Boreal forests represent a third of all forests on Earth, storing over 30% of all carbon present in the terrestrial biome, and play a critical role in global carbon dynamics, climate regulation, and for biodiversity. Despite their global ecological importance, boreal forests remain poorly understood, particularly in terms of high-resolution forest-type distribution, key drivers, and successional dynamics, all of which are crucial for predicting climate change impacts on boreal forests. Northern boreal regions' vast and remote nature limits inventory data and constrains field data collection, resulting in significant knowledge gaps. Remote sensing methods, coupled with field data, offer a powerful approach to addressing these challenges. In this thesis, I utilized multi-source and multi-scale remote sensing data streams and prediction analyses to enhance our understanding of the eastern Siberian and western North American boreal forests' overall composition, structure, and successional dynamics. The teams of the Alfred Wegener Institute investigated hundreds of forested sites throughout the circum-boreal (Alaska, Canada, Siberia) where they established forest inventories and acquired UAV-LiDAR (Unpiloted Aerial Vehicle; Light Detection and Ranging) data. These collected datasets serve as the basis for ground reference and validation for all studies in this thesis. This dissertation presents three studies applying forest ecology research using a unique combination of in-situ, UAV, and spaceborne remote sensing data at fine to coarse scales and different phenological phases. Through the three manuscripts and synthesis, I aim to answer the following research questions: - What are the dominant compositional patterns of boreal forests at their northern edges, and how do ecological gradients shape the distribution of different forest types? - How do structural characteristics of boreal forests vary across scales, and what do these patterns reveal about forest development and resilience? - How do seasonal phenological changes and spatial scales affect the detectability and interpretation of forest composition and structure? Motivated by said research questions, in the first study, I assessed the composition of the boreal forests in five regions of eastern Siberia, along a southwest-northeast transect. Using Sentinel-2 optical satellite late-summer imagery, I identified the drivers and ecological niches of the forest types using predictive models (Chapter 2). The resulting forest-type maps reached 78% accuracy, and provide higher accuracy and more thematic details than other available land cover maps (from ESA and Copernicus) in the boreal forests. Using generalized linear models (GLMs) coupled with bioclimatic, topographic, and ground surface temperature variables, I identified the main drivers of mapped forest types at the local and regional scales. Results show that mean annual temperature and mean summer and winter temperatures are the most influential predictors of forest-type distribution, and that topography has an impact at the local scale. Furthermore, I delineated potential versus realized forest-type niches and their applicability to other sites, and found that the filling of climatic environmental niches by forest types decreases with geographic distance. Following that, in a second study (Chapter 3), I characterized the structure and successional stages at 48 sites of the western North American boreal forest using high-resolution UAV-LiDAR-RGB and ecological network methods. I derived spectral and structural tree attributes from the UAV data and classified trees into plant functional types representative of boreal forest succession (here, needleleaf evergreen and deciduous broadleaf). I built an ecological network to characterize successional stages, their interactions, and assessed future stage transitions using forest patches of 20x20 meters and derived tree characteristics. Tree height and spectral variables are the most influential predictors of plant functional type in random forest algorithms, and the results reached high overall accuracies. The network-based analysis highlighted five interconnected successional stages that could be interpreted as ranging from early to late successional and a disturbed stage. Disturbed sites are mainly located in Interior and Southcentral Alaska, while late successional sites are predominant in the southern sites of our study region in Canada. Transitional stages are mainly located near the tundra-taiga boundary. These findings highlight the critical role of disturbances, such as fire or insect outbreaks, in shaping forest succession in Alaska and Northwest Canada. Lastly, in the third manuscript (Chapter 4), I explored the potential of upscaling UAV-LiDAR to Harmonized Landsat Sentinel-2 (HLS) optical satellite data to predict boreal forest structure. I used UAV-LiDAR across all our study sites in North America to derive canopy height and crown coverage variables. Then, I built random forest regression models with HLS spectral bands and indices as input to predict canopy height and crown cover throughout our North American sites at HLS satellite data 10 m resolution. The results show strong relationships between HLS spectral and UAV-LiDAR structural metrics, and a higher predictive performance for dense and sparse forests than for medium-density forests. I also compared the UAV-LiDAR crown cover estimates to the NASA ABoVE tree canopy cover product and identified overestimation of Crown Cover in the treeline ecotones. These findings demonstrate a scalable approach for predicting forest structure in the western North American boreal forests using UAV-LiDAR and spectral HLS satellite data and highlight the value of fine-scale structural data for algorithm building and assessments of satellite-derived products. For this study, I built an open-access HLS–forest structure dataset, containing HLS pixel-wise HLS spectral characteristics from peak and late summer together with labeled forest structure information. The current thesis therefore combines multi-source data (forest inventories, UAV-LiDAR and multispectral, medium resolution optical satellite) from different spatial scales to fill knowledge gaps concerning the distribution and structure of boreal forests, key drivers, successional dynamics, resilience, and scalability of forest structural variables.
Léa Chloé Marie Enguehard (Thu,) studied this question.