Wildfires play a key role on forest composition and structure in the Mediterranean biomes. Hence, Mediterranean species are adapted to fire, developing ecological strategies to naturally recover. Nevertheless, climate change impacts and land use changes are expected to increase the frequency and intensity of extreme wildfire events, endangering forest resilience to fire. Combining LiDAR and Landsat data provides a valuable opportunity to temporally extend detailed information on the forest structure. This study attempts to evaluate the feasibility of extrapolating LiDAR-derived canopy cover variables, as indicators of vegetation recovery, to Landsat time-series using Support Vector Regression (SVR) in a large forest fire. Canopy Cover (CC) and Canopy Cover above 2 m (CC2m) were derived from LiDAR data acquired in 2009 and 2016 from the National Plan for Aerial Orthophotography of Spain (PNOA) and time-series of annual Landsat composites for the period 1990-2020 were generated through the Google Earth Engine platform. We calibrated a SVR model from a stratified random sample using a 60% of the sample from 2016 for calibrating and the remaining 40% from both 2016 and 2009 for spatial and temporal validation, respectively. The two canopy cover variables yielded highly acceptable accuracy, with an R2 of 0.78 (CC) and 0.64 (CC2m), and an RMSE around 12.5-15% for the spatial validation, and with an R2 of 0.74 (CC) and 0.51 (CC2m), and an RMSE around 14-16.5% for the temporal validation. These results ensure the applicability of the extrapolation of the LiDAR-derived canopy cover variables to Landsat timeseries..
Wildfires are one of the most widespread disturbances of forest ecosystems. Countries of the Mediterranean basin registered the largest number of fires and burned area in the last decade. Optical remote sensing, particularly Landsat images, has been commonly used to characterise forest disturbance and subsequent recovery for long time series. Time series techniques such as temporal segmentation algorithms have been developed to facilitate the understanding of postfire vegetation recovery dynamics. This study aims to extract the main types of natural recovery trajectories from a Large Forest Fire occurred in 1994 from a thermophilous pine forests (Pinus Halepensis and Pinus Pinaster) in the long-term (1994-2018). We built annual composites from Landsat Surface Reflectance images and calculated Tasseled-cap components, which are sensitive to canopy moisture and structure (Wetness - TCW) and percent vegetation cover (Angle – TCA). We evaluated fire severity and fire recovery relationship. The differenced Normalised Burn Ratio (dNBR) was used as a fire severity proxy, whereas recovery processes were assessed from spectral profiles using LandTrendr temporal segmentation algorithm. TCW and TCA were used as inputs to LandTrendr and the outputs of fitting were subsequently used to classify recovery types based on a k-means classification with the optimum number of clusters based on the Elbow Method. Groups of continuous positive recovery, non-continuous recovery and continuous recovery with slope changes were identified. The proposed method could be an approach to model the long-term recovery for the Mediterranean areas and help decisionmakers in determining which areas could not recover naturally.