Multi-Sensor Remote Sensing of Forest Biodiversity - Novel Earth Observation Techniques for Forest Structure Analyses and Multi-Scale Characterization of Forests

Kurzfassung

About one third of the Earth's land surface is covered by forest providing habitats for numerous animals and plants. Forests are present in many different biomes, such as the tropical, subtropical, temperate, and boreal biome. Carbon sequestration, water filtration, as well as habitat creation are exemplary ecosystem functions of forests. The provision of those ecosystem functions is depending on structural characteristics of forests, e.g. vertical and horizontal properties, fragmentation, and temporal continuity. Temperature buffering, water storage, and recreational purposes are selected ecosystem services for human well-being, resulting from the interplay of ecosystem functions. In Central Europe, more than three centuries of forest management have altered the natural structure of forests. In Germany, about 95 % of the forests are managed for forestry purposes. In several regions, silvicultural management has led to a structural homogenization: few dominant tree species, age-class structures, as well as deadwood scarcity are typical characteristics. Furthermore, homogeneous forests are specifically susceptible to natural disturbances. Continuous information on forest structure at the national scale for multiple attributes of forest structure and several years is missing to better characterize the recent forest structure dynamics in Germany during times of large-scale disturbance events. In addition, the monitoring of novel forest management techniques aiming to enhance the structural complexity of forests is of high relevance for the development of more natural, biodiverse and resilient forests. The topic of this dissertation addresses those knowledge gaps by assessing recent forest structure changes in Germany at high spatial resolution for multiple years in the context of forest biodiversity. High spatio-temporal observations derived from EO data enable the monitoring of the Earth's surface dynamics. EO data comprises spaceborne derived information from different sensors, such as radar (e.g. Sentinel-1), multispectral (e.g. Sentinel-2), or Lidar (e.g. GEDI), to quantify forest structure and biodiversity. A systematic literature review on forest biodiversity monitoring with a focus on EO data, as part of this dissertation, showed that previous research mostly integrated single-sensor observations from multispectral satellites. In addition, the studies were often limited to sub-national data on forests. Furthermore, the potential of multi-sensor analysis of forests, as well as the integration of long time-series remains understudied. A novel machine-learning workflow making use of multi-sensor EO data was developed to model forest canopy height, total canopy cover, and above-ground biomass from 2017 to 2023 in 10 m spatial resolution for Germany. More precisely, satellite time-series from Sentinel-1 and Sentinel-2 serve as predictor variables for sampling data on forest structure attributes derived from GEDI in order to generate annual products. Both at the national and regional level, forest structure change dynamics were quantified. For two hotspots of forest canopy cover loss in Germany, namely the Harz and Thuringian forest, a reduction in canopy height of more than 20 m, total canopy cover exceeding 50 %, and above-ground biomass density of up to 200 Mg/ha was assessed. In the context of an integrative research project funded by DFG, BETA-FOR, experimental silvicultural treatments were implemented in German broad-leaved forests to assess forest structure-biodiversity relationships. Different arrangements of cuttings (aggregated: gap felling, distributed: selective removal) in combination with various deadwood structures were created to diversify the light conditions and habitats. The implementation of those small-scale treatment patches (50 m x 50 m) results in abrupt forest structure changes. A new methodological framework was set up to identify change points in Sentinel-1 and Sentinel-2 time-series. Based on a comprehensive catalog, various spectral indices were calculated and aggregated per patch and time-step as spatial statistics. Metrics (combination of spectral index and spatial statistic) from Sentinel-1 (n = 98) and Sentinel-2 (n = 903) were evaluated in order to identify metrics best assessing the change in forest structure (implementation of experimental silvicultural treatment). Overall, aggregated treatments were best assessed for both sensors using heterogeneity statistics. Only few distributed treatments could be identified, with benefits for Sentinel-2 metrics. Sentinel-1 VH polarization and Sentinel-2 NMDI determined the most treatment implementations accurately. By integrating forest structure heterogeneity information based on modeled attributes of forest structure, as well as satellite time-series metrics of Sentinel-1 and Sentinel-2, correlation analyses were carried out to in-situ remotely sensed forest structure indicators. MLS and TLS observations hold sub-canopy perspectives which are complementary to the top-of-canopy measurements of spaceborne sensors. The correlation analyses were conducted in the context of the experimental silvicultural treatments of the BETA-FOR project. Although different attributes of forest structure are quantified by the spaceborne and in-situ indicators, namely canopy cover and openness, structural heterogeneity, and structural complexity, strong correlations (|r| > 0.7) were identified among MLS box dimension, MLS canopy cover, TLS COI, Sentinel-1 VH, Sentinel-2 NMDI, and modeled GEDI total canopy cover. Both spaceborne and in-situ indicators of forest structure can accurately delineate among control (unaltered forest structure) and aggregated treatments, but not consistently among control and distributed treatments. Furthermore, a sensitivity towards the presence of standing deadwood in aggregated treatments was found for spaceborne, as well as in-situ indicators. Forest structure-biodiversity relationships were quantified by integrating forest structure heterogeneity indicators based on Sentinel-1 and Sentinel-2 time-series, as well as modeled attributes of forest structure from GEDI data. The indicators of forest structure consistently characterize the main difference in forest structure heterogeneity of control to aggregated treatments, and to a minor extent the structural differences among control and distributed treatments. Taxonomic diversity data of bats, birds, gastropods, hoverflies, insects, moths, spiders, and trees serves as biodiversity reference data. Linear correlations were calculated among forest structure heterogeneity indicators and biodiversity measurements reaching values greater than 0.4 for several taxa: birds, gastropods, hoverflies, insects, and tree species. To sum up, multi-sensor spaceborne data have a great potential for the multi-annual characterization of forest structure change dynamics at the national scale. Both radar and multispectral time-series are suited for the monitoring of novel experimental silvicultural treatments when conducted as gap felling. Correlation analyses among spaceborne and in-situ indicators of forest structure demonstrated the alignment of several metrics, thus confirming the potential of spaceborne indicators to up-scale in-situ remotely sensed indicators in space and time. Several forest structure-biodiversity relationships were identified based on spaceborne multi-sensor data and in-situ biodiversity measurements, suggesting a deeper integration of EO in ecology to better inform forest management for biodiversity preservation.