Conference Agenda

Terrestrial ecosystems are increasingly confronted with environmental changes such as climate change, natural disasters, or anthropogenic disturbances. Prolonged droughts, heat waves and increasing aridity are generally considered major consequences of ongoing global climate change and are expected to produce widespread changes in key ecosystem attributes, functions, and dynamics. Europe has been heavily affected by consecutive and increasingly severe droughts in the past decades, leading to large-scale vegetation die-offs and land degradation. This enhanced frequency in the past, combined with potential impacts of future climate change, makes it important to understand: How do droughts affect ecosystem stability and induce changes in ecosystem functioning? And what drives these changes?

As carbon gain in terrestrial ecosystems is a compromise between photosynthesis and transpiration, a ratio that is also known as water-use-efficiency (WUE), assessing changes in WUE plays a key role in assessing changes in terrestrial ecosystem functioning. Here, we use a remote sensing-based vegetation productivity index (MODIS EVI) together with transpiration data based on GLEAM to calculate changes in WUE across Europe between 2000 and 2023. We further investigate the response of WUE to individual drought events and model the impact of potential driving variables (e.g., drought severity, land management, soil texture, fire, etc.) using a machine learning (ML) approach.

Across Europe, we found regional differences in WUE over time with mainly positive trends in Northern Europe, aligning with less frequent and mild droughts, and negative trends in large parts of Central and Southern Europe aligning with more frequent and intense droughts. We found almost exclusively negative WUE anomalies under drought events, independent of the ecoregion, indicating increased transpiration or a loss in vegetation productivity, potentially due to die-offs and fire. Our ML model additionally highlight the impact of drought severity as well as ecosystem condition prior to a drought event on WUE and thus the ecosystems’ ability to respond to drought. We finally explored the link between ecosystem response to drought and ecosystem resilience in Southern European biodiversity hotspots.

Here I present recent advancements from our research in the field of spatial biodiversity modeling. The basic underlying concept is the utilization of spatially continuous data on the environment, originating from remote sensing and other data sources, for the purpose of making predictions of biodiversity or conservation value across the landscape. We utilize deep learning models as well as classic mechanistic statistical models to correlate a selection of biodiversity callibration points, e.g. produced via metabarcoding of environmental DNA (eDNA), with the environmental predictors that are available from public data sources. We demonstrate how models optimized/trained in this manner can help to fill our (spatial) gaps in our understanding of the spatial distribution of biodiversity, ranging from the identification of high-conservation value forests, to predictions of species diversity and other biodiversity metrics. These models can be applied to produce continuous rasters of biodiversity metrics (heatmaps) that can help decision makers and researchers to identify areas that are of particular biodiversity value. We demonstrate such data-products on national level on the example of Sweden. The talk will also cover the aspect of including the temporal component in such models, allowing us to predict the expected fluctuation of insect species richness throughout the year in a spatially explicit framework.

Despite being in the middle of a global biodiversity crisis, we still have comparably little knowledge of the spatial distribution of biodiversity for most organism groups. Such knowledge is crucial in making informed conservation priority decisions. Here we present a project where we develop deep learning biodiversity modelling tools that can predict the expected species diversity of any organism group, given a set of publicly available geospatial data-products. We train the model on biodiversity data of arthropods derived from a Sweden-wide metabarcoded bulk DNA inventory. The unique DNA barcode sequences were retrieved from over 4000 bulk DNA samples collected from 200 sites throughout one year. By combining this data with spatial information such as temperature, precipitation, elevation, NDVI, human impact indices etc., we can train a convolutional neural network (CNN) to predict the expected number of arthropods at any given location and month. One of the major advantages with CNNs is the direct interpretation of contextual data, in this case unedited tiff-files from 25 remotely sensed features. We compare the CNN suitability for biodiversity modelling tasks with other machine learning models. Even though CNN did not perform the best on this limited dataset, it holds promises for biodiversity monitoring at both spatial and temporal scales as the accessibility to larger biodiversity and remote sensing datasets increases.

Vegetation phenology, the study of recurring plant life-cycle events, is essential for understanding ecosystem responses to environmental changes, especially in the context of climate change. Remote sensing, particularly through vegetation indices like the Normalized Difference Vegetation Index (NDVI), has become a powerful tool for monitoring phenological events on large spatial and temporal scales. NDVI time series data can be used to derive key phenological metrics—including the start, peak, and end of growing seasons—providing valuable insights into vegetation health and productivity. However, current methods for extracting phenological metrics from NDVI data often fail to capture their biological and physiological significance. Additionally, while NDVI effectively tracks the vegetation growing season, it has limitations in detecting dormancy phases. This study presents SWELL (Simulated Waves of Energy, Light, and Life), a novel process-based phenology model designed to simulate the complete annual NDVI profile, from leaf unfolding to dormancy release, using photothermal response functions. SWELL aims to bridge the gap between remotely sensed phenological phases and underlying ecophysiological processes, providing a more comprehensive understanding of vegetation dynamics. When tested on European beech MODIS NDVI data, SWELL successfully reproduced seasonal profiles across years and ecoregions, showing similar performance in both calibration and validation and comparable accuracy to a benchmark statistical method fitted to annual NDVI series. Additionally, it demonstrated biogeographic consistency with beech responses to varying photothermal conditions. SWELL addresses current observational and conceptual limitations in phenology modeling, offering a novel tool for understanding and predicting vegetation phenology in the context of climate change.

Accurate, high-resolution data on global vegetation height distribution is essential for monitoring Earth's carbon stock, fluxes, and forest ecosystem dynamics. Additionally, the vertical structure of vegetation has been shown to predict biodiversity across various taxa. Given the critical importance of these tasks in the context of climate change and the biodiversity crisis, there is an urgent need for a reliable, high-resolution, and easily updatable global canopy height model (CHM).

Since 2018, two spaceborne laser altimeters, the Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2) and the Global Ecosystem Dynamics Investigation (GEDI), have been operational, collecting terrain and surface elevation data with near-global coverage. While ICESat-2 provides general elevation data, GEDI is specifically designed for vegetation mapping. Two global CHMs with resolutions of 30 m (Potapov et al. 2021) and 10 m (Lang et al. 2022) have been developed, utilizing machine learning models to fill gaps in sparse GEDI measurements based on optical satellite imagery. More recently, Tolan et al. (2024) integrated GEDI data with airborne LiDAR to produce a 1 m resolution global CHM. However, our recent comparative study has revealed significant and systematic biases in all of these products, indicating that accurate global mapping of vegetation height remains a challenge.

In this contribution, we address the fundamental limitations of GEDI-based CHMs arising from input data quality, as well as potential enhancements achievable by integrating ICESat-2 data. We then introduce an improved method that significantly increases accuracy over existing global models and provide a detailed analysis of the factors influencing this accuracy, including the relative importance of different predictors (e.g., optical, radar, or terrain variables). Finally, we discuss pathways for further improvement and demonstrate the method through case studies from three topographically diverse regions.

Effectively assessing plant species diversity across landscapes is essential for biodiversity monitoring and management amidst the current biodiversity loss crisis. Remote sensing research has recently advanced promising operational tools for estimating essential biodiversity variables over large scales from satellite spectral data. In particular, Féret & de Boissieu (2020) developed an R package (biodivMapR), that allows to derive alpha and beta diversity indicators from Sentinel‐2 data, based on the Spectral Variation Hypothesis and the concept of “spectral species”. This study aimed to assess the effectiveness of this tool in the context of a tropical African landscape by testing its spectral-derived indicators against ground truth data. Forest inventories were conducted at 1256 m² plots across a 4 km regular sampling grid throughout the Mabi-Yaya Nature Reserve, located in the southeastern Ivory Coast. Alpha and beta diversity indices were computed from the field measurements and confronted with the indicators derived from biodivMapR. Results showed a significant moderate positive correlation between the field- and spectral-estimated Shannon indices (R² = 0.46) and the Bray-Curtis dissimilarity matrices (R² = 0.44). These results highlight the potential of biodivMapR and its derived Sentinel‑2-based species diversity indicators as tools for monitoring biodiversity in key African conservation landscapes. Further research will extend to two protected areas in Cameroon, broadening the evaluation of this remote-sensing approach’s applicability for biodiversity research and decision support for conservation efforts across diverse regions.

The Przewalski’s horse (Extinct in the Wild in 1996) is currently listed as Endangered. It is a flagship species which could be used for conservation of the whole habitat. However, reintroduction into its former habitat and further conservation are fraught with challenges and require immense effort. First individuals were reintroduced to the Great Gobi B Strictly Protected Area (Gobi B), Mongolia, in 1997. We observed selected horse groups in the Gobi B between intra-annual (2019) selected periods in 2019 and used ecological niche models (ENMs) to: 1) model habitat preferences for feeding and resting with a binomial logistic regression; 2) identify the influence of origin (Wild-born vs Reintroduced); and 3) describe the potential influence of human presence on the habitat selected by the horses for these behaviours. We used three types of satellite-derived predictors: i) topography (ALOS); ii) vegetation indexes (Landsat); and iii) land cover (Copernicus). We assessed the spatial similarity between Reintroduced vs. Wild-born models with pairwise comparisons of the two response variables (feeding and resting). We found significant differences between the horses’ origin in habitat preferences. Predictors showed opposite signals for Wild-born and Reintroduced horses’ feeding behaviour (positive and negative, respectively). For the successful reintroduction of Przewalski's horses, habitat suitability, anthropogenic pressure, and reintroduced group size should be considered key factors. High spatial resolution remote sensing data provide robust habitat predictors for feeding and resting areas selected by Przewalski's horses.

In situ bio-optical datasets are essential for the assessment of the uncertainties of satellite ocean colour measurements and derived products. This is especially critical in coastal waters, where land adjacency effects, complex atmospheric aerosol mixtures, high loads of optically active components in particular high concentration of chromophoric dissolved organic matter and bottom reflectance effects contaminate the signal that reaches the satellite.

The Tara Europa expedition, the ocean component of the Traversing European Coastlines (TREC) program carried a comprehensive sampling of coastal ecosystems all along the European coast in 2023 and 2024. The Tara Europa expedition offered the unique opportunity of an oceanographic survey from a unique platform, using the same set of protocols, instruments, and sample analysis, collocated with a rich biological dataset describing the microbiologic diversity in detail.

Within the ESA-funded Hyperspectral Bio-Optical Observations Sailing on Tara (HyperBOOST) project, PML, CNR, LOV and UMaine extended the variables collected during the TREC integrated sampling by including bio-optical measurements relevant to present and future satellite ocean colour missions. This effort provided a comprehensive dataset encompassing in-situ hyperspectral radiometry, bio-optical properties, optically active components, biogeochemical and biodiversity relevant data for optically complex waters. This dataset will be useful to develop new algorithms and as validation data for several missions, products, and datasets.

This presentation will provide a summary of the bio-optical dataset collected on Tara and explore its relevance to present and future satellite missions in view of development and validation of coastal and oceanic biodiversity applications.

Vegetation diversity has been demonstrated to influence ecosystem function and to provide essential services. However, the biodiversity-ecosystem function relationships are very complex and still not fully accounted for at different spatial-temporal scales. Remote sensing is a viable method to monitor plant diversity at different scales that are relevant for management purposes. This is most commonly done by exploiting the spectral variability hypothesis, which relates spectral heterogeneity to plant diversity. This study examnined the relationship between spectral diversity (SD), functional diversity (FD), and water use efficiency (WUE) of the herbaceous understory of a Mediterranean tree-grass ecosystem using a combination of proximal sensing, namely field spectroscopy and unmanned aerial vehicles (UAVs), and satellite imagery from the Copernicus program (Sentinel-2 and Sentinel-3). A canopy-scale spectral library (2017-2023) coupled with destructive functional trait sampling was used to derive a reference ecosystem-level FD and SD dataset. Subsequently, UAV and Sentinel thermal and near-infrared imagery were used to ingest a coupled surface energy balance and carbon assimilation model to estimate evapotranspiration (ET), gross primary productivity and WUE. Preliminary results demonstrated a significant relationship (r > 0.6, p-value < 0.001) between SD and FD across different phenological stages. Along with this, high-resolution ET retrievals from UAV imagery showed a positive relationship with SD (r ~ 0.8) while a weaker relationship (r ~ 0.4) was found between WUE and FD. However, the few data points available from the UAV campaigns limit the generality of these relationships, which might be driven by other factors such as the vegetation traits themselves. As such, satellite-based ET and WUE were produced to obtain a dense time series between 2017 and 2023 to better isolate the relationship between diversity metrics and WUE at different temporal scales (monthly, seasonal and annual).

Measuring and monitoring global biodiversity requires accessible, reliable biodiversity data products. Next-generation remote sensing approaches, including imaging spectroscopy and lidar, when integrated with field data, can help create scalable biodiversity data products. However, despite their potential, the techniques to do this are still in development and their limitations are poorly understood. Addressing this need motivated the U.S.’s National Aeronautics and Space Administration’s (NASA) first integrated field and remote sensing campaign focused on biodiversity - the Biodiversity Survey of the Cape (BioSCape) - which took place in South Africa in late 2023.

Here, we present BioSCape, its expected research contributions, and its Open Access datasets. BioSCape’s airborne data includes 45,000km² of contemporaneous measurements from six instruments aboard three aircraft. Imaging spectroscopy measurements covering ultraviolet and visible to near-, shortwave- and thermal infrared regions were collected by NASA’s PRISM, AVIRIS-NG and HyTES instruments, while LVIS collected full-waveform lidar measurements. Additional discrete return lidar and high resolution RGB photography were collected by the South African Environmental Observation Network’s Airborne Remote Sensing Platform. Accompanying the airborne data are a range of coincident field measurements, from vegetation and phytoplankton community data to acoustic and environmental DNA sampling.

BioSCape’s Open Access dataset is unprecedented and will dramatically increase our ability to map multiple diversity indices, plant functional traits, kelp forest extent and condition, acoustic diversity, estuarine essential biodiversity variables, phytoplankton functional types, environmental DNA-derived diversity metrics, invasive species, phylogenetic traits, and many other biodiversity characteristics of terrestrial and aquatic ecosystems. In doing so, BioSCape is bringing us closer to measuring biodiversity from space.

Forests and other wooded lands cover almost 40% of the land area in the EU27 (Forest Europe, 2020). Forests are some of the most biodiverse ecosystems and at the same time provide a wide range of ecosystem services. They produce wood and non-wood products with a strategic economic and social relevance, remove and stock carbon dioxide and pollutants from the atmosphere, sequestering up to 60% of anthropogenic carbon emissions. Forests are relevant for purifying water, protecting against soil erosion and flooding, and serve as places of high recreational and spiritual value.

Forest resource monitoring by National Forest Inventories (NFIs) constitutes a crucial tool in many countries. Forest data by NFIs provide the basis for land management policy and decision-making, for in-depth assessment of forest health, and national evaluation and reporting of the current and future condition of forests, including their biodiversity status.

This contribution presents briefly the new Italian NFI (“Inventario Forestale Nazionale Italiano – IFNI”) is scheduled for the year 2025. In addition to traditional forest measures, new variables for biodiversity monitoring were introduced including the presence and abundance of tree-related microhabitat, epiphytic lichens and plant morphological groups. We then focus the presentation to the Earth Observation component of IFNI for wall-to-wall mapping of inventoried forest variables through the integration of ground and remote sensing data, as well as the implementation of advanced remote sensing tools and data to streamline fieldwork and improve estimators’ precision.

Coastal dunes are unique transitional dynamic ecosystems along sandy shorelines, highly threatened by human activities. Traditional monitoring of their temporal changes has relied on field resurvey campaigns with high costs and times. Very high spatial and temporal resolution of open-access remotely sensed (RS) data offers a promising cost-effective alternative.

Our study examines temporal changes in coastal dune vegetation within the Mediterranean protected area “Castelporziano Presidential Estate” (IT6030084) with restricted access. We analyzed floristic and landscape changes over a 25-year period of three habitat units: Herbaceous Dune Vegetation (HDV), Woody Dune Vegetation (WDV), Broadleaf Mixed Forests (BMF). We assessed whether plant diversity influences landscape dynamics by combining satellite imagery and resurveyed field data (through 58 resurveyed vegetation plots). Landscape changes were analyzed using a chord diagram, while floristic shifts were examined with Rank Abundance curves. Shannon diversity was calculated for floristic and landscape diversity, within 25, 75, and 125 m buffers around the plots. Linear Mixed Models were applied to explore the influence of floristic diversity on landscape changes.

Our results showed a reduction in artificial cover due to natural encroachment, accompanied by a vegetation succession at landscape scale. Additionally, in the analysis of floristic changes, we observed strong differences between T₀ and T₁, particularly in WDV, where Cistus sp. pl. dominance disappeared. The models explained variability well (R² > 0.82), especially for larger buffers, and indicated differences between the relationships at T₀ and T₁. Notably, landscape changes were linked with negative trends to increment in species dominance, such as for WDV at T₀, while positive trends reflected greater floristic equipartition.

To conclude, our RS approach represents an effective tool for assessing the relationship of plant diversity on landscape and for monitoring temporal changes, and it could represent a starting point for implementing conservation measures within Protected Areas accelerating resurvey times.

Exploring the intricate interplay between global biodiversity patterns and the looming impact of climate change stands as a paramount inquiry within the realm of earth system science. Furthermore, the acknowledgment of shifts in plant functional diversity emerges as a key catalyst, wielding substantial influence over pivotal ecosystem processes like the carbon cycle. Various essential plant traits, intricately tied to vegetation function—ranging from photosynthesis to carbon storage and water/nutrient uptake—underscore the significance of comprehensive global trait maps. These maps prove indispensable for unraveling environmental interactions, identifying threats to the biosphere, and fostering a profound understanding of our planet's intricacies. However, the sparse and non-representative nature of current trait observations poses a formidable challenge. Presently, global maps of vegetation traits are constructed by bridging observational gaps, primarily relying on empirical or statistical relationships between trait observations, climate and soil data, and remote sensing information. However, these approaches exhibit limited explanatory power, struggle to encompass a myriad of traits, and face constraints in ensuring ecological consistency in their extrapolations.

The VESTA (Vegetation Spatialization of Traits Algorithm) project emerges as a groundbreaking initiative aimed at refining our grasp on global above and belowground plant traits. This endeavor involves integrating a trait-based dynamic global vegetation model (DGVM) with Earth observation (EO) data. Trait-based DGVMs, rooted in a process-based foundation, forge a direct nexus between the environment, plant ecology, and emerging vegetation patterns. Leveraging insights from contemporary global trait databases, the model is initialized to mirror real-world conditions. Subsequently, EO data enters the equation to fine-tune the model through a calibration process, adjusting trait relationship curves having as reference satellite measurements of vegetation structure and productivity.

Drawing parallels to prior methods used in climate reanalysis, EO-constrained trait-based DGVMs yield a multivariate, spatially comprehensive, and coherent record of global vegetation traits. The resultant dataset encapsulates trait distributions, offering detailed insights into plant functional diversity metrics—mean, variance, skewness, and kurtosis—at specific locations. Notably, these trait maps extend beyond mere snapshots, evolving into a temporal series that affords a nuanced comprehension of the prevailing state of functional diversity and its temporal shifts. Ultimately, the fruition of this project manifests as an invaluable EO product, showcasing leaf, wood, and root traits and their change through time.

The scientific community working in remote sensing and biodiversity often faces challenges in integrating and analyzing diverse Earth observation data with biological and ecological measures for extensive monitoring and understanding of biodiversity changes. Additionally, assessing ecosystems' stress responses and changes in biodiversity using Earth observation data remains complex. The book "Biodiversity Insights from Space" aims to demonstrate the utilization of Earth observation data for biodiversity monitoring across different biomes through the assessment of biodiversity indicators and attributes within the EBV framework. It provides comprehensive guidelines and case studies that illustrate the benefits and challenges of using Earth observation data for detecting stress responses and changes in biodiversity, addressing biodiversity targets, and biodiversity management.

Microphytobenthos (MPB) are microalgae that form biofilms on sediment surfaces and play an important role in coastal ecosystems, particularly in supporting food webs, carbon (CO₂) fluxes, and stabilizing mudflats. Traditionally, MPB assessments have been conducted in situ; in recent years, remote sensors have increasingly been used for these evaluations. However, studying MPB using satellite data is challenging due to "scaling bias" – differences in observations based on the data's spatial resolution. For example, carbon flux estimates, derived from biomass, are calculated using a Gross Primary Production (GPP) model based on NDVI (Normalized Difference Vegetation Index). This scaling bias occurs due to non-linear conversions from NDVI to biomass associated with the spatial variability of MPB. This study aims to measure the scaling bias using drone data, which offer higher resolution than satellites. The drone data was collected over four sites during different seasons. It helps analyze MPB's spatial patterns and simulate what satellite pixels would capture at coarser resolutions. The NDVI data is modeled using a beta distribution, and the conversion from NDVI to biomass is handled by an exponential model to account for saturation at higher biomass levels. A linear resampling process is used to simulate satellite pixels from drone data, though this assumption is being further examined and discussed. The results show that biomass calculated at coarser satellite resolutions tends to be slightly lower than those from finer drone data, with a scaling bias of a few percent.

Mapping landscapes is essential to meet the challenges of climate change and the need for sustainable development while preserving biodiversity and ecosystems. Here we present a method for extracting essential landscape components solely from radiometric information derived from satellite imagery. This approach is based on the concept of Remote Sensing-based Essentiel Landscape Variables (RS-ELVs). The method was initially developed and tested in the context of central Madagascar, with its contrasting landscapes in terms of climate and agricultural practices. RS-ELVs are derived from MODIS time series for temporal and spectral variables, and Sentinel-2 and MODIS imagery for textural variables. The segmentation and clustering parameters used to determine the landscape units and their types (radiometric landscapes) are based on statistical optimisation methods. For Madagascar, six radiometric landscape types were identified. The landscape types were then characterised using independent remote sensing data, a land cover map and field observations. Finally, prospects for the future are presented with the operationalisation of the processing chain via a graphical interface and first results of applications in Central America (Costa Rica). These results highlight the potential application of the method to map landscape units in different geographical and ecological contexts.

Understanding forest dynamics is critical to biodiversity conservation and policy development, especially in regions such as the Italian Apennines, including the Matese Regional Park, where significant land cover changes have occurred over the last century. These changes, driven by new herding techniques, forest use and management, pasture abandonment, and climate change have led to decreasing grassland and increasing forested areas. While previous studies have examined these transformations, a significant gap remains regarding other drivers, such as changes in forest composition and climate-related stress.

This study addresses this gap by leveraging spaceborne remote sensing technologies to classify land cover, comparing historic imagery with recent multispectral and hyperspectral satellite data. Studies of large-scale forest dynamics have prevalently relied on the interpretation of images providing panchromatic data, such as those from the 1943 Royal Air Force flight or the Gruppo Aeronautico Italiano flights conducted between 1952 and 1954. Today, Sentinel satellites from the European Space Agency’s Copernicus program provide spatial resolutions of up to 10 m as well as multitemporal and multispectral information useful for more accurate land cover classification. Additionally, high spectral resolution (240 bands between 400 and 2500 nm) data from PRISMA and EnMAP satellites are now available, allowing for more accurate classifications and information on stress and changes in complex habitats such as grasslands, despite their limited acquisition availability and medium resolution (30m).

In this study, a ground truth database collected in the field was used to assess the accuracy of classification results based on these various sources in a case-study area of the Matese Regional Park in Campania, Italy. The findings allow us to compare the pros and cons of the various data sources and confirm an ongoing trend of diminishing grazed areas, which can lead to the proliferation of invasive species that threaten protected species and their habitats.

Functional traits determine how plants respond to the accelerating environmental change and affect ecosystem dynamics. In the context of global biodiversity loss and the ongoing degradation of ecosystems, understanding functional traits aids in biodiversity assessment, ecosystem functioning, and conservation planning. Tropical forests play a vital role in adjusting the global climate and atmosphere. Thus, accurately monitoring and tracking the spatiotemporal dynamics of their functional composition and structure is of high priority for mitigating and halting biodiversity loss. The main goal of this study is to demonstrate to what extent remotely sensed data and environmental variables can be useful to map and predict functional traits including morphology, nutrients, and photosynthesis across the tropics with artificial intelligence methods. For our analyses, we integrated multi-source remotely sensed data with in-situ plant trait measurements to map and predict 15 functional traits with Random Forests and Multilayer Perceptron algorithms at 10 m, and we obtained optimal predictive accuracies with mean R² scores being 0.40, 0.43, and 0.57 for predicting photosynthetic, morphological, and nutrient traits at pan-tropical scale. We explored the distribution and variation patterns of traits at multiple spatial scales, and further investigated main factors in driving the distribution and variation of each trait. We found that soil properties and climatic characteristics consistently contributed the most to the distribution and variation patterns of these functional traits. This study provides comprehensive and new approaches for mapping and predicting multiple key functional traits and underpinning the understanding of the relationships between biodiversity and ecosystem-function under environmental change in the most biodiverse terrestrial ecosystem.

Current and forthcoming spaceborne visible to shortwave infrared (VSWIR) imaging spectrometers have the potential to deepen our understanding of the relationships between plant trait composition and long-term ecosystem stability. Changing fire regimes and hotter droughts are impacting ecosystems globally. Identifying systems at high risk for declines in ecosystem functioning and biodiversity is crucial for effective land management, and is a promising use case for spaceborne VSWIR data. California is a global biodiversity hotspot that has recently experienced a multi-year megadrought and repeated high-severity fires, making it an ideal test case for studying the relationships between plant trait composition and ecosystem stability. This research presents preliminary results towards integrating long-term multi-spectral satellite data (Landsat 4-9) with plant trait maps derived from airborne VSWIR data to (1) identify historical drivers of fire recovery rates and drought sensitivity and (2) explore fire impacts on trait distributions across diverse field sites in California. For objective (1), we use Landsat vegetation index time series to quantify different metrics of ecosystem stability, including fire resistance, fire recovery time, and drought sensitivity. We then train random forest models to identify drivers of decreased ecosystem stability based on topography, climate history, disturbance severity and frequency, and vegetation type. For objective (2), we explore the relationships between changes in plant functional richness and each stability metric developed in aim (1). Next steps include testing the ability to scale this work to trait maps derived from NASA Earth Surface Mineral Dust Source Investigation (EMIT) data.

Forest ecosystems cover approximately one tenth of the Earth’s surface and provide numerous ecological functions and services, largely due to their high biodiversity and their critical role in climate regulation and biogeochemical cycles. However, climate change and human activities poses a significant threat to the conservation of these ecosystems. Essential biodiversity variables (EBVs) aggregate biodiversity observations collected through different methods such as in situ monitoring and remote sensing and aim at supporting environmental monitoring. The performance of Earth observation for biodiversity estimation largely depend on the type of forest, the type of EBV and the characteristics of the sensors in use.

This presentation aims to share results on the estimation of EBVs based on airborne imaging spectroscopy in two distinct forest types: a dense temperate forest and a sparse Mediterranean forest. The case study for the temperate forest is the Fabas forest located in the South of Toulouse (France). We highlight the advantage of using a 10 m Ground Sampling Distance (GSD) for species classification at the tree scale, followed by the estimation of biodiversity parameters (α- and β-parameters). Our results showed high correlations between spectral diversity and observed taxonomic diversity (Rho ranging from 0.76 to 0.82). Functional diversity was more variable (Rho ranging from 0.45 to 0.63).The case study for the Mediterranean forest is the Tonzi site in California (USA). For this dataset, we focus on the estimation of a set of leaf biochemical properties (pigment content equivalent water thickness and leaf mass per area) using radiative transfer modelling.

Understanding the varied responses of tropical forests to climate seasonality and global change requires comprehensive knowledge of the abundance, function, and demographics of tree species within these ecosystems. Unlike temperate forests, tropical forest phenology emerges from individual-level events, which are often poorly understood due to same-species asynchronous flowering and complex species distributions. New spaceborne tools offers promising opportunities to improve our understanding of tree species distribution, phenology and mortality.

PlanetScope (PS) imagery, with its daily global coverage at ~3m spatial resolution, provides a scalable and cost-effective means to monitor tropical trees, but its spatial and spectral limitations make it difficult to resolve individual crowns and detect species. We address this challenge by focusing on large tropical tree crowns that exhibit conspicuous phenological events, such as vigorous floral displays or significant leaf loss. These strong phenological signals enable resolving individual crowns otherwise difficult to detect in primarily “evergreen” tropical canopies.

Our project prototypes advanced Artificial Intelligence (AI) and Deep Learning (DL) models designed to process and interpret daily PS imagery time-series to monitor tree-level phenological events, including flowering and leaf shedding. We will discuss its potential and limitation to monitor short- and long-lived flowering events and the challenges of frequent cloud cover occlusion. Our trade-off study will identify what species are detectable from space based on their crown size, phenological traits (flower cover fraction and flowering temporal length) and timing (e.g. dry vs. wet season). Ultimately, our research aims to identify keystone species that can act as sentinel of tropical health, enhancing our scientific understanding of species distribution and develop automatic observing framework to monitor phenological responses and tree mortality in face of climate seasonality and global change.

Aquatic fungi (AF) are key parts of biodiversity in freshwater, marine and cryospheric ecosystems, where their ecosystem functions include decomposition of organic matter, nutrient cycling, and as parasites that may control populations of animals and plants. However, the biodiversity and ecological roles of AF have for a long time been underappreciated. AF are missing from all large-scale ecosystem monitoring initiatives, there are considerable knowledge gaps of AF ecology and taxonomy, and the public awareness of AF is limited at best.

With the rapid development of earth observation (EO) data and analysis, and their implementation in different monitoring frameworks, there may be untapped opportunities for the use of EO data in AF monitoring. Specifically, AF responses to environmental change may be indirectly visible by remote sensing of e.g. algal blooms, water turbidity, and different anthropogenic pressures.

As part of the Biodiversa+ EU co-funded project MoSTFun, we perform a study exploring which EO-derived variables best explain AF biodiversity patterns and drivers. We use two well-established field sites in the SITES monitoring network in Sweden as case studies; a freshwater system (lake Erken, 59.8 N 18.6 E) and a glacier system (Tarfala, 67.9 N, 18.6 E). These case studies will provide long-term in situ environmental and taxonomic data with high temporal resolution. The in situ data will be analysed in parallel with optical signals from medium-resolution (Sentinel-2) and very-high-resolution (CNES Pléiades) satellites.

The results from this study will be included in downstream development of Essential Biodiversity Variables (EBVs) for AF and to form recommendations for the use of EO-derived variables to inform AF monitoring

The Greater Cape Floristic Region is a biodiversity hotspot that harbors extraordinary plant diversity, with over 10,000 species, nearly 80% endemism, and exceptionally high β-diversity, or turnover in species composition among sites. Numerous studies have explored the use of remote sensing data to estimate different components of biodiversity, but few studies have examined the extent to which in-situ biodiversity observations can be integrated with high-dimensional remote sensing data from multiple instruments to quantify and map β-diversity. Here we use forest plot data from Garden Route National Park in South Africa to explore the relative importance of hyperspectral imagery and waveform lidar to quantify and map functional, phylogenetic, and taxonomic components of vegetation β-diversity. Based on previous studies that demonstrate that remote sensing mainly detects phenotypes, we hypothesized our ability to quantify vegetation composition using remote sensing should be greatest for functional, lowest for taxonomic, and intermediate for phylogenetic β-diversity. We calculated taxonomic, functional, and phylogenetic β-diversity for 47 forest tree species in 647 plots and used a reduced set of 20 of the original 339 hyperspectral and lidar variables to fit Generalized Dissimilarity Models for each dimension of β-diversity and assess the relative contribution of the 16 hyperspectral and four lidar variables. We found percent deviance explained was greatest for phylogenetic β-diversity (74.5%), intermediate for functional β-diversity (52.2%), and least for taxonomic β-diversity (40.0%). Lidar variables were the most important predictors for phylogenetic and functional β-diversity, while hyperspectral variables were most important for taxonomic β-diversity. Our results demonstrate the high explanatory power and relative strength of hyperspectral and lidar data to quantify and map taxonomic, phylogenetic, and functional β-diversity for tree species across large regions, especially using phylogenetic information and lidar data that distinguishes vertical structure among different tree species.

Satellite remote sensing data is key to improve our understanding of wildlife-environment interactions at large scale. It is a continent-wide data source, extensively used by researchers globally, for instance to link wildlife occurrences to habitat characteristics and facilitate extrapolation to larger areas. However, the accuracy of remotely sensed satellite data can vary depending on the land cover type and location. Therefore, it is crucial to estimate how large the classification error of land cover data is using ground truth data. Previous work have shown that images taken by camera traps can be used to measure variables such as snow cover and green-up of the vegetation. This research, part of the ‘Big_Picture’ project, recently funded by Biodiversa+, focuses on using camera trap images as ground truth data to refine satellite-derived measurements of land cover and vegetation phenology across Europe. We will use the Phenopix package in R to quantify the greenness from camera trap images and to automate the identification of the spring green-up. Next, we will link these measures to the Copernicus NDVI v2 product to estimate the timing of the vegetation green-up throughout Europe, which then in turn will be related to the timing of reproduction in a range of mammal species. Furthermore, we will manually classify 23 land cover types across Europe in camera trap images (as a ground truth) to assess the classification error in the Copernicus land cover product. These images will also serve as a training data set for deep learning models in order to automate this process for broader spatial coverage. This study will provide a novel approach to enhance the accuracy of remote sensing data for ecological applications, potentially benefiting large-scale wildlife monitoring efforts.

Emperor penguins are sea ice obligate species whose breeding cycle is intricately linked to the fluctuations of Antarctic fast ice. Predictions of their future populations, based on IPCC climate change driven sea-ice extent estimates are pessimistic, suggesting that almost all colonies will be extinct by the end of the century. However, as in recent years, sea ice extent has not declined in a linear way, some parties have called for extra evidence on the actual population demographic before extra conservation measures are put in place. Here we present a 15 year population index for the species using very high resolution satellite imagery to assess penguin populations. We use a maximum likelihood classification analysis to isolate penguin area and assess that area with a Markov model linked to Bayesian statistics. In this analysis, 16 colonies, in the sector between 0° and 90°W were assessed each year between 2009 and 2023. The results show that although regional patterns vary, the overall decrease for this sector is 22% over the period (1.47% per year), a rate of change significantly higher than that predicted by the demographic modelling in the “high emission” scenario. Several regional factors could have influenced this analysis, however these results show the importance of satellite population estimates on a species that is almost impossible to access on the ground and highlight the need for complete EO survey of the whole population and better understanding of the drivers of change linked to warming conditions.

As sea ice retreats in the Arctic, the future of walruses (Odobenus rosmarus) is uncertain. Understanding how the alteration in their habitat is affecting them is essential to predict and safeguard their existence. However, it is logistically challenging to monitor walruses via conventional research platforms (such as boats and planes), as they live in remote locations across the whole Arctic, limiting the areas where field surveys can be conducted, as well as restricting the regularity of such surveys. Satellite imagery could be a non-invasive solution to studying walruses, which have been successfully detected in both medium and very high-resolution satellite imagery. The Walrus from Space project, with partners around the Arctic, aims to monitor Atlantic walruses (Odobenus rosmarus rosmarus) using very high-resolution satellite imagery and the help from citizen scientists to review the large number of images (~500,000 image chips of 200 m x 200 m), every year for 5 years (2020-2024). Three citizen science campaigns have been completed so far, including two search campaigns with imagery from 2020 and 2021, one counting campaign with imagery from 2020. To date, 12,000+ citizen scientists took part reviewing more than a million image chips. They found small (< 5 walruses) and very large group of walruses (100+ walruses) hauled out on sandy and rocky shores, including in poorly surveyed locations, highlighting the potential to use satellite imagery to monitor walruses.

One of the effects of agricultural intensification is the removal of woody vegetation features from landscapes. These woody plants provide habitats for various plant and animal species and thus provide important ecosystem services that increase biodiversity in agricultural landscapes. The importance of the woody vegetation landscape features has also been recognised by governments, leading to programmes for their conservation. However, the programmes have encountered a problem arising from the lack of data on the extent and distribution of the woody vegetation landscape features.

We mapped woody vegetation by using national orthophotos as input. First, a convolutional neural network was trained to detect all tree canopies in the areas of interest. In order to obtain a nationally applicable model, 20 areas of interest representing different Slovenian landscapes were selected for training, validating and testing the model. Subsequently, the detected woody vegetation landscape features were vectorized and the resulting polygons were divided into seven different classes. These classes were: single trees, trees in rows, groups of trees and shrubs, orchard trees, riparian vegetation, hedges and forest. The geometric characteristics of the polygon and the positional relationships between the classified polygon and its neighbours were used in the classification.

While the detection of tree crowns with a Jaccard index of 79% in the agricultural areas works as desired, the subsequent classification is still a work in progress. The woody vegetation features are mostly correctly classified in areas with low feature density; however, numerous polygons that are close to each other remain a challenge. It is particularly difficult to correctly recognise trees in rows and orchard trees, with hedges also often being classified as groups of trees and shrubs.

Climate change is one of the most pressing environmental issues of our time, with significant implications across ecosystems, including inland freshwater systems.

As global temperatures rise due to greenhouse gas emissions, inland water bodies such as rivers, lakes, and wetlands are experiencing noticeable warming with an average temperature rise of 0.5 degrees per decade. This increase water temperature is causing widespread changes in aquatic ecosystems, altering species distribution, biological processes, and ecosystem resilience:

- Disruption of thermal stratification and mixing patterns

- Altered species distribution and biodiversity loss

- Enhanced eutrophication and algal blooms

- Reduced oxygen levels and metabolic stress

In the same time, climate change is increasing the frequency of extreme events such as floods and droughts.

The Adour Garonne Water Agency (France) has decided to launch a research and innovation project to study the functioning of aquatic environments that are being modified by climate change, in terms of both hydrology (flooding, low water) and quality (water temperature, turbidity, etc.), considering the two aspects to be intimately linked. To carry out this experiment, which aims to provide a better understanding of the impact of climate change on the basin, it is crucial to deploy a significant number of instruments to test the effectiveness of the system. To date, only the vorteX-io device allows simultaneous acquisition of real-time quantitative and qualitative measurements. For this reason, the Agency has commissioned vorteX-io to provide water temperature and metrics with 150 vorteX-io micro stations on the Garonne River Basin as part of this project.

The vorteX-io micro station is a device derived from space technology, innovative and intelligent, lightweight, robust, and plug-and-play. Water parameters are transferred in real-time through GSM or SpaceIOT networks. The micro stations are equipped with unprecedented features that allow them to remotely and in real-time measure water temperature, provide contextual images and floods metrics (water levels, flow, rain rates).

This instrument provides in situ datasets for calibration, validation and accuracy assessment of EO projects in space hydrology, i.e. in the ESA st3art project dedicated to the calibration and validation of Sentinel 3.

The long-term vision is to cover river basins in Europe with an in-situ network, to be used at large scale as earth-observation in situ component either for monitoring water quality parameters or for extreme hazards monitoring such as floods and droughts.

Characterizing the pathways from Earth Observation (EO) data to products to societal benefits is a complex but crucial task to understand the value of EO investments. The U.S. Group on Earth Observation (USGEO) Earth Observation Assessment (EOA) measures the effectiveness of EO systems in meeting high-level objectives identified within Societal Benefit Areas (SBA), including Biodiversity and Ecosystems. The first two EOAs, conducted in 2012 and 2016, assessed all 13 SBAs simultaneously. Future EOAs will instead assess two to four SBAs per cycle, updating all SBAs over a 5-year period. In the upcoming cycles, USGEO will convene a large group of U.S. Government federal scientists to design a value tree study and identify connections between EO data sources and thematic sub-areas under the Ecosystems SBA and Biodiversity SBA. Here, we showcase the results from the two SBA value trees presented in previous EOA studies and offer recommendations for enhancing future assessments.

This study presents the Connectivity, Climate, and Land use (CCL) Nexus approach, a comprehensive framework developed to assess the interactions among landscape connectivity, climate change, and land use/cover transformations in the Mediterranean context of Central Italy. The analysis incorporates Earth Observation (EO) data, integrating both high-resolution land use and climate information to provide a solid foundation for scenario-based modeling. Specifically, bioclimatic indicators, including the aridity index, were sourced from the Copernicus Climate Data Store (CDS) and utilized at their native 1 km spatial resolution to capture nuanced climate variables affecting vegetation productivity and ecosystem resilience. These EO-derived climatic data, combined with updated satellite-based land use maps, support a robust input dataset for PANDORA model simulations over the period from 2001 to 2100.

The PANDORA model, used in this study, leverages principles of landscape thermodynamics and bio-energy fluxes, offering a structured method to simulate the effects of climate and land use scenarios on landscape connectivity. Scenarios included both Business-as-Usual (BAU) and intervention-based projections, with particular attention to the effects of urbanization and naturalization on connectivity. The aridity index, along with land cover and soil characteristics, were assigned specific parameters to evaluate the bio-energy landscape connectivity (BELC) index across various climate models and land use scenarios, from present-day conditions to high-intensity change scenarios.

Results show that while climate change scenarios yield moderate impacts on connectivity, urban expansion presents the most significant disruption, with naturalization alone proving insufficient to counterbalance urban pressures. The findings advocate for the integration of EO data within multi-level planning frameworks to enhance the efficacy of land management, prioritizing actions that promote connectivity, biodiversity conservation, and resilience against future climate variability. This approach demonstrates the value of satellite-derived climate and land use data in supporting localized planning decisions and advancing sustainable regional development in complex socio-ecological systems.

Accurate mapping of vegetation’s 3D structure is essential for understanding ecological processes like biomass distribution, carbon sequestration, habitat diversity, and biodiversity. Satellite-based LiDAR missions, such as GEDI and ICESat-2, have significantly advanced the measurement of canopy height, cover, density, and vertical heterogeneity metrics. However, the sparse data collection nature of these missions requires combining GEDI/ICESat-2 measurements with multispectral (e.g., Sentinel-2) and synthetic aperture radar (SAR) datasets (e.g., Sentinel-1 and ALOS-2) to achieve spatially continuous mapping. This integration supports robust, spatially explicit mapping of critical vegetation structure indicators. By integrating LiDAR with optical and SAR data, we demonstrate an effective approach to overcoming the limitations of single-source datasets. This presentation includes a comparative analysis of GEDI- and ICESat-2-derived wall-to-wall vegetation structure maps, highlighting the primary strengths and limitations of GEDI/ICESat-2 data for generating accurate and ecologically relevant vegetation metrics.

Pretraining deep neural networks in a self-supervised manner on large datasets can produce models that generalize to a variety of downstream tasks. This is especially beneficial for environmental monitoring tasks where reference data is often limited, preventing the application of supervised learning. Models that can interpret multimodal data to resolve ambiguities of single-modality inputs may have improved prediction capabilities on remote sensing tasks.

Our work fills an important gap in existing benchmark datasets for geospatial models. First, our benchmark focuses on the natural world, whereas many existing datasets focus on the built-up world. Second, existing datasets tend to be local or cover relatively small geographic regions in the global North. However, evaluating and distinguishing performance among pretrained models that aim to contribute to planet-scale environmental monitoring requires downstream tasks that are distributed around the globe. Third, existing datasets include only a few modalities as input (e.g., RGB, Sentinel-1 (S1) SAR, and Sentinel-2 (S2) optical images), even though many additional data modalities are relevant to environmental prediction tasks.

We present MMEarth-Bench, a collection of datasets for various global-scale environmental monitoring tasks. MMEarth-Bench consists of five downstream tasks of high relevance to climate change mitigation and biodiversity conservation: aboveground biomass, species occurrence, soil nitrogen, soil organic carbon, and soil pH. Each downstream task dataset is aligned with the twelve modalities comprising the MMEarth dataset, designed for global multimodal pretraining, including S2 optical images, S1 SAR, elevation, canopy height, landcover, climate variables, location, and time. We use MMEarth-Bench to evaluate pretrained models, often called “foundation models,” that make use of multiple modalities during inference, as opposed to utilizing just a single modality such as optical images. We demonstrate the importance of making use of many modalities at test time in environmental monitoring tasks and also evaluate the geographic generalization capabilities of existing models.

Is wildlife trafficking truly visible from space? Can satellites reliably detect where sustainable land management practices are being implemented? Prior research indicates that remote sensing data combined with machine learning approaches can estimate these, along with other Sustainable Development Goal (SDG) indicators, with impressive accuracy. However, considering the capabilities of modern spaceborne sensors, it seems more plausible that models are capturing correlations between these practices and observable environmental factors rather than the practices themselves.

Of the 14 indicators that are used to measure progress towards SDG 15, ‘Life on Land,’ we identify those that satellite imagery may conceivably be able to estimate with greater spatial and temporal precision than existing data products, enabling well-informed local interventions previously considered infeasible. We then explore the geospatial metrics that machine learning models might actually be detecting based on causal links established in existing literature. By visualising these connections in a network graph, we argue that while satellite-based instruments hold enormous potential to monitor the SDG indicators at scale, it is essential to consider which features these techniques can genuinely detect and use this understanding to inform reasonable uncertainty bounds for the predicted indicators. We further propose broadly applying this methodology to space-based predictions to enhance interpretability.

Satellite data bears opportunities to quantify and study trait-based functional diversity in forest ecosystems at landscape scales. The high temporal frequency of multispectral satellites like Sentinel-2 allows for capturing changes in canopy traits and diversity metrics over time, contributing to global biodiversity monitoring efforts. Until now, satellite-based studies on trait-based functional diversity have mostly focused on the state of vegetation during peak greenness or during the absence of clouds.

We present an approach using Sentinel-2 time-series data to map and analyze spectral indices related to physiological canopy traits and corresponding functional diversity metrics on 250 km² of temperate mixed forests in Switzerland throughout multiple seasonal cycles. Using composites that were compiled every seven days, we assessed the variation of the indices (CIre, CCI, and NDWI) and the corresponding diversity metrics functional richness and divergence over the course of five years (2017 – 2021). We describe the seasonal and inter-annual variations of trait-related indices and diversity metrics among different forest communities and compare their deviations from values at peak greenness with measurements from other times during the growing season.

We found that, although peak greenness (end of June, beginning of July) was a stable period for inter-annual comparison, for the indices and traits investigated, a period of a few weeks before peak greenness (mid to end of June) might be better. In contrast, for capturing rapid trait changes due to meteorological events, periods closer to the start or end of the season should be considered.

Based on our findings, we provide suggestions and considerations for inter-annual analyses, working toward large-scale monitoring of functional diversity using satellites. Our work contributes to understanding the temporal variation of trait-related spectral indices and functional diversity measurements at landscape scales and presents the steps needed to observe functional diversity over time.

Climate change-induced drought stress is increasingly subjecting Scots pine (Pinus sylvestris) to environmental pressures, making them more susceptible to diseases and pests. The recent devastation of Norway spruce (Picea abies) by the European spruce bark beetle has raised concerns that Scots pine may face a similar fate. Efficient and scalable monitoring of Scots pine vitality is therefore crucial for early detection and management of potential large-scale mortality events. Currently, Flanders uses the forest vitality monitoring network to assess the health of various tree species, including Scots pine. However, this method is labor-intensive and challenging to implement over extensive areas. In this study, we take a first step toward developing a method for the spatially explicit monitoring of Scots pine vitality using multispectral satellites.

To address this challenge, we investigated the use of satellite-based multispectral remote sensing to detect vitality loss in Scots pine at the stand level. Ground reference data on tree vitality were collected with an RGB-NIR drone over 100 hectares of Scots pine stands across Flanders. These drone images were binary classified into vital and non-vital pixels. Drone pixels representing undergrowth and soil were effectively masked out by using a digital elevation model derived by time-for-motion from the drone images.

We compared the performance of Sentinel-2 and PlanetScope satellite data in classifying Scots pine vitality. Sentinel-2 offers higher spectral resolution with bands in the blue, green, red, red edge, and near-infrared (NIR) parts of the spectrum, while PlanetScope provides higher spatial resolution but with fewer spectral bands. Our analysis showed that with a single Sentinel-2 summer image, a classification accuracy of 80% was achieved for distinguishing between vital (<10% discolored or absent needles) and non-vital Scots pine pixels. Moreover, models based on Sentinel-2 data substantially consistently outperformed those based on PlanetScope data, even when using a set of corresponding spectral bands. However, the classification results exhibited a substantial omission error for the non-vital class, possibly due to the subtle symptoms associated with the early stages of vitality loss.

These findings suggest that Sentinel-2 satellite data, when calibrated with accurate ground reference data, can be used to detect vitality loss in Scots pine at a regional scale. This study represents a first step toward developing an efficient, scalable method for monitoring Scots pine vitality using multispectral remote sensing, enhancing proactive forest management strategies to mitigate the impacts of climate change-induced stressors on Scots pine populations.

The Biodiversity Survey of the Cape (BioSCape) campaign was an airborne and field campaign focused on biodiversity in South Africa. Airborne data were acquired via four sensors on two aircraft: PRISM (visible to near infrared wavelengths) and AVIRIS-NG (visible to shortwave infrared wavelengths) on a Gulfstream III and HyTES (thermal infrared wavelengths) and LVIS (full waveform lidar) on a Gulfstream V. Coincident field data were acquired across aquatic and terrestrial ecosystems. All of BioSCape’s data will be Open Access, and the campaign is making significant efforts to ensure the data is also Findable, Accessible, Interoperable, and Reuseable (FAIR). BioSCape is doing this in the following ways:

- Creating an Open Access data portal, supported by NASA’s Multi-Mission Geographic Information System (MMGIS). This portal allows users to download airborne data through an easy-to-use graphical interface.

- The complexity of the airborne data products prompted BioSCape to harmonize data from the four sensors to produce common gridded orthomosaics. This first-of-a-kind analysis-ready dataset can easily be integrated with field data. This will maximize scientific impact and lower barriers to using the data.

- BioSCape also has a centralized webpage where all archived data (field and airborne) can be easily found. Underlying this webpage’s utility is a careful data curation process coordinated through controlled project keywords and NASA’s Common Metadata Repository which ensures that users can easily access a comprehensive listing of BioSCape data collections. This is coordinated and executed by the Oak Ridge National Laboratory Distributed Active Archiving Center (ORNL DAAC).

- BioSCape, in collaboration with Goddard Space Flight Center and Amazon Web Services, has set up a cloud computing environment. This facilitates easy access to the data and to computing resources, which is especially important for South African users.

- BioSCape is running several capacity building events, including locally in South Africa, and creating free online resources to ensure maximum impact of the data.

A significant spread of evergreen broad-leaved (EVE) species has been observed in southern European forests, driven by global change dynamics. Prolonged growing seasons and milder winters, coupled with land-use change are reshaping species composition of forests. In this context large-scale spatial analysis of EVE species distribution and cover in Italian forests is lacking. The main goal of the study is to seamlessly map keystone EVE species abundance and overall EVE cover in Italian broad-leaved forests. The modelling approach involves time series classification and regression based on a modified InceptionTime model. Transfer learning is used to overcome generalizability issues concerning the sparsely available training data from plot observations and the large study area. Annual aggregates of Sentinel-2 L2A bands and derived indices serve as input to the time series models to integrate phenology information in the mapping process. For pretraining an Italian forest vegetation database containing information about forest type with ~16,000 plots is used. During field campaigns in 2023 and 2024 1,440 plot observations were conducted within five protected areas in Italy (Sibillini, Gran Sasso, Gennargentu, Cilento, Nebrodi), that are used for finetuning. Generalizability of the resulting models is evaluated through cross-validation across these areas. The resulting maps contain abundance of key species and overall EVE cover. RMSE values for cover range between 0.17 and 0.22, which shows the challenge in mapping large areas with heterogeneous forest types from few plot observations. Preliminary model results and mapping also reveal that the lack of valid satellite observations during winter and leaf-off season in higher elevations due to snow and extensive cloud cover is the largest error source in broad-leaved forest areas. The study offers insights into challenges and opportunities of Deep Learning in large-scale forest research and mapping applications.

Acknowledgements

This research has been conducted within the project “TRACEVE - Tracing the evergreen broad-leaved species and their spread” (I 6452-B) funded by the Austrian Science Fund (FWF).

Southeast Asia is a global biodiversity hotspot, and yet it has some of the highest rates of habitat loss in the planet. Furthermore this is a region with limited data, and whilst multiple private and government sources of data exist, these are rarely available for the mapping and monitoring of biodiversity. Here we assess the availability of biodiversity data for Southeast Asia, how representative is it, and how might it be used, and combined with other forms of geospatial data to map and monitor biodiversity in systems across the region. Furthermore we assess the ability to map the EBVs for the Asian region, what do we have the data for, and what else do we need to develop and use the EBVs effectively?

Lastly we review recent innovations in monitoring within Asia, such as the use of bioacoustic monitoring paired with deeplearning to automatically and continuously monitor bird diversity across many sites across China. I review the innovations and changes in the biodiversity data landscape across Asia, and discuss where we need to go next.

Forests play an important role in the global carbon cycle as they store large amounts of carbon. Understanding the dynamics of forests is an important issue for ecology and climate change research. However, relations between forests structure, biomass and productivity are rarely investigated, in particular for tropical forests.

Using an individual based forest model (FORMIND) we developed an approach to simulate dynamics of around 410 billion individual trees within 7.8 Mio km² of Amazon forests. We combined the simulations with remote sensing observations from Lidar in order to detect different forest states and structures caused by natural and anthropogenic disturbances.

Under current conditions, we identified the Amazon rainforest as a carbon sink, gaining 0.5 Gt C per year. We also estimated other ecosystem functions like gross primary production (GPP) and woody aboveground net primary production(wANPP), aboveground biomass, basal area and stem density.

We found that successional states play an important role for the relations between productivity and biomass. Forests in early to intermediate successional states are the most productive and carbon use efficiencies are non-linear. Simulated values can be compared to observed values at various spatial resolutions (local to Amazon-wide, multiscale approach). Notably, we found that our results match different observed patterns.

We conclude that forest structure has a substantial impact on productivity and biomass. It is an essential factor that should be taken into account when estimating carbon budgets of the Amazon rainforest.

The East China Sea (ECS) experiences the formation of low-salinity water (LSW) plumes every summer, driven by substantial freshwater input from the Yangtze River. These plumes extend towards Jeju Island and the southern Korean Peninsula, areas rich in aquaculture activity, causing significant damage to fisheries. Monitoring these plumes is critical to mitigating their ecological and economic impacts. Traditional sea surface salinity (SSS) monitoring tools, such as the L-band microwave sensor on the Soil Moisture Active Passive (SMAP) satellite, are limited by low spatial (25 km) and temporal resolution (2–3 days) and inability to capture coastal dynamics.

Given that LSW contains high levels of colored dissolved organic matter (CDOM) closely correlated with salinity, ocean color sensors capable of estimating CDOM are widely used to monitor coastal LSW. In the ECS, the Geostationary Ocean Color Imager (GOCI) has provided essential hourly observations at a 500 m resolution for SSS monitoring. With the end of GOCI’s mission in 2021, its successor, GOCI-II, offers improved spatial resolution (250 m) to enhance coastal monitoring.

This study focuses on ensuring the continuity of SSS monitoring across the two satellite generations (GOCI and GOCI-II) and analyzing the relationship between LSW and essential marine variables, such as sea surface temperature, CDOM, and chlorophyll. This enables the assessment of the impact of LSW on the marine environment.

• This research was supported by the National Research Foundation of Korea (NRF) grant funded by the Ministry of Science and ICT of Korea (MSIT) (RS-2024-00356738).

The pedunculate oak (Quercus robur) is a vital species in Croatian forestry due to its high-quality timber and ecological importance. Located between the Sava and Danube rivers and their tributaries, Spačva forest is among the largest lowland pedunculate oak forests in Europe, spanning over 40,000 hectares. This forest plays a critical role in regional biodiversity and hydrological stability; however, it faces mounting threats from climate change. Increased storm intensity, prolonged droughts, and declining groundwater levels, coupled with lace bug infestations, have all contributed to tree stress and mortality within the forest. Monitoring tree transpiration can serve as an early indicator of such environmental stress, as it reflects water exchange processes between the atmosphere and biosphere.

In this study, we analyzed xylem sap flow as a proxy for transpiration in pedunculate oaks at four sites within Spačva forest, with two of these sites situated at slightly higher elevations. Data from Sentinel-2 satellite imagery, collected during the vegetation periods of 2019 and 2020, were used to assess transpiration rates in relation to several vegetation indices, including EVI, MSI, NDVI, NIRv, and SELI. Among these indices, SELI demonstrated a strong potential to detect seasonal peaks in daily transpiration and accurately capture seasonal dynamics. These findings suggest that Sentinel-2 imagery offers significant potential for monitoring oak forest transpiration patterns and could be instrumental in planning hydrological interventions to mitigate climate change impacts in sensitive forest ecosystems like Spačva.

Forest phenology, i.e. the timing and pattern of natural events, is crucial as it serves as an important indicator of environmental change and helps to assess the impact on the many ecosystem functions of forests. We have analysed a wealth of scientific articles dealing with post-2000 forest phenology using both optical and radar satellite data. The aim of our contribution is to summarize what has been done in the field of forest phenology, highlight areas where further research is needed, and assess how current studies present their results and validate them against ground-truth data. We aim to provide clear directions for future research and to improve the accuracy of using satellite imagery to study forest phenology.

Our contribution shows that satellite-based studies of forest phenology are, firstly, geographically unevenly distributed with notable global and regional imbalances. Second, they focus on temperate and boreal forests, with deciduous forests dominating phenological studies, while mixed and evergreen forests receive less attention. This also reveals a significant gap in tropical forest research. Although tropical forests play a crucial role in climate regulation and biodiversity, they are still underrepresented in phenological studies. Expanding research in these regions is essential for a balanced, global understanding of forest phenology.

The exponential growth of forest phenology studies since 2008 is due to the policy of open access to satellite data, technological advances and data processing platforms such as Google Earth Engine and Copernicus. MODIS remains the most important sensor due to its daily coarse-resolution data, which is ideal for large-scale events. Higher resolution satellites, such as Sentinel-2 and PlanetScope, support finer spatial analysis, but their lower temporal frequency and cost constraints pose a challenge, especially in cloudy regions where radar data, although underutilised, offer the possibility of penetrating clouds (Belda et al., 2020; Kandasamy et al., 2013). Currently, LSP mapping relies heavily on optical sensors to capture vegetation indices that reflect canopy characteristics. NDVI, EVI and EVI2 are the most commonly used vegetation indices in LSP studies. In recent years, radar-based indices have increased, reflecting a shift in phenological research methods. Each index is sensitive to environmental variables such as background noise, which emphasises the need for researchers to choose indices that are appropriate for specific regions and forest types. Combining indices with other variables, such as climate data, increases the accuracy of vegetation condition assessments and ecosystem function analyses.

LSP metrics are extracted by different methods, including threshold-based and inflection point-based approaches (De Beurs and Henebry, 2010; Tian et al., 2021). The choice of method has a significant impact on phenological metrics, with optimal models depending on the region, vegetation type and research objectives. Studies recommend that phenological products include quality assurance data that consider factors such as time of observation, image quality and appropriate model selection to reduce uncertainties in phenological metrics (Radeloff et al., 2024).

Ground-based observations, including citizen science initiatives, phenocams, and flux towers, remain crucial for validating satellite data and improving LSP accuracy, but data quality and detailed documentation (e.g. data acquisition protocols and observation precision) are essential. A significant proportion (25%) of studies still lack ground validation and transparency in terms of uncertainties and validation standards, emphasising the need for better integration. Visual observations, such as those from the USA National Phenology Network, dominate validation efforts, while phenocams provide low-cost, high-resolution data but have limited spatial coverage. Improved synergy between ground-based and satellite-based data, coupled with standardised protocols, will be crucial to advance phenological research and improve large-scale ecosystem monitoring.

To optimise regional LSP studies, researchers should prioritize key tree species that shape local forest dynamics. This focus provides insights into the phenology of dominant species and supports ecosystem-level understanding. Detailed LSP studies can also help to produce accurate species maps, which are essential for monitoring forest biodiversity, estimating biomass and assessing climate impacts. Remote sensing can improve the mapping of tree species by identifying unique phenological signatures under different conditions, reducing reliance on costly field surveys.

Belda, S., Pipia, L., Morcillo-Pallarés, P., Rivera-Caicedo, J.P., Amin, E., De Grave, C., Verrelst, J., 2020. DATimeS: A machine learning time series GUI toolbox for gap-filling and vegetation phenology trends detection. Environmental Modelling & Software 127, 104666. https://doi.org/10.1016/j.envsoft.2020.104666

De Beurs, K.M., Henebry, G.M., 2010. Spatio-temporal statistical methods for modelling land surface phenology, in: Phenological Research: Methods for Environmental and Climate Change Analysis. Springer Link, pp. 177–208. https://doi.org/10.1007/978-90-481-3335-2_9

Kandasamy, S., Baret, F., Verger, A., Neveux, P., Weiss, M., 2013. A comparison of methods for smoothing and gap filling time series of remote sensing observations &ndash; application to MODIS LAI products. Biogeosciences 10, 4055–4071. https://doi.org/10.5194/bg-10-4055-2013

Radeloff, V.C., Roy, D.P., Wulder, M.A., Anderson, M., Cook, B., Crawford, C.J., Friedl, M., Gao, F., Gorelick, N., Hansen, M., Healey, S., Hostert, P., Hulley, G., Huntington, J.L., Johnson, D.M., Neigh, C., Lyapustin, A., Lymburner, L., Pahlevan, N., Pekel, J.-F., Scambos, T.A., Schaaf, C., Strobl, P., Woodcock, C.E., Zhang, H.K., Zhu, Z., 2024. Need and vision for global medium-resolution Landsat and Sentinel-2 data products. Remote Sensing of Environment 300, 113918. https://doi.org/10.1016/j.rse.2023.113918

Tian, F., Cai, Z., Jin, H., Hufkens, K., Scheifinger, H., Tagesson, T., Smets, B., Van Hoolst, R., Bonte, K., Ivits, E., Tong, X., Ardö, J., Eklundh, L., 2021. Calibrating vegetation phenology from Sentinel-2 using eddy covariance, PhenoCam, and PEP725 networks across Europe. Remote Sensing of Environment 260, 112456. https://doi.org/10.1016/j.rse.2021.112456

Tree mortality rates are rising across many regions of the world. Yet the underlying dynamics remain poorly understood due to the complex interplay of abiotic and biotic factors, including global warming, climate extremes, pests, pathogens, and other environmental stressors. Ground-based observations on tree mortality, such as national forest inventories, are often sparse, inconsistent, and lack spatial precision. Earth observations, combined with machine learning, offer a promising pathway for mapping standing dead trees and potentially uncovering the driving forces behind this phenomenon. However, the development of a unified global product for tracking tree mortality patterns is constrained by the lack of comprehensive, georeferenced training data spanning diverse biomes and forest types.

Aerial imagery from drones or airplanes, paired with computer vision methods, provides a powerful tool for high-precision, efficient mapping of standing deadwood on local scales. Data from these local efforts offer valuable training material to develop models based on satellite data, enabling continuous spatial and temporal inference of standing deadwood on a global scale. To harness this potential and advance global understanding of tree mortality patterns, we have developed a dynamic database (https://deadtrees.earth). This platform allows users to 1) upload and download aerial imagery with optional labels of standing deadwood, 2) automatically detect standing dead trees in uploaded imagery using a generic computer vision model for semantic segmentation, and 3) visualize and download spatiotemporal tree mortality products derived from Earth observation data. With contributions from over 150 participants, the database already contains more than 1,500 orthoimages covering more than 300,000 ha from diverse continents and biomes.

With contributions from over 150 participants, the database already contains more than 1,500 orthoimages covering all biomas with approximately 300,000 ha in more than 60 countries, with the highest density of data in Europe and the Americas, emphasizing the need for core contributions from Asia and Africa.

This presentation will provide a comprehensive overview of the deadtrees.earth database, discussing its motivation, current status, and future directions. By integrating Earth observation, machine learning, and ground-based data, this initiative seeks to fill critical knowledge gaps in global tree mortality dynamics and create an accessible, valuable resource for researchers and stakeholders.

Ecosystem structure and structural complexity are crucial for biodiversity, carbon storage, ecosystem resilience and recovery after disturbances. Most large-scale assessments of terrestrial ecosystem change and resilience, however, are based on passively measured indicators of greenness. Spaceborne LiDAR (light detection and ranging) instruments are active measurement devices that provide high-resolution three-dimensionally resolved data on ecosystem structure. Currently, their usage is constrained by short time series and discontinuous spatial coverage. Here, we address this problem by extending existing large-scale LiDAR measurements from GEDI (Global Ecosystem Dynamics Investigation) backward in time using predictive machine learning models based on passive optical satellite data, static location data, and climate variables. We conduct rigorous assessments of prediction accuracy and analyze the footprints of disturbances and ecosystem degradation in the extended time series. This approach allows us to investigate long-term changes and trends in ecosystem structure and provides a method for using recently developed sensors to assess past changes.

Monitoring biodiversity through the integration of optical and in-situ data requires a suite of specifications to deliver good biodiversity metrics and products. Airborne imaging spectroscopy has shown to be effective in monitoring biodiversity and understanding the processes of its change. Yet, novel improvements in airborne imaging spectroscopy sensors in terms of sensor characteristics and the quality of the data delivered hold the promise to enhance our ability to detect, monitor and predict biodiversity processes. Here, we show a first application of the new airborne imaging spectrometer AVIRIS-4 data for mapping and monitoring biodiversity in alpine regions of Switzerland. AVIRIS-4, operated by the Airborne Research Facility for the Earth System (ARES) at the University of Zurich, provides data at 7.5 nm bandwidth across the 380-2490 nm range, thus AVIRIS-4 enables detailed environmental analysis, including assessing biodiversity in grassland ecosystems.

This study presents the preliminary results of an initial quality assessment of AVIRIS-4 data by comparing airborne-derived hyperspectral data with in-situ field measurements aimed at measuring biodiversity. We acquired a cloud-free set of flight lines with a spatial resolution in the lower meter range during the summer of 2024 over the Swiss National Park as well as in-situ data collected from approximately 80 grassland plots where we measured canopy spectral reflectance, leaf optical properties, and biomass. We present a comprehensive workflow for data processing, including atmospheric and bidirectional reflectance distribution function (BRDF) corrections, and evaluate the correlation between hyperspectral imagery and field measurements. These results enhance the understanding of AVIRIS-4's potential for biodiversity monitoring and offer valuable insights for optimizing remote sensing techniques in future conservation efforts.

Phenotypic plasticity is likely to play a crucial role in ensuring the persistence of plant species in a rapidly warming world. While many studies have shown that plastic responses evolve in reaction to environmental heterogeneity, the relative influence of different landscape features, each subjected to varying degrees of human pressures, remains poorly understood. In this study, we use high-resolution (10-meter) remote sensing data combined with data from greenhouse experiments testing thermal responses of European populations of three Hypericum species to assess how compositional and configurational land cover heterogeneity, along with topographic roughness, influence the degree of thermal plasticity. We germinated and cultivated seeds collected from natural habitats and obtained from European managed seeds banks in four temperature treatments within greenhouse compartments and growth chambers. We estimated population-level thermal plasticity in five key life-history traits using Random Regression Mixed Models (RRMMs) and analyzed the effects of landscape features across five spatial scales. Our preliminary results show variation in the importance of different landscape features for different traits and species. Overall, this study highlights the various mechanisms through which human activities can influence the ability of species to respond to climate change and how remote sensed data can be combined with traditional experiments to gauge such patterns.

Plant phenology is increasingly recognized as a critical indicator of ecological processes and responses to environmental change. The advent of remote sensing technologies has enhanced our ability to study phenology over space and time. Still, their temporal and spatial resolution influences their effectiveness in capturing detailed phenological changes in highly heterogeneous ecosystems, such as coastal wetlands. We used Sentinel-2 Enhanced Vegetation Index time series to characterize the main plant phenological types in the Suisun Marsh, California, USA. Our remotely sensed phenological patterns and cluster-based typologies reveal the nuanced interplay between vegetation types, phenology, elevation, and hydrology. The nine phenological clusters were sensitive to elevation and hydrological regimes. Strong inter-cluster variation in landscape phenological metrics—timing and magnitude of greenness—along with varying proportions of vegetation types across clusters suggests that these interacting factors influence seasonal vegetation cycles, indicative of photosynthesis and productivity. Furthermore, our study demonstrates that phenological metrics such as the start, peak, and end of the growing season are effective tools for distinguishing between wetland vegetation types with similar above-ground functions. We highlight the potential of remotely sensed phenology to enhance landscape-scale accounting of ecosystem benefits and identify wetland-upland transition zones. Our findings showed that different vegetation types exhibit similar phenological behavior across the landscapes, likely due to hydrological, microclimatic, and other factors that need further studies. However, these differences might also be affected by the limitation of moderate-resolution multispectral sensors. Hence, further improvements should explore data fusion and higher spectral and/or spatial resolution.

Insect migration is a major natural phenomenon, transferring vast amounts of biomass and energy globally, often spanning intercontinental scales. However, their migratory patterns remain underexplored, despite their substantial ecological impacts. Tracking the movements of migratory insects present unique challenges, mainly due to the multigenerational nature of their migrations, where successive generations may occupy breeding ranges with vastly different ecological conditions. Satellite remote sensing offers a powerful tool for monitoring insect habitats across space and time, as well as to analyze environmental cues that may trigger their migratory behavior. Here, we explore the use of time-series of remote sensing data in dynamic spatio-temporal models to characterize the transient reproductive habitats of migratory insects. Key variables such as the Normalized Difference Vegetation Index (NDVI) for herbivorous insects and the Normalized Difference Water Index (NDWI) for aquatic species, show highly informative to delimit ecological niches supporting immature development. Using these models, we examine the case of the trans-Saharan painted lady butterfly (Vanessa cardui) to: 1) track shifts in ecological niches throughout its annual cycle, indirectly inferring seasonal movements; 2) identify spatial and/or temporal hotspots important for migratory population dynamics; 3) assess insect’s ability to follow “green-waves” and adapt migratory timing to vegetation phenology; 4) link insect demographic fluctuations and outbreaks to anomalies in primary productivity; and 5) infer future trajectories in migratory patterns under global environmental change. Our research underscores the transformative potential of remote sensing - using phenological metrics and vegetation indices- to advance the field of insect migration. Our ultimate goal is to provide a robust framework applicable across migratory species, aiding in the development of conservation strategies and in the prediction, monitoring, and management of migratory insect impacts on ecosystems, agriculture, forestry, and health.

Functional diversity has been recognized as a key driver of ecosystem resilience and resistance, yet our understanding of global patterns of functional diversity is constrained to specific regions or geographically limited datasets. Meanwhile, rapidly growing citizen science initiatives, such as iNaturalist or Pl@ntNet, have generated millions of ground-level species observations across the globe. Despite citizen science species observations being noisy and opportunistically sampled, previous studies have shown that integrating them with large functional trait databases enables the creation of global trait maps with promising accuracy. However, aggregating citizen science data only allows for the generation of relatively sparse and coarse trait maps, e.g. at 0.2 to 2.0 degree spatial resolution.

Here, by using such citizen science data in concert with high-resolution Earth observation data, we extend this approach to model the relationships between functional traits and their structural and environmental determinants, providing global trait maps with globally continuous coverage and high spatial resolution (up to 1km). This fusion of ground-based citizen science and continuous satellite data allows us not only to map more than 20 ecologically relevant traits but also to derive crucial functional diversity metrics at a global scale. These metrics—such as functional richness and evenness—provide new opportunities to explore the role of functional diversity in ecosystem stability, particularly in response to climate extremes associated with climate change.

Our approach presents a scalable framework to advance understanding of plant functional traits and diversity, opening the door to new insights on how ecosystems may respond to an increasingly variable and extreme climate.

Recent advances in remote sensing, including drones, multispectral sensors with high spatial and spectral resolution, and LiDAR, have opened up new possibilities for ecological studies, providing valuable tools for monitoring and understanding ecosystem processes. Promising applications of remote sensing in ecology include the ability to identify the functional traits of plants, which is crucial for understanding community dynamics and assessing the impact of environmental changes on the resilience and functioning of ecosystems. In this study, we utilized multispectral imagery at different spatial and spectral resolutions—gathered by satellites and drones—as well as high-resolution drone LiDAR data to investigate the potential of remote sensing in capturing fine functional characteristics of trees in 100 m² plots. Our analysis was based on an extensive dataset containing precise locations and functional characteristics—morphological, nutritional, and structural—of over 20,000 trees in a temperate forest community (Wythamwoods, UK). Our results indicate that taxonomic and functional diversity (RaoQ) were the biodiversity metrics most effectively explained by remote sensing data. Among the individual functional traits, nutritional traits (e.g., phosphorus and potassium) and structural traits exhibited the highest explanatory power. The importance of predictor variables varied according to the response variable; however, LiDAR-derived metrics, such as Leaf Area Index (LAI) and canopy rugosity, as well as spectral band vegetation indices and texture indices derived from higher spatial and spectral resolution imagery (drone), consistently emerged as the most important predictor. By linking remote sensing data to functional traits at a fine spatial scale, our results emphasise the potential of remote sensing to improve our understanding of plant functional diversity and ecosystem structure, and thus contribute to monitoring ecosystem resilience in response to environmental change at the local scale.

Phytoplankton is an essential component of marine ecosystems, constituting the basis of the marine trophic chain and supporting key biogeochemical processes such as nitrogen fixation, carbon sequestration, remineralization under both oxic and anoxic conditions, and pH regulation. This study focuses on analysing phytoplankton diversity across the Mediterranean Sea relying on both satellite observations and model outputs provided by the Copernicus Marine Service. Namely, the L4 Ocean Colour gap-less product (OCEANCOLOUR_MED_BGC_L4_MY_009_144) and the multi-year Mediterranean Sea physics reanalysis product (MEDSEA_MULTIYEAR_PHY_006_004) are used. Based on chlorophyll concentration values, relative abundances of phytoplankton functional types (PFTs) of interest, i.e., haptophytes, dinoflagellates, diatoms, cryptophytes, prokaryotes and green algae, are derived by applying the algorithm described in [Di Cicco et al., 2017]. Temporal evolution of PFTs in the last 25 years is analysed by dividing the Mediterranean Sea into nine zones according to their level of trophic activity [Basterretxea et al., 2018]. While a general decrease of bulk phytoplankton biomass is reported, the various regions exhibit different trends of the PFTs relative abundances. These are related to key physical variables such as sea surface temperature, salinity, and mixed layer depth. Finally, the impact of the change in PFT distribution on ecosystem functions such as nitrogen fixation, carbon sequestration, or ocean acidification is discussed.

Di Cicco, Sammartino, Marullo, Santoleri, “Regional Empirical Algorithms for an Improved Identification of Phytoplankton Functional Types and Size Classes in the Mediterranean Sea Using Satellite Data”, Frontiers in Marine Science, vol. 4. 2017

Basterretxea, Font-Muñoz, Salgado-Hernanz, Arrieta, Hernández-Carrasco, “Patterns of chlorophyll interannual variability in Mediterranean biogeographical regions”, Remote Sensing of Environment, vol. 215, p. 7-17. 2018.

Multitemporal and multispectral Sentinel-2 (S2) imagery were used to assess the effects of two most widespread invasive trees species in Central Europe, Prunus serotina and Robinia pseudoacacia, on spectral eco-physiological traits of forests in Poland. The effects were analyzed across two forest habitats: nutrient-rich forests dominated by oaks Quercus robur and Q. petraea and nutrient-poor forests dominated by Scots pine Pinus sylvestris. We established 160 study plots (0.05 ha), including 64 plots with P. serotina, 64 with R. pseudoacacia, and 32 control (not-invaded) plots. In each plot, we measured diameter at breast height (DBH) of all invasive trees, and using allometric models we calculated the aboveground biomass of non-native species. From S2 imagery, a set of spectral eco-physiological indices to map the photosynthetic rate, light use efficiency and leaf chlorophyll/carotenoid content was calculated. The monthly differences between not invaded and invaded oaks and Scots pine forests were analyzed using linear mixed models (LMMs), one-way ANOVA, and Estimated Marginal Means. Furthermore, the effects on eco-physiological traits due to the presence of P. serotina and R. pseudocacia were analyzed along the invasion gradient by LMMs. Our results highlighted the effectiveness of the methodology applied on S2 to assess the effects of invasion on spectral eco-physiological traits in oak and Scotes pine forest (marginal R2 range: 0.295-0.808; conditional R2 range: 0.653-0.885). In general, Scots pine forests were more sensitive to invasion with higher impacts during springer and summer months, while in oaks forests the impacts of invasion were observed mostly during springer months. The invaded plots highlighted changes in photosynthetic rate and light use efficiency compared to not invaded plots. Thus, multitemporal, multispectral satellite image analysis is an effective tool to assess the effects of non-native invasive tree species on spectral eco-physiological traits.

Amidst accelerating biodiversity loss and ecosystem degradation, the GEO Indigenous Alliance stands as a transformative force, advocating for the integration of Indigenous knowledge with Earth Observation (EO) technology to safeguard our planet’s biodiversity. In this session, Diana Mastracci, founder of Space4Innovation and international strategic liaison for the GEO Indigenous Alliance, will share insights into how the Alliance fosters collaboration among Indigenous communities, scientists, and policymakers to create a more inclusive and robust approach to biodiversity monitoring and conservation.

This presentation will showcase the Alliance’s pivotal role in elevating Indigenous voices, championing data sovereignty, and co-developing solutions that harmonize traditional ecological knowledge with cutting-edge EO methodologies. Through real-world case studies, attendees will learn how Indigenous perspectives have enriched scientific understanding of ecosystem dynamics and fortified conservation strategies, paving the way for resilient, adaptive policies.

Attendees will leave with a deeper appreciation of the potential unlocked by bridging knowledge systems, underscoring the essential role of Indigenous-led stewardship in protecting biodiversity and building sustainable environmental policies.

We present the roadmap from the conceptualization to the beta-release of the digital platform of the Italian National Biodiversity Future Centre (NBFC), a project in the framework of the National Recovery and Resilience Plan (NRRP). The initial steps involved reviewing the current scientific, technical, and political aspects, as well as the interconnections among major global and European biodiversity platforms designed to tackle the biodiversity crisis. This review aimed to assess options with the highest potential for providing services, data, and models to the scientific community and other stakeholders, ultimately leading to improvements in biodiversity. Following this, we identified key priorities in applied ecology and conservation that need to be addressed to enhance the effectiveness of the Nature Biodiversity Future Center platform.

On-site and online workshops, peer-to-peer discussions, and dedicated questionnaires were utilized to gather information on data, models, projects, and networks (such as LTER) involving all scientists participating in the National Biodiversity Framework Consortium (NBFC) activities. The scientific needs and ideas of the NBFC were thoroughly discussed with CINECA, a center of excellence in the Italian and European ecosystem for supercomputing technologies. Currently, the NBFC digital platform is organized into four thematic areas: (1) digitization of Natural History Collections; (2) molecular biodiversity; (3) biomolecules, biosources, and bioactivity; and (4) biodiversity and ecosystem function (BEF).

In November 2024, an international symposium held in Alghero, Italy, brought together experts from around the world to discuss important aspects of the relationship between Biodiversity and Ecosystem Functions (BEF) in the context of Global Change. The symposium specifically focused on the fourth thematic area of the digital platform, essential biodiversity variables, and how digital platforms, digital twins, and international monitoring networks can help address the challenging NBFC commitment to monitor, conserve, restore, and enhance biodiversity and ecosystem functions in a fast-changing world.

Synthetic Aperture Radar (SAR) data, particularly from Sentinel-1, offer significant potential for high-resolution soil moisture monitoring due to their insensitivity to daylight and atmospheric conditions. However, soil moisture retrieval in forested areas remains challenging with Sentinel-1’s C-band radar, as its wavelength limits vegetation penetration. This study addresses soil moisture estimation within forest ecosystems using Sentinel-1 SAR data, focusing on capturing soil moisture variability under dense vegetation cover. By analyzing long-term time series across various forest types and combining SAR data with in situ soil moisture measurements at different depths, we demonstrate that, despite limited penetration, reflections from vegetation can reveal partial soil moisture variability. This approach highlights the utility of SAR data for monitoring soil-vegetation interactions and contributes to essential biodiversity variables related to ecosystem functions and forest hydrology.

While urban populations grow, cities are ultimately confined in space, needing to accommodate diverse social, ecological, and economic functions. Cities worldwide face the challenge of creating integrated urban environments that balance growth ambitions with new standards for green growth, promoting biodiversity, mitigating climate change, and supporting inclusiveness and quality of life.

Urban Nature-Based Solutions (NBS) offer a multifaceted approach to addressing complex urbanization challenges. As cities grapple with limited space amidst burgeoning populations, NBS emerge as indispensable tools for fostering sustainable development. Monitoring and evaluating the impact and potential of NBS activities are inherently challenging due to the complexity of urban environments and the dynamic nature of these solutions. Herein lies the value of EO technology, offering a bird's-eye view of urban landscapes and facilitating continuous monitoring at various scales. EO enables the systematic collection of high-resolution spatial data, providing insights into vegetation dynamics, land use changes, and environmental conditions over time. EO enables near real-time responsiveness to environmental shifts and evaluation of NBS effectiveness, enhancing the resilience of NBS interventions in the face of urban challenges such as climate change and population growth.

Based on the results of a UNEP funded urban NBS activity, we will illustrate how EO enables near real-time responsiveness to environmental shifts and evaluation of NBS effectiveness, hence enhancing the resilience of NBS interventions in the face of urban challenges such as climate change and population growth. We will shed light on the technology and provide practical use cases from around the world for the applied use of EO to underpin urban green management and planning, emphasizing how modern EO technology can be used to create and maintain an accurate and updated urban information.

This paper examines the integration of indigenous knowledge and community involvement in biodiversity conservation and Nature-Based Solutions (NBS) monitoring and reporting, particularly as a complement to Earth Observation (EO) data across remote surfing communities of Indonesia. Indigenous communities hold vast ecological knowledge rooted in centuries of direct interaction with their natural environment, offering valuable insights for effective biodiversity monitoring and adaptive management practices. Recognizing indigenous knowledge systems and empowering these communities as active participants in data collection, analysis, and interpretation can bridge data gaps and enrich EO datasets with localized, nuanced insights often missing from satellite and remote sensing technologies and increase the uptake and understanding of scientific methodology.

Our study highlights strategies for fostering equitable partnerships with Indonesian indigenous communities to collaboratively develop monitoring frameworks that reflect both traditional and scientific knowledge. These frameworks enable the monitoring of coral reef surf break ecosystems, including biodiversity, species migration, habitat changes, ecosystem health and coastal erosion within the context of traditional coastal and marine practices. By empowering indigenous communities through capacity-building and funding, we can also promote sustainable livelihoods through the development of surf tourism while improving biodiversity outcomes. Moreover, we explore the role of digital platforms, mobile applications, and community-based monitoring tools that facilitate the seamless integration of field observations from indigenous monitors with EO data, enhancing the accuracy and resolution of environmental datasets.

Through case studies and best practices, this paper demonstrates how indigenous knowledge can be systematically incorporated into NBS monitoring and reporting, fostering co-created solutions that align with global biodiversity targets. Leveraging this knowledge base enhances EO data's value by grounding it in field realities, creating a robust, participatory approach to environmental stewardship. Ultimately, integrating indigenous knowledge with EO data advances a more inclusive, comprehensive approach to biodiversity conservation and climate resilience.

Predicting vegetation Ecosystem Functional Properties in different EU ecosystems from space: opportunities and challenges

Gaia Vaglio Laurin¹, Lorenza Nardella¹, Alessandro Serbastiani², Carlo Calfapietra¹, Bartolomeo Ventura³, Dario Papale⁴.

¹ National Research Council, Research Institute on Terrestrial Ecosystems, Montelibretti, Italy

² ENEA Agenzia Nazionale - Centro Ricerche Casaccia, Italy

³ EURAC Research, Bolzano, Italy

Selected Ecosystem Functional Properties, calculated from data collected by 15 flux tower stations of the Integrated Carbon Observation System network in Europe, were linked to several vegetation indices extracted by satellite PRISMA hyperspectral data and Sentinel 2 data. Fifth-teen ICOS stations in five different ecosystems including various forest types, grasslands, and wetlands were considered, together with multitemporal images collected during the vegetation growing period. Several challenging pre-processing steps, for both flux and especially for PRISMA data, were needed prior to test Random Forest regression. Gross Primary Productivity, Net Ecosystem Exchanges, Water Use Efficiency, Light Use Efficiency, and Bowen Ratio were predicted, with results indicating in most cases a very good capacity to predict EFPs from space at high spatial resolution. Additional insights were derived for forest ecosystems alone. The results helps to clarify the vegetation indices and the satellite data having higher prediction power. This research effort shows the potential to upscale the ecosystem functional dynamics derived at flux tower stations to larger extent using with different satellite datasets, providing a contribution to improved functional biodiversity monitoring.

Intertidal mudflats, covering just 0.036% of the ocean's surface, host microphytobenthic biofilms that play an important role in the global carbon cycle, responsible for approximately 500 Mt of gross carbon uptake per annum. Despite their significance, the temporal dynamics of biofilm formation and factors driving carbon capture by mudflats remain poorly understood. Our study focuses on two mudflats in the upper Bay of Fundy in New Brunswick, Canada, known for the world’s highest tides and expansive intertidal zones. We use remote sensing data from three platforms (satellite, drone (UAV), and spectroradiometer) to monitor the microphytobenthos over seasonal and tidal cycles, while bi-weekly surface sediment sampling provides ground-truth data for estimating its biomass, quantified through fluorescence measurements of chlorophyll and phaeophytin, and High-Performance Liquid Chromatography (HPLC) for xanthophylls. Preliminary results show chlorophyll a biomass ranging from 20 to 60 mg m-2 for May to mid-August 2024, and from 20 to 130 mg m-2 for mid-August to October 2024 in the top 2 mm of sediment, with increased patchiness observed in September–October. Eddy-covariance measurements in June 2024 indicated CO2 fluxes varying with tidal state, wind direction, and time of day, with estimated uptake reaching 0.38 mg CO2 m-2 s-1 at midday (for comparison, ~half the average annual uptake observed in daytime tropical forests). We plan to integrate Sentinel-2 satellite data with CO2 flux measurements to link microphytobenthic abundance and distribution to carbon capture at peak sunlight conditions, while accounting for variations in tidal cycles. This research advances knowledge on blue carbon sequestration, thereby contributing to ecological and climate models, and offering practical insights for coastal management, particularly in New Brunswick’s extensive soft-sediment intertidal ecosystems.

Remote sensing of tree diversity is crucial for addressing biodiversity loss. Yet, pixel level approaches have limitations in capturing structural details and species-level variation. We hypothesize that fusing spectral information from Sentinel-2 imagery with high-resolution semantic features from freely available aerial orthophotos can enhance the accuracy of tree diversity assessments. These semantic features —such as canopy edges, textures, and structural patterns— provide unique spatial information that can support regression tasks for estimating tree diversity indices. To test this, we employ a two-stream deep learning architecture trained and validated on more than 50,000 National Forest Inventory (NFI) plots from Spain. One stream processes Sentinel-2 multispectral data to extract spectral attributes, while the other analyzes 25-cm resolution orthophotos from the Spanish National Plan of Aerial Orthophotography (PNOA) to capture detailed semantic features. Our approach estimates tree diversity indices at the patch level (50m x 50m), including species richness, Shannon index, Simpson index, and Pielou’s Evenness, among others, at the national scale. Our preliminary results show significant accuracy improvements for all indices compared to using Sentinel-2 data alone. Furthermore, interpretability methods reveal which features most influence model predictions, offering insights into the ecological drivers of diversity. By integrating both spectral and semantic information, our study present a framework for scalable, patch-level tree diversity assessments, especially valuable in regions where high-resolution imagery is available.

Nowadays two remote sensing techniques allow the realization of 3D forest structure measurements over large areas overcoming spatial and temporal limitations of field inventory plots and terrestrial laser scanning: Lidar (in full-waveform and high-density discrete-return airborne or spaceborne configurations) and Synthetic Aperture Radar (SAR). In particular, for SAR configurations, (Polarimetric) SAR Interferometry ((Pol-)InSAR) [1] and SAR Tomography (TomoSAR) [2] are two techniques that can extract 3D structure information related not only to height, but also to structure intended as the 3D size, location and arrangements of trees, trunks and branches. (Pol-)InSAR has been demonstrated in several experiments for the estimation of forest height and horizontal structure parameters associated e.g. to stand density index especially for high-frequency data [3]. TomoSAR is an imaging technique that reconstructs the full 3D distribution of the radar reflectivity. Despite the lack of a clear physical interpretation of the reconstructed reflectivity and its (ambiguous) dependency on the electromagnetic properties of the forest elements, a framework for qualitative and quantitative forest structure characterization from (low frequency) tomographic SAR measurements has been proposed recently in [4]-[5] in correspondence of structure indices already established in forestry and ecology studies. In this context, the availability of Pol-InSAR and TomoSAR measurements within the BIOMASS mission is a unique opportunity for a low-frequency, spatially continuous, 3D structure characterization at a global scale by exploiting a fully resolved information along the height dimension.

Supported by experimental results from dedicated airborne campaigns and spaceborne acquisitions, this presentation critically reviews and discusses the current understanding and the open questions in (Pol-)InSAR / TomoSAR structure characterization in terms of the ecological significance of the defined indices, their sensitivity to different ecological structure types and gradients as a function of the implemented resolutions, and the robustness to reflectivity variations not relevant to structure (e.g. induced by spatial changes of the dielectric properties of the forest volume caused by rain or temperature gradients). Potentials for characterizing structure changes in time are addressed as well.

References:

[1] K. Papathanassiou, S. Cloude, “Single-baseline polarimetric SAR interferometry,” IEEE Transactions on Geoscience and Remote Sensing, vol. 39, no. 11, pp. 2352-2363, Nov. 2001.

[2] A. Reigber and A. Moreira, "First demonstration of airborne SAR tomography using multibaseline L-band data," IEEE Transactions on Geoscience and Remote Sensing, vol. 38, no. 5, pp. 2142-2152, Sept. 2000

[3] C. Choi, M. Pardini, M. Heym and K. P. Papathanassiou, "Improving Forest Height-To-Biomass Allometry With Structure Information: A Tandem-X Study," IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 14, pp. 10415-10427, 2021.

[4] M. Tello, V. Cazcarra-Bes, M. Pardini and K. Papathanassiou, “Forest Structure Characterization From SAR Tomography at L-Band,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 11, no. 10, pp. 3402-3414, Oct. 2018.

[5] M. Pardini, M. Tello, V. Cazcarra-Bes, K. P. Papathanassiou and I. Hajnsek, “L- and P-Band 3-D SAR Reflectivity Profiles Versus Lidar Waveforms: The AfriSAR Case,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 11, no. 10, pp. 3386-3401, Oct. 2018.

The Great Western Woodlands (GWW), located in south-western Australia, is the largest temperate woodland ecosystem in the world, comprised of a mosaic of mallees, shrublands and grasslands dominated by eucalypt woodland. This region is of significant ecological and conservation importance due to its unique biodiversity, and for being an important sink of carbon. Despite the minimal human intervention in this ecosystem, the GWW faces threats related to climate change, particularly increases in fire frequency. Projected alterations in the disturbance regime raise concerns about possible conversion of obligate-seeder eucalypts woodlands, which are highly sensitive to fire, into base resprouting mallee stands. Such transformation would have important implications for biodiversity, carbon budgets and ecosystem functions. For these reasons, monitoring ecosystem extent in the GWW is highly relevant for informing management strategies and characterizing temporal ecological change.

In this study, we aimed to produce high accuracy multitemporal maps of ecosystems extent for the GWW region using remote sensing imagery, with focus on improving the separation between eucalypts woodland and mallee stands. Whilst some of these vegetation communities were distinguishable using optical imagery alone, subtle differences in vertical structure and growth patterns required the exploration of radar signal responses. As such, we incorporated optical and Synthetic Aperture Radar imagery from different sources in our analysis, to take advantage of spectral and structural differences of our target classes.

We found that optical and SAR data fusion resulted in overall accuracy of over 87%, with both user and producer accuracy for all ecosystem classes over 70%. In this presentation we also discuss the shortcomings and benefits of different methodologies for incorporating multi-sensor Earth Observation imagery for ecosystem classification. Furthermore, we present our approach for tracking disturbance events and correctly assigning ecosystem classes to recently disturbed areas, using CSIRO’s Earth Analytics and Science Innovation (EASI) platform.

The increasing frequency of climatic anomalies, such as extreme drought events and high temperatures, impacts habitat diversity and functioning, driving biodiversity loss. The correlations among satellite-based vegetation indices (e.g. NDVI, EVI, LAI) and climatic data such as drought indices (e.g., SPI and SPEI) can detect the relationship between vegetation functioning and precipitation availability, identifying the spatial and temporal impact of extreme climatic events on specific ecosystems.

As part of the "DigitAP" project, which goals to support the monitoring of Italian protected areas through advanced technological tools, this study aims to provide a service to help local authorities in timely identify the areas most sensitive to climatic anomalies within Italian protected areas. With this aim, a monitoring system combining climate, vegetation indices, and ground-truth data collection will be implemented.

Climatic anomalies were derived from the monthly Standardized Precipitation Evapotranspiration Index (SPEI), obtained from the BIGBANG model at a 1 km resolution, covering the national level from 1952 to 2023. Vegetation indices were derived at different spatial scales from MODIS and Sentinel-2 using the longest available temporal series. Corine Land Cover (CLC) products were used to assess the temporal distribution of ecosystems and discriminate ecosystem types. The significance of the correlations between climatic data and vegetation indices, as well as the time lag between critical events at different integration times (e.g. 3,6,12 months), was evaluated.

The high heterogeneity of Italian protected areas resulted in different distribution patterns in both climatic and vegetation indices. In turn, each ecosystem responds to different thresholds in terms of event’s intensity and duration, showing different correlations dynamics between the analyzed indices.

These analyses show the potential of such a service to actively monitor the impact of critical events on ecosystems and support local authorities in the management of protected areas.

Mountain ecosystems are particularly vulnerable to global change, including rising temperatures, deforestation, and loss of biodiversity. Understanding the relationship between plant diversity and ecosystem stability is a complex challenge, as stability depends not only on species composition but also on environmental factors. In this study, we examine how gradients of environmental heterogeneity and plant taxonomic and phylogenetic diversity, generated by the complex topography of mountain ecosystems, affect the spatio-temporal stability of ecosystems in the Mediterranean Andes of central Chile. Due to its high plant diversity and remarkable climatic and topographic variation, this is an ideal system to assess the extent to which plant diversity mediates the effects of environmental heterogeneity on ecosystem stability across spatio-temporal and ecological scales. Using a fractal sampling design, we analyzed the direct and indirect effects of topography on plant taxonomic and phylogenetic diversity in relation to the temporal stability of vegetation productivity. Stability was calculated by the normalized difference vegetation index (kNDVI) using Sentinel-2 satellite data over six years (2017-2024), generating the temporal series D-index, while topographic variables were derived from a digital elevation model (DEM; 30 m resolution) of the Advanced Land Observing Satellite (ALOS-PALSAR) L-band synthetic aperture radar instrument. Our results show that the spatio-temporal stability of ecosystems is negatively influenced by lower species turnover, suggesting that dominant species play a crucial role in community temporal stability due to their functional traits. Although environmental variability promotes species turnover in different habitats, we found that phylogenetic diversity has no significant relationship with ecosystem stability. This highlights that ecosystem functionality is more closely related to functional diversity and community structure than to evolutionary proximity among species. We recommend that future research integrate measures of functional diversity and community structure to better understand the interaction between abiotic factors and spatio-temporal stability, and to support the design of conservation strategies based on the interaction between the environment and community diversity structure.

Biodiversity monitoring is essential for ecosystem conservation and management, yet high costs and labour intensity often limit traditional field methods. Earth observation is increasingly looked at as a key tool for monitoring ecosystem biodiversity, enabling free access to high-resolution, uniform, periodic data with improved imagery processing possibilities. Among the potential approaches to relate the remotely sensed data to ground biodiversity, the Spectral Variation Hypothesis (SVH) assumes a positive correlation between spectral diversity from optical remote sensing and biodiversity based on the premise that areas with high spectral heterogeneity contain more ecological niches. Over the past two decades, the SVH has been rigorously tested across various ecosystems using diverse remote sensing data, techniques to analyze them, and addressing different ecological questions, revealing its potential and limitations. Through a systematic review of more than 130 publications, we provide a comprehensive and up-to-date state-of-the-art on the SVH and discuss the advances and uncertainties in using spectral diversity for biodiversity monitoring. In particular, we provide an overview of the different ecosystems, remote sensing data characteristics (i.e., spatial, spectral and temporal resolution), metrics, tools, and applications for which the SVH was tested and the strength of the association between spectral diversity and biodiversity metrics reported by each publication. This study is meant as a guideline for researchers navigating the complexities of applying the SVH, offering insights into the current state of knowledge and future research possibilities in biodiversity estimation by remote sensing data.

Terrestrial ecosystems are cardinal pieces for biodiversity, and their qualitative and quantitative estimation are crucial for its conservation. Earth Observation (EO) data offer new opportunities for ecological sciences, and their monitoring capacity opened the way to the assessment of critical processes in terrestrial ecosystems. This research shows the results of a spatially explicit forest ecosystem mapping in Italy that has been employed to estimate the amount of forest identification in burned areas with a special focus on protected areas. The procedure integrates forest habitat data in Italy from the European Vegetation Archive (EVA), with Sentinel-2 imagery processing (vegetation indices time series, observations of spectral bands, and spectral indices) and environmental data variables (i.e., climatic and topographic), to feed a Random Forest (RF) classifier. The obtained results classify four forest ecosystems according to the EUNIS legend. EUNIS (European Nature Information System) is a system tool for habitat identification and assessment. The classification model predicted 4 forest classes at II and III levels: broadleaved deciduous (T1), broadleaved evergreen (T2), needleleaved evergreen (T3) and needleleaved evergreen forest (T34) achieving an overall accuracy of 90%. Successively, the forest map has been employed to estimate the amount of the different forest classes present in all the burned areas detected by the European Forest Fire Information System (EFFIS) from 2019 to 2024 inside and outside the Italian protected areas systems. The estimates obtained could be used for evaluating the impact of wildfires on forest distribution and supporting ecosystem conservation efforts through the detection of disturbances and consequential forest ecosystem changes in space and time.

Species distribution models (SDMs) estimate species distributions by analyzing the relationships between species occurrences and environmental variables. Their efficacy largely depends on the selection of ecologically relevant predictors, and remote sensing (RS) data have been shown to enhance SDM performance. However, RS imagery reflects temporal changes in vegetation and environmental conditions, resulting in dynamic predictors that vary over time. Despite this, the impact of seasonality on RS predictors is often overlooked. This study aimed to assess how seasonality in RS predictors affects SDM performance for bird species.

The study was conducted across the Czech Republic, using presence-absence data from the Breeding Bird Survey (2018–2021), covering 147 survey squares and 104 bird species. We used Sentinel-2 satellite imagery to derive monthly and full-season composites of vegetation indices and reflectance bands from March to September (hereafter "periods"). Additionally, we included bioclimatic variables, topography, and vegetation structure as predictors. SDMs were constructed using Lasso-regularized logistic regression, and model performance was assessed with AUC and R². Linear mixed-effects models were employed to evaluate model performance, temporal prediction stability, and predictor importance stability across all species.

Our results show that model performance depended on the period from which the predictors were derived, and this varied significantly among species. This variation can be partially attributed to species' habitat preferences and prevalence. Differences in model performance across periods aligned with shifts in predictor importance, as seasonal changes in vegetation and habitat conditions caused different RS predictors to become significant throughout the year.

In conclusion, seasonal changes in vegetation, as reflected in the temporal variability of RS predictors, significantly affect SDM performance and predictor selection. Although species’ ecological characteristics played a role, the effects remained species-dependent, making it difficult to develop universal recommendations. Nevertheless, accounting for seasonal variations in RS predictors could enhance model accuracy across species.

Since Alexander von Humboldt's discovery of condensed life zones on tropical mountains, these areas have attracted significant attention from biologists, as they are believed to hold vital clues about life-forming processes. However, they remain one of the most enigmatic subjects in natural sciences. This study identifies the causal mechanisms driving plant ecology and evolution along the elevational gradient of tropical mountains. By utilizing satellite remote sensing data of plant pigment traits, moisture levels, and surface temperature, analyzed across five mega-diverse tropical mountain regions in combination with field data, key ecological insights were uncovered.

The findings reveal that ancient clade species are filtered out below the condensation zone, a major ecological turnover point that suggests the world's phylogenetically richest terrestrial plant edge, driven by the Mass Elevation Effect. Another significant edge corresponds to the ever-wet zone, the habitat of bryophytes. Dendrograms of species traits and phylograms exhibit similar structures, demonstrating that plant species and communities exhibit niche conservatism, reflecting the environmental conditions of their initial evolution.

The study elucidates the traits of major forest and plant communities, explaining the soil-vegetation interactions that determine their locations and evolutionary dynamics. Using an unprecedented volume of data, the research tests several macro-ecological and remote sensing hypotheses through essential or potential Earth Observation-derived Essential Biodiversity Variables (EBVs) from Sentinel 1-2 and Landsat data. The extensive dataset allowed for the identification of causal mechanisms influencing plant physiology and morphology along the elevational gradient, and highlighted major clades such as angiosperms, gymnosperms, ferns, epiphytes, orchids, and bryophytes. Additionally, the study provides new insights into the Mass Elevation Effect, the mid-elevation species hump, niche conservatism, cloud forests, speciation, species cradles and museums, as well as the Spectral Variability Hypothesis.

Marine biodiversity, especially submerged aquatic vegetation (SAV) like seagrass, is increasingly prioritized on the international biodiversity agenda, recognized now as a distinct Essential Biodiversity Variables (EBV’s). Satellite Remote Sensing (SRS) offers crucial tools for assessing SAV; however, the presence of phytoplankton communities, dissolved or suspended matter, and water column effects complicate remote sensing applications in marine ecosystems. Currently, no effective mid-resolution multispectral index exists to reliably isolate photosynthetic components in the marine environment, particularly in inshore ecosystems. Here, I present a novel Marine Photosynthesis Index (MPI) specifically designed to penetrate deep into the water column while capturing high variability in photosynthetic activity. The MPI leverages three spectral bands within the visible light spectrum (450–675 nm), optimized for mapping macrophytes, and demonstrates strong sensitivity to photosynthetic activity from phytoplankton—the foundational level of the marine food web. Tested under estuarine and offshore conditions in Denmark and Sweden using radiometrically, sun-glint, and atmospherically corrected Landsat OLI data, the MPI significantly outperforms traditionally employed indices for SAV mapping. Beyond this, the MPI effectively differentiates photosynthetic activity between algal and plant SAV, with high responsiveness to substrate variations on both soft and hard bottoms. Additionally, it captures early stages of phytoplankton presence, including pre-bloom upwelling events in the visible water column. The MPI’s robust performance across deep water column penetration, sensitivity to macrophyte and phytoplankton dynamics, and resistance to noise, phenology effects, and seasonal variability, was further enhanced with multitemporal analysis. This capability makes MPI a promising SRS index for continuous monitoring and habitat mapping in coastal marine ecosystems, addressing a key need for effective inshore marine ecosystem assessment.

Mapping the spatial distribution of biodiversity is crucial for prioritising and optimising conservation and restoration efforts to mitigate ongoing biodiversity loss. Satellite-based remote sensing is the most accessible method for detecting the spatial patterns of ecosystem characteristics including biodiversity over large extents, but despite active research, relationships between spectral signatures and on-the-ground vegetation diversity patterns remain contested. Specifically, high-resolution maps of Arctic and sub-Arctic biodiversity are lacking. Thus, using machine learning methods, we examine the relationships between (1) spectral diversity metrics, as well as other spectral indices and traits, derived from Sentinel-2 and WorldView-3 satellite images, and (2) taxonomic, functional, and phylogenetic diversity, and indicator-based biodiversity relevance, of plant communities across a northern boreal landscape spanning ca. 160 km2. Relying on a survey of over 1800 1-m2 vegetation plots, we address the validity of the spectral variability hypothesis in peatlands, boreal forests and oroarctic tundra and assess the abilities of multispectral satellite sensors to predict diversity metrics across the whole northern boreal terrestrial landscape. Our tentative results indicate that while there are correlations between spectral and other diversity metrics, the strengths of these relationships vary across different ecosystems and different metrics. Thus, models for estimating on-the-ground diversity should address different dimensions of diversity and different ecosystem types separately.

Understanding vegetation dynamics in alpine protected areas is essential for assessing the impacts of climate change and land use. This study employs a comprehensive remote sensing approach utilizing Landsat 4–9 time series data, pre-existing park maps, and auxiliary datasets to monitor vegetation changes in an alpine protected area. Initially, terrain correction was applied to all satellite images to mitigate topographic distortions. A best available pixel (BAP) technique was then used to construct cloud-free annual composite images for both the growing and senescence seasons. Through statistical tests, an optimal combination of predictors—including spectral bands, vegetation indices, and topographic variables—was selected to enhance classification accuracy. Training pixels were extracted from the pre-existing park mapping using a z-statistic approach to ensure statistical representativeness. Eight land cover classes were, then, classified using a Random Forest approach. Post-processing involved applying time series-based rules to refine classification results. Validation against an independent dataset derived from historical orthophotos demonstrated high accuracy, with Kappa coefficient values ranging from 0.94 to 0.98 and overall accuracy between 0.95 and 0.99. Change analysis identified stable pure pixels, mixed pixels, and pixels exhibiting transitions between land cover classes. The results revealed vegetation change trends globally and within specific sub-areas of the park. This methodology provides valuable insights into vegetation dynamics influenced by climate and land use changes, offering a robust framework for long-term ecological monitoring in alpine and subalpine environments.

Forest structure is the result of forest dynamics and biophysical processes that affect their function and diversity. It can be understood as the arrangement of trees and their components in space, but also as the 3D distribution of biomass [1]. The challenge remains in the definition of 3D forest structure optimized for remote sensing measurements. In this sense, this contribution aims at establishing a framework for the joint exploitation of two remote sensing techniques known for their sensitivity to 3D forest structure and dynamics: LiDAR and SAR data.

LiDAR sensors provide high resolution but discrete measurements of vegetation reflectance profiles (i.e. waveforms) acquired in a nadir-looking geometry. SAR systems, however, provide lower (though still high) resolution, continuous measurements in a side-looking geometry that allows large-scale coverage and short revisit times. They measure interferometric coherences (InSAR) and radar reflectivity profiles (TomoSAR) related to the physical vegetation structure. The combination of LiDAR and SAR data requires a physical or statistical link between them at different scales and spatial resolutions [2].

Here, different applications and methods aiming at characterizing forest structure at different scales by exploiting the synergies and complementarities of these two types of information are presented and discussed. The need for spatial correlation between vertical reflectivity profiles becomes crucial to capture structural heterogeneity present in disturbed forests. Natural growth versus logging or fire forest scenarios can be simulated with prognostic ecosystem models, e.g. FORMIND [3], and evaluated through multi-scale analysis e.g. by using a wavelet frame [4] with X-band InSAR data. The sensitivity of both LiDAR and SAR data to forest structure has also been proven by using structural horizontal and vertical indices derived from correlating vertical reflectivity profiles [5]. Using LiDAR GEDI waveforms in combination with TanDEM-X interferometric coherence allows enhanced large-scale forest height estimation [6], which can be then used to analyze relative height changes of different temporal periods. At last, GEDI waveforms have proven suitable for the generation of a basis representative of forest structure information that allows the reconstruction of X-band reflectivity profiles [7].

[1] T. A. Spies, P. A. Stine, R. A. Gravenmier, J. W. Long, M. J. Reilly, “Synthesis of science to inform land management within the Northwest Forest Plan area,” Gen. Tech. Rep. PNW-GTR-966, Portland, OR: U.S. Department of Agriculture, Forest Service, Pacific Northwest Research Station. 1020, p. 3 vol., 2018, DOI: 10.2737/PNW-GTR-966.

[2] M. Pardini, J. Armston, W. Qi, S. K. Lee, M. Tello, V. Cazcarra-Bes, C. Choi, K. P. Papathanassiou, R. O. Dubayah, L. E. Fatoyinbo, “Early Lessons on Combining Lidar and Multi-baseline SAR Measurements for Forest Structure Characterization”, Surveys in Geophysics, vol. 40, no. 4, pp. 803–837, 2019, DOI: 10.1007/S10712-019-09553-9/TABLES/2.

[3] R. Fischer, F. Bohn, M. Dantas de Paula, C. Dislich, J. Groeneveld, A. G. Gutiérrez, M. Kazmierczak, N. Knapp, S. Lehmann, S. Paulick, S. Pütz, E. Rödig, F. Taubert, P. Köhler, A. Huth, “Lessons learned from applying a forest gap model to understand ecosystem and carbon dynamics of complex tropical forests”, Ecological Modelling, vol. 326, pp. 124–133, 2016, DOI: 10.1016/j.ecolmodel.2015.11.018.

[4] L. Albrecht, A. Huth, R. Fischer, K. Papathanassiou, O. Antropov and L. Lehnert, “Estimating forest structure change by means of wavelet statistics using TanDEM-X datasets”, in Proceedings of the European Conference on Synthetic Aperture Radar, EUSAR, pp. 658-662, VDE, April 2024, Munich, Germany.

[5] M. Tello, V. Cazcarra-Bes, M. Pardini and K. Papathanassiou, “Forest Structure Characterization from SAR Tomography at L-Band,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 11, no. 10, pp. 3402-3414, Oct. 2018, DOI: 10.1109/JSTARS.2018.2859050.

[6] C. Choi, M. Pardini, J. Armston, K. Papathanassiou, “Forest Biomass Mapping Using Continuous InSAR and Discrete Waveform Lidar Measurements: A TanDEM-X / GEDI Test Study”, in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 16, pp. 7675-7689, 2023, DOI: 10.1109/JSTARS.2023.3302026.

[7] R. Guliaev, M. Pardini, K. Papathanassiou, “Forest 3D Radar Reflectivity Reconstruction at X-Band Using a Lidar Derived Polarimetric Coherence Tomography Basis”, Remote Sensing, vol. 16, no. 2146, 2024, DOI: 10.3390/rs16122146.

The insurance hypothesis suggests that there is an urgent need to create biodiverse forests to effectively manage the rising threat from climate extremes such as drought. However, previous research comparing tree species mixtures and monocultures has shown that species mixing does not necessarily result in higher drought resilience. Instead, forest 3D structure has been suggested to play an important and overlooked role in shaping how forests respond to drought. Here, using National LiDAR datasets and Sentinel-2 time series, we quantify the structure of forests and woodlands in England and Wales and their response to recent drought events. We investigate how the relationship between structure and resilience varies between broadleaf, conifer, and mixed forests, and present a national assessment of drought risk based on forest structure. Drawing from our preliminary findings, we explore whether diversifying forest structure could be a promising strategy for sustainable, climate-smart forest management.

Abiotic conditions strongly shape population and community dynamics across the world’s forest biomes. Thus, ecosystem function at the transitional zones of forests, the edge of a biome’s climate space, should be less resilient to ongoing environmental change. Those places may have a decreased recovering ability and may thus be more vulnerable to shifts in forest communities. Evidence for this vulnerability comes mostly from experimental studies and biogeographical observations. We still lack an understanding of whether the vulnerability at the forest transitional zone is related to their resilience at large scale. Understanding the dynamics of those systems is key for protecting and restoring them. Here, we assess globally the resilience patterns across forest biomes and test whether resilience decreases towards the edge of their climate space. We measure resilience using detrended and deseasonalised lag-1 temporal autocorrelation and variance in remotely sensed estimates of net primary productivity from 2001 to 2022. Our preliminary results indicate that especially in boreal, temperate broadleaf and tropical moist forests resilience decreases towards the biome’s edge. In boreal and temperate forests this pattern is strongly driven by temperature constrains at the extreme hot and cold edges. In tropical moist forests the extreme hot edge of the biome’s climate space appears to have a strong effect on the resilience decline at the biome’s transitional zone. Our findings offer a comprehensive view of ecosystem resilience at transitional hot and cold edges, with divergent patterns across the world’s forest biomes. This framework provides a powerful backdrop for predicting spatiotemporal shifts in global forest communities to ongoing environmental change.

Wetlands are dynamic ecosystems essential for biodiversity conservation. Wetland classification traditionally relies on two primary approaches: the floristic and hydrogeomorphic (HGM) methods, which are often applied in isolation. The floristic approach emphasizes plant diversity and composition, while the HGM approach focuses on hydrological and geomorphological characteristics. While tracking changes in wetland vegetation from space has become increasingly feasible with advances in satellite-based remote sensing, vegetation alone may not fully capture wetland biodiversity. The hydrogeomorphic methods provide an additional perspective by considering hydrological and geomorphological factors that shape species distribution and ecosystem processes. Given the distinct focuses of each method, it is unclear whether either approach, when used alone, sufficiently captures the full range of ecosystem functional groups (EFGs) necessary to reflect wetland ecosystem functionality.

This study aims to compare the effectiveness of both classification methods by applying them to the same wetland regions, assessing which functional groups are captured by each approach, identifying any critical groups that may be overlooked, and exploring the potential benefits of an integrated classification system for enhanced biodiversity monitoring and conservation. Our findings will highlight the limitations and strengths of each classification system in capturing the full spectrum of biodiversity, offering a foundation for more nuanced wetland monitoring.

This comparative analysis provides valuable insights for global frameworks such as the Global Biodiversity Framework (GBF) and the Convention on Biological Diversity (CBD) by identifying which classification approach or combination of approaches most effectively supports biodiversity monitoring and reporting. These insights will enable more comprehensive and informed recommendations for global wetland conservation efforts, ensuring that reporting captures both ecological diversity and the functional roles that wetlands play in supporting biodiversity.

Research into extreme climate events (ECEs) in the ocean has primarily focused on abiotic parameters, with less attention on biogeochemical properties, despite their significant impact on marine ecosystem functioning and services. In particular, the occurrence of extreme chlorophyll-a values, measured from satellite platforms for over two decades, reflects the occurrence of intense phytoplankton blooms that may sometimes entail adverse events such as eutrophication, toxic events produced by harmful algae blooms (HABs), or changes in the natural phytoplankton dynamics and phenology.

This study presents two novel extreme indices, estimated from the satellite MODIS-AQUA v2018 reprocessed dataset for the period 2003-2021, for all European seas. These two indices combine the 90th percentile (P90) and the monthly 90th percentiles (mP90). The "Extreme Highest" (EH) exceedances index (greater than P90 and mP90) accounts for the extreme observations predominantly produced during the primary interannual spring growing season, while the "Extreme Anomalous" (EA) exceedances index (greater than mP90 and lower than P90) encompasses the extreme chlorophyll observations during periods of low phytoplankton growth. The latter reflect a range of extreme events, including unexpected episodic anomalous blooms, extreme values occurring during the autumn secondary seasonal bloom, and extremes registered outside of the anticipated timing of the spring season.

The statistics and maps of these indices over the European seas reveal that EH and EA have distinct (almost complementary) seasonal and spatial distribution: EH prevail in mesotrophic and euphotic waters during the main interannual bloom season whilst EA are more abundant in oligotrophic waters out of the main seasonal bloom. Significant increasing and decreasing trends have been estimated in different European regions, reflecting different climate-driven physical and ecological changes. While these results are encouraging, further work is required to account for their uncertainties, mostly related to data representativeness and the performance of the chlorophyll-a estimation algorithms.

The European Space Agency (ESA) is committed to reducing its environmental impact as a key player in the space sector and is contributing to the sustainable development of the society. ESA’s Green Agenda proposes a holistic approach to tackle sustainability matters at ESA and in the space sector, considering, on one hand, the great benefit ESA programmes bring to the sustainable development of the society, and, on another hand, the measurement and mitigation of its own environmental footprint. While climate change has been a central focus of our environmental sustainability efforts, Climate and Sustainability Office aims to enlarge EGA’s scope to other planetary boundaries for our assessments. To drive meaningful environmental progress, we decided to consider second most critical boundary, biosphere integrity.

In collaboration with scientists from the Wild Business at the University of Oxford, our team is expanding its focus to assess ESA’s environmental impact by starting to analyse the impact on biodiversity, currently the second most affected planetary boundary. This involves evaluating factors such as changes in endangered species populations and the restoration of habitats like forests, grasslands, and wetlands. For large organizations like ESA, it is crucial to identify which activities have the greatest impact on biodiversity so that we can mitigate these effects in the future.

As a starting point, we are conducting a pilot biodiversity assessment focused on the Scope 1 and Scope 2 impacts of one ESA site and one ESA project. This initial study allows us to evaluate the space sector's ability not only to contribute to biodiversity monitoring but also to assess and potentially mitigate its own broader environmental impacts. By identifying best practices in this pilot, we aim to inform the future assessment of Scope 3 activities, address gaps in currently developed methodology, and lay the groundwork for broader, more comprehensive biodiversity study that would also cover downstream applications.

Riparian forests are crucial biodiversity hotspots, providing habitats for a wide range of bird species. In this study, we explored the relationship between bird biodiversity and habitat structure within four riparian biotopes in South Tyrol (Italy). These biotopes have been designated as important areas due to their high avian diversity. To investigate the structural characteristics of these forests and their influence on bird populations, we combined high-resolution LiDAR data and multispectral Sentinel-2 imagery to extract detailed information on vegetation structure, canopy complexity, and phenological changes. Bird data were collected using acoustic loggers strategically placed across the study areas, capturing a comprehensive set of avian soundscapes throughout the seasons. We utilized buffers of varying sizes (10m, 30m, 50m, 70m, and 90m) around the loggers to extract structural vegetation metrics and spectral information, helping us determine the spatial extent at which habitat variables most strongly correlate with biodiversity patterns. By integrating these datasets, we analyzed how variations in habitat structure and phenology influence bird species richness. Our findings provide insights into how forest management and conservation efforts can enhance biodiversity within these sensitive riparian ecosystems and help guide conservation strategies for maintaining biodiversity and habitat quality in these riparian forests.

Pigments provide helpful information for assessing health and functioning of marine ecosystems. Accurate phytoplankton pigment measurements in fact allow for the evaluation of total phytoplankton biomass and functional diversity, contributing to the understanding of ecosystem processes and diversity changes. This research presents a novel machine learning-based approach to retrieve pigments from multispectral radiometry data developed relying on an in-situ dataset of concurrent radiometric and High-Performance Liquid Chromatography (HPLC) measurements collected in the Mediterranean Sea and the Black Sea between 2014 and 2022.

Based on the in-situ dataset, a Random Forest algorithm has been trained, tested and cross-validated. Predictors preprocessing included logarithm transformations of both input and output data, as well as scaling and PCA transformations. The core model framework employs cross-validation to evaluate performance, balancing the model's sensitivity to low pigment values and minimizing the risk of overfitting. According to the cross-validation, the model retrieves pigments with a relative error lower than 45% and reaches, on average, an r2 metric of 0.6. While the nominal model is optimized for the Copernicus Sentinel 3 Ocean and Land Colour Instrument (OLCI) using 13 bands, another model has been trained for legacy wavelengths (5 bands) to analyze temporal trends.

The study, developed within the framework of the Biodiversa+ PETRI-MED project, advances the use of diagnostic pigment analysis (DPA) for inferring Phytoplankton Functional Types (PFTs) from remote sensing data aiming at contributing to ecosystem health monitoring, restoration and biodiversity conservation. The integration of machine learning with open radiometry datasets offers a scalable solution for monitoring biodiversity indicators from space.

Future work will involve integrating additional environmental variables (e.g., temperature, salinity, nutrients, and turbulence indicators) to enhance model accuracy.

Fast and potentially irreversible changes in tropical regions due to climate and anthropogenic changes threaten the persistence of these ecosystems of global significance. Tropical ecosystems hold the highest biodiversity and provide some of the largest rates of ecosystem functioning, contribute substantially for the functioning of biogeochemical cycles, water and carbon cycle as well as contributing to regulating Earth’s energy balance. Moreover, tropical systems support an amazing cultural diversity with a mixture of indigenous, traditional, community and other governance structures, and provide fundamental ecosystem services, economic benefits and social processes that scale from local to global scales. Yet, the same interactions that maintain the social-ecological systems that developed over centuries in tropical ecosystems have been seldom studied and are faced by a set of pressures that may destabilize or lead to potential system collapse. Within PANGEA - The PAN tropical investigation of bioGeochemistry and Ecological Adaptation (PANGEA): Scoping a NASA-Sponsored Field Campaign – we examined and developed a set of outstanding questions on the processes that maintain SES resilience in tropical ecosystems and how to study them using remote sensing capacities. Here we present the process we undertook in PANGEA, and which were the set of questions that were prioritized. We expect that through addressing these questions we move beyond and are able to understand the drivers and processes of biodiversity changes in tropical regions globally.

Biodiversity loss and climate change pose significant threats to human existence on Earth. Through the Natural Climate Protection Action Programme (ANK), the German government seeks to address both natural climate protection and the enhancement of Germany’s ecosystems with 69 measures across ten key action areas (e.g. moors, wilderness and protected areas, forest ecosystems, oceans and coasts, urban and transport areas, rivers, floodplains and lakes).

To assess the effectiveness of the ANK in biodiversity protection, standardised, long-term biodiversity data must be collected and analysed from both within and outside of ANK project areas. For this purpose, the applicability of remote sensing-based methods in combination with field monitoring data, is being evaluated. A standardised protocol including computational routines for recording, classifying and assessing selected biodiversity parameters in ANK areas using remote sensing technologies is being developed, tested and applied for biodiversity monitoring in relevant regions. The goal is to enable regular and long-term, and (partially) automated assessments of biodiversity changes at reasonable costs, using this evaluation protocol. Over time, this monitoring should also support other existing nationwide biodiversity monitoring programmes.

Here, we provide an overview of the recently initiated project, which focuses in particular on the opportunities and limitations of various remote sensing-based methods for conducting large-scale to nationwide biodiversity and habitat parameter surveys across diverse landscapes with relatively high temporal resolution. Key biodiversity parameters for the project, which will be used to describe the long-term effects of ANK measures on biodiversity, include aspects such as the diversity, heterogeneity, and development of habitat types and vegetation structures.

Since biodiversity changes due to ANK measures may be subtle, slow, complex, or unforeseen, long-term monitoring may present unique challenges for satellite-based monitoring approaches.

Many models and metrics in remote sensing biodiversity research draw on the existence of large optical datasets. Acquiring such datasets however can be a complicated and difficult task.
This paper looks into using a class of generative models called Denoising Diffusion Models to create and augment optical satellite datasets. Aggregating a dataset for a specific domain can be a difficult task for some regions given satellite fly-by times and environmental factors such as cloud probability, and providing an unlimited amount of artificial data can significantly increase efficiency and robustness of a training process by the mitigation of biases due to unavailability of data. A good generative model can further be used to create datasets for specific tasks and objects rather than geographical regions, interesting use cases for instance being the observation of wildfires or fisheries. Finally, creating artificial datasets could also immensely decrease the effort needed for classification tasks, a common method suggests pretraining models on artificially created classified samples, refining the training on a small number of manually annotated samples later on.
In this paper, we study which biomes can be realistically synthesised using our model and if we can impaint existing data with objects of scientific interest such as fisheries or wildfires.
We validate our results using statistical measurements such as the Fréchet inception distance (FID) but furthermore also measure the usability of our datasets by employing comparatively them in real-life scenarios.

Invasive aquatic plants, or macrophytes, are a threat to shallow aquatic ecosystems by outcompeting native species and causing considerable ecological and economic harm. This study examines two widely distributed species in the Northern Hemisphere: Nelumbo nucifera (sacred lotus, native to East Asia) and Ludwigia hexapetala (water primrose, native to Central and South America), comparing their phenological traits and productivity across different environmental gradients: native vs. non-native ranges and different climatic regions.

Sentinel-2 satellite data covering years from 2017 to 2022 were used to generate time series for Water Adjusted Vegetation Index (WAVI), a proxy for canopy density and biomass, at seven study sites: Mantua lakes and Lake Varese (humid subtropical climate, non-native range for both species), Lake Fangzheng, Lake Bayangdian, and Lake Xuanwu (respectively humid continental, cold semi-arid, and humid subtropical climate, native range for N. nucifera), Lake Grand-Lieu and Santa Rosa Lagoon (respectively temperate oceanic and warm-summer Mediterranean climate, non-native range for L. hexapetala). Seasonal dynamics parameters (phenological metrics and productivity) were extracted from WAVI time series, and their meteo-climatic and environmental drivers were analysed using parametric models (GAMs).

The results indicate that N. nucifera exhibits higher productivity in non-native sites compared to the native ones, while in the subtropical native sites, the growing season starts earlier than in the non-native sites. For L. hexapetala, meteo-climatic factors were found to be the main drivers of its phenology, especially temperature and solar radiation.

As this approach can be easily extended in terms of spatio-temporal scales and to other macrophyte species, using operational data and available archives, it can benefit studies on the variability of the eco-physiological characteristics of invasive macrophyte species under climate change scenarios that may guide the management and restoration of aquatic ecosystems.

The international consensus on the urgent necessity to act to protect a vulnerable environment and endangered biodiversity raises key challenges, including the need to improve and accelerate estimating carbon stocks and changes in coastal ecosystems on a global scale. Remote sensing methods, combined with ground truthing and modelling, are essential for addressing this challenge cost-effectively.

The ESA Coastal Blue Carbon project is an unprecedented effort to review, assess, and attempt to provide key elements for the sustainable management of Blue Carbon Ecosystems (BCEs) through diverse case studies. Over two years, a multidisciplinary consortium is investigating the mangrove, seagrass, and tidal salt marsh BCEs in France, Canada, Spain and French Guiana. The project aims to develop innovative tools and methods based on Earth Observation (EO) to estimate and monitor changes in carbon stocks, and brings together a community of end-users, to ensure the tools meet the operational needs, including:

- Conservation stakeholders aiming to enhance the impact of their actions.

- Decision-makers looking to integrate blue carbon into national carbon accounting and set ambitious mitigation targets.

- The financial sector seeking reliable blue carbon investment opportunities.

Our rationale is to capitalise on existing data and multi-scale resolution imagery to assess the potential for global replicability of the space-based methodologies from highly representative pilot regions of the main BCEs across three different continents. The project consists of two phases: the first focuses on developing and consolidating requirements to create new methods on test areas, while the second emphasizes upscaling demonstration, and impact assessment. We aim at producing maps of carbon storage estimates for three different years from 2015 to 2025, with a spatial resolution no coarser than 10m while ensuring active participation from Early Adopters.

Ecosystem dynamics and change are inherently slow processes that are difficult to characterise using time-limited studies of vegetation. Furthermore, anthropogenic pressures from land use, alien species and climate change alter vegetation dynamics. This study aims to assess the changes in vegetation and their main drivers on a small Mediterranean island in the Tuscan Archipelago, Pianosa, over 18 years. The first vegetation surveys were carried out in 2005 and the most recent ones in 2023. The analysis used a combination of techniques, matching data from field surveys with different remotely sensed information for both sampling times, including land cover types and the widely employed Normalised Difference Vegetation Index (NDVI). The land cover classification was used to describe landscape-scale changes in vegetation patterns, while the differences in NDVI values were used to extract information on plot-level vegetation change. Land cover types classification was carried out on 20 cm resolution RGB orthophotos of the study area for the two sampling times, with the aid of textural metrics, using Neural Networks and validated internally. Landscape fragmentation metrics were retrieved for each plot within a buffer. NDVI was calculated using composite Landsat-7 and Sentinel-2 imagery for the two sampling times. Significant differences in values between 2005 and 2023 were assessed for different vegetation types. The main processes identified as responsible for detected changes in species composition include the spread of alien species, the encroachment of typical shrub species on grasslands, accompanied by a transition from open areas with herbaceous species to Mediterranean marquis, and a reduction in the abundance of species characteristic of rocky cliff communities. Changes in vegetation species composition were also observed at the taxonomic and functional level, probably due to changes in vegetation physiognomy. These findings can contribute to our understanding of the main drivers of change in small island contexts and may provide crucial insights for conserving habitats in the Tuscan Archipelago.

Biodiversity models are important tools to understand the drivers of biodiversity patterns and predict them in space and time, providing operational tools for conservation and restoration actions. Biodiversity models benefit from the easy access to remote sensing data allowing the assessment of habitat and landscape change at high resolution over time. However, integrating the large volume of remote sensing data within biodiversity models in a meaningful way is still an outstanding question.

Remote sensing foundation models (RSFMs) are deep neural networks trained on large datasets, typically using self-supervised learning tasks, to extract generalist representations a.k.a embeddings of the landscape without supervision.

In this study, we evaluate the contribution of RSFMs features in biodiversity models at large scale over Europe across realms. Focusing on radar (Sentinel-1) and multi-spectral (Sentinel-2) data, we select RSFMs built with different training strategies (reconstruction problems, knowledge distillation, contrastive learning) and computer vision architectures (CNNs, ViTs).

First, we use unsupervised analysis tools to assess redundancies and contrast in the spatial and environmental structure of learnt representations across models. Preliminary analysis showed that models agree over broad patterns separating ecosystem types but tend to differ in their ability to capture fine-grained habitat characteristics. Second, we evaluate different data fusion (early, stagewise, late) architectures to combine environmental (climate, soil, terrain) and remote sensing predictors to optimize predictions on three datasets: soil trophic groups diversity, bird community composition and habitat classes. Finally, using explainable AI, we quantify the relative contributions of landscape features learnt by RSFMs amongst other environmental features and its variability across target groups.

Through this study, we aim to offer guidelines for the choice of RSFMs from an increasing constellation of models and their use within biodiversity models at large scale.

Forest ecosystems, which cover approximately one third of the Earth's land area, are essential for the provision of essential ecosystem services, but their extent and health are increasingly threatened by climate change. Mapping functional traits of forests, such as leaf chlorophyll content (LCC), leaf nitrogen content (LNC), leaf mass per area (LMA), leaf water content (LWC) and leaf area index (LAI), is crucial for understanding their responses to environmental stressors and for managing these vital resources. Although remote sensing has significant potential to assess forest health and functionality, methodological and technological challenges have limited the accurate quantification of forest traits from remotely sensed data. The advent of next-generation satellites and advanced retrieval schemes offers a great opportunity to overcome these limitations. In this study, we addressed the opportunities and challenges of mapping functional traits from hyperspectral and multispectral satellite imagery in forest ecosystems using state-of-the-art retrieval schemes. In summer 2022, we conducted extensive field campaigns synchronised with PRISMA and Sentinel-2 satellite overpasses in mid-latitude forests of the Ticino Park (Italy) to collect trait samples for calibration and validation of the retrieval models. Our results highlighted the ability of PRISMA imagery to accurately quantify key forest functional traits, including LWC (R²=0.97, nRMSE=4.7%), LMA (R²=0.95, nRMSE=5.6%), LNC (R²=0.63, nRMSE=14.2%), LCC (R²=0.44, nRMSE=18.3%) and LAI (R²=0.91, nRMSE=8.3%). A comparison of the trait values between June and early September revealed a significant decrease in leaf biochemistry and LAI, attributed to the stress of the severe drought that affected the Ticino Park during the summer of 2022. This underscores the critical role of hyperspectral satellite monitoring in assessing forest health and dynamics, and highlights the importance of mapping functional characteristics to better understand and manage these ecosystems amid ongoing environmental changes.

European forests phenology by MODIS Leaf Area Index and GEDI Plant Area Index

Alexander Cotrina-Sanchez^1,2, David A. Coomes², James Ball², Amelia Holcomb³, Carlo Calfapietra⁴, Riccardo Valentini¹ and Gaia Vaglio Laurin^4*

¹ Department for Innovation in Biological, Agro-food, and Forestry Systems, Tuscia University, Viterbo, Italy.

² Conservation Research Institute, Department of Plant Sciences, University of Cambridge, Cambridge, UK.

³ Department of Computer Science, University of Cambridge, Cambridge, UK

⁴ Research Institute on Terrestrial Ecosystems, National Research Council, Montelibretti Research Area, Italy

An accurate characterization of the timing of phenological events, such as the start of season and end of the season, is critical to understand the response of terrestrial ecosystems to climate change. In broadleaf deciduous forests, there are known discrepancies in patterns measured from the ground and space. Light detection and ranging (lidar), can penetrate canopy and is potentially useful to solve some of the challenges in remote sensing phenology. Here a comparison of phenology time series from active lidar (plant area index from the Global Ecosystem Dynamics Investigation) and passive optical (leaf area index from the Moderate Resolution Imaging Spectroradiometer) was carried out. The results evidence clear differences in the detection of the senescence phase in broadleaved European forests at different latitudes, that can be explained by the different sensors detection mechanisms, with GEDI Plant Area Index estimating a longer end of season phase, and the capability to detect phenology changes along the vertical profile too. The passive and active data here tested see two different moments of the senescence: the color change of leaves and the fall of leaves and branch exposure, respectively. During the growing season, MODIS Leaf Area Index better captures fine greenness variations. Sensor integration is recommended to provide a comprehensive representation of the phenology phases, contributing to advancements in ecological and climate change research.

Considering the global threat to freshwater ecosystems, the conservation of aquatic plant diversity has emerged as a priority area of concern. In the last decade, remote sensing has facilitated the measurement of biodiversity, particularly across terrestrial biomes. The combination of spectral features with additional information derived from community phylogeny can further advance the accurate characterisation of plant functional diversity across scales.

In this study, we investigated the potential of using spectral features extracted from centimetre-resolution hyperspectral imagery collected by a drone in conjunction with phylogenetic features derived from a fully resolved supertree to estimate functional diversity (richness, divergence, and evenness) in communities of floating hydrophytes and helophytes sampled from different sites. To this end, we employed non-linear parametric and machine learning models.

The results demonstrate that all three functional diversity metrics can be estimated from spectral features using machine learning models (random forest; R² = 0.90–0.92), whereas parametric models exhibit inferior performance (generalised additive models; R² = 0.40–0.79), particularly in the estimation of community evenness. The integration of phylogenetic and spectral features enhances the predictive capacity of machine learning models for functional richness and divergence (R²=0.95-0.96), although this benefit is significant for estimating only community evenness when parametric models are employed.

The conjunction of imaging spectroscopy and phylogenetic analysis offers a quantitative means of capturing the diversity observed in plant communities across scales and gradients, which is valuable to ecologists engaged in the study and monitoring of biodiversity and associated processes.

The expansion of remote sensing applications has advanced the study of vegetation function and diversity, mainly focusing on terrestrial plants, but more recently including aquatic species. However, the relationship between spectral characteristics and plant diversity, especially in land-water interface ecotones, remains underexplored. To address this, new empirical data were collected from study sites in Italy and China to develop methods for estimating species and functional diversity from spectral data covering highly heterogeneous plant communities ranging from terrestrial to aquatic ecosystems.

The reference data collection in the Italian study site was carried out in June-August 2024 in the Mantua lake system (wetland ecosystem), Parco del Mincio wet meadows (grassland ecosystem) and Bosco Fontana (forest ecosystem) from 30 target plant communities (10 each for the three ecosystem types), ranging from aquatic (floating and emergent hydrophytes, riparian helophytes) to terrestrial (wet grasslands and floodplain forests): community composition, functional traits, spectral response, drone-based hyperspectral and LIDAR data, and synthetic parameters characterising environmental conditions (e.g., trophic status, substrate).

Spectral features extracted from centimetre resolution imaging spectroscopy data were used to estimate plant species diversity based on optical species clustering and parametric models fed with multidimensional spectral features. In addition, the functional diversity of sampled communities was modelled and mapped from centimetre resolution imaging spectroscopy data using diversity metrics based on spectro-functional traits covering target plant groups and spectral hypervolumes (richness and divergence).

Further work will be carried out to integrate the data collected in both study sites (Italy and China) into a unique dataset, from which quantitative comparisons of the results obtained will be made to explore which approach is effective for both aquatic and terrestrial vegetation, and to assess the ecological relevance of spatial patterns of plant traits and diversity assessed from remote sensing data across scales and sites.

Monitoring and reporting on biodiversity and land cover is an important global need that requires diverse techniques and innovative approaches. The Alberta Biodiversity Monitoring Institute (ABMI) integrates advanced remote sensing technologies—including satellite data—with species observations to create a robust monitoring framework in Alberta, Canada. Cross-sector collaboration and strong knowledge translation programs are key to ensuring that the data collected and the insights generated are effectively shared and used. Here we showcase examples of how we've worked collaboratively to develop accessible and innovative biodiversity and land cover information products, utilizing space-based information in our workflows and overall framework.

For nearly two decades, we have monitored changes in wildlife and habitats across Alberta's 661,848 km², delivering relevant, scientifically credible information about the province's living resources. Geospatial approaches provide direct insights on the status of landscape features and serve as key covariates for modelling species distributions. We use geospatial approaches to derive datasets such as human footprint inventories, wide-area habitat mapping, and post-disturbance forest recovery. These datasets combine with species observations in modelling pipelines to report on biodiversity intactness for hundreds of species—offering invaluable insights for evidence-based natural resource management.

A key step in our monitoring cycle is enhancing accessibility and application of results through knowledge translation. We share data and results via multiple online information products, including status reports, an Online Reporting for Biodiversity tool, a Mapping Portal, and other product-specific web browsing tools, all using satellite-derived data. These resources ensure biodiversity information is available and actionable for policymakers, resource-sectors, Indigenous communities, and the public.

The integration of satellite data, remote sensing, and species observations, combined with a strong focus on multi-sector collaboration and knowledge translation, provides a strong template for biodiversity monitoring programs. This comprehensive approach not only informs environmental decisions but also supports meaningful conservation outcomes across Alberta.

BioSCape, a biodiversity-focused airborne and field campaign, collected data across terrestrial and aquatic ecosystems in South Africa. BioSCape was largely funded by NASA, a US federal institution and many U.S.-affiliated researchers lead projects on the BioSCape Science Team. However, BioSCape’s 150+ person Science Team is intentionally diverse, with over 150 members from both the U.S. and South Africa and spanning scientific disciplines, proximity to end-users, field experience, local knowledge, technical capacity, and culture.

Being aware of the risk of parachute science, BioSCape has made progress towards developing best-practices to prevent it. Here, we will present our lessons learned and the ways in which BioSCape promoted co-design of the research and worked towards achieving Open Science, capacity building, and outreach goals. We present how BioSCape’s co-designed research agenda increased the potential for local impact and how BioSCape may contribute towards South Africa’s tracking of progress towards the goals and targets set out in the Kunming-Montreal Global Biodiversity Framework (“The Biodiversity Plan”). We review the ways that BioSCape incorporated local expertise into the design of the campaign and how an ethical and inclusive atmosphere was fostered across the team.

Savanna ecosystems play a crucial role in the global carbon cycle, serving as important yet increasingly sensitive biodiversity hotspots. Recent studies have emphasized the importance of monitoring the spatial and temporal dynamics of the vegetation layer to better understand changes that alter its composition and structure. However, the dynamic and heterogeneous nature of savanna vegetation presents unique challenges for satellite remote sensing applications. This study aims to address some of these challenges and presents our progress towards the development of a framework for monitoring woody vegetation in savanna ecosystems.

We integrate Synthetic Aperture Radar (SAR) data from Copernicus Sentinel-1 with spaceborne lidar data from the Global Ecosystem Dynamics Investigation (GEDI) to model vegetation structural variables across the Kruger National Park, South Africa. Our analysis focuses on GEDI-derived variables, particularly relative height (98th percentile), canopy cover, foliage height diversity index, and total plant area index. To address savanna-specific challenges, we apply an extended quality-filtering workflow for GEDI shots, incorporating MODIS Burned Area data and a Copernicus Sentinel-2 derived permanent bare vegetation mask. SAR time series data between 2018 and 2024 are processed to monthly composites using a local resolution weighting approach, capturing seasonal backscatter dynamics.

Preliminary results demonstrate the effectiveness of this multi-sensor approach. Clustering of GEDI vegetation structural variables from the leaf-on period reveals distinct structural classes, with corresponding SAR backscatter time series showing high separability during dry season months. Additionally, the study highlights the superior capacity of radar in distinguishing structural characteristics compared to optical vegetation indices.

This research contributes to the development of an open-source, reproducible framework for wall-to-wall mapping of vegetation structure variables and diversity over time in heterogenous savanna landscapes. The findings have significant implications for biodiversity monitoring and conservation in these ecologically important and dynamic ecosystems.

Urban trees are essential for supporting biodiversity, as they provide habitats for various species and help regulate water storage and temperature, and sequester CO₂ in urban ecosystems. Urban forests have been proposed as a nature-based solution to fight climate change and provide ecosystem services to citizens. Mapping and monitoring urban trees is vital as it facilitates conservation strategies for both flora and fauna, early diagnosis of plant pathogens, and zoning and urban development. However, mapping trees has proved difficult for urban planners since they rely on in situ surveys or community-led projects that may not cover all areas; one such case is London, where the official survey only accounts for ~10% of the estimated 8 million trees in the city. Moreover, the geographic coordinates of trees are surprisingly unreliable due to a lack of precision of measuring devices (e.g. phones or commercial GPS).

We propose a method for calibrating urban tree locations using physical ground sensors as "anchors". These sensors help reconcile spatial mismatches across various spatial datasets, including high-resolution satellite and aerial imagery and tree surveys collected by city councils or in open-data projects like OSM. These low-power sensors can also collect microclimate and other biodiversity-related data, such as passive acoustic animal activity monitoring, providing a richer picture of tree and urban ecosystem health and enabling high resolution maps not previously possible.

Our ultimate goal is to combine remote sensing information with ground-based measurements to support reliable data that can be used in geographic-based foundation models to help better urban planning strategies around trees that maximise their benefit to humans and nature.