A link between (remotely sensed) spectral variation and plant biodiversity (mostly expressed as species counts) has been suggested in the form of the spectral variability hypothesis (SVH). According to the hypothesis, higher spectral variation in canopy reflectance is related to (1) variation in habitats or linked vegetation types or plant communities which each having their own specific optical community traits or (2) variation in the species themselves according to their specific optical traits. Because optical community-traits are a summarized signal of the optical species traits of plant individuals, it is mostly a matter of scale which of both drives variation across pixels. The spectral variability hypothesis was examined in several empirical remote sensing case studies over the last years. These studies often report some correlation between biodiversity (species counts) and spectral variation; however, the strength of the observed correlations varies strongly between studies. In contrast, studies attempting to understand the causal relationship between (plant) species counts and spectral variation are still scarce. In this study, we address this research gap by critically discussing the link between spectral variation and biodiversity metrics. We support our perspectives through simulations and experimental data. Our results show that in many situations the spectral variation caused by species or functional traits is subtle when compared to other factors including for example physiological status and seasonality. Moreover, the methodical approach to calculate spectral variation was found to have a notable effect on the discussed link. We find that the degree of contrast in reflectance (used in many earlier studies examining the spectral variation hypothesis) has little to do with the number but rather with the identity of the species or communities involved. Hence, we recommend that spectral variability should not be quantified based on contrast rather than on manifoldness. The latter can for example by obtained using unsupervised clustering. While we describe cases where a certain link between spectral variation and plant species diversity can be expected, we believe that as a scientific hypothesis (which suggests a general validity of this assumed relationship) the SVH is flawed and requires refinement. Hence, we call for more research identifying the drivers of spectral variation in vegetation canopies and their link to plant species diversity and biodiversity in general. Such research will enable understanding under which conditions spectral variation is a meaningful indicator for biodiversity and how it could be implemented as part of a corresponding monitoring network.

Climate change and human activities have triggered an accelerated loss of biodiversity on a global scale. The Convention on Biological Diversity (CBD) commits nations to stem this loss by the end of this decade. In this context, Remote Sensing (RS) emerges as a potential tool for providing continuous, multi-scale, and systematic information that helps to adequately implement the post-2020 Global Biodiversity Framework, and other conservation actions.
The capability of remote sensing to inform about biodiversity on terrestrial surfaces relies on the Spectral Variation Hypothesis (SVH), which states that the variability of the spectroradiometric signals captured by remote imagers (spectral diversity) correlates with the variation in species on the ground. In vegetated environments, this variability could be influenced by taxonomic diversity (e.g., species richness) or the variability of vegetation functional properties (functional diversity). Researchers have explored for which metrics, methods, and sensors the SVH works and thus can be used to estimate plant biodiversity reliably from spectral diversity. However, these studies are limited to a few metrics, methods, and sensor features, usually evaluated in a single or a few vegetation communities. Therefore, most of their results are not easily generalizable and lack the information necessary to understand the differences in the performance found within and between studies.
To overcome these issues, we developed a modeling framework that allowed us to evaluate the capability of different functional diversity metrics (FDMs) to link spectral and vegetation functional diversities and assess the role of various sources of uncertainty inherent to remote sensing. This work focuses on plant “functional diversity,” which represents the variability of plant functional traits. Using radiative transfer models, we simulated 81.000 vegetation communities separated into 1.000 regions or images. Then we evaluated the information content of several FDMs derived from optical reflectance factors (R) or estimates of vegetation parameters (via model inversion). Some of these parameters are Essential Biodiversity Variables (EBVs) designed to monitor biodiversity change. Then we evaluated the effect of the spatial resolution, the spectral characteristics of the sensor, and signal (random) noise on the relationships between vegetation and remote sensing-based FDMs. Simulations allowed us to control and know all the factors influencing these relationships; however, they cannot represent all the complexity of the real world. For this reason, we tested the plausibility of the simulations using observational field data acquired in the forest biodiversity plots of the FunDivEUROPE network and Sentinel-2 (multispectral, 10 m pixel size) and DESIS (hyperspectral, 30 m pixel size) imagery.
The simulations demonstrated that variability of both reflectance factors and vegetation parameter estimates were correlated with plant functional diversity under ideal conditions (noiseless, high spatial, and spectral resolutions). Nonetheless, not all the functional diversity metrics were suited for studying plant biodiversity from space, and not all the aspects of functional diversity can be remotely captured. Only Rao’s quadratic entropy (Q), functional dispersion, and richness provided consistent relationships between remote sensing and plant trait-based diversities within and between communities (r² ∈ [0.61, 0.92]). Functional evenness and divergence were only weakly correlated with plant functional diversity (r² ∈ [0.03, 0.24]). We also found that the characteristics of the remote sensing data affected relationships of metrics based on reflectance factors or vegetation parameter estimates differently; and that the degradation of the spatial resolution had substantial effects on both methods (r² < 0.40). However, the degradation of spectral information (from hyperspectral to multispectral) mainly affected the parameter retrieval approach. The dimensionality reduction applied for the computation of functional diversity metrics minimized the effects of the noise in plant and spectral variables. Still, the noise increased the uncertainty for retrieving vegetation parameters when the spatial resolution was high. The analysis of remote sensing imagery was in agreement with the simulations. Sentinel-2 imagery provided better estimates of plant diversity from the ground plots than DESIS, since its higher spatial resolution could capture the internal variability of these plots. DESIS spatial resolution equals ground plot size, and therefore, it could only characterize the spectral diversity of the plot surroundings. Still, DESIS provided significant relationships with taxonomic diversity metrics. However, when Sentinel-2 spatial resolution was resampled to DESIS’s one, no correlations were found with field data. This result suggests that DESIS’s higher spectral information (despite not covering the entire optical domain) still constitutes an advantage when spatial resolution and signal noise are suboptimal.
The SVH has been tested systematically for the first time by consistently evaluating several metrics based on different types of information over a wide range of synthetic vegetation communities and under 27 different configurations and noise levels of the remote sensor. These simulations are coherent with the results found with remote sensing imagery and field observations. Our work clarifies the strengths and limitations of optical RS to monitor plant functional diversity exploiting the SVH. Furthermore, it draws valuable guidelines for the design of remote sensing products of plant functional diversity and missions dedicated to monitoring biodiversity change.

The increasing need for continuous global information on the biodiversity of Earth's forests calls for new approaches using satellite data. Advancing our scientific understanding of biodiversity states and changes on regional to global scales will contribute to assessing the impact of environmental change on ecosystems and facilitate the prediction of future ecosystem functioning. As an important dimension of biodiversity, functional diversity of plant traits links ecosystem functioning and biodiversity. Approaches to mapping functional diversity from space using satellite data build the base for large-scale analyses, opportunities to study vegetation throughout the phenological cycle, include time-series of multiple years, and access remote areas on large extents.

We investigate how functional diversity maps, derived from satellite imagery, link to in-situ measurements of biodiversity and ecosystem functioning, such as resilience and stability. We present an approach to upscale, map, and quantify functional diversity from physiological forest traits derived from Sentinel-2 data in a temperate forest ecosystem. We used two complementary data sources, APEX airborne imaging spectroscopy data, and Sentinel-2 satellite data. The high-resolution 2 m APEX data were spatially resampled to ensure our methods support spatially consistent mapping of functional diversity from space.

We derive functional diversity metrics from physiological forest traits, chlorophyll content, carotenoid/chlorophyll ratio, and water content in the forests of Canton Aargau, Switzerland, encompassing around 500 km2 of forests, using Sentinel-2 multispectral data. Based on the physiological forest traits related to forest health, stress, and potential productivity, we derive a cantonal map of forest functional diversity working towards comparing forests across landscapes and providing the base for analysis of changes over time. Comparing biodiversity monitoring data from in-situ measurements to functional diversity maps from Sentinel-2 data will provide answers to the question of how in-situ biodiversity measurements link to functional diversity captured from space. We use data from the Swiss biodiversity monitoring program, a long-term governmental program initiated in 2001 that monitors species richness of multiple taxa and at different spatial scales. These data allow studying relationships between plant functional and taxonomic diversity and how plant functional diversity links to the biodiversity of other taxa.

Diversity is coupled to ecosystem functioning. Linking Sentinel-2 derived functional diversity to forest resilience and stability provides information on the relationship between functional diversity and ecosystem functions. Understanding the relationships of functional diversity and ecosystem functions at the landscape scale is critical to monitoring and understanding forest ecosystems. We used drought resilience data derived from relative temporal changes in canopy water content to gain new information on the links of functional diversity from satellite data and ecosystem stability.

Our results will provide important information about the potential to monitor biodiversity and ecosystem function from space and open up new possibilities for large-scale interpretation. This paves the way towards a continental or global assessment of the functional diversity of forests using satellite data.

Given the accelerating global change, the protection of plant biodiversity is of major importance, which inevitably requires understanding its global distribution its relation to the earth system as a whole. Such plant biodiversity-environment relationships can be understood in terms of functional rather than taxonomic diversity. Still, knowledge on the global distribution of functional diversity and how it is shaped by individual functional traits remains sparse. In a joint effort, the scientific community is constantly gathering plant trait observations and making it available through the TRY database, which provides more than 11.8 million records of 2,091 traits across 280,000 plant taxa (TRY v. 5). Several studies attempted to spatially extrapolate these sparse trait observations using globally available predictors, including data on climate, soil properties, or landcover. However, so far there are only global maps for a few traits with hardly measurable and probably high uncertainties. Also, Earth observation satellite data is becoming a key technology for generating global products on plant functional traits. However, the bird-eye view from Earth observation primarily informs on upper canopies and, as a result, may only represent functional traits of the most competitive plants and not the actual plant community.

Here, we take a completely different perspective: an ever-growing plethora of geocoded plant photographs with species information crowdsourced by already more than a million citizens around the globe, i.e., the citizen science project iNaturalist. With the human eye, we immediately see that such plant photos, even if they are very heterogeneous, can provide information about morphological plant characteristics - and thus also about functional traits. To effectively harness this data treasure, we make use of deep learning and Convolutional Neural networks (CNN). Training CNN to accurately and robustly capture plant functional traits from plant photographs requires an excessive amount of training data for model training, consisting of pairs of plant photographs and plant traits. Such datasets are not readily available; however, while, iNaturalist data informs on plant species in the photographs, the above-mentioned TRY database contains enormous amounts of species-specific plant trait records. Therefore, we investigated how joining these two databases (iNaturalist, TRY) by the species affiliation enables a weakly supervised learning of CNN models for predicting functional plant traits from simple photographs. Our results show that morphological image features indeed suffice to predict several traits representing the main axes of plant functioning, including growth height, leaf area, leaf mass. Lower but still promising accuracies were obtained for traits that relate not directly to visible morphological features (nitrogen concentration and stem specific density). The accuracy was enhanced when using CNN ensembles (combinations of different CNN architectures), incorporating prior knowledge on trait plasticity and contextual information on climate (WorldClim). Our results suggest that these models generalised across growth forms, taxa, and biomes. We did not find any phylogenetic signal in the residuals (e.g., bias towards taxonomic groups), indicating that the models indeed learned to see plant traits in the photographs. The trait predictions were robust over heterogeneous crowdsourced photographs (image quality, and other image acquisition settings). Spatially aggregating such CNN-based trait predictions from 185,000 independent iNaturalist photographs via their geolocation enabled the generation global trait distribution maps, which reflect known macroecological patterns. A quantitative comparison with previous global trait distribution maps revealed significant correlations with Pearson’s r > .5 concerning growth height, seed mass, specific leaf area and stem specific density. Still, global trait distribution maps also commonly show large discrepancies, which we can also partly confirm with our products. Note, that most of the global trait distribution maps typically feature substantial uncertainty as trait expressions are commonly spatially extrapolated across large extents from sparse TRY records. In contrast, the extremely high and steadily increasing sampling density in the iNaturalist database enables the generation of global products without spatial extrapolation. All trait products derived this way are freely available, including the first-ever published map on the global distribution of leaf area.

The presented approach presents an alternative way to understand macroecological patterns. There exists a great potential of fusing crowd-sourced data with spaceborne remote sensing data or to use the respective products for comparison. Yet, biases in citizen science-based data still remains largely unknown and until now there does not exist a comprehensive reference dataset to evaluate global trait maps. Our results show that integrating crowd-sourced data may enable to further close the gap between actual and intended spatial coverage, which ecology has been falling short of for decades. Moreover, these findings demonstrate the potential of exploiting big data, with its volumes, sparsity, and variety, derived from professional and citizen science in concert with deep learning for assessing Earth’s plant functional diversity.

Grasslands are highly important ecosystems for biodiversity, carbon sequestration and soil protection. They cover significant parts of the European Union and are one of the largest biomes on Earth. Unfortunately, monitoring the changing status of grasslands, even in protected sites, is challenging due to limited conventional data collection resulting in inefficient management. The European Commission has developed a pilot operational monitoring service, known as EU Grassland Watch, for the grassland dominated Natura 2000 sites which exploits Earth observation data.
The key enabler for this service is the Copernicus programme and the Sentinel satellites which represent a step change in observing capability and capacity that can be directed to environmental monitoring. The European land area can now be monitored at a scale and frequency where habitat patch extents and changes can be identified and phenological and episodic events related to landscape processes can be followed. The Copernicus Land Monitoring Service (CLMS) also provides a range of support products including some directly related to Natura 2000 sites and grasslands.
The first step is to map the annual status and changes of land cover / land use (LC/LU) in and around (the site areas plus a 2 km buffer) each selected Natura 2000 site. Optical images from Copernicus Sentinel-2 plus the US Landsat 8 are used to both allow gap filling caused by clouds and snow and to extend the time series back to 1994. Copernicus Sentinel-1 synthetic aperture RADAR (SAR) data are also used to support the mapping and monitoring processes from 2014 onwards.
The EO data for the classification process are combined over short time periods as separate optical and SAR monthly (3-monthly and optical only prior to 2014) composites. The optical and SAR composites are classified separately using a Support Vector Machine (SVM) approach based on training data selected from the existing CLMS Natura 2000 and CORINE land cover products to give 17 classes over a four-level hierarchy. The two separate optical and SAR results are then merged by a simple rule based on the highest SVM probability to give a single LC/LU map for each year. The raster-based LC/LU maps are then aggregated to the level of the reference objects from the CLMS Natura 2000 product, with the output vector layer including the fractions of each class per object plus biophysical properties calculated from the optical data. In addition annual summaries of phenology, productivity, mowing and ploughing are provided along with indicators of the performance to the management within the sites (e.g. trends in total grassland area).
The EO-based products add up to a considerable volume of data which needs to be easily accessible to a broad range of users and stakeholders. A geospatial database has therefore been developed to store all the above products and indicators and provide easy access to the information via an interactive public webtool, the user-facing part of the EU Grassland Watch service. The webtool displays the results at a range of reporting levels to support stakeholders at local to European scales and will be updated at regular intervals once fully operational and visualises the results in ways most appropriate to the application, e.g., heat maps, trend indicators, temporal plots or thematic maps.
The EU Grassland Watch service will support EU Member States to take appropriate steps to proactively manage this important network of protected sites and prevent the deterioration of species and habitats. This service is a clear demonstration of how frequent spatially detailed EO data can be combined with exiting geospatial information to provide an effective tool for environmental management at local to continental scales. It is also a further example of a downstream service being built on the free and open data policy of Copernicus.

Grasslands cover about one third of the global land surface and are the most cultivated biome on Earth. They provide numerous ecosystem services, such as carbon sequestration and food production, and have a high importance for conservation as they are often species rich. Biodiversity, ecosystem services and functions in grasslands are strongly affected by the management regime, i.e. grazing, fertilisation, timing and frequency of mowing events, and its quantitative variation. The optimisation of farming efficiency towards yield maximisation usually leads to an increase in these treatments and thus land-use intensity (LUI), which may have numerous negative implications for the environment, such as biodiversity loss, water pollution, land degradation and increased carbon emissions. Information on grassland LUI is thus crucial for understanding trends and dynamics related to biodiversity, ecosystem functioning, earth system science and environmental monitoring.

However, large extent, high resolution information on grassland LUI is rare. New satellite generations, such as Copernicus Sentinel-2, enable a spatially comprehensive detection of the mainly subtle changes induced by differences in land-use intensity based on their comparably fine spatial and temporal resolution. We developed a methodology calculating a continuous LUI index by quantifying key parameters of grassland LUI such as grazing intensity, mowing frequency and fertiliser application across Germany using Convolutional Neural Networks (CNN) on Sentinel-2 satellite data with 20 m x 20 m spatial resolution. An unprecedented LUI validation and its components using large scale in situ grassland management data of the DFG (Deutsche Forschungsgemeinschaft - German Research Foundation) Biodiversity Exploratories program was performed. This dataset included spatially explicit information about livestock, fertilisation and mowing events in three regions across Germany from 2006 to 2018 which differ in soil as well as plant species composition and also land-use intensity.

In contrast to many past studies we employed 2 methodological aspects which allow an advanced result interpretation. First, a feature contribution analysis using Shapley values, summarising explanations of individual predictions to gain information about the global model structure and to robustly extract variable importances across different underlying models. Thus, the contributions of the predictor variables (spectral S2 bands) over time (one year) could robustly be quantified. Second, we assessed the model's spatial transferability by delineating the area of applicability (AOA) based on the feature space given by the training data and the model's variable importances. This provides valuable information especially for large scale predictions. Further, We evaluated the methodology's robustness with a spatial 3-fold cross-validation by training and predicting on geographically distinctly separated regions.

Shapley values substantiated our method's plausibility revealing a high relevance of springtime satellite observations and spectral bands related to vegetation health and structure. This was in line with the timing of management activities and overall vegetation dynamics as well as published management effects on grassland condition. Regarding LUI components, we achieved an overall classification accuracy (R2) of up to 66% for grazing intensity, 68% for mowing, and 85% for fertilisation. The subsequent overall LUI derivation was achieved with an accuracy of R2=0.82. The loss in predictive power when training and predicting on geographically distinctly separated regions varied between 3-20 %. AOA analysis showed differences between investigated vegetation periods potentially related to the increasing satellite data availability. The pixel number outside AOA differs substantially between regions revealing spatial patterns most likely reflecting fields or parcels. This points towards regional differences in land management practices, abiotic conditions or vegetation compositions not reflected by our training data.

The presented methodology enables a high resolution, large extent mapping of land-use intensity of grasslands.