Enhancing post-fire forest recovery monitoring through a remote sensing perspective

Abstract

Forests are essential for human well-being, providing a wide range of benefits like climate regulation, biodiversity conservation, watershed protection and prevention of soil erosion. Although forests have historically been modulated by a multitude of disturbances, they now face unprecedented challenges due to changes in climate and land use. In Mediterranean forests, fire is one of the most common disturbance agents, shaping their structure, composition and functioning. Mediterranean species exhibit adaptive mechanisms to resist and recover, thus being considered fire-resilient ecosystems. Nonetheless, recovery may be hampered by the expected increasing exposure to more frequent and severe fire events. Yet, estimating recovery poses a challenge as it is a dynamic process spanning different spatial and temporal scales. Providing systematic and spatio-temporally explicit information is therefore pivotal to better understand changes in vegetation dynamics in response to fire disturbance. The overall objective of this thesis is to contribute to the understanding of postfire forest recovery in Mediterranean ecosystems using remotely sensed data from active and passive sensors. The main goal is conducted through the following specific objectives: 1. To obtain the post-fire recovery trajectories from Landsat time series. 2. To appraise recovery rates and driving factors of forest recovery. 3. To analyse forest structural changes along the post-fire recovery process by combining LiDAR and Landsat data. 4. To quantify changes in cover composition at the subpixel level from unmixing Landsat data. These objectives are addressed through three papers that have been published in relevant scientific journals. In Paper I we approached the objectives 1 and 2. Through two case studies in Mediterranean pine forests in Spain, we characterised post-fire spectral recovery dynamics at successional stages. We identified different categories of spectral recovery trajectories using temporal segmentation of Landsat time series (1994–2018) and K-means clustering. LandTrendr algorithm was used to derive trajectory metrics from Tasseled Cap Wetness (TCW), sensitive to canopy structure, and Tasseled Cap Angle (TCA), related to vegetation cover gradients. Different categories of post-fire trajectories revealed processes of continuous recovery (continuous recovery, continuous recovery with slope changes, continuous recovery stabilised) and non-continuous recovery. As fire-prone ecosystems, vegetation quickly colonised the space after fire by displaying higher recovery rates in the short-term, but this does not imply the recovery to the pre-fire forest conditions two decades after fire. We further evaluated the influence of environmental and contextual factors on recovery rates. The modelling results indicated that recovery rates were strongly related to fire severity in the short term, whereas climatic conditions in relation to drought were more determinant in the long-term. In paper II we approached the third objective and combined LiDAR data and Landsat imagery to provide insights on the return of forest structure after fire in fire-prone Mediterranean pine forests in the SE of Spain. We addressed the extrapolation of forest structural variables (Vegetation Cover, Tree Cover, Mean Height and heterogeneity) over three decades (1990-2020) using a Support Vector Regression model (SVR). Model performances to estimate LiDAR-derived structural variables using Landsat images and topographic variables was high, showing stability of the estimations both temporally and spatially. Time-series of structural recovery underlined that less than 50% of burned pixels completely recovered to a pre-fire structure 26 years after fire, suggesting an ongoing recovery process. In paper III we approached the fourth objective and developed a methodology to quantify changes in woody-vegetation (tree and shrub) cover composition using a regression-based unmixing approach from Landsat Spectral Temporal Metrics (STM). We used synthetically mixed training data from Landsat STM as input for a SVR model to disentangling tree and shrub cover dynamics in Mediterranean forests, yielding spatio-temporally explicit information on post-fire forest compositional recovery. Our findings suggest that successional dynamics of tree and shrub strongly depended on pre-fire conditions since the majority of the burned areas tended to the pre-fire composition. However, areas shifting from tree to shrub dominance were found 26 years after fire, indicating ongoing transitions that may constitute a successional stage or would prevail in a mature stage. Our results emphasise the utility of unmixing Landsat data to gather information on shifts in composition along the recovery process. Providing retrospective information on post-fire recovery dynamics can potentially support post-fire forest management by acknowledging the spatio-temporal patterns of forest recovery. Enhancing forest resilience and adaptation pose a challenge for forest managers because Mediterranean forests highly subjected to fire occurrence are also those that face changes in fire regimes along with susceptibility to other disturbances. Estimations of post-fire recovery from remotely sensed data can therefore provide a basis for forest management strategies to better cope with climate change and facilitate decision-makers the selection of management alternatives.