Conference Agenda

Biodiversity is under pressure by anthropogenic and climate change, but it is difficult to measure, monitor and predict changes across the globe. We face large knowledge gaps in terms of the spatial distribution and temporal dynamics of biodiversity and related ecosystem functions. A new suite of current and upcoming remote sensing instruments is providing large-scale measurements of plant canopy structure, plant functional traits and diversity, and ecosystem functioning from space. For example, spaceborne lidar, such as from the GEDI instrument, can provide us with a new view on the three-dimensional plant canopy structure and its diversity at the landscape scale. I will present new results and challenges for mapping forest structural diversity in California and Central Africa using GEDI at scales from 1 to 25 km, which provides insights on a range of complex and diverse Mediterranean and tropical forest ecosystems. We found that GEDI’s RH98, Cover and FHD metrics were most effective to capture variation in forest canopy height, density and layering, and that GEDI captured the variation of canopy structure generally well in closed forests in flat terrain, while challenges emerged in open forests and in complex terrain. We found high structural diversity in mid-elevation and coastal forests in the US and in volcanic ranges and forest-savanna transitions in Africa. GEDI revealed spatial patterns of structural diversity that aligned with known ecological processes, including the influence of wildfire in the western US and topographic variation in central Africa. Besides ecosystem structure, we developed new methods using imaging spectroscopy to map the distribution of leaf biochemical and biophysical traits and derived patterns of plant functional diversity at the landscape scale. We developed and tested the methods using large-scale airborne imaging spectroscopy data acquired using AVIRIS Classic across a diverse elevation gradient in the California Sierra Nevada mountains to test the application to spaceborne instruments such as EnMAP, PRISMA or the future NASA SBG and ESA CHIME missions at 30 m spatial resolution. I will present results that give insights into mapping foliar traits at large spatial scale and the role of trait-trait relationships in mapping plant functional diversity. We found that there are at least three relevant functional axes of variation that should be represented in functional diversity analyses, and that the relationship among those axes and functional plant strategies is context dependent. We also found that patterns of functional diversity were related to elevation gradients and disturbance patterns, especially related to wildfire. Combining these new measurements with ground-based data will help to better understand biodiversity patterns and change over time. I will present examples of new analyses of remotely sensed patterns of plant functional and structural diversity, and their relationship to other dimensions of biodiversity and ecosystem functions, that demonstrate the value and potential of new remote sensing instruments and methods for biodiversity monitoring from space.

Plant-pollinator interactions are critical to terrestrial ecosystem functioning and global food production but are experiencing increasing pressures from land use and global environmental changes. Environmental conditions, such as climate and vegetation cover, influence both foraging resources and nesting habitat for pollinators. Yet, little is known about the role of vegetation structure and functional traits in determining the organisation of plant-pollinator networks, nor on methods to predict such networks at broad spatial scales. Here, we take a novel approach and evaluate how plant functional traits and vegetation structure influence plant-pollinator interaction patterns. Plant-pollinator network data analysed comprised a total of 209 networks from across the tropics, with vegetation structure and functional trait information extracted using spectral and LiDAR remote sensing datasets. We found that pollination network metrics responded to plant functional traits along a spectrum of plant resource use acquisition and conservation strategies, where networks were more modular with lower vegetation height and leaf nutrient content, while higher leaf photosynthetic capacity and nutrient contents were associated with higher levels of network connectance and complementary specialization. Additionally, networks were more nested with increasing trait variability. Our findings reveal that plant functional strategies, captured by remote sensing, play an important role in structuring biotic interactions such as those between plants and pollinators, paving the way to predict these interactions at scale.

Imaging spectroscopy missions like Earth Surface Mineral Dust Source Investigation and Surface Biology and Geology (SBG) provide valuable opportunities for assessing plant traits. Current empirical approaches, such as Partial Least Squares Regression (PLSR) and various machine learning methods, often lack interpretability and rigorous uncertainty quantification, and typically cannot transfer models across different sensors. To address these limitations, we propose a Bayesian framework to estimate multiple plant traits directly from spectra without requiring transformations like those used in PLSR.

Our Bayesian framework includes four models: a linear model (comparable to PLSR), a non-linear model (utilizing kernel transformation), a hierarchical model (accounting for trait variation across broadleaf and needleleaf trees), and a phenological model (allowing regression parameters to vary temporally). Additionally, we introduce a projection technique that reduces fitted trait models to submodels with fewer spectral bands while maintaining predictive accuracy. This technique identifies the optimal bands necessary for accurate trait estimation and enables flexible model adaptation across different spectral configurations.

We apply these models to predict leaf-level traits using a global dataset and extend this to the airborne scale using AVIRIS-NG data and trait measurements from the 2022 SBG High-Frequency Timeseries campaign. At both scales, the linear Bayesian model performs comparably or slightly better than PLSR, while the other Bayesian models show varying degrees of improvement depending on the specific trait. The reduced models identify between 6 and 30 essential bands.

To test the framework’s adaptability across sensor configurations, we resample AVIRIS-NG spectra to different resolutions and add synthetic errors. Our projection algorithm successfully adapts the AVIRIS-NG model to these simulated sensors without requiring spectral resampling. This approach demonstrates a robust, interpretable, and sensor-agnostic method for plant trait estimation, enabling consistent and reliable large-scale trait mapping across multiple missions.

Understanding the dynamics of forests is crucial for ecology and climate change research.
Reliably estimating the extent and properties of existing forests on a global scale remains essential for planetary carbon balance investigations.
However, high local variability in forest structure, biomass, and productivity poses significant challenges.
These elements are influenced by successional states and disturbances, both natural and anthropogenic.
Remote sensing methods from orbital measurements can capture local features on a global scale.
Among large-coverage satellite missions, the TanDEM-X Radar offers high spatial resolution and temporal consistency.
Its bistatic backscattering methodology allows for interferometric analysis and detailed investigation of vertical forest structure.
We present different interferometric concepts to invert indicative forest properties such as vegetation canopy height and forest structure.
These techniques excel in providing extensive and consistent spatial and temporal coverage, enabling the monitoring of forests worldwide.
On the other hand, the individual-based forest model FORMIND offers detailed simulations of forest dynamics by accounting for individual tree growth, competition, and mortality.
This model captures the complexities of forest structure and successional stages with high ecological detail.
We integrated TanDEM-X X-band radar coherence measurements with coherence properties based on structure-specific simulations on top of the forest model FORMIND.
This synergy combines the broad reach and efficiency of remote sensing with the detailed, process-based understanding from ecological modeling.
We were able to improve local estimation precision while retaining large spatial coverage over the study sites of Barro Colorado Island (Panama) and Paracou (French Guyana).
Our approach enhances the accuracy of global carbon balance investigations and contributes to a better understanding of forest dynamics in the context of climate change.
By leveraging the strengths of both remote sensing and model-based methodologies, we demonstrate that their combined application provides a more comprehensive and precise assessment of forest ecosystems.
This integrated strategy is essential for effective climate change mitigation and the sustainable management of forest resources.

Estimating forest biodiversity is essential for effective conservation and ecosystem management. Traditional field surveys, while valuable, are often time-consuming and labor-intensive, challenging the collection of comprehensive and accurate biodiversity data. Over recent decades, various methods have emerged to assess forest structure and tree species diversity using remote sensing technologies. One notable indirect approach is the "Height Variation Hypothesis" (HVH). This hypothesis states that greater heterogeneity in tree height, as measured by LiDAR data, indicates higher complexity in forest structure and greater tree species diversity.

The HVH is based on the relationship between variations in canopy height and tree species diversity, using the forest's vertical structure as a biodiversity indicator. This hypothesis has garnered significant attention in recent literature, with numerous studies exploring its applications. Researchers have tested the HVH using airborne laser scanning LiDAR data and, more recently, GEDI LiDAR data, demonstrating how space-borne LiDAR can identify biodiversity patterns through variations in tree canopy height.

The approach has also been applied to forests affected by extreme wind events, which cleared entire areas, to investigate the role of tree height heterogeneity in forest stability and biodiversity. Beyond forest ecosystems, the HVH has been extended to agricultural landscapes, integrating LiDAR and photogrammetric data with ecological modelling to assess vertical heterogeneity at the landscape level. This integration has provided valuable insights into conserving avian and bee diversity in human-dominated landscapes.

In summary, the HVH presents a promising method for estimating biodiversity in different natural ecosystems, using LiDAR data. By synthesizing findings from recent studies, we highlight the potential of LiDAR technology to enhance our understanding of biodiversity patterns and support effective conservation and management strategies.

Soil accounts for up to a third of the total Amazonian forest carbon stocks ; however, uncertainties in soil organic carbon (SOC) stocks are very large compared to above-ground stocks. It is important that we learn more about SOC stocks and their management to learn about the functioning of the land carbon sink under continued climate and land use change. This study investigates the relationship between canopy structure and SOC in tropical forests, with the goal of improving SOC predictions across the landscape using satellite remote sensing.

We took soil samples in 142 locations up to a depth of 30 cm, with corresponding measurements of canopy structure using field hemispherical photography, airborne lidar and spaceborne lidar (within footprints of the Global Ecosystem Dynamics Investigation). These were analysed using open source software to ensure the methods are readily accessible. SOC in our study sites ranged from 0.34% to 9.04% and Plant Area Index between 2.28 and 9.59. We use statistical inference from Generalised Linear Models (GLMs) to develop understanding of mechanistic relationships between soil carbon concentrations and indicators of forest canopy structure (e.g. Plant Area Index, rumple index, vertical complexity index). These results inform modelling strategies for predicting soil carbon on landscape scales using spaceborne sensors such as GEDI and Landsat.

Our research offers a novel approach to refining landscape scale predictions of SOC in tropical ecosystems, providing further insights into the variation in carbon storage. This ultimately contributes to global efforts to understand terrestrial carbon dynamics and the land carbon sink under climate change conditions . Furthermore, our work demonstrates the value of openly available global data products, and methods that use this appropriately.