Trends and patterns in post-disturbance forest recovery estimated from Landsat and Sentinel-2 data using regression-based spectral unmixing

Abstract

<p>Natural disturbances and post-disturbance recovery are principal drivers of forest ecosystem dynamics and both are sensitive to climate change. While disturbances and their causes and consequences have received considerable attention from the scientific community in recent years, there is – however – a substantial lack of knowledge on post-disturbance recovery. Recovery is considered an essential measure of forest resilience to climate change, especially with regard to ecosystem service provision (e.g., protection from avalanches, water purification). Disturbances remove the top tree canopy, exposing the forest floor composed of different land cover types, such as bare soil, grassland and shrubby vegetation, which will gradually transition to treed vegetation over succession. The assessment of forest recovery by means of medium resolution optical remote sensing data (i.e., ~20 m spatial grain) poses some challenges in analyzing those spatially and temporally heterogenous recovery trajectories. To tackle this problem, we employed a temporally generalized regression-based spectral unmixing approach to dense time series of Landsat and Sentinel-2 data with the aim of characterizing the post-disturbance recovery trajectories across a large study site covering the eastern Alps (~125,000 km²). For training the spectral unmixing approach, we developed a multi-year spectral library for three endmembers: treed vegetation, non-treed vegetation and bare soil. Selection of pure endmembers was based on the LUCAS database, a pan-European disturbance map and Google Earth imageries. Applying the generalized regression-based spectral unmixing approach to a dense time series of Landsat and Sentinel-2 images results in annual fraction maps for the three endmembers, which can be used to characterize recovery trajectories after major disturbance events. Each pixel’s post-disturbance trajectory can thereby be described in a three-dimensional space composed of variable fractions of treed vegetation, noon-treed vegetation and bare soil. To facilitate interpretation of recovery trajectories, we focus on specific disturbance events covering the storms Kyrill (2007), Uschi (2003), and Vaia (2018). This allows for identifying (dis-)similarities between recovery trajectories of the same disturbance event and thus to investigate the full breath of potential recovery patterns after natural disturbances.</p>