Hannes Feilhauer

The Arctic is undergoing rapid environmental change, with vegetation responses that feed back into surface energy balance, permafrost dynamics, and biogeochemical cycling at global scale. Yet existing satellite-based monitoring products, including broadband vegetation indices and categorical land cover maps, lack the sensitivity to detect early, process-driven shifts in vegetation function that precede visible compositional change. Plant functional traits retrieved from space-borne hyperspectral imagery offer a more physiologically based alternative, enabling spatially continuous characterisation of ecosystem function independently of species identity or land cover class. Here, we use EnMAP hyperspectral imagery and a convolutional neural network to retrieve 20 plant functional traits across Arctic tundra spotlight acquisitions spanning the North American Arctic and Greenland from 2022 to 2024. We characterise the structure of the resulting trait space, introduce pixel originality as a spatially continuous measure of functional uniqueness derived from multivariate trait distributions, and assess the capacity of trait-based predictors to differentiate vegetation types from the Circumpolar Arctic Land Use map using Random Forest models. Two principal components captured ≈ 86 % of total trait variability, organised along axes corresponding to the leaf economics spectrum and pigment composition. Pixel originality was almost entirely predictable from the selected traits (R² ≈ 0.97), with LAI, LMA, and carbon content as the dominant drivers, and showed smooth spatial gradients consistent with microtopographically structured tundra landscapes rather than discrete functional hotspots. Trait-based Random Forest classification of vegetation types achieved a mean accuracy of ≈ 73 %, demonstrating that functional traits alone carry substantial discriminatory power for broad vegetation class differentiation. These findings establish EnMAP-derived functional traits as a robust basis for landscape-scale monitoring of Arctic tundra diversity and function, with pixel originality offering a promising tool for tracking functional change across rapidly transforming ecosystems.

The Arctic is undergoing rapid environmental change, with vegetation responses that feed back into surface energy balance, permafrost dynamics, and biogeochemical cycling at global scale. Yet existing satellite-based monitoring products, including broadband vegetation indices and categorical land cover maps, lack the sensitivity to detect early, process-driven shifts in vegetation function that precede visible compositional change. Plant functional traits retrieved from space-borne hyperspectral imagery offer a more physiologically based alternative, enabling spatially continuous characterisation of ecosystem function independently of species identity or land cover class. Here, we use EnMAP hyperspectral imagery and a convolutional neural network to retrieve 20 plant functional traits across Arctic tundra spotlight acquisitions spanning the North American Arctic and Greenland from 2022 to 2024. We characterise the structure of the resulting trait space, introduce pixel originality as a spatially continuous measure of functional uniqueness derived from multivariate trait distributions, and assess the capacity of trait-based predictors to differentiate vegetation types from the Circumpolar Arctic Land Use map using Random Forest models. Two principal components captured ≈ 86 % of total trait variability, organised along axes corresponding to the leaf economics spectrum and pigment composition. Pixel originality was almost entirely predictable from the selected traits (R² ≈ 0.97), with LAI, LMA, and carbon content as the dominant drivers, and showed smooth spatial gradients consistent with microtopographically structured tundra landscapes rather than discrete functional hotspots. Trait-based Random Forest classification of vegetation types achieved a mean accuracy of ≈ 73 %, demonstrating that functional traits alone carry substantial discriminatory power for broad vegetation class differentiation. These findings establish EnMAP-derived functional traits as a robust basis for landscape-scale monitoring of Arctic tundra diversity and function, with pixel originality offering a promising tool for tracking functional change across rapidly transforming ecosystems.

The Arctic tundra is warming faster than any other biome, and its biodiversity, ecosystem functions, and Indigenous land use are increasingly shaped by interacting climate- and human-driven stressors. Conservation planning must therefore move beyond static protected-area targets toward approaches that explicitly account for future change. Here, we synthesize the major stressors that will shape tundra futures (woody plant expansion, permafrost thaw and associated disturbance dynamics, accelerating industrial development and infrastructure), and explain why their differing rates and interactions create a moving target for protection. We then identify key data and scenario gaps that currently limit circumpolar prioritization, including insufficient understanding of biodiversity responses to core stressors, uneven monitoring capacity to detect change and locate refugia, and limited integration of biodiversity change with ecosystem functions and culturally important areas. Building on this, we outline the concept of a step-by-step, scenario-based decision-support workflow for dynamic systematic conservation planning. Finally, we describe co-design and governance pathways for implementation and iteration with Indigenous peoples and local communities, stakeholders, and policymakers to reduce conflicts, increase legitimacy, and enable adaptive updates as conditions change. Together, these steps provide an actionable foundation for proactive tundra conservation under rapid Arctic change.

Large-scale mapping of plant biophysical and biochemical traits is essential for ecological and environmental applications. Given their finer spectral resolution and unprecedented data availability, hyperspectral data, in concert with machine and particularly deep learning models, have emerged as a promising, non-destructive tool for accurately retrieving these traits. However, when deploying these methods on a large scale, reliably quantifying the associated uncertainty remains a critical challenge, especially when models encounter out-of-domain (OOD) data, i.e., samples that differ substantially from those of the training data, such as unseen geographical regions, species, biomes, data acquisition modalities, or scene components (e.g., clouds and water bodies). Traditional uncertainty quantification methods for deep learning models, including deep ensembles (deterministic and probabilistic) and Monte Carlo dropout, rely on the variance of predictions but often fail to capture uncertainty in OOD scenarios, leading to overly optimistic and possibly misleading uncertainty estimates. To address this limitation, we propose a distance-based uncertainty estimation method (Dis_UN) that quantifies prediction uncertainty by measuring the dissimilarity in the predictor space (spectral inputs) and embedding space (features learned by the deep model) between the training and test data. Dis_UN leverages residuals as a proxy for uncertainty and employs dissimilarity indices in data manifolds to estimate worst-case errors via 95-quantile regression. We evaluate Dis_UN using a pretrained deep learning model to predict multiple plant traits from hyperspectral images, analyzing its performance across OOD data, such as pixels containing spectral variations from urban surfaces, bare ground, water, clouds, or open surface waters. In this study, we target six leaf and canopy traits: leaf mass per area, chlorophylls, carotenoids, nitrogen content, equivalent water thickness, and leaf area index. Compared to scaled variance-based methods, Dis_UN provides (1) a superior estimation of uncertainty in OOD scenarios, achieving 36 % higher contrast (KS distances: 0.648 vs. 0.475) between non-vegetation pixels, particularly under mixed-pixel conditions at medium resolution (30 m); (2) uncertainty quantification without requiring normality or symmetry assumptions, accommodating asymmetric error patterns; (3) enhanced interpretability of uncertainty sources, as uncertainty is directly linked to sample dissimilarity from the training data; and (4) computational efficiency at inference (2.6–7.7× faster), requiring only a single forward pass compared to multiple passes for ensemble-based methods. Challenges remain for traits that are affected by spectral saturation. These findings highlight the advantages of distance-aware uncertainty quantification methods and underscore the necessity of diverse training datasets to minimize sampling biases and enhance model robustness. The proposed framework improves the reliability of uncertainty estimation in vegetation monitoring and offers a promising approach for broader applications.

In response to solar forcing, hydrometeorological variability, and ecosystem properties, a global seasonal oscillation of vegetation greenness can be observed from space. This “green wave” plays a central role in regulating carbon uptake, energy exchange, and biosphere–atmosphere interactions. Yet, the dynamics of this large-scale macrophenological signal have so far been described mainly through summary statistics of local or pixel-based indicators. A concise global descriptor of the green wave is still lacking. Here, we introduce a metric that characterizes global vegetation seasonality by computing the spatio-temporal center of mass of terrestrial greenness from satellite observations. The resulting global three-dimensional trajectory provides an intuitive and integrative representation of seasonal and interannual macrophenological variability. Based on earlier evidence of widespread greening in the northern hemisphere, we expected a strong poleward displacement during boreal summer and a weaker compensating shift during austral summer. Instead, we find a consistent northward displacement during both seasonal phases. Across independent datasets, the austral summer shift exceeds the boreal one. This asymmetry leads to a contraction of the annual latitudinal amplitude of the green wave, a tendency that further strengthens in future projections using CMIP6 models. The proposed trajectory framework offers a new way to quantify and communicate large-scale biosphere change in physically interpretable units, and provides a basis for linking vegetation dynamics to climate forcing and human land use.

Arctic landscapes are very sensitive to warming with changes happening much faster than in other regions. The investigation on circumarctic Arctic vegetation change is carried out in the framework of the Federal Ministry of Research, Technology and Space (BMFTR) funded project SQUEEZE (Protection of the Disappearing Arctic Tundra: Potential, Planning, and Communication) in a large consortium. This study presented here focusses on the region close to Inuvik in the Mackenzie delta area in northwest Canada, which holds many different habitat types important to Indigenous peoples. The habitat diversity is important for ecosystem health and should be monitored as well as protected. In the region north of Inuvik, habitats range from tundra with low shrub structure, over forest tundra with sparse spruce forests, to taiga with dense needleleaf forests south of Inuvik and wetlands, lakes and river floodplains distributed over the area. These environments can be used for hunting, fishing, foraging of food, medicinal plants, firewood and construction material or as grazing grounds for caribou. However, those regions are facing changes due to climate change. Most dominant processes are increased permafrost thaw, shrubification of the tundra, northward shift of the treeline, more fires and pests in forests and changed waterways.Remote sensing offers valuable insights into the current state of this region and can help to track changes. Airborne remote sensing provides high resolution and allows to cover large areas. The airborne data used in this work was acquired with the AWI Perma-X flight campaign in the summers 2023 and 2025. We use the Modular Airborne Camera System-Polar (MACS-Polar) optical data. The MACS-Polar camera was developed by the German Aerospace Centre (DLR, Adlershof) specifically for challenging, contrasting light conditions in the polar region. MACS images were processed to four-band (visible and near-infrared, VNIR) orthomosaics and digital surface models with spatial resolution of 15 cm and 3D point clouds with point densities of up to 25 points per m2. Features of the VNIR images as well as structural features of the surface will be used to classify the habitat types. The analysis of the data for the years 2023 and 2025 in this work allows for tracking of changes between the years. The outcomes are classified maps of habitats, such as wetland, tundra, forest tundra and different forest types, in the area around Inuvik. Those will be made publicly available to the Indigenous communities in northwest Canada. MACS optical orthomosaics can be challenging because of changing illumination during flight times and the data derivation from Structure from Motion can hold inaccuracies. However, the resulting maps of the current state of vegetation structure are valuable products. Future work can build upon those by looking at longer timescales and upscaling with Sentinel-2 satellite data.

Numerous studies have shown the immense potential of Sentinel-2 satellite data for tree species classification across temperate forest ecosystems. However, key challenges such as the classification of minor tree species and detection of admixture species remain. In particular, when mapping tree species classes at a national scale, heterogeneous landscapes, different forest management practices and differences in phenology introduce additional uncertainties. Nevertheless, detailed information on tree species composition, including minor species, is crucial for nature conservation, biodiversity monitoring, and the development of forest management strategies opting for enhanced resilience. At the same time, spatially explicit information on model uncertainties and domains are often lacking in state-of-art national tree species maps. Here, we aim to provide new insight in mapping minor tree species relevant for nature conservation and forest resilience at the landscape scale in Germany by combining diverse reference data sources such as forest management maps, biotope mapping data, and regional forest inventories with Sentinel-2 timeseries and periodic forest masks. The obtained tree species maps are complemented by spatially explicit uncertainty maps and areas of model applicability, allowing for a transparent and species-specific assessment of the classification results. Subsequently, these outputs will be used to derive tree species composition and tree species diversity measures across different spatial scales. We compare our findings to existing tree species maps and aim to quantify tree species diversity measures in relation to in-situ inventory plots across different landscape regions in Germany. Additionally, the generation of periodic tree species maps enables the analysis of temporal changes and shifts in tree species composition and biodiversity, especially in view of the drastic changes and stresses that forests in Germany have been exposed to in recent years as a result of climate change. This study contributes to a more comprehensive understanding of the potential and the limitations of remotely sensed tree species maps for monitoring biodiversity and supporting conservation planning across spatial and temporal scales.

Viewed from space, a “green wave” seasonally traverses Earth’s surface, from the north in boreal summer to the south in austral summer. This wave represents vegetation phenology, driven primarily by solar irradiation and modulated by climate variability and ecosystem dynamics. Despite its significance for multiple Earth system processes, we lack a unified metric to characterize and understand its dynamics. Here, we propose a concept to quantify global phenology by tracking the green wave’s centroid using satellite and Earth system model data. The resulting trajectory summarizes global phenological dynamics and directional trends. Earlier reports on global greening led us to hypothesize a rapidly northward shifting trajectory during boreal summer and a moderate southward shift during austral summer. Contrary to this expectation, we find that the centroid moves northward during both summer periods, with the austral summer shift consistently exceeding the boreal shift across datasets. As a consequence, the amplitude of the green wave trajectory is decreasing, a trend projected to intensify throughout this century. We also detect an accelerating eastward shift, a phenomenon not previously reported. Tracking the green wave’s centroid reveals how regionally changing land dynamics affect the global functioning of Earth’s terrestrial biosphere.

Drivers of non-native plant success in drylands are poorly understood. Here we identify functional differences between dryland native and non-native perennial plants and assess how biotic, abiotic and anthropogenic factors shape the success of the latter. On the basis of plant community and functional trait data from 98 sites across 25 countries, we report a total of 41 non-native plant species at 31 sites. Non-natives tend towards faster growth strategies than natives. Non-native plant richness is higher at sites with greater grazing pressure and under environmental conditions associated with higher soil fertility, decomposition and fungal richness—conditions that tend to occur in less arid regions—and lower where native plant and herbivore richness are greater. Non-native plant cover correlates positively with grazing pressure and negatively with native plant richness. Taken together, our results suggest that non-native plant success in drylands is facilitated when high grazing pressure coincides with elevated resource availability. Such context-dependence of non-native plant success and linkages with native plant and herbivore diversity highlight the need for managing grazing and conserving biodiversity across the world’s drylands.