Conference Agenda

Although many long-distance migratory birds select a stable set of wintering sites and intermediate stopover points, facultative migrants exhibit notable interannual variability in their migratory patterns, typically in response to food availability along their route. Using spatial data from the Open Data Cube alongside census data collected from three estuaries in central Chile between 2006 and 2024, we analyzed variations in the summer populations (December-February) of Franklin's Gull (Leucophaeus pipixcan) in relation to indicators of food availability, such as the mean and standard deviation of chlorophyll-a concentration (chl a) and sea surface temperature (SST) across different latitudinal ranges (0-40°S) along their migratory route.

The most robust model (GLM with temporal autocorrelation) to predict the number of Franklin's Gulls arriving at central Chilean estuaries during the austral summer incorporated a negative effect of chl a standard deviation off the Peruvian coast (0-10°S) during spring (November-December). This suggests that in years when primary productivity is high along the Peruvian coast, the gulls find sufficient resources at lower latitudes, reducing their visits to central Chile. This hypothesis is supported by the negative correlation between species abundance observed in central Chile and an eBird abundance index for Peru. Our findings illustrate how Earth Observations and spatial data integration through this platform enable robust, scalable insights into migratory species responses to ecosystem productivity shifts. Our results emphasize that primary productivity along migratory routes directly influences the range extent of these gulls, providing valuable input for conservation and monitoring frameworks reliant on space-based biodiversity data.

Habitat mapping offers a crucial visual representation of the spatial distribution and characteristics of habitats within ecosystems, supporting biodiversity conservation and ecological monitoring. This process typically combines remote sensing data, such as satellite imagery and airborne data, with advanced geographic information systems and high-resolution environmental layers to create detailed and dynamic maps of habitat distribution. Incorporating updated field survey techniques and certified open-access databases is essential for generating comprehensive, accurate habitat maps that enable temporal and spatial analyses of habitat change.

Advancements in computer science and data analysis further enhance habitat mapping by enabling "computational biodiversity," a user-centric approach that leverages sophisticated computational methodologies to assess conservation status. Cutting-edge satellite technologies for pixel-level detection have strengthened ecosystem monitoring, filling critical knowledge gaps in habitat distribution and phenological trends. However, a recent review of European user and policy requirements, particularly under the Habitats Directive, has identified significant limitations in current monitoring techniques, which slow down effective conservation at national and continental scales. Establishing standardized procedures for habitat mapping and monitoring is therefore essential to meet institutional reporting requirements and steer conservation efforts.

A rigorous evaluation of current data collection methodologies and spatial analysis techniques, along with the integration of emerging tools like next-generation satellite products and AI algorithms, is paramount. Additionally, a meticulous assessment of the urgency, feasibility, and constraints of these approaches is necessary to ensure timely, effective conservation actions and to address the evolving challenges in habitat and biodiversity management.

This study presents a novel approach for classifying and mapping macroalgae species along the Portuguese coastline using high-resolution Sentinel-2 satellite imagery combined with ecological data from citizen science initiatives (i.e. iNaturalist). A dataset containing 79,767 marine-forest observations across Europe was filtered to select high-quality records of macroalgae specific to the Portuguese coast, yielding 487 data points after spatial and spectral clustering. To distinguish macroalgae species with similar spectral responses, we computed mean spectral values for each species and used k-means clustering to categorize them into four distinct groups, ultimately focusing on the two largest clusters for further analysis. Leveraging remote sensing indices such as the Normalized Difference Water Index, Normalized Difference Vegetation Index, and Normalized Difference Mangrove Index as features, we implemented a Random Forest model trained on labeled data to classify macroalgae and map their distribution. The accuracy and consistency of the classification were validated by comparing outputs to an independent dataset across multiple years (2023–2024), revealing consistent spatial patterns. The study delivers a replicable method for monitoring coastal macroalgae, to support marine biodiversity conservation and environmental management along the Portuguese coast.

The black grouse (Lyrurus tetrix) is a galliform species emblematic of the European Alps, currently threatened by habitat change. In this study, we attempted to map black grouse Brood Habitat Suitability (BHS) at the scale of an Alpine bioregion, coupling a Species Distribution Model (SDM) with multi-source remote sensing data. To extract landscape composition features likely to influence BHS, Convolutional Neural Networks (CNNs) were employed utilising Very High Spatial Resolution (VHSR) SPOT6-7 imagery. Altitude, phenological indices derived from Sentinel-2 time series (NDVImax, NDWI1max) and a texture feature derived from the SPOT6-7 images (Haralick entropy) were used to refine the landscape characterisation. Finally, an SDM based on a Random Forest ensemble model was used for the mapping of black grouse BHS. Consistent with the ecological needs of black grouse, altitude, ericaceous heathland and NDVImax emerged as the three most important variables. In particular, the proportion of ericaceous heathland reflects the foraging needs of female black grouse, which is the main ecological determinant of habitat suitability for brood rearing with sufficient vegetation cover. This study highlights the effectiveness of integrating VHSR and multispectral time series, together with the advantages offered by Machine Learning techniques, in extracting species-specific information tailored to conservation issues.

We presented a methodology based on the SEEA-EA statistical framework to develop condition accounts for urban ecosystems. Urban condition is obtained from satellite information and remote sensing and GIS techniques, using Euclidean distance to calculate the condition index. This allows for a spatial and explicit assessment of urban condition, which is calculated for each pixel. However, the reference area is obtained through an object-based assessment, since the reference value for each variable is considered within a real territory rather than individual pixels.

This methodology involves achieving the following steps: 1. Delimitation of the urban categories to be evaluated; 2. Selection of the variables that characterise the abiotic and biotic environment; 3. Establishment of the reference polygon with which to compare the condition values; 4. Calculation of weighted condition indicators; 5. Generation of a single condition index from the aggregation of the indicators.

In the city of Madrid, it has been observed that the areas with the highest condition levels are characterised by a significant density of trees and bird species richness. In contrast, areas with the lowest condition levels are defined by high levels of contamination, impervious surfaces, built-up areas and major communication routes.

This innovative approach to calculating urban conditions represents an advancement in local-scale urban condition accounting and offers a potentially compatible tool with current urban policy frameworks.

The methodology offers several advantages over existing metrics, including object-based analysis, reduced operational costs, an integrated ecosystem perspective, simplicity and methodological flexibility, lower reliance on human judgment, the capacity to capture complex urban dynamics and easily interpretable results. Potential applications include identifying critical action points, evaluating the effectiveness of plans and policies, assessing urban resilience and guiding green infrastructure planning, all of which are relevant to the city of Madrid’s Green Infrastructure and Biodiversity Plan 2020–2030.

In recent decades, European forests have faced an increased incidence of disturbances. This phenomenon is likely to persist, given the rising frequency of extreme events expected in the future. As forest landscapes fulfill a variety of functions as well as provide a variety of services, changes in severity and recurrence of disturbance regimes could be considered among the most severe climate change impacts on forest ecosystems. Therefore, estimating canopy recovery after disturbance serves as a critical assessment for understanding forest resilience, which can ultimately help determine the ability of forests to regain their capacity to provide essential ecosystem services.

This study examines the impact of varying forest fire disturbance frequencies, a key attribute of disturbance regimes, on the recovery of European forests. Forest fire data were acquired from the Copernicus EFFIS service. A remote sensing based approach, using MODIS time series data of a canopy cover structural variable like Leaf Area Index (LAI), was developed to evaluate recovery dynamics over time, from 2000 to the present, at a spatial resolution of 500 meters. Recovery intervals were determined from the tree cover time series as the duration required to reach the pre-disturbance canopy cover baseline, using the previous forest status as a reference. Severity was defined in relative terms, by comparing forest conditions before and after disturbances. Additionally, this study analyzed severity and recovery indicators in relation to forest species distribution and productivity metrics across Europe, offering valuable insights into the effects of disturbances on the interactions between bundles of ecosystem services.

This work was conducted within the ongoing EU ECO2ADAPT project, funded by Horizon Europe, to develop sustainable forest management practices that enhance biodiversity and resilience in response to the challenges of climate change.

LULC monitoring is key to understanding biophysical variables and its link with human management of the territory, especially in the context of global change. Copernicus Land Monitoring Service’s portfolio provides a comprehensive set of ready-to-use LULC multiannual products. From the 90’s, CORINE Land Cover has continuously shown the evolution of the surface at a European level every 6 years. Complementary, within the last decade, CLMS has developed Priority Area Monitoring layers, which are actual LULC products focused on different key areas: urban, riparian, protected and coastal spaces.

Traditionally, LULC information has been manually derived by expert photo-interpreters over a satellite image. This pipeline shows limitations inferring in quality: (i) satellite coarse spatial resolution, (ii) unique moment, (iii) bias from different operators (impact on comparability through year) and (iv) cost-effectiveness.

In this work, we propose a novel methodology to retrieve high-resolution PA LULC through a semi-automatic and operative workflow by using time-series super-resolved Sentinel-2 imagery feeding Artificial Intelligence models. Furthermore, this study aims to use valuable previous CLMS information to feed models by applying a thorough filtering based on the spectro-phenological behavior of each class when compared to the EO data predictors.

The first results reflect an accuracy at Level 1 higher than 90% for all classes. Moreover, several classes at more detailed levels (types of forests, managed vs natural grasslands, vineyards, etc.) turned out to be captured by this approach. The use of ARD super-resolved Sentinel-2 imagery and models focused on time-series information improves the results by (i) reducing noise, (ii) capturing unseen elements in original imagery (e.g. small roads, individual houses) and, more importantly, (iii) giving sufficient spatial detail to derive ready-to-use vector information, key to reduce the manual effort.

These results suggest the capability of the solution to be reproducible in broader areas and more frequent time steps. This product, via crosswalks between PA LULC and EUNIS candidates at levels 3 and 4, gives the necessary information to design a correct stratification of in-situ surveys through Europe and, hence, the generation of future habitat mapping.

Despite a growing knowledge on processes underlying wetland restoration, our ability to predict restoration trajectories is still limited. Temporal monitoring of vegetation changes is a tool to better understand these trajectories and identify their potential drivers. We present an innovative approach for monitoring the restoration of wetlands using satellite remote sensing, applied to a site in Bordeaux Metropole. Between 2019 and 2023, annual vegetation maps were produced, with a high degree of spatial and typological detail. For each year, a field campaign was carried out to compile a reference database of vegetation types. An automated method for processing Earth Observation data, based on the use of ensemble classification methods was then applied to produce annual maps. This mapping process, called “Biocoast”, has been developed by i-Sea for around 8 years, and has been successfully applied on numerous and various sites. For each year, a set of at least 4 Pléiades images (2 m) were acquired during the main period of vegetation development (from spring to early fall), ensuring the discrimination of phenological changes. The accuracy obtained for each map is very satisfactory, with overall accuracies over 85% for all years, with a 16-class typology.

Vegetation trajectories, both in space and over time, were analyzed by the means of transition matrices produced between each pair of years to provide a step-by-step understanding of changes in vegetation surfaces. In order to characterize the influence of flooding patterns in vegetation dynamics, the spatio-temporal variability in surface moisture was analyzed using Sentinel-2 time series. These patterns were produced by unsupervised approaches, making it possible to produce annual clusters of the most frequently flooded / moistest areas. The results showed a high degree of relevance in observing these changes, thus opening up the possibility of working on vegetation trajectories prediction in wetlands using remote sensing.

Habitat fragmentation is a major threat to biodiversity across the globe, but existing literature largely ignores naturally patchy ecosystems in favor of forests where deforestation creates spatially distinct fragments. We use savannas to highlight the problems with applying forest fragmentation principles to spatially patchy ecosystems. Fragmentation is difficult to identify in savannas because (1) typical patch-based metrics are difficult to apply to savannas which are naturally heterogeneous, (2) disturbance is a key process in savannas, and (3) anthropogenic pressures savannas face are different than forests. The absence of data on fragmentation makes it extremely difficult to make conservation and mitigation strategies to protect these biodiverse and dynamic ecosystems. We suggest that identifying fragmentation using landscape functionality, specifically connectivity, enables better understanding of ecosystem dynamics. Tools and concepts from connectivity research are well suited to identifying barriers other than vegetation structure contributing to fragmentation. Opportunities exist to improve fragmentation mapping by looking beyond vegetation structure by (1) incorporating other landscape features (i.e., fences) and (2) validating that all landscape features impact functional connectivity by using ecological field datasets (genetic, movement, occurrence). Rapid advancements in deep learning and satellite imagery as well as increasingly accessible data open many possibilities for comprehensive maps of fragmentation and more and nuanced interpretations of fragmentation.

In the context of ever-increasing human impacts and accelerating climate warming, a more nuanced understanding and accurate prediction of species occurrences and abundances across space and time is essential. Recently, new types of Species Distribution Models (SDMs) based on deep learning—referred to as deep-SDMs—have shown considerable success in predicting species occurrences. Studies demonstrate that deep-SDMs outperform conventional SDMs in occurrence prediction, and their architecture holds promise for tackling abundance prediction challenges. However, deep-SDMs require millions of observations for training and, consequently, have not been widely trained for abundance prediction due to the limited availability of abundance data, which is generally much smaller than presence-only datasets.

To address this limitation, we propose using transfer learning to adapt an occurrence deep-SDM for use in an abundance deep-SDM. This approach is based on the hypothesis that the neural network layers from a model trained on presence-only data can capture general patterns and information that are transferable to abundance predictions. As a case study, we focused on coastal fish species in the Mediterranean Sea. We assessed the efficacy of a deep-SDM trained on 406 fish counts in predicting fish species abundance by utilizing transfer learning from a deep-SDM trained on 62,000 presence-only records.

Our findings reveal that this approach significantly enhances the abundance prediction performance of deep-SDMs, with an average improvement of 35% (based on the D2 Absolute Log Error score). Consequently, deep-SDMs become 20% more efficient than conventional SDMs on average. These improvements are primarily due to better predictions of rare species abundances. This result underscores the capacity of deep-SDMs to leverage presence-only data to predict species abundances—a new and unexpected capability. This advancement paves the way for a broader application of deep learning in predicting species abundance and biodiversity patterns, especially for rare species.

Seaweed assemblages are essential components of coastal ecosystems, providing numerous ecological, economic, and social benefits, such as serving as nursing grounds that support complex trophic webs, playing vital roles in nutrient cycles and carbon storage, and constituting a valuable resource for tourism, pharmaceuticals and biofuel industries. Unoccupied Aerial Vehicles (UAV) with different sensors, have been increasingly applied in the recent years to mapping seaweed coverage and habitats worldwide allowing resolutions at the centimetric scale of relatively small areas compared to satellite coverages. Satellite multispectral data, on the other hand, covers wide areas but has coarser resolutions which limits their use in the narrow and complex intertidal zones. Our methodology combines UAV multispectral data, with in situ precise georeferencing of independent training and validation areas for the application of supervised classification techniques of intertidal seaweed assemblages. The resultant high-resolution UAV-derived seaweed extension raster can be combined with the coarser resolution satellite imagery. Sentinel satellite images were obtained for the same day of the UAV acquisition and pre-processed to mask ocean, land, clouds and other features. For each satellite pixel, the associated pixels in the UAV-derived seaweed map are extracted. A classification model is created between the reflectance data and spectral indexes of each satellite pixel and the associated seaweed extent from the UAV imagery. Model validation is performed with a subset of the labelled satellite data. Such methodology is tested on assemblages dominated by Ascophyllum nodosum and Fucus spp. at northern Portugal and with the recent Sentinel-2 satellite imagery which currently stands as the multi-spectral dataset with highest resolutions of free access. The methodology can potentially be applied to monitoring and detecting changes in intertidal seaweed habitat types and extents, as well as on the assessment of Atlantic standing carbon stocks and the effectiveness of seaweed restoration actions over time elsewhere.

By 2050 there will be approximately 10 billion people on the planet, most of whom will reside in cities. Vegetation in urban areas provide a vast array of ecosystem services, including biodiversity protection. Following landscape ecology approach, vegetation spatial distribution can be analyzed to derive information on the level of connectivity of the urban green spaces (UGS) and to support future nature-based solutions finalized to increment their potential ecological functionality. In this context, the application of network theory for assessing landscape connectivity is a promising approach to support a more sustainable urban development. This approach helps to safeguard biodiversity by addressing the challenges of habitat degradation and fragmentation posed by urbanization. To address this task, we presented a standardized and comparable assessment of landscape connectivity of UGS in 28 European Capital cities. To do so we first created an innovative European Urban Vegetation Map (EUVM) – which classifies the urban vegetation classes into trees, shrubs, and herbaceous, with a spatial resolution of 10 m for the year 2018. The EUVM was successfully validated against field surveys acquired on the basis of 2210 field observations collected by the Land Use and Coverage Area Frame Survey (LUCAS), obtaining an average overall accuracy of 83.57%. Based on the EUVM we created a model of the ecological network connectivity using a graph-based approach for calculating several landscape connectivity metrics for each city (Probability of Connectivity - PC), Equivalent Connected Area - ECA, and Integral Index of Connectivity – IIC), several more traditional landscape metrics were calculated on the same EUVM for comparison. The database of all the indicators (both from graph theory as well as from traditional landscape metrics) calculated for all the cities were analyzed in order to assess the relevance, redundancy and usefulness of the different approaches.

Remote sensing is a valuable tool for spatial/temporal analysis of inland water environments. However, the use of a single sensor can be limiting in highly dynamic environments, such as Lake Trasimeno in Italy, where wind and temperature significantly affect the lake conditions. The dynamic nature of this environment has been confirmed by continuous measurements from a fixed spectroradiometer placed in the lake (WISPStation). In this context, in the frame of Space It Up project, the aim of this study is to use a combination of hyperspectral and multispectral sensors to understand the intra- and inter-daily dynamics of Lake Trasimeno. The dataset includes 20 different dates between 2019 and 2024 and a total of 125 remotely sensed images from 14 different sensors. Specifically, six hyperspectral sensors (PRISMA, DESIS, ENMAP, EMIT, PACE and AVIRIS) and eight multispectral sensors (Landsat-8, Sentinel-2A/B, Sentinel-3A/B, MODIS-Aqua/Terra, VIIRS-SNPP/JPSS) were used. The images were downloaded as Level-2 and used as input to the bio-optical model (BOMBER) to generate.

The maps of water quality parameters (total suspended organic and inorganic matter and chlorophyll-a) were generated from Level-2 images using the bio-optical model (BOMBER) parametrised with the sIOP of lake Trasimeno. A comparison was then made at both spectral and concentration levels between the remotely sensed images and the in situ data. The spectral analysis showed a strong overall agreement between the remotely sensed images and the WISPStation data (MAPE=28.8%, SA=11.6°). Preliminary results on the concentrations of water quality parameters confirmed that the multi-sensor analysis was crucial to detect rapid changes in the lake, mainly due to variations in temperature and wind, which would have been impossible to detect with a single sensor analysis. In particular, during the late summer period, the high growth of phytoplankton in the waters during the day emerged, with maximum values recorded in the afternoon.

Atlantic bluefin tuna (Thunnus thynnus, ABFT) and albacore tuna (Thunnus alalunga, ALB) are temperate tuna species widely distributed and targeted since ancient times. Both species are known for their capability to perform transoceanic migrations as well as by their endothermic adaptations. Their movements vary seasonally and annually, occupying a variety of habitats with a wide range of environmental conditions.

The Bay of Biscay is a seasonal feeding area for juveniles of both species, where an intense artisanal fishery is developed. However, their presence throughout the year in the area is quite variable. With the electronic tagging of juvenile individuals for more than 15 years, we have gathered key information concerning the horizontal and vertical behaviour of ABFT and ALB in the Atlantic Ocean. Combining this tagging data with satellite telemetry, we built a three-dimensional habitat model and characterized the spatial and temporal distribution of these species in the Atlantic Ocean. This allowed to characterize their migration phenology across Atlantic ecoregions.

The integration of the habitat preferences and three-dimensional distribution of ABFT and ALB into spatially structured population dynamics models and ecosystem models can improve the management of these species as well as the characterization of their top-down effects across different ecoregions of the Atlantic Ocean.

The Baltic Sea, with its strong salinity gradient, large areas of anoxic bottom water and intensive anthropogenic use, is characterised in large parts of its biosphere by low biodiversity, both naturally and due to anthropogenic pressures. Changing climate and increased frequency of extreme events exert further pressure on this delicate ecosystem, leading to changes in phenology of phytoplankton communities and mismatches in food web interactions, with unclear consequences for trophic transfer and uncertainty about its future stability. In response to this challenge, a concept to enhance ecosystem monitoring in the Baltic Sea is underway at the Leibniz Institute for Baltic Sea Research Warnemünde. The concept builds on traditional biological monitoring techniques and established programmes and integrates hyperspectral in situ and remotely sensed observations with bio-optical modelling, organismal data from eDNA, phytoplankton functional types, and lipid biomarkers for phytoplankton biomass for different ecological applications within the Baltic Sea. Our focus is on workflows which leverage reflectance-based approaches to develop indicators of change in phytoplankton biodiversity in response to climate change as well as anthropogenic influences (e.g., eutrophication, marine heatwaves) by empirically associating diagnostic reflectance features to the taxonomic and functional composition of phytoplankton assemblages. By including biogeochemical proxy records from past climate periods in our analysis, we connect across different temporal and spatial scales, and look to unravel drivers of past changes and how these may inform present and future changes. The aim is to establish a holistic ecosystem observing system which optimizes the use of existing data with new satellite data sources and provides a framework towards operationalising indicators for management directly relevant for implementing, e.g. the Marine Strategy Framework Directive (MSFD) and the HELCOM Baltic Sea Action Plan, thus significantly enhancing our capacity to rapidly detect changes in the state of phytoplankton communities, emerging invasive species and pathogens.

Mediterranean dunes and salt marshes are home to a wide range of organisms and unique and fragile plant species assemblages. These plant communities are highly threatened by human activities and extreme climatic events. To help preserve dunes and salt marshes the assessment of their vulnerability status relies on the accurate mapping of different habitats, together with the identification of major local drivers of habitat and species loss. Here, we focus on dunes and salt marshes habitats of the Tyrrenian coast of Central Italy to accurately map habitat types and predict each habitat patch ‘risk status’ according to major environmental drivers and anthropic stressors. We perform a supervised habitat classification at 10 m scale based on plot surveys data using Artificial Neural Networks (ANNs) on Sentinel-2 imagery, Normalized Difference Vegetation Index (NDVI) data and textural metrics. Secondly, to assess habitat patches risk status we retrieve a series of indicators related to coastal erosion, flood risk, distance to infrastructures, and landscape fragmentation metrics in buffers around sampled localities to obtain an overall index of vulnerability. We tested the accuracy of the habitat map with an internal and an external validation, using plot data from various sources, and assess habitat patches actual conservation status in relation to their risk status with field-based indicators such as functional and taxonomical composition and community completeness. The results of this study can help shedding light on dunes and salt marshes conservation along the Thyrrenian coast of Italy while providing valuable information for decision makers to implement protection efforts across most vulnerable habitat patches of dunes and salt marshes of central Italy.

The anthropogenic alteration of natural forests in many tropical and subtropical ecosystems is one of the most significant drivers of biodiversity loss and global change. Among the most affected regions is the Chaco forest, the largest dry forest in the Americas. This threat has prompted the United Nations to include sustainable forest management as a key target in the 15th Sustainable Development Goal (SDG), emphasizing the need for updated indicators and monitoring tools. Remote Sensing (RS) provides cost-effective, multi-temporal data across various spatial scales, making it a valuable tool for assessing forest degradation and management. This study combines RS spectral indices with field data on forest structural alterations to differentiate between sites with varying management regimes and sustainability levels. Using a representative area of the Chaco forests—the Chancaní Provincial Reserve and surrounding areas in the West Arid Chaco—as our study area, and implementing a phenological analysis of a wide set of RS spectral eco-physiological traits derived from Sentinel-2 images we aim to answer the following questions: a) do forests with different management regimes and dominant species exhibit different spectral phenology?, b) Which indices are most effective in differentiating forests with distinct levels of ecosystem structural alteration? Forest structure types and conservation levels were related to monthly spectral indexes behavior using Linear Mixed Models and Random forest analysis. The phenology of spectral indices varied significantly across low, intermediate, and high conservation levels. BI2, NDWI, and MCARI were the Remote Sensing indices that effectively distinguished forest stands with varying conservation levels and degrees of structural degradation. The proposed procedure, which combines Remote Sensing with field data, proved effective in detecting and characterizing forests with varying conservation and sustainability conditions. It could be considered as one of the Remote Sensing indicators for monitoring progress towards the SDG established by the United Nations

Coral reefs in tropical or subtropical environments are known to be indicators of global warming and have provided information that is important for the monitoring of pollution and environmental change. We present quantitative estimations of changes in the areal extent of coral reef habitats at Weno Island, Micronesia, using high-spatial-resolution remote sensing images and field observations. Coral reef habitat maps are generated from QuickBird satellite images for 2011 and 2024, and the difference between the number of pixels occupied by each seabed type is calculated, revealing that the areal extent of living corals changes between 2011 and 2024. In the process of satellite-based mapping, water column correction is performed to eliminate the effect of the light attenuation within the water column from the satellite image, employing a band combination approach known as the depth invariant index (DII) transformation. The combination of the new images generated by the DII transformation are used for image segmentation for the application of object-based image classification. This study can be used as a basis for remediation planning to diminish the impact of changes in coral reefs.

Climate change and local human impacts are causing detrimental effects across marine ecosystems. However, at present, it is still difficult to make projections on their future trends, because there is a substantial lack of data on present-past spatial distribution and extent of both habitats and human activities. The use of satellite technology for large-scale, long-term monitoring and mapping of seagrasses and macroalgae habitats have been limitedly used, despite the despite the potential of this approach. Here, we discuss the pros and cons of using satellite technology in this ecological framework together with the training the Artificial Intelligence (AI) to delineate autonomously the boundaries of the habitats from satellite images. In Italy, two Marine Protected Areas have been chosen as case studies: the MPA of Porto Cesareo (Apulia, Ionian Sea) and the MPA of Torre Guaceto (Apulia, Adriatic Sea). In these areas there are different habitats such as Posidonia oceanica, Cymodocea nodosa and Cystoseira spp, and a lot of data have been collected in the past. In Spain, a Fishery Protected Area in Vilanova i la Geltrù (Catalonia, Mediterranean Sea) has been chosen as another case study, in order to monitor and map a Posidonia oceanica meadow, located near the OBSEA, a cabled seafloor observatory. The “Essential Biodiversity Variables” is a set of variables required for the maintenance of biodiversity. The EBV monitored in this study is the “Ecosystem structure”, which is the measure of the condition of the ecosystem’s structural components. An important objective of this effort is also to document the existence of relationships between global (e.g. temperature rise), and local anthropogenic pressures (e.g. water turbidity) and visible changes in the two habitats through the time. Changes in temperature and water turbidity variables are expected to affect growth and photosynthetic efficiency of seagrasses and macroalgae.

The Salish Sea, a dynamic system of straits, fjords, and channels in southwestern British Columbia, Canada, is home to ecologically and culturally important bull kelp forests. Yet the long-term fluctuations in the area and the persistence of this pivotal coastal marine habitat are unknown. Using very high-resolution satellite imagery to map kelp forests over two decades, we present the spatial changes in kelp forest area within the Salish Sea before (2002 to 2013) and during/after (2014 to 2022) the ‘Blob,’ an anomalously warm period in the Northeast Pacific. The total area of bull kelp forests from 2014 to 2022 has decreased compared to 2002 to 2013, particularly in the northern sector of the Salish Sea. Further comparison with 1850s British Admiralty Nautical Charts shows that warm, less exposed areas experienced a considerable decrease in the persistence of kelp beds compared to the satellite-derived modern kelp, confirming a century-scale loss. In particular, kelp forests on the central warmest coasts have decreased considerably over the century, likely due to warming temperatures. While the coldest coasts to the south have maintained their centennial persistence, the northern Salish Sea requires further research to understand its current dynamics.

Phytoplankton play a vital role in marine ecosystems, driving primary productivity and influencing global biogeochemical cycles with significant implications for climate regulation. However, assessing phytoplankton community composition (PCC) in coastal environments poses unique challenges due to complex optical conditions influenced by variable particle and dissolved organic matter concentrations.

This study explores how different combinations of phytoplankton and detrital particles contribute to total particle absorption, reflecting diverse coastal water conditions. Our primary aim is to improve pigment retrieval for PCC in complex coastal environments using absorption-based bio-optical algorithms. Specifically, we assess the performance of the Gaussian decomposition method from Chase et al. (2013) across a wide range of particle concentrations and adapt it to these distinct conditions. In such areas, detrital absorption can significantly impact the algorithm’s analytical accuracy. Additionally, variation in turbidity levels may indirectly influence phytoplankton taxonomy and absorption characteristics as they respond physiologically to changes in light availability. Accordingly, this study seeks to increase pigment concentration estimation accuracy to provide clearer insights into phytoplankton community composition and the environmental conditions relevant to algorithm application.

To achieve these goals, we leverage a comprehensive dataset from the 2023-2024 Tara Europa expedition, comprising punctual data as High-Performance Liquid Chromatography (HPLC) and filter-pad-derived absorption measurements of phytoplankton and particles from 200 stations. Additionally, continuous hyperspectral absorption and attenuation data were collected from a WETLabs AC-S instrument.

Given the importance of satellite observations in large-scale ecological monitoring, this research also aims to refine and validate this absorption-based bio-optical algorithm to support hyperspectral missions such as EnMAP, PACE, and PRISMA. By integrating this algorithm with in situ hyperspectral absorption data, we aim to enhance PCC retrieval accuracy, ultimately advancing our understanding of coastal phytoplankton dynamics across various optically complex environments.

Fire is widely acknowledged as a key factor in shaping vegetation structure and function in Mediterranean ecosystems, which are generally resilient to fire. However, climate change is projected to increase the severity and frequency of fires in these regions, leading to longer fire seasons and making management efforts aimed at ecosystem restoration more challenging. In this study, we used MODIS satellite data (MOD09A1 V6.1) from 2003 to 2023 to observe post-fire vegetation recovery from the 2007 fire season in the Peloponnese region of Greece, which experienced some of the largest fires on record. Utilizing the Normalized Difference Vegetation Index (NDVI) as a proxy for vegetation greening, we identified patterns of vegetation recovery by calculating differences between the NDVI values at 5, 10, and 15 years after the fire season and (i) the NDVI before the fire (dNDVI_pre), i.e., 2006, and (ii) the NDVI just after the fire (dNDVI_post), i.e., 2007. These patterns were subsequently compared across different land cover types in relation to burn severity. Our results demonstrated that over time, dNDVI_post increased with fire severity (positive slope of the linear model between fire severity and dNDVI_post) across all land cover types, indicating that the higher the burn severity, the faster the regreening—likely due to the greater initial reductions in vegetation cover that allowed pioneer plants to rapidly recolonize the burned area. Additionally, our results demonstrated that over time, dNDVI_pre decreased with fire severity (negative slope of the linear model between fire severity and dNDVI_pre). The dNDVI_pre highlighted significant differences between pre- and post-fire conditions, especially in areas with high burn severity. In contrast, low-severity fires showed greater resilience, with ecosystems returning to near pre-fire NDVI values within five years. Notably, in agricultural land cover types, recovery appeared to be very rapid and less influenced by burn severity. Conversely, in pastures and sparsely vegetated areas, recovery was highly dependent on burn severity; in the former, it took almost 15 years to restore original greenness conditions, while in the latter, recovery was still incomplete even after 15 years.

Upland habitats provide vital ecological services, yet they are highly threatened by natural and anthropogenic stressors. Monitoring these vulnerable habitats is fundamental for conservation and involves determining information about their spatial locations and conditions. Remote sensing has evolved as a promising tool to map the distribution of upland habitats in space and time. However, the resolutions of most freely available satellite images (e.g., 10-m resolution for Sentinel-2) may not be sufficient for mapping relatively small features, especially in the heterogeneous landscape—in terms of habitat composition—of uplands. Moreover, the use of traditional remote sensing methods, imposing discrete boundaries between habitats, may not accurately represent upland habitats as they often occur in mosaics and merge with each other. In this context, we used high-resolution (2 m) Pleiades satellite imagery and Random Forest (RF) machine learning to map habitats at two Irish upland sites. Specifically, we investigated the impact of varying spatial resolutions on classification accuracy and proposed a complementary approach to traditional methods for mapping complex upland habitats. Results showed that the accuracy generally improved with finer spatial resolution data, with the highest accuracy values (80.34% and 79.64%) achieved for both sites using the 2-m resolution datasets. The probability maps derived from the RF-based fuzzy classification technique can represent complex mosaics and gradual transitions occurring in upland habitats. The presented approach can potentially enhance our understanding of the spatiotemporal dynamics of habitats over large areas.

Pacific Herring (Clupea pallasii) is a pelagic species present in the North Pacific Ocean’s inshore and offshore
waters including the Salish Sea in British Columbia (BC, Canada) (DFO, 2021). As an important forage fish, it is
known for its role as a key species for the Salish Sea. Herring presence is considered an indicator of ecosystem
health, as it provides important ecological, economic, and social benefits (Morin et al., 2023; DFO, 2021). Many
species of the SaS rely upon the spawning of this forage fish for nutrition (marine mammals, birds, salmonids,
etc.) and plays a significant cultural role for coastal communities in BC since time immemorial (DFO, 2024).
However, First Nations in Southern BC have reported instability in the population stock as well as in spawning
timing, magnitude and locations. Those observations have been corroborated by recent reports by the Ministry
of Fisheries and Ocean (DFO), but there is a chance that the current monitoring framework, based on in-situ
observations, could actually miss a significant portion of the spawning events, preventing a proper evaluation of
the magnitude of the changes.

To address this knowledge gap, we propose the development of a satellite cloud-based method using Google
Earth Engine to investigate herring spawning presence in the Salish Sea over the past 40 years using a com-
bination of Landsat (EM, ETM+, OLI & OLI-2) and Sentinel-2 (MSI) imagery collections. To achieve this, we
developed a Spectral Herring Spawning Index (SHSI), a novel index using Rsr to highlight the spawning events.
We are comparing the accuracy of this index in a threshold-based detection technique and as a feature in classi-
fication algorithms. The outcomes will allow a broad spatiotemporal analysis and will provide near-real time tools
to First Nations to monitor spawning events on their territories.

References:

DFO (2021). Integrated Fisheries Management Plan Summary - Pacific Herring (Clupea pallasii) Pacific Region 2021/2022. Technical report.
URL: https://waves-vagues.dfo-mpo.gc.ca/library-bibliotheque/41097877.pdf.
DFO (2024). Stock Status Update with Application of Management Procedures for Pacific Herring (Clupea pallasii) in British Columbia: Status
in 2023 and Forecast for 2024. Sci. Advis. Sec. Sci. Resp. 2024/001, Ministry of Fisheries and Oceans. URL: https://publications.
gc.ca/collections/collection_2024/mpo-dfo/fs70-7/Fs70-7-2024-001-eng.pdf.
Morin J., Evans A. B., Efford M. (2023, April). The Rise of Vancouver and the Collapse of Forage Fish: A Story of Urbanization and the
Destruction of an Aquatic Ecosystem on the Salish Sea (1885–1920 CE). Human Ecology 51(2), 303–322. URL: https://doi.org/10.
1007/s10745-023-00398-w, doi:10.1007/s10745-023-00398-w

Common ragweed (Ambrosia artemisiifolia) is an invasive, allergenic species originating from North America that has spread widely across Croatia, particularly in Zagreb, Poreč, and Slavonia. Its rapid spread poses a threat to biodiversity, public health, and agriculture, with economic losses in Europe reaching up to €130 million annually. Although numerous local and national initiatives aim to control ragweed, traditional methods like field inspections and citizen reporting are limited in effectiveness.

The ESA funced project conducted by LIST LABS and their partners proposes a novel framework for automated ragweed monitoring using Earth Observation (EO) data, machine learning models, and existing field and phenology data. The primary objective is to develop a prototype architecture that enables the detection, classification, and prediction of ragweed growth locations, focusing on high-risk areas in Zagreb and Poreč. By integrating high-resolution satellite imagery with spatial data from local institutions, the system aims to achieve 90% detection accuracy with less than 30% commission error for areas exceeding 100 m² with >30% ragweed cover.

The framework includes a web-based GIS application for visualizing detected and predicted ragweed locations, providing public authorities and citizens with transparent, near real-time information. This solution promises significant cost reductions in field inspections and improved responsiveness in ragweed management. Furthermore, it highlights the advantages of space technology for invasive species control, supporting a more effective fight against the spread of ragweed and its health impacts in Croatia and beyond.

The poster will present the results of the projects funded under ESA’s Third Call for Outline Proposals under the Implementation Arrangement with the Government of Croatia.

Climate change and land-use changes are key drivers of global biodiversity loss. Many species are shifting to higher elevations or latitudes in response to global warming, leading to reduced ranges and increased extinction risks, particularly for species confined to narrow, high-altitude habitats such as those in mountain ecosystems. Predicting future distributions of mountain species requires not only an understanding of their climate responses but also integrating detailed remote-sensing data, such as topographical data, land-use patterns, and species' dispersal capacities. The latter is critical for accurately predicting species ability to colonize new habitats, which may be constrained by both natural barriers and human-altered landscapes.

In this study, we projected the future distribution of 33 mountain mammals and 345 non-migratory mountain bird species by 2050 under different emission scenarios (SSP-RCP 1-2.6 and SSP-RCP 5-8.5). Using Species Distribution Models (SDMs) that incorporated topography, climate, and land-use data, we assessed the impacts of global change on species' ranges across mountain regions worldwide, accounting for realistic dispersal scenarios.

Under the high-emissions scenario, species were projected to experience significantly greater range loss compared to the low-emissions scenario, with an average loss of 16.59% for birds and 14.98% for mammals. The highest range losses were projected for species located in tropical mountain ranges and Oceania, while European and North American mountains showed the lowest losses, highlighting substantial regional differences in species vulnerability. When land-use changes were included in the models, projected range losses increased further, particularly under the low-emissions scenario.

These findings emphasize the importance of considering both climate and land-use changes when assessing biodiversity risks in mountain regions. Our results highlight the urgency of mitigating climate change and managing land use to preserve the unique biodiversity of these areas. Moreover, we identified species and regions most at risk, providing essential insights for developing targeted conservation strategies to mitigate the effects of global environmental change on mountain ecosystems.

With increasing threats to biodiversity due to climate change and other human-induced disturbances, understanding the dynamic patterns of how species are distributed on a global scale is crucial for effective conservation and management strategies. While species distribution models (SDMs) have been applied extensively in conservation, SDMs most often focus on single species and/or regional scales, which hinders their utility in global biodiversity assessments. To address this limitation, we are developing a global joint species distribution modeling approach that leverages deep learning, remote sensing data, and species occurrences to model the global distributions of plant species across all ecosystems worldwide.

Vegetation over the landscape and plant leaves themselves interact with visible light in ways that are unique and distinctive to many species. This information is captured in spectral imagery from spaceborne sensors – such as in the Sentinel and Landsat series – and is increasingly being used to indicate plant features and functions at local at global scales. Our model incorporates multi-spectral and multi-temporal satellite imagery alongside the more standard array of SDM feature layers – e.g., environmental, and bioclimatic variables at a medium to high resolution – in a convolutional deep neural network trained on hundreds of millions of plant occurrence data points. The final joint SDM will be developed to simultaneously model the distributions of thousands of plant species, accounting for environmental factors, and biogeographical patterns.

In an initial phase, we built a national-level model for Switzerland and successfully predicted the distribution of over 4,000 plant species. We are now scaling this approach to produce global scale joint SDMs.

Although results are forthcoming, this approach is anticipated to provide novel insights into global vegetation patterns and contribute to biodiversity conservation on a global scale by offering scalable, data-driven solutions.

Interoperability allows ecosystem restoration platforms or databases to share a common language and exchange data, contributing to transparent and effective tracking of ecosystem restoration efforts. The Framework for Ecosystem Restoration Monitoring (FERM) developed by FAO to support the implementation and monitoring of ecosystem restoration facilitates the registration of restoration initiatives and good practices while ensuring interoperability with other platforms and databases collecting restoration data. FERM aims at developing interoperability frameworks with restoration monitoring sources for facilitating the process of reporting Target 2 of the KM-GBF.

At the global scale FERM has worked with SDG custodians, Rio conventions, such as UNCCD, Ramsar and FRA to identify related information already collected for restoration and facilitate data exchange. The partnership with the UNCCD will bring into place the use of satellite remote sensing to assess the degradation of ecosystems as reported by countries.

At the regional and national scales, FERM has worked with AFR100, Initiative 20x20 and the Great Green Wall and with pilot countries to coordinate reporting and identify linkages and synergies between regional/national restoration and Target 2 reporting.

FERM offers an innovative interoperability solution to reporting towards Target 2 of the KM-GBF providing different ways of disaggregating total area under restoration (i.e. by ecosystem, by Protected Area and Other Effective Conservation Measures, by Indigenous and Traditional Territories, and by type of restoration activity) but also aims at creating a global map to showcase restoration project areas (as polygons or points) and good practices, supporting the monitoring of global progress of ecosystem restoration. Making precise data on restoration projects publicly available can significantly enhance scientific research on monitoring the long-term effectiveness of restoration efforts using remote sensing technologies.

Target 8 and 11 of the Global Biodiversity Framework aim to use ecosystem-based approaches to build resilience to climate change and restore or enhance nature’s contributions to people. In the SONATA project for Serbia, we will create a detailed EUNIS-classified habitat map (by combining Remote Sensing and non-EO data) and focus on the habitats surrounding several farmers’ land to evaluate the implementation of certain nature-based solutions (NbS) considering the occurring habitat types. The goal is to discover how NbS can contribute to optimizing ecosystem services: food production, pollination potential and carbon sequestration. We will link ecosystem condition to capacity for provision of ecosystem services as we will analyze grassland condition indicators for the grasslands neighboring the farmer’s fields. A spatial tool will be created that allows scenario analysis for optimizing the ecosystem services based on alternative NbS methods and their spatial distribution. The aim is to identify the optimal spatial configurations of NbS within a farmer’s land to maximize the ecosystem services, according to his priority. To create the optimization models for the scenario analysis, many on-site experiments will be set up among which a pollination experiment to estimate the pollination potential and to derive yield estimates. This project will attach great value to the all-round task of ‘knowledge and skill transfer’ between the partners. The main goal is to implement a sustainable service of habitat mapping that can be used by the Serbian partners, and the spatial optimization tool and scenario analyses will explore the often diverging interests of different stakeholders. This will allow farmers to gain insights in the potential benefits of NbS for their businesses and it will allow policymakers to be informed on the value of NbS in targeting conservation and safeguarding the longer-term viability of agricultural activities (under climate change).

The MarineBasis Monitoring Programmes in Nuuk and Disko Bay, West Greenland, have conducted monthly sampling of hydrography, water chemistry, and phytoplankton for > 15 and 7 years, respectively, as part of the Greenland Ecosystem Monitoring Program (GEM). However, this long-term sampling at single stations may miss phytoplankton community dynamics occurring at finer temporal and spatial scales. To address these limitations, we assess the performance of CMEMS GlobColour chlorophyll-a (Chl-a; product ID=cmems_obs-oc_glo_bgc-plankton_my_l3-olci-300m_P1D) estimates (2016-2022), hereafter CMEMS, against in situ data from Nuup Kangerlua (Godthåbsfjord) and Disko Bay. Our goal is to explore the potential of CMEMS data to enhance both spatial and temporal coverage and support phenology studies. The CMEMS product demonstrated strong performance, with Chl-a estimates significantly correlated with in situ measurements (r=0.57; p>0.001; RMSE=1.2 µg/L). The resulting Chl-a maps reveal considerable spatial and temporal variability, reflecting the complex dynamics of these regions. Time series derived from selected locations captured seasonal patterns well, with Disko Bay showing better agreement due to its simpler water composition. In Nuup Kangerlua, discrepancies were observed: following ice break-up, when low sun angles led to Chl-a overestimations by CMEMS; during spring, when in situ measurements report the highest Chl-a values that are underestimated by CMEMS; and in late summer and autumn, when CMEMS overestimated Chl-a, likely due to glacier flour (silt) interference. Future work will focus on analyzing the phenology of major spring/summer phytoplankton blooms in both regions, investigating interannual variability, and exploring potential links to environmental changes and extreme events.

The German Federal Statistical Office (DESTATIS) reports on the extent, condition, and services of ecosystems in Germany every three years since 2015, following the international "System of Environmental Economic Accounting" (SEE-EA) framework. The Federal Agency for Cartography and Geodesy (BKG) supports DESTATIS by providing geospatial data.

One of the ecosystem classes mapped is "Riparian Forest," which is difficult to define using conventional methods due to its complexity. To establish riparian zone boundaries, the "Delineation of Riparian Zones" from the "Copernicus Riparian Zones High Resolution Layer" is used. This dataset is combined with land cover data from the German Digital Land Cover Model (LBM-DE) to identify riparian forests. However, since the "Delineation of Riparian Zones" was discontinued after 2012, we developed a time- and cost-efficient way to update it from 2018 onwards.

Various geodata and remote sensing data are used to derive the product “Delineation of Riparian Zones”. In the calculation, the product is subdivided into Potential Riparian Zones (PRZ), Observable Riparian Zones (ORZ) and Actual Riparian Zones (ARZ). PRZ is the maximum potential extent of riparian zones without anthropogenic influences and is retained from the original Copernicus dataset. ORZ is the observed extent of riparian features from remote sensing data and ARZ is the result of a combination of PRZ and ORZ.

The main difference from the Copernicus product is the data used to define ORZ and the focus on Germany. Freely available data for German authorities is prioritized. To adapt this method for other European countries, Corine Land Cover Data can replace German land cover data, and a Europe-wide Sentinel-2 mosaic can be used for object extraction.

Indonesia is one of the countries which have abundant wetlands, especially peatlands. Peatlands in South Kalimantan contribute to securities of water, food, species, and climate change. Especially for climate change, they have carbon-rich stored in their organic soils. However, instead of storing carbon, distributed or drained peatlands due to human-caused environmental change produce greenhouse gas emissions and harm the habitat of endangered species in South Kalimantan. We explored the space-borne Synthetic Aperture Radar (SAR) using Sentinel-1 to monitor surface displacement and surface soil moisture (SSM) in peatlands. A small Baseline InSAR time series was processed to find peatland subsidence. For the value of SSM, we used the technique of SAR backscattering and low pass filter classification. We found the highest peat subsidence rate up to -48 mm/year in the district of Landasan Ulin. The total area suffered by peatland subsidence was estimated at 4,636.98 hectares and it produced a total CO2 emission of 1.699 tC hectares/year. The result confirmed that peatlands in South Kalimantan have been degraded in the districts of Bumi Makmur, Beruntung Baru, Gambut, Liang Anggang, Landasan Ulin, and Cempaka. The highest degraded peatland was found in the Bumi Makmur Subdistrict which the SSM algorithm identified as an area of 217.55 hectares while the Wosten model estimated 254.88 hectares.

Monitoring biodiversity for the longer term requires ongoing knowledge of both landscape disturbance factors, and post-disturbance recovery. Plant and animal communities will change with time following temporary disturbances (e.g., wildfire, natural resource exploration), leading to shifts in local and landscape-level biodiversity. A multitude of factors influence post-disturbance recovery: the nature of the disturbance itself, local ecosite, topographic and climatic conditions, and further human or animal usage (e.g., use of off-road vehicles). Tracking recovery in support of maintaining up-to-date knowledge of biodiversity therefore requires a more sophisticated approach than simply tracking time since disturbance.

To help fill this knowledge gap, the Alberta Biodiversity Monitoring Institute (ABMI) is leveraging the long-running Landsat image archive and Google’s Earth Engine platform to create public datasets that characterize spectrally-based regeneration of disturbed forest stands. Time series of the normalized burn ratio (NBR) are processed to extract metrics reflecting the rates and status of spectral signals as they return to pre-disturbance levels. Current public datasets focus on forest harvested for timber across Alberta, Canada, and include spectral regeneration information for >70,000 harvested areas. On average it takes 8.7 years for harvest areas to reach 80% of pre-disturbance NBR signals, and >70% of those in the dataset have reached 100%. Large-scale analysis reveals that boreal areas recover their spectral signals more quickly than those in the foothills or mountainous areas.

Work to adapt the developed approach for extracting spectral regeneration metrics to other industrial human footprints (e.g., well sites) is ongoing. Early results show average spectral regeneration rates of 73% for reclaimed oil sands mines, and average rates of 67% to 76% for active and abandoned well sites, respectively. This work can provide a broad overview of trends in spectral regeneration on disturbed forest areas over large scales, improving our understanding of current landscape conditions.

Accurate assessment of the variability and distribution of phytoplankton community composition (PCC) significantly influences better comprehension of biological carbon cycles and marine ecosystem dynamics. Although conventional empirical algorithms remain robust, their reliance on linear combinations limits their ability to achieve high-precision PCC retrieval. Recent advancements in deep learning using a huge number of ocean observation data offer a promising approach for more accurate PCC quantification. In this context, we proposed a novel estimation method that utilizes transformer-based deep learning (DL) to accurately retrieve both the chlorophyll concentration and the most representative PCC, such as diatoms, dinoflagellates, haptophytes, pelagophytes, cryptophytes, green algae, prokaryotes, and prochlorococcus. Our proposed DL takes into account various factors: optical properties from multi-ocean color satellite composited data (i.e., OC-CCI and GlobClour), physical properties from a numerical model (i.e., GLORYS), and in situ measurement collected by BioGeoChemical-Argo and high-performance liquid chromatography. The proposed DL model features a novel structure capable of simultaneously performing inverse and forward processes, allowing efficient and robust estimation. The proposed DL model reveals generalization capability and superior robustness over the global ocean through comprehensive validation. Finally, the proposed DL model was utilized to produce global monthly chlorophyll concentration and PCC, and it demonstrated better performance than conventional PCC products.

Terrestrial ecosystems around the world have been losing resilience to stressors over the past decades. Impacts of climate change and anthropogenic land use changes interact, modifying disturbance regimes and putting increasing pressure on ecosystems’ capacity to resist to disturbances, recover from them and adapt. Global assessments of ecosystem resilience often rely on simplifying assumptions for low-dimensional systems and frequently exclude anthropogenic impacts, focusing instead solely on intact natural areas. Here, we assess ecosystem resilience globally based on remotely sensed time series on vegetation productivity from MODIS using a range of different early warning signals (EWS). We evaluate the performance of different EWS for predicting both in-situ recorded ecosystem collapses and remotely sensed disturbances. Finally, we train explanatory machine learning models to disentangle climatic and anthropogenic drivers of the occurring resilience losses at global and local scales. Our approach contributes to a better understanding of the drivers of ecosystem resilience losses and supports a critical evaluation of EWS assessments.

Satellite imagery is commonly used for deriving snow cover metrics, i.e. snow cover duration and melt-out, at high resolution for large areas, which are important determinants of plant species distributions in cold environments. This has fostered their use as predictors in species distribution models (SDMs) of alpine and arctic plants. Despite their widespread use, little is known about how well remotely-sensed snow cover metrics perform in SDMs compared to those of other sources

Here, we evaluate the use of Sentinel-2 (S2) derived melt-out dates, compared to soil temperature derived, webcam derived and modeled melt-out dates at Schrankogel in the Stubaier Alps (Tyrol, Austria) as predictors in SDMs. The SDMs are based on a set of topographic and climatic predictors (slope, topographic wetness index, potential solar radiation and mean summer air temperature) alongside the melt-out date of one of the four data sources. We compared the impact of melt-out dates on the predictions of the distribution of 70 plant species among models to assess the value of S2 melt-out date as a predictor in SDMs and to identify the most powerful source of snow cover data for species distribution modeling.

Acknowledgements: This work has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (Grant agreement No. 883669).

Aerial hyperspectral and multispectral satellite data are the two most commonly used datasets to identify high conservation values open habitats. This study aimed to analyze the difference in classification accuracy of Natura 2000 habitats representing: meadows, grasslands, heaths, and mires between data with different spectral resolutions and the results utility for nature conservation compared to conventional maps. The analysis was conducted in five study areas in Poland. The classification was performed on multispectral Sentinel-2 (S2) and hyperspectral HySpex (HS) images using the Random Forest algorithm. Based on the results, it can be stated that the use of HS data resulted in higher classification accuracy, on average 0.14, than using S2 images, regardless of the area of the habitat. However, the difference in accuracy was not constant, varying by area and habitat characterization.

The HS and S2 data make it possible to create maps that provide a great deal of new knowledge about the distribution of Natura 2000 habitats, which is necessary for the management of protected areas. The obtained results indicate that by using S2 images it is possible to identify, at a satisfactory level, alluvial meadows and grassland. For heaths and mires, using HS data improved the results, but it is also possible to acquire a general distribution of these classes, whereas HS images are obligatory for mapping salt, Molinia, and lowland hay meadows.

Understanding marine ecosystem responses to increasing temperatures is crucial, especially in rapidly warming regions like the Mediterranean Sea. Phytoplankton are key indicators of ecosystem shifts, forming the foundation of the marine food web, playing a significant role in carbon cycling and marine productivity. The Rhodes Gyre, an 'oasis' within the oligotrophic Levantine Basin (Eastern Mediterranean), is notable for its high primary productivity and as a major formation area of Levantine Intermediate Water—an important feature of the Mediterranean's circulation. However, previous studies on phytoplankton dynamics have been constrained by sparse in-situ data and the surface-only coverage of satellite observations, limiting insights into long-term subsurface changes. Here, we use a Global 3D Multiobservational oceanographic dataset, which combines satellite ocean colour observations and Argo-derived in-situ hydrological data to provide depth-resolved biological information, enabling the estimation of ecological indicators across temporal, spatial, and vertical scales over a 23-year period (1998–2020). Our findings reveal a marked rise in surface temperatures after 2009, likely linked to broader oceanic warming, accompanied by declines in Chlorophyll-a (Chl-a) and Particulate Organic Carbon (POC). This warming has intensified stratification, contributing to a shallower Mixed Layer Depth (MLD) and reduced deep mixing. By analyzing Chl-a vertical distribution we show that higher concentrations of Chl-a now occur below the MLD during summer, suggesting nutrient entrapment in subsurface layers, that coincides with an increase in oligotrophy in the mixing zone (surface to MLD). Phenology indicators show a shortening of the phytoplankton blooming period by approximately five weeks in the upper 150 meters and ten weeks in the mixing zone, suggesting a weakening of vertical mixing, potentially linked to reduced winter wind speed. Our results highlight the Rhodes Gyre's increasing vulnerability to climate-driven changes and the utility of long-term 3D observational data in revealing ecosystem responses that might be overlooked by satellite-derived datasets.

Semi-natural dry grasslands are home to extremely diverse plant and animal communities, also providing invaluable functions relevant to the preservation of agricultural and natural ecosystems. Yet, dry grasslands are among the most endangered terrestrial ecosystems worldwide, due to several changes associated with natural and anthropogenic factors, and often occur in small and fragmented patches. By integrating the current knowledge on ecological requirements of plant communities with multi-seasonal VHR satellite images, we considered a Geographic Object-Based Image Analysis (GEOBIA) approach combined with a data-driven classification for the identification of grassland habitats protected by European Habitats Directive, in the Alta Murgia National Park, southern Italy. We tested machine learning object-based classification algorithms in the Orfeo Toolbox environment, by assessing the performance of Support Vector Machine and Random Forest classifiers applied to Pléiades and Worldview-2 satellite images. Based on field vegetation surveys, we implemented a land-cover nomenclature that combines the definition of three protected habitat categories (EU codes: 6210, 62A0, 6220) with information regarding their structural and compositional variability in the study area. As a direct result, we obtained a fine-scale map of grassland communities occurring in the area, including different combinations of protected habitat categories, and their successional stages associated with anthropogenic pressures (e.g., overgrazing, fire) and natural factors (e.g., encroachment, drought). In addition to the value of a detailed quantification of local habitat distribution, the adopted methodology represents a useful tool for the assessment of habitat quality, in turn potentially indicating ongoing changes in environmental conditions. With the view of application to image time series, the proposed automatic classification procedure is particularly suitable for the monitoring of habitat conservation status over time, as also required by the European Habitats Directive.

Chilean Mediterranean-type forest ecosystems harbor an unique biodiversity, are highly diverse and significantly exposed to climate change impacts, particularly severe drought events. Since 2010, a megadrought has established in this region, with a declining trend in annual tree growth and productivity, even affecting drought-tolerant evergreen forests. Projections suggest a reduction in tree growth by 2065, potentially causing drastic changes in the functioning of forests. To assess resilience of forests during two major drought events detected by spectral indices, we analyzed 23 growing seasons using MODIS Vegetation and Evapotranspiration data (MOD13Q1 EVI and MOD16A2 ET) in Central Chile. Resilience was estimated by the number of days per growing season with extreme anomalies, indicating time under perturbation. We assessed resilience across Central Chile and focused on (1) evergreen sclerophyllous forests, representative of natural ecosystems of Central Chile, and (2) deciduous Nothofagus macrocarpa forests, dominated by N. macrocarpa, an endemic species with a restricted range, and the northernmost distribution of the Nothofagus genus. Since 2010, we observed an increasing trend in extreme negative anomalies for both EVI and ET, with a peak during the 2019-20 growing season, when 16,210 km² of vegetation was affected. Evergreen forests showed lower resilience, experiencing longer periods under perturbation during the Megadrought (2010-2015) for both EVI and ET. In contrast, we found little significant decline in the productivity of N. macrocarpa forests during this event, with ET indicating consistently low impact levels affecting 5000 km² over time. During the 2019-20 season, both forest types experienced over 200 days of extreme anomalies. Evergreen forests were most affected with 98% of their distribution impacted, while N. macrocarpa forests were affected by 80%, showing latitudinal differences in resilience, as southern forests were more resilient than the northern ones.

Biodiversity is under pressure due to a variety of environmental disturbances, making its monitoring essential for effective conservation action. Herein, we present GeoPl@ntNet, an advanced satellite remote sensing (SRS) and deep learning-based workflow designed to map and monitor European plant species (over 10,000 organisms) and ecosystems (over 200 EUNIS habitats) while providing biodiversity indicators, all at very-high resolution (50m). GeoPl@ntNet leverages both computer vision (convolutional neural networks) and natural language processing (transformers) to integrate multiple biodiversity and environmental data streams, using millions of heterogeneous presence-only records combined with hundreds of thousands of standardized presence-absence surveys.

The framework is composed of three components: (i) image classification, where satellite imagery (i.e., patches and time series) and environmental rasters (e.g., bioclimatic rasters and soil rasters) are used to predict plant assemblages; (ii) fill-mask modeling, which gets a syntaxic understanding of vegetation patterns; and (iii) text classification, which uses the predicted assemblages to identify habitat types. These tasks enable GeoPl@ntNet to produce very high-resolution maps of individual species and habitats across Europe, and derive key biodiversity metrics, including species richness, presence of invasive or threatened species, and ecosystem health indicators.

In addition, we will discuss the validation of all steps (i.e., the spatial block hold-out approach to address spatial autocorrelation), the interpretability of the maps (i.e., how they can offer insights into the dynamic interactions between environmental drivers and biodiversity patterns), and the results obtained (i.e., our model outperforming MaxEnt and expert systems). Finally, we will dive into the potential of GeoPl@ntNet as a powerful tool for understanding and monitoring biodiversity dynamics and see if the integration of SRS technologies and deep learning can enable us to enhance our comprehension of ecosystems. We will reflect on how it could help guiding conservation efforts and supporting policy frameworks aimed at reversing biodiversity loss in Europe.

Despite the prevailing assumptions about the detrimental impacts of human activities on alpha, beta, and gamma diversity, as key measures of biodiversity, there is a lack of empirical research investigating these effects, with trends in beta diversity receiving particularly little attention. Besides the existing literature on species homogenization, there is no study that compares the turnover patterns in regions with varying human influence. In this research we start by describing large scale patterns of plant beta diversity, by using the sPlot global vegetation dataset and timeseries of Sentinel2 data. We then combine these patterns with proxies that capture human footprint, to investigate their impacts on the observed patterns.

Satellite-derived observations of ocean colour provide continuous data on Chlorophyll-a concentration (Chl-a) at global scales but are limited to the ocean's surface. So far, biogeochemical models have been the only means to generate continuous vertically resolved Chl-a profiles, on a regular grid. A new multi-observations oceanographic dataset provides depth-resolved biological information, based on merged satellite- and Argo-derived in-situ hydrological data (MULTIOBS). This product is distributed by the European Copernicus Marine Service, and offers global multiyear, gridded Chl-a profiles within the ocean’s productive zone at a weekly temporal resolution. MULTIOBS addresses the scarcity of observation-based vertically resolved Chl-a datasets, particularly in less sampled regions like the Eastern Mediterranean Sea (EMS). Here, we present an independent, in-depth evaluation of the updated MULTIOBS Chl-a product for the oligotrophic waters of the EMS using in-situ Chl-a profiles. Our analysis shows that this new product accurately captures key features of the Chl-a vertical distribution, including seasonal changes in profile shape, absolute Chl-a across depths and its seasonal/interannual variability, as well as the depth of the Deep Chlorophyll Maximum. At the same time, we identify conditions where discrepancies can occur between MULTIOBS-derived and in-situ Chl-a. We conclude that MULTIOBS is a valuable dataset providing vertically resolved Chl-a data, enabling a holistic understanding of euphotic zone-integrated Chl-a with an unprecedented spatiotemporal resolution, spanning 25 years, eventually paving the way for a more accurate assessment of marine ecosystems productivity. This merged product mitigates some of the limitations associated with satellite and Argo float data, and its long-term observations within the water column will advance our understanding of oceanic productivity in a warmer Earth.

Phytoplankton are at the base of the aquatic food chain and of global importance for ecosystem functioning. Effective and reliable monitoring of phytoplankton taxonomic groups is crucial to understand how lake ecosystems will respond to climate change. Inland waters phytoplankton diversity mapping from space has evolved in the past years. Today, hyperspectral sensors provide high spatial and temporal resolution, enabling detailed tracking of phytoplankton bloom evolution. However, robust and scalable retrieval methods are missing.

In this study, we investigate the potential of retrieving phytoplankton taxonomic groups in a eutrophic lake from multiannual in-situ remote sensing reflectance data (Rrs) by validating with a phytoplankton abundance dataset from an underwater camera.

We used a time series of Rrs data acquired with a WISP in situ spectroradiometer installed on a research platform in Greifensee, Switzerland. Using the freely available radiative transfer model Water Colour Simulator (WASI), we retrieved the relative abundance of four phytoplankton taxa (green algae, cryptophytes, cyanobacteria, and diatoms) from these Rrs measurements. We validated our results against data from the Aquascope phytoplankton camera installed on the same platform since 2018. Immersed at 3m depth, the camera acquires hourly photos of aquatic particles in an automated manner. Around 100 phytoplankton taxa are classified in these images using machine learning algorithms.

Our approach successfully estimates the relative abundance of the selected phytoplankton taxa during selected good weather days. Inversions conducted over several months revealed that WASI can also track the evolution of different phytoplankton blooms throughout the season. Among the abovementioned taxa, diatom blooms are the hardest to identify, which may be attributed to the quality of the Rrs data, particularly given that those blooms occur in winter.

By upscaling this method to Earth observation data from the PACE or CHIME missions, phytoplankton taxonomic groups in inland waters could be globally monitored.

The Mediterranean Sea hosts unique marine and coastal habitats whose resilience relies on complex bio-physical interactions. The adaptative capacity of these habitats to cope with climate change and extreme weather events is closely linked to biodiversity, as higher diversity provides broader genetic pools for adaptive traits. Photosynthetic plankton forms the foundation of the marine food web, driving primary production and nutrient cycling while supporting higher trophic levels, including invertebrates, fish, and marine mammals. Consequently, plankton diversity serves as a crucial bio-indicator for assessing ecosystem functioning. The omics-based Shannon index is an effective data tool for intuitively summarizing the alfa diversity within plankton communities by accounting for species richness and evenness. However, the challenge of measuring this parameter across vast oceanic areas using in situ samples can hinder effective environmental monitoring. In contrast, the broad spatial and temporal coverage of satellite ocean color data, combined with outputs from physical-biogeochemical models, holds great potential for identifying and monitoring key surface characteristics of the Mediterranean Sea, helping to fill gaps left by traditional oceanographic sampling methods. Integrating Earth Observation (EO) datasets with in situ omics measurements could thus enhance our understanding of plankton biodiversity and dynamics at high spatial and temporal resolution. To achieve this goal, in the framework of the Biodiversa+ PETRI-MED project, we used the omics-based Shannon index as the target variable for plankton diversity and a suite of satellite- and model-derived predictors from associated matchups to train a supervised machine learning algorithm, uncovering nonlinear relationships. This approach might lead to developing an EO-based index to help map spatiotemporal patterns and monitor trends in plankton communities across the Mediterranean Sea.

Climate change constitutes one of the main threats to global biodiversity. Changes in intensity, frequency and length of drought periods and heatwaves, have contributed to substantial spatio-temporal variability in the hydrological cycle and water availability to ecosystem functioning. The last few decades have witnessed exceptional droughts and heatwaves on records Meanwhile, increasing tree mortality in drought-prone forest has been detected in Mediterranean areas. The holm oak forest dominated by Quercus ilex L., among the most emblematic forest in Mediterranean, has been subject to intense impact of enhanced drought period leading to productivity losses and in some cases to high mortality rates. Consequently, it is crucial to validate and assess the impact on productivity and mortality rates of Mediterranean holm oak forest following prolonged summer drought periods, and provide innovative tools of early detection through remote sensing data. In this study, we investigated the effect of summer drought periods on the productivity and mortality rates of holm oak forest in Sardinia (Italy) combining multispectral Sentinel-2 satellite and with very-high spatial resolution PlanetScope imagery, together with meteorological ERA5 dataset. Our results highlighted a decrement of summer precipitation and an increment of summer temperature between 2–4 °C over the last couple of decades in Sardinia compared to climate normal over 1971-2000. Furthermore, the differences of summer Normalized Difference Vegetation Index (NDVI) values between 2022 and 2024, and validated through visual inspection of coeval PlanetScope imagery allowed to identify with high accuracy holm oak forests impacted by the effects of recent climate change. The majority of productivity losses and mortality rates on holm oak both in terms of intensity and extension was highly correlated with the increment of climate anomalies registered in Sardinia. This study supplies an efficient management tool for the early detection and mapping of holm oak response to climate change.

The dwarf pine (Pinus mugo ssp. mugo Turra) is a key species in the dynamics of treeline ecotones within alpine environments. Understanding the factors driving growth and changes in land cover is crucial for accurately assessing current biomass levels and developing effective management strategies for this species. This study aims to create a historical mapping of dwarf pine in the Sarntal Valley,where it has gained significant economic interest in recent decades due to the growing demand for its essential oil. Additionally, there is an urgent need to establish a sustainable management plan for this species, which has yet to be subjected to regulatory measures concerning harvesting practices.

Effective monitoring of forests, particularly in response to climate and land-use changes, requires the analysis of long-term data. While advanced deep learning techniques have shown success with short time series of satellite imagery, utilizing extensive aerial imagery presents challenges, including variations in imaging technologies, sensor characteristics, and irregular data collection intervals. This study addresses these challenges by conducting multi-temporal mapping of dwarf pine over the past 75 years. We compare black and white aerial images with RGB orthophotos from 1945 to 2020, using an Artificial Neural Network supervised classifier. This classifier is augmented with textural measurements to develop a robust training layer for classification, followed by fine-tuning with a deep learning approach using a U-Net classifier. Our findings indicate that combining deep learning algorithms, grounded in problem-specific prior knowledge, can effectively monitor landscape changes through long-term remote sensing data.

Mountains serve as biodiversity hotspots owing to their island nature as high-altitude habitats in a sea of lowlands that renders them important evolutionary labs, the concentration of a wide range of environmental conditions in a relatively small area, and their unique climatic history. Considering their evolutionary and ecological importance along with their sensitivity to climate change and land degradation, mountain environments are in dire need of conservation. A first step towards this direction is the cost-effective monitoring of mountain ecosystems’ extent and condition trends using consistent remote sensing data time series. However, widespread adoption of such datasets still imposes certain challenges and more case-studies are needed to showcase their value, enhance capacity building and provide more detailed information to relevant stakeholders and policy makers. To this end, we estimated the Mountain Green Cover Index (SDG 15.4.2), developed by the United Nations under the 2030 Sustainable Development Agenda, at national and sub-national level for Greece in years 2000, 2005, 2010, 2015, 2018, 2021 utilizing land cover data and a digital elevation model. While the index remained almost stable at national level throughout the years, its further disaggregation shows higher fluctuation in specific regions indicating the uneven distribution of pressures on mountain ecosystems, like urbanization and wildfires, within the country. Our results reiterate the need for localization of SDG reporting and further incorporation of earth observation data in ecosystem monitoring in order to facilitate the design and implementation of effective policy and conservation measures for mountain ecosystems.

During 2022, Bohemian Switzerland NP was affected by the largest wildfire in the Czech Republic throughout its modern history. This landscape of sandstone towers, traditionally occupied by pine and beech forests, was a subject of massive plantation of Norway spruce and non-native Pinus strobus since the 19th century. A series of weather extremes in the last years caused an exceptional drought and consequent massive bark beetle outbreak and spruce die off, followed by the catastrophic wildfire event, being a rather uncommon phenomenon in Central Europe. The area serves as a perfect model situation to study the role of species composition, bark beetle and water availability on the fire dynamics, impact on biodiversity and natural regeneration. Pre-fire vegetation state, fire severity and post-fire regeneration were assessed using a combination of remote sensing sources (satellite, aerial and drone multispectral and Lidar data) and field surveys (species composition, fire severity). High resolution remote sensing data enable us to study both disturbance and post-fire regeneration in detail relevant for the underlying ecological processes. Our research revealed relationship between pre-fire forest conditions (composition and health) and both fire disturbance and regeneration, disturbance being the lower at native deciduous tree stands and waterlogged sites, severe at standing dead spruce and the strongest at dry bark-beetle clearings covered by a thick layer of litter. Derived information on fire severity, detailed 3D stand structure and health status are to be used as a proxy of the fire disturbance impact on biodiversity and to explain regeneration patterns.

Well-functioning coastal marine environments provide a wide range of environmental services, such as habitats for marine life, fishing opportunities supporting local livelihoods, recreation, biodiversity and climate change resilience. Many human societies across the globe are located in the coastal region and consequently coastal regions are subject to significant human impact and many places coastal marine environments have been destroyed or depleted leading to significant reduction in biodiversity and consequently a drop in environmental services provided for. Simultaneously as the consequences of climate changes become ever more apparent as an increasing part of coastal societies face an increasing risk of enduring floods, coastal erosion with the risk of loosing homes and lives associated with it.

Increasing awareness regarding the importance of marine habitats is picking up. In turn, this calls for innovative solutions to monitor and provide decision support regarding management and restauration of marine habitats supporting both biodiversity and mitigating coastal risk.

DHI has developed a range of innovative remote sensing-based tools and services, now wrapped into an online tool called Coastal Mapper. This platform uses state-of-the-art satellite technology, AI and machine learning for mapping and monitoring coastal changes as they happen offering decision makers a science-based approach managing and restoring marine habitats as well as mitigating the impacts of climate changes reducing risk for many local communities.

Counting large animals traditionally relies on observations from airplanes or helicopters, which are both time-consuming and expensive. Recently, there has been significant advancement in satellite technology, with images achieving much higher resolution in both time and space. Additionally, more spectral bands have become available. Consequently, satellite images are becoming a cheaper and often better alternative, offering greater spatial and temporal resolution than both drones and aerial surveys, covering entire regions. This technological progress, coupled with a growing need to monitor global biodiversity—especially in remote areas—highlights the urgent requirement to explore and benchmark the capabilities of satellite data and modern computing power for developing biodiversity monitoring tools.

This poster provides an overview of methods and results from two projects—SpaceOx and SmartWhales—aiming to detect and count large animals from space. We explore the cutting-edge application of Very High Resolution (VHR) satellite imagery for Arctic and marine biodiversity conservation, presenting results from pilot study sites in Zackenberg, Greenland (Arctic), and Crystal River, USA (marine). VHR imagery was successfully utilized to monitor muskoxen and manatee populations, providing critical data for conservation efforts.

This poster highlights the pivotal role of advanced technology in protecting Arctic and marine life and fostering a sustainable future for global biodiversity. It demonstrates the transformative impact of Earth Observation data and modeling technologies on large animal conservation and biodiversity sustainability, especially in remote areas.

Giraffe populations have declined by around 40% in the last three decades. Climate change, poaching, habitat loss, and increasing human pressures are confining giraffes to smaller and more isolated patches of habitats. In this study, we aimed to identify; (1) suitable Masai giraffe (Giraffa tippelskirchi) habitats within the transboundary landscape of Tsavo-Mkomazi in Southern Kenya and Northern Tanzania; and (2) key connecting corridors in a multiple-use landscape for conservation prioritization. We combined Masai giraffe presence data collected through a total aerial survey with moderate resolution satellite data to model habitat suitability at 250 m resolution using species distribution models (SDMs) implemented in Google Earth Engine (GEE). Model accuracy was assessed using area under precision recall curve (AUC-PR). We then used the habitat suitability index as a resistance surface to model functional connectivity using Circuitscape theory and cost-weighted distance pairwise methods. Human habitat modification, rainfall, and elevation were the main model predictors of Masai giraffe habitat and corridors. On average, our 10-fold model fitting attained a good predictive performance with an average AUC-PR = 0.80 (SD = 0.01, range = 0.79–0.83). The model predicted an area of 15,002 km2 as potential suitable Masai giraffe habitat with17% outside protected areas. Although Tsavo West National Park formed a key habitat and a key connecting corridor, non-protected areas connecting Tsavo West and Tsavo East National Parks are equally important in maintaining landscape connectivity joining more than two Masai giraffe core areas. To maintain critical Masai giraffe’s habitats and landscape functional connectivity, especially in multiple-use landscapes, conservation-compatible land use practices, capacity building, and land use planning should be considered at the outset of infrastructure development. This modeling shows the potential of utilizing remotely sensed information and ground surveys to guide the management of habitats and their connecting corridors across important African landscapes, complementing existing efforts to identify, conserve, and protect wildlife habitats and their linkage zones.

In the recent decades, the Northern Adriatic Sea (NAS), one of the most productive areas of the Mediterranean Sea, faced several changes in both the trophic status and phytoplankton community structure related to anthropogenic and meteoclimatic pressures. Among the latter, ocean warming and marine heatwaves (MHW) are expected to have an important impact. The aim of this study was to highlight the trends of Sea Surface Temperature (SST) and chlorophyll-a (chl-a, proxy of phytoplankton biomass) and analyse the effect of ocean warming and marine heatwaves on phytoplankton biomass in the Northern Adriatic Sea. Increases and decreases of SST and chl-a were observed in the entire NAS, respectively, with a marked seasonal variability. Chl-a trends showed a strong spatial variability, with the highest decrease along the western coast. Spatial and seasonal variability of MHWs mean values and trends were also observed. Lagged correlations highlighted a different response of chl-a to SST anomalies along time, with a spreading of negative correlations throughout the NAS with subsequent lags, and positive correlations in eutrophic lagoonal areas. Different case studies and cluster analysis were used to assess the effects of ocean warming, also related to MHWs, on phytoplankton biomass. The relationships varied based on the background trophic conditions: in oligotrophic regions, marine heatwaves and extreme heat conditions led to reduced chlorophyll-a concentrations, while in eutrophic areas, such as the western coast and lagoons, an increase in phytoplankton biomass was observed. Our results indicated that MHWs and SST increases, are among the factors that are negatively affecting the phytoplankton communities of the NAS, although the interpretation of the effects is complicated by the fact that local phytoplankton dynamics are shaped by the relevance of many other factors more or less T dependent, such as air-sea heat fluxes, water column stability, rain regime, river discharge.

An accurate spatial distribution of forest species composition is essential for regional biodiversity monitoring. When combined with filed structural metrics (e.g., basal area and canopy height), species distribution data enhance estimations of ecosystem functions their relationship with biodiversity, supporting territorial planning, biodiversity assessment, and forest management. Advancements in high spatial resolution (HSR) remote sensing have demonstrated the effectiveness of deep-learning classification models, furthered by rapid artificial intelligence (AI) developments. In addition to AI-based methods, research has significantly advanced thanks to mechanistic Joint Species Distribution Models (JSDM) and community assembly models, particularly when integrating functional traits and phylogenetic data. These models offer relevant insights into environmental filtering and competitive interactions within ecosystems. Here, we present an approach that combines JSDM with AI-based algorithms to map forest species composition, relative abundance, and basal area across Italy. Our model incorporates Earth Observation (EO) data inputs such as maximum, minimum, median, 10^th, and 90^th percentile NDVI retrieved using Sentinel2 satellite images, phenological patterns, and canopy height, as well as species functional traits, Community Weighted Means, Functional Diversity Index, and phylogenetic distances. Additionally, it integrates pedo-climatic variables to enhance predictive accuracy. An additional Machine Learning algorithm based on association rule learning will be investigated. Association rules can provide additional insights due to their inherently explainable structure, allowing for clearer interpretation of relationships within the data. Preliminary results include a comparison between a pure AI approach and a hybrid model integrating process-based and AI methodologies, demonstrating the strengths of each approach in modeling complex forest ecosystems.

The expansion of unpaved roads followed by poorly planned crossings disrupts the eco-hydrological connectivity of streams. Road-stream crossings impact the flow of water and sediment, instream habitat, and species movement. So far, the number of crossings in the Amazon are largely underestimated due to challenges in accurately mapping these small structures. The aim of this study was to analyze the historical impact of road-stream crossings on the eco-hydrological connectivity of Amazonian streams. We calculated land use and land cover data from 1987 to 2023 from the MapBiomas project. We used Planet satellite imagery to manually map ca. 16,000 km of roads to identify intersections with hydrography data of headwater streams in the municipalities of Santarém and Paragominas, Brazil. We pre-identified 2,205 intersections, most of which located in agriculture landscapes. We then drove more than 12,000 km on unpaved roads to validate the intersections and characterize the associated infrastructure (e.g. structure type, alterations in channel morphology, habitat lentification). On average 27% of the mapped intersections were absent in the field, highlighting the importance of ground-truthing the estimates. The most common crossing structures found were culverts (56% Santarém and 47% Paragominas) followed by single span crossings (28% and 38%, respectively). These validated data were used to adjust the calculation of the Dendritic Connectivity Index and the predominance of culverts led to steep falls in eco-hydrological connectivity. While this was expected in highly deforested catchments (57% loss), catchments with high forest cover also experienced 30% loss of connectivity over the study period. Our results show that road-stream crossings need to be recognized as a threat to the eco-hydrological connectivity of Amazonian streams. Given the essential value of connectivity to freshwater biodiversity, crossings should be managed through better-planned structures. Moreover, the removal of abandoned or underutilized crossings could help restoring connectivity, benefiting freshwater biodiversity.

Monitoring microbial plankton abundance and diversity provides valid indications for assessing the health of the marine pelagic habitat. Photosynthetic plankton is responsible for almost 50% of the primary production of the planet, being fundamental for the functioning of marine food webs and biogeochemical processes in marine ecosystems. Ubiquitous highly-diverse heterotrophic microbes are essential to metabolise the diverse compounds that constitute the dissolved and particulate organic matter pools, participate in the biological carbon sequestration and contribute to the biogeochemical cycles. However, the effective assessment of microbial plankton diversity is suffering from lacking observations at high spatial and temporal coverages that are not achievable by in situ sampling. The PETRI-MED project, funded through the European Biodiversity Partnership BIODIVERSA+, aims to develop novel strategies to synoptically assess status and trends of plankton biodiversity in coastal and open waters of the Mediterranean Sea.

This is achieved following a multidisciplinary approach capitalizing on the large potential offered by the past and ongoing satellite missions (e.g., Copernicus Sentinel-3), complemented with field measurements of OMICS-based taxonomy, biogeochemical models and emerging Artificial Intelligence technologies. PETRI-MED is thus going to: 1) develop a novel observation system to assess marine plankton biodiversity status and trends, and ecological connectivity among areas, that deals with specific user needs identified within the project and European policy indications; 2) enhance our fundamental understanding and predictive capabilities on plankton biodiversity controls and sensitivity to natural and environmental stressors; 3) contribute towards science-based solutions in support of decision making for sustainable marine ecosystem management strategies.

The ocean, covering about 72% of the Earth's surface, plays a critical role in global biodiversity and climate systems. Consistent changes in ocean biodiversity can have irreversible impacts on marine food webs and climate feedback mechanisms. Such changes demand urgent attention in fisheries management and ecosystem sustainability. Climate change induces various alterations in ocean environments, including frequent extreme warming events, increased stratification, altered river discharges, and accelerated polar ice melt. To understand how a warming climate impacts marine biodiversity, long-term satellite ocean color data are indispensable for detecting these changes.

This study aims to distinguish the changes in ocean color due to anthropogenic climate factors from those resulting from natural variabilities (e.g., seasonal cycle, ENSO, etc.). We introduce a novel approach, the Ocean Physical Modes projection to Ocean Color, which utilizes the Extended Reanalysis Sea Surface Temperature to define climate-related ocean physical modes. This analysis helps identify the natural variability signals in ocean color that may obscure climate change trends.

Our findings indicate continuous optical shifts in the global ocean due to climate change. In the Northern Hemisphere, the water appears bluer in less productive tropical oceans and greener in more productive, high latitudes. These changes likely have significant impacts on ecosystems and fisheries.

Crop pollination is one of the most important ecosystem services for the food industry, as approximately 80% of global pollination is dependent on wild bees. However, the expansion of agricultural land has led to a decline in native bee populations, resulting in a pollination deficit for both native plants and agricultural crops. Improving connectivity in agricultural landscapes is essential to achieving sustainable agricultural production. To address this, it is necessary to assess pollination services by analyzing the functional connectivity of the landscape using multiple spatial dimensions. Field sampling often fails to capture floral resources at different spatial scales. Quantifying floral resources in both agricultural and non-agricultural habitats provides insight into what constitutes high quality habitat for bees, and creates opportunities to assess pollination availability over time and space. Therefore, in this study, we aim to 1) predict spatial variation in floral density and bee abundance at multiple spatial scales in agricultural landscapes and 2) assess the relationship between functional connectivity and bee abundance in these landscapes. To achieve this, Sentinel-2 data and time series of phenology and community composition were processed through predictive models to estimate bee abundance, floral density, and phenological diversity. The Omniscape model was then used to calculate movement fluxes and generate a connectivity map. Finally, priority areas for restoration and conservation were identified by categorizing pixels based on their intervention potential. The results of this research provide insights for land use planning and natural resource management in central Chile, contributing to the conservation of pollination services and improving landscape connectivity to increase agricultural productivity.

Coastal areas are transitional environments between land and sea, which are important biodiversity hotspots. Numerous threats put this fragile ecosystem at risk. Remote Sensing provides valuable support for describing and modelling landscape dynamics.

We conducted a multi-temporal landscape analysis focusing on the main processes of change that have shaped the Central Adriatic coast over the last 70 years, emphasizing the statistical assessment of these changes. We compared the dynamic processes and landscape changes inside and outside Long Term Ecological Research (LTER) sites.

The study area includes the Molise coast (central Italy) that hosts two LTER-protected sites (IT20-003-T: Foce Saccione-Bonifica Ramitelli and IT20-002-T: Foce Trigno–Marina di Petacciato) that are part of the N2K network (IT7222217 and IT7228221), along with comparably sized non-protected areas.

We digitized land cover maps at a scale of 1:5000 for the years 1954, 1986, and 2022, and calculated transition matrices denoting 16 dynamic processes (e.g. Urbanization, Agriculture Expansion, Forestation, etc.). We then compared changes between two time periods (1954-1986, 1986-2022) and analyzed the differences between LTER and non-LTER sites using a Random Forest model.

Most changes occurred during the first time step (1954-1986), while the landscape was less dynamic during the second time step (1986-2022). The LTER sites initially changed due to Agriculture Expansion, Urbanization, and Forestation, followed by a shift toward Naturalization in the second time step. Non-LTER sites, underwent more urbanization initially, followed by urban stability. This suggests that LTER sites are becoming more natural and rural, whereas urbanization has had a greater and lasting impact on non-LTER sites.

Our finding confirms the general trends of change occurring on Mediterranean coasts with clear differences inside and outside LTER protected areas. The implementation of machine learning procedures seems a promising quantitative approach to be implemented and tested across other landscapes and protection regimes.

Understanding the spatial structure of urban environments is critical for formulating spatial planning strategies, preserving ecosystem services, and maintaining biodiversity. Urban habitats differ substantially from natural habitats, subject to the pervasive influence of human activities and infrastructure, and to continuous transformation, due to the expansion and densification of urban areas and human activities. Urban green spaces are becoming smaller and more isolated, but are often still rich in biodiversity. We developed a tailored and innovative approach to provide a comprehensive representation of habitats across urban environments in Switzerland based on remote sensing data. By integrating ALS point clouds, aerial imagery and Planet satellite imagery with object-based image analysis (OBIA) and machine learning algorithms, we were able to map 8 functional urban green types (FUGT) based on vegetation height, density, structure and seasonal dynamics: three types of grass; shrubs and bushes; two types of trees; buildings with green roofs; and sealed surfaces. We analyzed the composition and spatial configuration of the FUGT patch mosaics in 3 large Swiss cities (Zurich, Geneva, Lugano) in randomly selected test areas. The structural metrics were calculated using FRAGSTATS software for each test area and for each FUGT within the test area. Finally, we compared the structural diversity within each city, and between the three investigated cities. The presented approach may support biodiversity conservation and effective land management strategies, in particular development and implementation of targeted conservation measures to mitigate the impacts of habitat fragmentation in urban environments.

The beginning of 2024 marked the publication of Croatia’s official map of coastal and benthic marine habitats, covering the national coastal sea and Croatian Exclusive Economic Zone (EEZ). One of the most comprehensive projects of its kind in Europe, this map spans 51% of the Adriatic Sea under Croatian jurisdiction, or approximately 30,278 km². The map is available in three scales (1:25,000, 1:10,000, and 1:5,000), varying among different marine areas based on protection levels and other criteria.

The mapping primarily relied on Remote Sensing, integrating Satellite-based Earth Observation and Aerial Photogrammetry with spatial analytics tools. Remote Sensing was used for habitat mapping down to 20 meters, while deeper areas were mapped using acoustic methods, supplemented with data from over 4,000 in-situ transects.

To achieve high spatial resolution and detailed content (up to the 5th level of the National Classification of Marine Habitats), advanced Remote Sensing data processing methodologies were employed, including Pixel-Based Image Analysis (PBIA) and Object-Based Image Analysis (OBIA). OBIA enabled detailed segmentation and habitat delineation using ortho-maps from aerial photogrammetry at 0.5 m resolution. PBIA utilized 110 seasonal multispectral Sentinel-2 images to analyze seasonality and classify key species, particularly Cymodocea nodosa and Posidonia oceanica.

The fusion of these datasets was achieved using GIS tools and spatial statistics. The final product, which includes up to three habitat types per spatial feature, was generated using a custom-developed cartographic generalization algorithm, ensuring spatial, topological, and content accuracy and resulting in a high-resolution map and extensive database.

This map serves as a critical tool for future Natura 2000 site and protected area management, ecological network suitability analysis, marine resource management, and spatial planning. Its methodology also provides a replicable model for Mediterranean and global marine conservation, offering critical insights for biodiversity stakeholders addressing climate and anthropogenic pressures.

From the late 19^th century until the satellite ocean colour era, the Forel-Ule colour scale (FU) and the Secchi disk depth (Zsd) were used widely to characterize water colour and clarity. By using algorithms that transform satellite remote-sensing reflectances to FU and Zsd, these historical datasets can be combined with satellite records to confidently track long-term changes in ocean surface chlorophyll. Here, we apply this approach to compare ocean colour dynamics in the Red Sea between three periods: the historical Pola expedition (1895-1898), the Coastal Zone Color Scanner (CZCS) era (1978-1986), and the more recent continuous satellite ocean colour period (1998-2022). Specifically, we combined historical in-situ FU and Zsd measurements with FU and Zsd derived from CZCS and Ocean Colour Climate Change Initiative (OC-CCI) reflectance data, using algorithms tailored specifically to these two products. Our analysis reveals that the Northern Red Sea (25^o–28^oN) is becoming greener in response to environmental changes. This observed increase in productivity is linked to a deeper mixed layer in the cyclonic gyre prevailing in the region, associated with increased ocean heat loss. Additionally, we report an extended phytoplankton bloom season in the recent period (~three weeks longer duration) following stronger mixing in early spring. Our findings suggest that, despite the upward trend of ocean warming documented in the region, expected to strengthen thermal stratification and decrease productivity, dynamic features such as gyres can significantly enhance vertical mixing, evoking unforeseen impacts on nutrient distribution and phytoplankton growth.

Canada´s forests are being affected by a changing climate in many ways including insect infestation, tree dieback and increased fire activity across Canada. Early springs and longer summers are impacting trees phenology cycle, and vulnerability of certain tree species versus adaptation of others will determine the future of forest composition and productivity, and ultimately its resilience to climate change.

Access to up-to-date information about tree-species composition, spatiotemporal variability, and response to natural and anthropogenic disturbances, is needed to enable sustainable management for current and future generations. However, visual interpretation of aerial photography remains the basis of tree species mapping in forest inventories. This tedious process faces various challenges, such as a long processing time, budget constraints, limited skilled personnel, data availability and quality of aerial photography.

Advances in machine learning and the growing number of hyperspectral space missions (e.g., ASI/PRISMA, DLR/EnMAP, Planet/Tanager-1 and the future ESA/CHIME), providing higher spectral, temporal, and radiometric resolutions, offer a unique opportunity towards time-efficient mapping of tree species from space.

Within this context, this work addresses the development of a tree-species mapping methodology that leverages deep learning and multi-temporal hyperspectral data.

Two airborne data collections using the Fenix 1K hyperspectral instrument, conducted in summer 2019 over a test site located in Quebec, and LiDAR-based elevation data were used for this purpose. A hybrid model based on the integration of autoencoder deep learning and Random Forest was developed. Forest inventory and ground data, available through the Quebec Forestry department, were leveraged to support model training/testing and accuracy assessment of the tree species classification.

The effect of multi-date data on classification accuracy was assessed using: 1) a July data collection corresponding to a peak season scenario, and 2) both July and October data collections as a bi-temporal scenario, where the senescence effect is also included.

Ecological connectivity is a fundamental trait of ecosystems, essential for maintaining their integrity and resilience. Therefore, within the global biodiversity framework, the importance of maintaining and restoring connectivity has been emphasized, which becomes especially relevant given the accelerated loss of natural areas. In this study, we develop a methodology based on circuit theory, where species movement is modeled as an electrical flow that propagates through the landscape. The landscape is represented as nodes connected by resistors, which are electrical components that conduct current with varying efficiency. The likelihood of a species moving from one node to another depends on the landscape's resistance, which is modeled based on various spatially explicit covariates, some obtained directly from remote sensors and others from secondary data.

The methodology was applied to the Lipa wetland system, located in northeastern Colombia, an area rich in biodiversity and important for connectivity between the Andes and the Orinoquia. For the functional analysis, six species classified as vulnerable and endangered on the IUCN Red List were identified, representing different environments within the wetland system and various biological groups, including three mammals, two birds, and one reptile. Covariates affecting species mobility were evaluated by experts on each prioritized species to ultimately obtain the resistance specific to each species. Based on this information, a connectivity algorithm was applied using the Circuitscape package in Julia, with peripheral nodes used to model the probability of species movement in all directions (omnidirectional connectivity).

Finally, an extension of the method is proposed using a Principal Component Analysis (PCA), which synthesizes the connectivity information produced for different species and highlights strategic areas for connectivity, facilitating its interpretation to efficiently guide biodiversity conservation decisions.

In this study, we present DeepMaxent, a new approach for species distribution modelling (SDM) that extends the traditional maximum entropy framework (Maxent) by integrating it into a neural network for representation learning. DeepMaxent takes advantage of the flexibility of neural network learning to capture the complex, non-linear relationships in species-environment interactions, while retaining the probabilistic underpinnings of Maxent. A very recent study has already shown the promising effectiveness of this approach on the dataset used extensively to compare SDM methods (Elith et al. 2020). In this presentation, we explore its application on larger-scale datasets, in particular a dataset called GLC2024 dataset, which includes environmental covariates derived from Landsat data. Our model is trained using presence-only data and evaluated on presence-absence data using the area under the curve (AUC) metric to compare performance. We are also conducting an in-depth ablation study to assess the impact of model depth, batch size and other hyperparameters, particularly in the context of large datasets. Our results indicate that DeepMaxent performs well when dealing with large amounts of data, underlining its potential for SDM.

Currently, usable data on changes in historical habitat parameters over time is lacking in order to easily integrate them into biodiversity analyses, e.g. to relate recorded changes in the occurrence of species (groups) to environmental changes. Also, it is currently difficult to create predictive models with available environmental and particularly land use datasets that are able to predict past species occurrences. However, such information is important for supplementing biodiversity monitoring programmes and to allow improved statements on the causes of observed biodiversity trends. To fill this gap, we present an innovative joint project between the German Environment Agency’s Application Laboratory for Artificial Intelligence and Big Data and the German Federal Agency for Nature Conservation, including initial results. Here, a prototype tool will be developed for deriving and quantifying relevant habitat changes from historical aerial photographs and satellite data, using the example of grasshoppers in Germany. By analysing Essential Biodiversity Variables (EBVs) in multimodal and -temporal manner, we want to gain a better understanding of the population trends in grasslands. The results will enable better integration of land-use change and ecosystem dynamics into retrospective analyses of grasshopper diversity. For example, historical habitat parameters such as the structural diversity of an area (habitat heterogeneity) could be calculated with a pixel-based analysis of historical aerial photographs and satellite data. Other relevant parameters include land sealing, scrub encroachment, vegetation height or open patches. Both Germany-wide satellite images and heterogenous aerial images from different federal states and years will be analysed. The future algorithm shall analyse these as individual images and as time series in order to quantify temporal changes in the habitat parameters. Overall, there is great potential to strategically improve the data basis and evaluation options for historical land use by remote sensing so that they can be better combined with biodiversity data.

Modelling species distribution is critical for the management of invasive alien species, as reliable information on habitat suitability is essential for effective conservation and rehabilitation strategies. This study aims to model the suitable habitat and potential distribution of the notorious invader Lantana camara (Lantana), in the Akagera National Park (1 122 km2), Rwanda, a savannah ecosystem. Spatio-temporal patterns of Lantana from 2015 to 2023 were predicted at 30-m spatial resolution using a presence-only species distribution model in the Google Earth Engine, implementing the Random Forest classification algorithm. The model incorporated remote sensing predictor variables, including Sentinel-1 SAR, Sentinel-2 multispectral data, and socio-ecological parameters, as well as in situ presence data. A maximum of 33 % of the study area was predicted to be suitable Lantana habitat in 2023. Habitat suitability maps indicated higher vulnerability to Lantana invasion in the central and most northern, and southern parts of the study area compared to the eastern and western regions for most years. Change detection analysis revealed an increase in habitat suitability in the northeastern sector and a decrease in the southwestern part of the park over the study period. The predictive performance of the model was robust, with AUCROC values ranging from 0.93 to 0.98 and AUCPR values ranging from 0.79 to 0.94. Key factors influencing Lantana habitat suitability in the Akagera National Park included the road network, elevation, and soil nitrogen levels. Additionally, the red edge, shortwave and near-infrared spectral bands were identified as important variables within the Random Forest classification, highlighting the effectiveness of combining remote sensing and socio-ecological data with machine learning techniques to predict invasive species distributions. These findings offer valuable guidance for developing effective conservation strategies to protect savannah ecosystems and mitigate Lantana spread in the future.

1. Introduction
Land use is the main driver of biodiversity loss (Díaz et al., 2019). This study investigates the impact of land use intensification on biodiversity from 2005 to 2022, a period marked by increasing global food demand. Using remote sensing-derived products, such as data on land use, N-fertilization, water use, and harvest intensity, we measure changes in land use and intensity to assess their effects on biodiversity loss. Our analysis identifies critical biodiversity hotspots and emphasizes the need for refined impact assessments through enhanced characterization factors.

2. Methods
We compiled a global dataset on land use intensities from satellite sources like HILDA+ and various spatial datasets for crop water use and fertilization (e.g., Winkler et al., 2020; Adalibieke et al., 2023; Mialyk et al., 2024). This data enabled the evaluation of land use intensification across different land types, including:

• Crops (fertilization, irrigation, harvest intensity)
• Pasture (N-input)
• Plantations (size, fertilizer use)
• Managed forests (size, harvest intensity)
• Urban areas (size)

We applied characterization factors from Scherer et al. (2023), covering five species groups (plants, amphibians, birds, mammals, and reptiles) and five broad land use types across three intensity levels (minimal, light, and intense). These factors allowed us to calculate the potential species loss (PSL) per ecoregion.

3. Results
Initial findings reveal that biodiversity loss due to land use is approximately 1.9 times higher than previously estimated. We identified biodiversity loss hotspots in regions such as Brazil and Eastern Africa, where intense land use correlates with substantial biodiversity declines. In 2015, the potential species loss (PSL) was around 17%. Regions with underestimated PSL, such as South America, Southeast Asia, and parts of Africa, indicate the need for improved assessments.

Land use types and regions that showed significantly higher PSL-values considering land use intensities are pasture, cropland and plantations, especially in South America and Southeast Asia. Furthermore, our data show that biodiversity impacts have risen over the last 20 years due to the intensification of agriculture. These findings suggest that models excluding land use intensities may underestimate biodiversity impacts, particularly in regions experiencing rapid agricultural expansion and trade-driven changes.

4. Discussion
Our findings underscore the critical need to refine biodiversity impact assessments by accounting for land use intensities and incorporating additional remote-sensing products. Identifying biodiversity hotspots through improved characterization factors supports targeted conservation efforts in areas most affected by land use intensification. Additionally, shifts in ecosystem structure, detectable through changes in land use and vegetation indices, highlight the complex relationship between land use, trade, and biodiversity loss. This research provides valuable insights for policy development aimed at mitigating biodiversity impacts, especially in high-trade regions.

5. Conclusion
This study emphasizes the importance of integrating remote-sensing data and land use intensities into biodiversity assessments. Our findings indicate significant biodiversity losses linked to land use intensification, underscoring the need for accurate indicators to inform effective conservation strategies in response to growing food demand and environmental pressures.

Adalibieke, W., Cui, X., Cai, H., You, L., and Zhou, F. (2023). Global crop-specific nitrogen fertilization dataset in 1961-2020. Scientific Data, 10(1):617.

Díaz, S., Settele, J., Brondízio, E. S., Ngo, H. T., Agard, J., Arneth, A., Balvanera, P., Brauman, K. A., Butchart, S. H. M., Chan, K. M. A., Garibaldi, L. A., Ichii, K., Liu, J., Subramanian, S. M., Midgley, G. F., Miloslavich, P., Molnár, Z., Obura, D., Pfaff, A., Polasky, S., Purvis, A., Razzaque, J., Reyers, B., Chowdhury, R. R., Shin, Y.-J., Visseren-Hamakers, I., Willis, K. J., and Zayas, C. N. (2019). Pervasive human-driven decline of life on earth points to the need for transformative change. Science, 366(6471).

Mialyk, O., Schyns, J. F., Booij, M. J., Su, H., Hogeboom, R. J., and Berger, M. (2024). Water footprints and crop water use of 175 individual crops for 1990-2019 simulated with a global crop model. Scientific Data, 11(1):206.

Scherer, L., Rosa, F., Sun, Z., Michelsen, O., de Laurentiis, V., Marques, A., Pfister, S., Verones, F., and Kuipers, K. J. J. (2023). Biodiversity impact assessment considering land use intensities and fragmentation. Environmental Science & Technology, 57(48):19612–19623.

Winkler, K., Fuchs, R., Rounsevell, M. D. A., and Herold, M. (2020). Hilda+ global land use change between 1960 and 2019.

The need to identify and control invasive species to protect native biodiversity is a major challenge for ecologists and conservationists. The European water lily (Nymphoides peltata) has established itself as a neophyte in Swedish waters, competing with native species for habitat and potentially disrupting the ecological balance. This can impact biodiversity and human activities such as fishing, swimming, and boating. Detecting and preventing its spread is therefore crucial for the protection of aquatic ecosystems and the species they support. Traditional field surveys for water lily detection have been conducted at selected areas but are expensive and time-consuming, creating a demand for more efficient methods to monitor its distribution and prioritize management efforts. A big challenge is to detect the occurrence of water lily where is not known yet because of remote and non-monitored lakes.

This is where Earth Observation can help and support water managers. For detecting the water lily, we use a Random Forest algorithm, a supervised machine learning method suitable for regression and classification tasks. Sentinel-2 data helps track the spread of invasive species over large areas. Nymphoides peltate develops very characteristic yellow flowers and provides therefore a unique spectral signature which facilitates remote sensing detection distinguishing it from other plants in aquatic ecosystems. The identified spots from our analysis have already been utilized by local authorities, benefiting from the advantages of this approach. The use of remote sensing supports the development of more effective management strategies by Swedish county administrations, aiming to minimize the impact of the European water lily on local biodiversity. This case serves as a model for monitoring neophytes that exhibit spectral differences from native ecosystems.

Forest-savanna transitions are among the most widespread ecotones in the tropics, supporting substantial unique biodiversity and providing a variety of ecosystem services. At the same time, both forests and savannas are experiencing rapid changes due to global change, potentially endangering both biodiversity and ecosystem services. However, forest-savanna transition zones have received relatively little focus from researchers compared to the core areas of these biomes, limiting our ability to understand change and act to conserve these areas effectively. A comprehensive understanding of the distribution and drivers of change within the forest-savanna transitions is therefore a key step for their successful conservation. Here we conducted the first satellite data-driven mapping of natural forest-savanna transition zones on a global scale using vegetation structural variables. By calculating rate of change of tree cover through space across the tropics, we identified 22 unique savanna-forest transition zones – three in Australia and Asia, eight in Africa, and eleven in South America. Next, we described the climatic space in which these transition zones occur and quantified environmental drivers which have been shown to influence forest-savanna coexistence such as fire occurrence, hydrological dynamics and soil properties to understand the relative importance of these drivers across the different zones. We also quantified the degree of patchiness and pattern formation to assess how common mosaics are within these zones. Finally, we evaluated how existing maps used for conservation planning overlap with our mapping of forest-savanna transitions. This work represents the first step towards understanding the distribution and ecosystem processes within the forest-savanna transition zones on a global scale. The mapping will serve as a basis for further investigation into the spatiotemporal dynamics of forest-savanna transition zones and help inform ecosystem conservation efforts in the tropics.

A Data Space is a framework that supports data sharing within a data ecosystem defined by a governance framework. It facilitates secure and trustworthy data transactions, emphasising trust and data sovereignty. The Green Deal Data Space is the EC solution to support Green Deal policies with relevant data and to contribute to better environmental transparency and better decision-making. The European Green Deal is a package of policy initiatives with the ultimate goal of reaching climate neutrality by 2050, which in the case of the biodiversity strategy 2030, aims to create and integrate ecological corridors as part of a Trans-European Nature Network to prevent genetic isolation, allowing for species migration and to maintaining and enhancing healthy ecosystems, among other goals.

Taking Terrestrial Habitat Connectivity in Catalonia as a policy driven testbed, some solutions are explored to derive connectivity from a pixel-based LULC approach combined with on the field information such as GBIF in-situ data and sensor camera trapping. Special care is being put in semantic tagging uplift using Essential Biodiversity Variables, as well as standard APIs to manage data and metadata. Entrusted and secured mechanisms are also carefully considered when sharing sensible species information.

This work is done under AD4GD EU, Switzerland and United Kingdom funded project (nº 101061001).

Habitats are a combination of abiotic factors and biophysical structures that host biodiversity and various nature’s contributions to people. As they become more and more damaged by human activities, they are also common conservation and restoration targets. Yet, their spatial extent and potential areas for restoration are not well mapped yet, which hinders their effective preservation and restoration.

This lack of knowledge stems from two main issues. First, modeling multiple habitats that are mutually exclusive at fine scale but that can co-occur at the landscape level is a challenging task. Second, there are strong imbalances in habitat extent (some habitats are very rare while others are very common) that further complicate well-parameterising a single multi-class model. Consequently, current habitat maps are limited in either their spatial or thematic resolution and extents so far.

Harnessing high resolution remote sensing data with Artificial Intelligence techniques has the potential to address these challenges and improve the quality of habitat models. Here, we aimed to compare various options to optimize the use of remote sensing and AI tools to model habitats at high thematic and spatial resolution and at large spatial extent. To illustrate that, we model EUNIS habitats at level 3 across Europe using the European Vegetation Archive. We validate the modeling strategies and compare them on independent habitat observations and regional maps.

We modeled habitat classes based on climate, terrain, hydrology, soil predictors, and incorporated ecosystem descriptors from high-resolution remote sensing products. Using deep learning algorithms, we explored the extraction of additional features from raw multi-spectral images. We evaluated various classification strategies, including binary, multi-class, and hierarchical approaches, each varying in their constraints on habitat co-occurrence.

Our results revealed distinct recall and precision trade-offs with different classification strategies. The integration of remote sensing-based predictors significantly improved the overall predictability of habitat models, with varying impacts across habitat classes. Additionally, the inclusion of multi-spectral images enhanced the recall of most habitats, emphasizing the importance of spatial landscape structure for habitat suitability. In conclusion, we advocate the use of high-resolution remote sensing imagery alongside AI for habitat mapping at large extents.

Two projects led by the Swedish Forest Agency and the Swedish Environmental Protection Agency have tested methods for mapping two groups of woodland Annex I habitats, each with unique challenges. Annex I habitats have detailed descriptions that are often difficult to capture in remote sensing data or models. However, for these habitat groups, careful feature engineering, neural networks, and expert-curated reference data have enabled effective mapping.

In this approach, feedforward neural networks (FNNs) were trained to classify habitat types by integrating Sentinel-2 imagery, lidar data, topographic information, soil maps, and land cover data. By using continuous tree species composition data previously modeled from Sentinel-2 time series, the model was made lightweight and transferable, providing pixel-wise probability scores for habitat occurrence from 0 to 100. Monte Carlo dropout was also implemented to improve output gradients and boost model performance.

Feature engineering helped translate domain expertise into indicators the network could interpret, such as remapping soil classes and constructing hydrological models. Careful reference data selection and iterative updates based on intermediate results were vital for model accuracy. Validation with local and habitat mapping experts demonstrated promising accuracy, supporting its use in conservation planning. This approach not only makes habitat monitoring more efficient but also offers a scalable, cost-effective solution for Annex I habitat mapping, aiding decision-makers in biodiversity conservation and land management.

Grasslands are crucial globally for their ecosystem services. They are essential for the meat industry, providing the main food source for animals like cows. However, grasslands are rapidly disappearing due to woody plant encroachment (WPE), one of the leading causes of grassland loss after conversion to cropland. WPE is a subtle and challenging threat to reverse, posing significant risks to grassland species and habitats, ranchers, the economy, and society.

Our project leverages machine/deep learning and cloud computing on multi-source satellite imagery (Landsat, Sentinel 1 & 2, Radarsat, etc.) to better detect WPE. Over the past four years, we have assessed optical remote sensing methods for WPE detection using field data and aerial imagery in Saskatchewan's grassland ecoregions (Canada). We aim i) to upscale this approach using multi-source satellite imagery to enhance early detection and ii) investigate factors driving WPE, iii) identifying the most vulnerable regions in Western Canada.

Our research will significantly enhance fundamental understandings of ecosystem dynamics. By investigating the drivers of WPE and its impacts, we will contribute to a deeper knowledge of grassland ecosystems, which is crucial for developing effective management strategies.

Sustainable grasslands are characterized by low woody plant cover. With growing consumer interest in sustainably produced goods, satellite remote sensing can provide an accurate and timely depiction of grassland sustainability with respect to WPE. Therefore, this project also aims to iv) assess the price premiums that ranchers can obtain by proving their products are produced on sustainable grasslands. Most importantly, we try to v) assess the environmental benefits related to biodiversity and climate change mitigation resulting from accurate WPE detection.

By aligning with the Kunming-Montreal Global Biodiversity Framework, we strive to provide results that support policy implementation for grassland biodiversity conservation. In our presentation, we will report on current work related to this project.

Global trends in marine turtle nesting numbers vary across regions, influenced by both environmental and human-induced factors. This study examines the potential impact of past temperature fluctuations on these trends, focusing on whether warmer beaches are linked to increased nesting activity due to higher female hatchling production, as sea turtles exhibit temperature-dependent sex determination (TSD), where warmer incubation temperatures produce more females. We chose the loggerhead turtle (Caretta caretta) for its wide distribution, strong site fidelity, and sensitivity to environmental changes. Using modelled air temperature data and satellite-derived land surface temperature (LST) data from the past four decades (1979 - 2023) across 35 key rookeries worldwide, we performed trend and correlation analyses to evaluate how temperature changes are reflected in the nesting activity trends of loggerhead turtles in these locations. Our findings suggest that rising temperatures are contributing to increased nesting activity in parts of the Caribbean, Atlantic, and Mediterranean (for instance, Cayman, Mexico, Brazil, Cyprus, and Turkey). While some regions experience short-term benefits, ongoing warming could lead to long-term population declines. This regional variability highlights how loggerhead turtles may respond to continued climate change, with current global increases in nest counts already reflecting the short-term effects of rising temperatures.

Bark beetle (Ips typographus, L.) outbreaks have become a major threat to forest ecosystems worldwide, exacerbated by climate change and resulting in significant economic and environmental damage. To minimize the impact of outbreaks it is crucial for forest management to implement ear-ly-detection measures. Remote sensing methods are a quantitative approach for monitoring the tree vitality and change. High spatial and temporal resolution satellite imagery, including multispec-tral data from platforms like Sentinel-2, allow for the inference of stress symptoms in trees, such as reduced photosynthetic activity and reduced vitality.

The objective of this project is to use satellite remote sensing data to reconstruct bark beetle out-breaks in South- and East Tyrol (Italy/Austria) since the Storm Event VAIA in summer 2018. The aim is to identify infestation “Hotspots”. Hotspots are areas in which bark beetle infestations were first identified and from which further spread is determined. The end product is a dispersion map with which the spread of the bark beetle infestation in this area is traced. Together with this project, an additional project is being carried out in which the focus is on physiological changes in the green-attack phase, which occur immediately after the infestation of the spruce, instead of structural changes, in order to detect an infestation earlier.

Satellite remote sensing (SRS) is essential for addressing several biodiversity-related challenges. It is suitable for detecting changes in ecosystem structure and highlights the impacts of bark beetle outbreaks for ecosystem functioning. Furthermore, SRS can contribute to an improved understand-ing of forest disturbances against the backdrop of climate change.

Wetlands and salt marshes are critical components of agricultural landscapes, supporting biodiversity, providing ecosystem services, and helping to mitigate the impacts of drought and flooding. However, since the 1970s, these habitats have been increasingly threatened by agricultural intensification, drainage, and mismanagement of water resources. This research focuses on the restoration of degraded wetland habitats in Moravian Pannonia, assessing both habitat heterogeneity and vulnerability to climate change using Earth Observation (EO) data.

The heterogeneity is assessed using the Spectral Variability Hypothesis, with satellite data from PlanetScope used to calculate Shannon entropy as a measure of spectral diversity. The analysis reveals higher spectral heterogeneity near ponds and along linear vegetation, whereas areas dominated by expansive species exhibit lower heterogeneity. These results emphasize the importance of promoting mosaics of smaller, diverse habitats to increase ecological resilience.

The climate change vulnerability assessment incorporates EO data from Landsat missions, meteorological data, hydrological and terrain modelling, and expert knowledge, following IPCC guidelines (exposure, sensitivity, and adaptive capacity). The findings indicate increasing exposure to rising air temperatures and prolonged droughts. The sensitivity is highest in water-dependent habitats and regions with sparse vegetation, while those with well-established water retention features demonstrate greater adaptive capacity. As the exposure and sensitivity to these climate stressors are expected to increase, enhancing adaptive capacity through improved water retention, supporting diverse plant communities, and promoting natural hydrological functions will be critical.

These insights will support adaptive management strategies and inform policy decisions to ensure the long-term sustainability of wetlands in the region.

Insects, highly diverse and abundant, regularly migrate long distances, connecting distant ecosystems and impacting global-scale processes. They play crucial roles in ecosystem functions like pollination and nutrient transfer, while also pose risks as agricultural pests and disease vectors. Migration and dispersal have also shaped evolutionary history, influencing current biogeographic distributions and species assemblages. Yet, accurately quantifying insect movement remains a challenge due to the dearth of reliable methods for tracking long-distance movements of these small, short-lived organisms. Additionally, our understanding of their taxonomy, biology, and distribution remains incomplete for many groups. Consequently, the true diversity of migratory insect species - and the full extent of their migratory behaviors - remains largely unknown. Here, we outline a methodological roadmap that integrates multiple disciplines to create probabilistic maps predicting potential migratory patterns of insects. Unlike vertebrates, which can often be tracked with real-time devices, insect migration research relies primarily on indirect geolocation methods to infer migratory origins and paths. We show the potential of combining complementary approaches to quantify i) spatial connectivity and ii) habitat dynamics. Spatial connectivity can be inferred through the analysis of stable isotopes, wind patterns, or genetic markers. Monitoring habitat dynamics, on the other hand, benefits from time-series remote-sensing satellite imagery, enabling us to model shifting habitat suitability over time. Applying this approach, we present case studies of notable long-distance insect movements. Ultimately, we envision a unified framework that combines diverse data sources to infer insect migratory dynamics, with the potential to scale up to automated monitoring systems for real-time ecological insights.

Coastal benthic habitats worldwide are increasingly affected by global environmental change, such as ocean acidification (OA) and marine heatwaves, alongside local stressors like pollution, habitat loss, bioinvasions, and overfishing. These stressors drive rapid shifts in biodiversity, community structure, and ecosystem functioning, particularly in ecosystems such as macroalgal forests, seagrass meadows, and rocky habitats. Integrating emerging remote sensing technologies into coastal benthic habitat mapping offers a much-needed opportunity to develop geospatial databases and quantify structural changes in these communities over long-term scales. In particular, the combination of close-range Structure-from-Motion (SfM), a powerful photogrammetric technique, coupled with recent image classification methods, has shown great potential for finely mapping complex benthic habitats, providing valuable insights for marine biodiversity conservation. This research focuses on coastal marine benthic habitats near the unique volcanic CO2 vent systems along the coast of Ischia Island (Naples, Italy). These CO2 vents cause local acidification and represent natural analogues to study potential future responses to OA across various ecological levels, habitats, and depths. We present preliminary data from aerial and underwater SfM-based imagery acquired through autonomous vehicles and SCUBA. Some examples of georeferenced raster datasets include orthomosaics and Digital Elevation Models (DEMs). Subsequently, the image analysis performed on these outputs will enable fine-scale mapping of the CO2 vent habitats in Ischia. As a further step, we aim to link the structural and topographic parameters (e.g., coral percent cover, colony size, and surface rugosity) derived from high-resolution imagery with ecosystem processes (e.g., photosynthesis, respiration, and calcification), providing novel insights into how benthic habitats respond to global environmental change.

Climate models project increasing frequency and intensity of droughts in the Mediterranean Basin, posing ecosystems under threat. Although adapted to water scarcity, Mediterranean ecosystems may be particularly vulnerable to extreme droughts as resource-limited systems. Furthermore, the Mediterranean region is a biodiversity hotspot, which, under normal conditions, provides resilience to ecosystems. However, biodiversity benefits may cease in more severe drought conditions.

The objective of this research is to examine the impact of diverse drought regimes on the response and resilience of Mediterranean ecosystems. We expect to detect a nonlinear relationship between drought regimes and vegetation response and the time since the last event to emerge as an impactful drought attribute. To this end, we employed an event-based approach to drought regime analysis, encompassing duration, intensity, severity, and time since the last event as drought attributes. Drought is evaluated through the Standardized Evapotranspiration-Precipitation Index at medium and long aggregation scales, with data retrieved from global downscaled re-analyses of the CHELSA database. We have analyzed the response of vegetation to drought events by extracting the temporal components of resistance, recovery, and resilience. The vegetation response is evaluated using the NDVI, EVI, NDWI and NIRV spectral indices from the MODIS multispectral sensor as vegetation functioning proxies.

We examined the 2001-2018 timeseries for the Tyrrhenian-Adriatic sclerophyllous and mixed forests ecoregion, to detect the functional shape of the vegetation response curve for this region. Our preliminary results suggest that drought detection can capture drops in vegetation productivity, yet not all of them, and that vegetation response components can depict different features of ecosystem response. With this research, we aim to contribute to a deeper understanding of the mechanisms that determine ecosystem resilience to climate change, providing insights that could inform conservation strategies and climate adaptation efforts in the Mediterranean.

The growing demand for scalable biodiversity monitoring methods has fuelled interest in remote sensing data, due to its widespread availability and extensive coverage. Traditionally, the application of remote sensing to biodiversity research has focused on mapping and monitoring habitats, but with increasing availability of large-scale citizen-science wildlife observation data, recent methods have started to explore predicting the presence of bird and plant species directly from satellite images. Here, we present a new data set for predicting species presence from sentinel-2 satellite data for a new taxonomic group -- butterflies -- in the United Kingdom, using the UK Butterfly Monitoring Scheme citizen-science data set. We experimentally optimise a convolutional neural network model to predict species presence directly from sentinel-2 satellite imagery, and find that this model especially outperforms the mean rate baseline for locations with high species biodiversity. To improve performance, we develop a soft, supervised contrastive learning loss that is tailored to probabilistic labels (such as species-presence data), and demonstrate that this improves prediction accuracy. Our method improves the model embeddings by aligning the similarity in species with the similarity in satellite images for pairs of locations. In summary, our new data set and contrastive learning method contribute to the open challenge of accurately predicting species biodiversity from remote sensing data, which is key to realising efficient biodiversity monitoring.

Land cover (LC) maps are used extensively for nature conservation and landscape planning, but low spatial resolution and coarse LC schemas typically limit their applicability to large, broadly defined habitats. In order to target smaller and more-specific habitats, LC maps must be developed at high resolution and fine class detail using automated methods that can efficiently scale to large areas of interest. In this work, we present a machine learning approach that addresses this challenge. First, we developed a multi-stage semantic segmentation approach that uses Convolutional Neural Networks (CNNs) to classify LC across the Peak District National Park (PDNP, 1439 km2) in the UK using a detailed, hierarchical LC schema. The entire PDNP was then mapped at 12.5 cm ground resolution using RGB aerial photography. High-level classes were predicted with 95% accuracy and were subsequently used as masks to predict low-level classes with 72% to 92% accuracy. Next, we used these predictions to analyse the degree and distribution of fragmentation of one specific habitat—wet grassland and rush pasture—at the landscape scale in the PDNP. We found that fragmentation varied across areas designated as primary habitat, highlighting the importance of high-resolution LC maps provided by CNN-powered analysis for nature conservation.

Sand Tracer is an innovative tool that utilises satellite remote sensing to enable precision management of sand dunes, addressing critical drivers of biodiversity changes and enhancing coastal protection against sea-level rise.

Sand Tracer integrates high-resolution satellite imagery and LiDAR data, leveraging artificial intelligence (AI) to provide detailed insights into dune dynamics. By monitoring and estimating sand displacement volumes across both space and time, Sand Tracer provides a near-monthly depth estimate at approximately 1x1m resolution. This granular data surpasses traditional, coarse radar-based approaches, allowing for precise assessment of the impacts of dune management practices on island and coastal biodiversity and the protective function of dunes.

Incorporating abiotic factors such as wind conditions further refines the analysis, enabling stakeholders, including provincial authorities, land managers, and national water management agencies, to develop targeted management strategies based on robust biodiversity indicators. This frequent and detailed monitoring capability empowers stakeholders to adapt practices, supporting Nature Based Solutions (NBS) for dune ecosystems and coastal defenses.

The integration of citizen science through the "Adopt Your Own Blowout" initiative will further enhance Sand Tracer by collecting on-the-ground sediment and photo data, correlating with satellite-derived insights.

This presentation will showcase: (1) the technical aspects of data fusion, (2) case studies demonstrating Sand Tracer’s application, and (3) the implications for future dune management and coastal resilience initiatives, highlighting the potential for informing policy decisions related to coastal protection and biodiversity conservation.

The GUARDEN Project aims to enhance biodiversity monitoring through the integration of satellite remote sensing data and species occurrence records. This study focuses on a case study in France, using the GeoLifeClef2024 database to analyse the distribution of plant species. By exploiting Sentinel-2 satellite imagery, we assess essential biodiversity variables (EBVs), including ecosystem structure, focusing on species interactions and species distribution. The study uses a novel approach by analysing two datasets (cubes) with and without species interactors to explore the relationship between species co-occurrence and remote sensing data. The presence-absence data for the flora in the study area constitute the ground truth for assessing model performances.

Initial findings will be presented at Biospace25, highlighting the integration of species occurrence data with Earth Observation (EO) data to monitor species diversity. The approach underscores the importance of satellite remote sensing in understanding and mitigating the impacts of climate change, habitat fragmentation, and invasive alien species on biodiversity.

Accurate tree species mapping is crucial for biodiversity conservation and sustainable forest management. This study integrates hyperspectral data from EnMAP (Environmental Mapping and Analysis Program) and PRISMA (PRecursore IperSpettrale della Missione Applicativa) with Sentinel-2 multispectral data to classify tree species in the biodiverse and topographically varied landscapes of Tuscany, Italy.

To address the challenge of limited data availability due to the narrow swath widths of hyperspectral satellites, we leveraged dual hyperspectral datasets alongside multispectral imagery. We used 10 Sentinel-2 images captured throughout the year to leverage phenological changes for species identification. 6 EnMAP images were taken on August 6th, 2024, while PRISMA images were acquired on different dates and years due to data availability constraints. Although not all images cover the same area, common areas were identified for training and testing.

The datasets were co-registered using AROSIC for PRISMA and pixel-based co-registration for EnMAP with Sentinel-2 data. Essential vegetation indices such as AFRI_1600, CCCI, CIgreen, CIrededge, EVI, NDVI_MIR, NDVI, SAVI, and NDMI were calculated from Sentinel-2 dataset. The Sentinel-2 data was downscaled to 30 meters to match the resolution of EnMAP and PRISMA.

For training, we used the Tuscany regional map and orthophoto map from the Tuscany Regional Geoportal. Polygons with more than 80% of a single species were selected and visually confirmed using the orthophoto map. We drew our own polygons to extract spectral signatures for training, focusing on 14 tree species that had sufficient training data.

Random Forest (RF) and Support Vector Machine (SVM) algorithms were employed for classification, with Independent Component Analysis (ICA) used to reduce data dimensionality. The resulting species maps were validated against ground truth data on areas where the images from both datasets overlap. Accuracy was evaluated using traditional metrics such as the F1 score, the Kappa coefficient, and individual class scores.

The derived species maps were further used to calculate key biodiversity indices: Shannon-Wiener Index, Simpson’s Diversity Index, Species Richness, and a custom biodiversity index. This custom index was calculated based on the resolution of the biodiversity map (90 meters), where each pixel corresponds to 9 pixels of the classified map. The index varies from 1 (if all 9 pixels are different species) to 1/9 (if all 9 pixels belong to the same species).

Nitrate leaching from agricultural fields can lead to elevated nitrate levels in water bodies, putting pressure on aquatic ecosystems. Catch crops are a nature-based solution to reduce nitrate leaching from agricultural fields and are grown from late summer to early spring, bridging the gap between main cropping seasons. In addition to reducing nitrate leaching, catch crops also improve soil health and its biological quality. Because of these benefits, catch crops are promoted under the EU’s Nitrate Directive and the Common Agricultural Policy (CAP). Monitoring their adoption is therefore crucial for understanding their impact on nitrate leaching and soil health and for supporting these policies.

Monitoring catch crop adoption currently often relies on field visits by authorities, which does not provide a comprehensive overview to what extent catch crops are adopted across a region. In contrast, satellite remote sensing offers large-scale coverage and high spatio-temporal resolution. We therefore explored the use of Sentinel time series data to classify catch crops at the field level in Flanders (Belgium), using the temporal dynamics of catch crops to differentiate them from other vegetation types. We compared both traditional machine learning and time series-specific deep learning methods, evaluating Random Forest (RF), Time Series Forest (TSF), and 1D-Convolutional Neural Networks (1D-CNN) in their ability to handle temporal data. The time series inputs included monthly, dekadal and daily frequencies, with features including NDVI and two biophysical variables, generated at such high frequency using the CropSAR service which combines Sentinel-1 and Sentinel-2 imagery. The results demonstrated that RF showed the highest adaptability to different input features, achieving a median F1-score of >88% on the best performing dataset and that high temporal resolution time series improved classification accuracy. Future work could explore transfer learning to address the challenge of limited training data while taking advantage of deep learning algorithms.

Information on grassland sustainability is important to understand the condition and stability of grassland ecosystems and can be used to guide conservation and management actions. It speaks about the consistency with which grassland is maintained as grassland over a longer period of time. From an ecological perspective, the persistence of grassland contributes positively to the richness of plant species and resilience to disturbances such as climate variability and thus serves as an indicator of the quality of the biodiversity of a landscape or grassland ecosystem. Our aim in this study was to determine the persistence of permanent grassland in Slovenia as a function of age (i.e. years in which the grassland remains undisturbed by other land uses) and to reveal spatio-temporal patterns associated with conservation or signs of change. We used time series of Sentinel-2 and Landsat 5/8 satellite imagery for the period between 2000 and 2021 to identify the annual presence of bare soil rather than tracking the continuous presence of grass. Using a machine learning-based bare soil marker (developed as part of the EU CAP activities), we detected ploughing and similar events by observing exposed bare soil on grassland. The results, presented as national statistics aggregated by administrative region, indicate that 98% of all permanent grassland in Slovenia has remained unchanged over time. However, there are significant regional differences: In some areas, changes of less than 0.3% were observed, while in others almost 5% of permanent grassland was lost. We found that information on grassland permanence is of particular interest to official national statistics and nature conservation stakeholders.

Efficient and cost-effective monitoring of forest biodiversity is an important endeavor, more so considering how climate change is affecting terrestrial habitats. Several metrics have been developed to estimate alfa- and beta-diversity from space through remote sensing technologies, and in recent years, Rao’s Q diversity index has proven to be a valuable tool for assessing biodiversity at various scales and using different datasets, as, unlike Shannon’s species diversity index, it doesn’t overestimate biodiversity based on optical imagery digital numbers (DN) values. However, research on how biodiversity measured from Rao’s Q diversity index estimated from remote sensing compares to the capability to map certain terrestrial habitat types, and how sensors’ characteristics influence both aspects, is still lacking. Integrating the two aspects is important to monitor both taxonomic diversity (through habitat mapping) and functional diversity (through Rao’s Q index). For this reason, we evaluated the ability of vegetation indices (VIs) computed from three sensors (PRISMA, Sentinel-2, PlanetScope), with the addition of a Canopy Height Model (CHM) to infer biodiversity through Rao’s Q diversity index, in a Mediterranean Natural Reserve presenting a complex pattern of distinct forest types. The metrics obtained are compared to results on habitat mapping obtained on the same area from previous studies and disclose the relationship between functional diversity and classification accuracy between and within the considered habitat types.

Protected areas (PAs) are essential for restricting human pressure on natural environments, such as habitat loss and overexploitation, and halting biodiversity loss. The effective expansion of PAs is critical for achieving global biodiversity targets, but it generates trade-offs between biodiversity conservation, food security, and economic development goals. The locations of PAs determine the level of human pressure they face and, ultimately, affects their effectiveness at conserving biodiversity. PAs located in regions with intense human activity are considered to be crucial for conserving local biodiversity, but are more exposed to anthropogenic pressure. With the intensification of human activities, and under increased need to expand PA coverage to conserve biodiversity, it is essential to understand how the expansion of PAs overlaps with existing human pressure. Satellite Remote Sensing can help monitor the overlap between human pressure and PAs, and its change through time. Here, we measure the changing overlap of PAs with three human pressure layers globally, during 1975-2020: human population, human settlements, cropland areas. We define a set of “control” areas with similar biophysical characteristics to PAs, using a matching method based on satellite-borne maps. We then compare the level of human pressure between PAs and control sites, at the time of PA establishment. Our aim is to understand whether more recently established PAs are facing increasing challenges from human pressure, when compared to control sites. Our hypothesis is that as the global coverage of PA increases the risk of trade-off with human activities will increase accordingly.

Harmful algal blooms (HAB) in coastal waters are expected to increase in frequency in the coming decades. Current monitoring programs rely mainly on in situ sampling, while multi-spectral satellite images offer a broader view of Chlorophyll-a concentration, aiding in HAB mapping and bloom tracking. However, their limited number of spectral bands limits the identification of bloom-dominant species. Hyperspectral satellite data, which provide narrow and spectrally contiguous reflectance signals, holds promise for detecting diagnostic pigments and improving HAB monitoring. This study developed line height (LH) algorithms based on in situ hyperspectral Remote Sensing Reflectance (Rrs) data collected over dense HAB areas, i.e., where water is dominated by one or few species, behaving as a “massive open-air culture”. The presence of Chl-b was indicated by a positive LH using bands at 628, 646 and 665 nm (named LH646), while Chl-c was detected using bands at 601, 628 and 646 nm (LH628). These algorithms were applied to PRISMA, EMIT, and PACE satellite images during summer HAB events along the French Atlantic coast, dominated by dinoflagellates such as Lepidodinium chlorophorum (LEPI), containing Chl-b, and Lingulodinium polyedra (LINGU) or Alexandrium spp. (ALEX), which contain Chl-c. The LH646 algorithm effectively detected Chl-b in LEPI-dominated blooms, while the LH628 algorithm identified Chl-c in ALEX or LINGU blooms. The results of this study have a two-fold aim: firstly, to enhance the monitoring of HAB events and their dominant species, and secondly, to showcase the potential of hyperspectral data for this application. It underscores the value of integrating additional spectral bands, particularly in the red region, for more precise detection of key pigments, ultimately advancing species-specific HAB tracking.

Wild rivers are an invaluable resource that play a vital role in maintaining healthy ecosystems and providing ecosystem services. These rivers provide habitat for a wide variety of plant and animal species. However, the increasing pressure of human activities has been causing a rapid decline of biodiversity and ecological function. But there is currently no map available that identifies the river segments that remain under good conditions, which would be worth protecting and conservation.

The quality of the river in terms of wildness is multidimensional and difficult to measure with existing remote sensing products such as land cover and human modification products. However, by using remote sensing images with citizen science and machine learning methods, we were able to better improve our abilities to provide a detailed map of river wildness with high spatial resolution. We built a reference database of annotated images thanks to the contribution of citizen scientists through a web application (https://lab.citizenscience.ch/en/project/761). The application asks each participant to rank two images based on their wilderness for multiple rounds. Then, the rankings were then used to assign a wildness score to each image using the true skill algorithm. Finally, we used this dataset to train a convolutional neural network to identify the wildness of river sections.

By providing a detailed map of river wildness at a much higher spatial and temporal resolution than current products, this study will improve our understanding of how these rivers evolve under the pressure of human activities. This knowledge can inform critical downstream analyses, including biodiversity monitoring, hydrological modeling, and conservation planning. Moreover, our findings reveal an alarming trend: red-listed fish species are increasingly exposed to degraded river environments.

Seasonally dry waterways serve as energetically efficient movement corridors for many wildlife species, thereby shaping important ecological patterns. Because climate change causes many waterways to become less predictable, understanding the linkages between these and wildlife behavior is critical for biodiversity conservation. Unlike optical imagery, radar remote sensing offers an opportunity to detect these riparian movement corridors at fine scales, even under forest vegetation and cloud cover. Here, I evaluated the use of a NASA Shuttle Radar Topography Mission-derived hydrological elevation model, Height Above Nearest Drainage (HAND), to predict the movement behaviors of endangered tigers (Panthera tigris) in the Himalayan watershed in lowland Nepal. In this region characterized by riverine forests and a seasonal monsoon, I hypothesized that HAND (30 m resolution) would perform better than OpenStreetMap river and stream maps to predict tiger traveling behaviors. I piloted this approach on three individual tigers that were GPS collared for 6-13 months. I first fit two-state Hidden Markov Models to identify traveling movements. Then, I estimated tigers’ selection for HAND (m) and distance to mapped rivers and streams (m) using integrated step selection functions. Two tigers (male and female, respectively) in core national park lands demonstrated a small but highly significant selection towards locations closer to channel bottoms, and no relationship with distance to rivers and streams. One male tiger that inhabited more developed areas in an open floodplain instead showed a slight tendency towards larger rivers and streams. These results indicate that the hydrography models outperform existing maps for identifying energetically efficient movement pathways for wildlife that depend on minor, under-canopy waterways. Thus, high-resolution space-based imagery can reveal previously unobserved biophysical processes and fine-scale ecological connectivity that are key to habitat conservation.

To protect nature and reverse the degradation of ecosystems, strategies and policies are introduced at the national and supra-national levels. Examples include the United Nations Convention on Biological Diversity (CBD), Kunming-Montreal Global Biodiversity Framework and the EU’s Biodiversity Strategy for 2030, all setting strategic goals and specific targets, along with a set of indicators for supporting progress of the implementation. Recognizing the limitations and the challenges related to data collection for these indicators, the scientific community suggested the use of Remote Sensing (RS) as a complementary or an alternative source.

The recent development of the Earth Observation Data Cubes (EODC) framework facilitates EO data management and information extraction, enabling the mapping and monitoring of temporal and spatial patterns on the Earth’s surface.

This submission presents the ELBIOS EODC, specifically developed to support the biodiversity management and conservation over Greece. Based on the Open Data Cube (ODC) framework, it exploits multi-spectral optical Copernicus Sentinel-2 data and provides a series of Satellite Earth Observation (SEO) biodiversity products (Green Fractional Vegetation Cover, Annual net primary productivity, Leaf Area Index, Intra-annual relative range, Plant Phenology Index, Date of Annual maximum) linked to EBVs, from January 2017 onwards. Six SEO biodiversity products are included in the EL-BIOS EODC along with three spectral indices. In total the ELBIOS cube includes currently 12.400 data sets and approximately 7 TB of data.

Last, but not least, the ELBIOS EODC, to our knowledge, is the first and only EODC in Greece right now.

The corncrake (Crex crex) is a vulnerable species that relies on undisturbed grasslands during its breeding season. Early or intensive mowing presents a significant threat to the corncrake's habitat, leading to population declines. To address this issue, we first developed a reliable method for detecting mowing activities in the intermittent Lake Cerknica using optical satellite imagery time series from Sentinel-2 and PlanetScope, focusing on the Normalised Difference Vegetation Index (NDVI) and Normalised Difference Water Index (NDWI) for the period 2017–2023. Building on this method, we now assess how mowing affects corncrake populations by integrating spatial reference data on corncrake locations from 2017–2023. The analysis correlates the mowing detection results with field data provided by the Notranjska Regional Park (NRP), examining the spatial overlap between mowed areas and known corncrake habitats. Preliminary findings indicate a substantial impact of early mowing events on the availability of suitable breeding grounds for corncrakes. This study offers valuable insights into the timing and frequency of mowing and its effects on corncrake populations, contributing to biodiversity management strategies in Lake Cerknica and other Natura 2000 areas. The results can guide future conservation practices, helping balance land use with the protection of critical habitats for endangered species.