. These variables were modeled following a Bayesian Model Averaging approach on the basis of spectral indices and texture features derived from WorldView-2 satellite imagery at 2 m spatial resolution. We assessed model interpolation and transferability within each mega-fire, as well as model extrapolation between mega-fires along the climatic gradient. Texture features were the predictors that contributed most in all cases. The woody species cover model had the best performance regarding spatial interpolation and transferability within the three study sites, with predictive errors lower than 25% for the two approaches. Model extrapolation between the Transition and Mediterranean sites had low levels of error (from 6% to 19%) for the three field variables, because the landscape in these areas is similar in structure and function and, therefore, in spectral characteristics. However, model extrapolation from the Atlantic site achieved the weakest results (error higher than 30%), due to the large ecological differences between this particular site and the others. This study demonstrates the potential of fine-grained satellite imagery for land managers to conduct post-fire recovery studies with a high degree of generality across different geographic/climatic contexts.