Earth Observation (EO) provides crucial spatiotemporal data for agricultural monitoring, yet integrating EO into statistical frameworks to deliver reliable, policy-relevant statistics remains challenging. Using four agricultural use cases, we explain how EO data, combined with advanced statistical techniques, fit within the stages of the policy cycle. The first use case addresses the problem identification or agenda setting stage. Biomass and canopy nitrogen content were modelled over fields in Luxembourg from Sentinel-2 data, supplying detailed inputs for nitrate leaching models. Unlike previous static seasonal averages, this dynamic approach enables more accurate groundwater protection assessments. The second use case exemplifies the policy implementation stage. It focuses on pesticide risk reduction efforts in line with the German National Action Plan on Sustainable Use of Plant Protection Products and the EU Farm-to-Fork strategy by integrating high-resolution EO data into operational day-to-day agricultural prognostic models and decision support systems. This enables farmers to identify local risk zones and optimise pesticide use, promoting sustainable plant protection through reliable, practice-ready digital tools. Two more case studies reflect the evaluation stage of the policy cycle. First, multi-sensor EO data is used to delineate crop management zones, providing an objective, quantitative assessment of treatment effects and yield variability that supports robust spatial statistics. Second, Bayesian neural networks were applied to EO imagery to identify weeds in maize fields, coupling predictions with uncertainty measures for sound statistical inference and generalization to new fields. These studies demonstrate the production of reliable, spatially explicit indicators, supporting different agricultural policy stages. This integration allows the generation of timely, accurate statistics that inform policymakers, advancing monitoring and compliance in the food security domain. We recommend that future work should focus on real-time EO data streams and enhancing ground validation efforts, ensuring accuracy and operational relevance of EO-based agricultural statistics for sound policy applications.

Producing reliable, timely and spatially detailed agricultural statistics is essential for national authorities and for monitoring European policy frameworks such as the Common Agricultural Policy (CAP), the EU Biodiversity Strategy for 2030, the European Green Deal and the newly adopted EU Soil Monitoring and Resilience Directive. Until recently, Europe lacked a harmonized, annually updated, high-resolution cropland dataset capable of supporting these needs at continental scale. The newly released Copernicus High-Resolution Layer Croplands (HRL Croplands) addresses this gap. Developed under the EEA-commissioned HRL-VLCC project, the product provides annual 10-m resolution mapping of crop types and cropping patterns across Europe from 2017 onwards. The suite consists of two primary components: Crop Type Layer (CTY): Distinguishes 17 annual and permanent crop types using a multi-source input stack (Sentinel-1, Sentinel-2, meteorological and topographic data) processed with a transformer-based machine-learning model trained on harmonized GSAA reference data. Annual updates and interannual consistency checks enable robust area statistics, crop distribution analyses and long-term trend monitoring. Cropping Pattern Layers (CP): Describe how cropland is managed throughout the year, providing field-level indicators derived from Sentinel-1 and Sentinel-2 time series. These include emergence and harvest dates, bare-soil periods, fallow land, secondary crops and crop-rotation information. Together, the HRL Cropland layers provide an operational and harmonized dataset enabling national authorities to monitor cropland dynamics, support MRV processes, evaluate agricultural practices and verify policy compliance across Europe, national and sub-national. With their combination of Europe-wide coverage, fine spatial detail and yearly updates, these layers meet Tier 3 standards, enabling rigorous, evidence-based decision making for agricultural, soil and environmental policies. During the conference, we will highlight how these HRL Cropland layers can be applied to support policy implementation and generate agricultural statistics at regional, national and European scales.

Earth Observation has been long seen as an enabler in the production of agriculture statistics, mostly within the domain of crop statistics. An important part of investigating the EO capabilities is understanding and being forthright regarding the limitations with respect to quality and quantity. The presentation will outline the statistical requirements stemming from EU Regulation 2022/2379 on statistics on agricultural input and output (SAIO) and the related Commission Implementing Regulation (EU) 2023/1538 laying down rules for the application of the SAIO main act. Within the domain of crop statistics, there are two data collections where EO is a potential data source: Crop area and production and the Management of grasslands. As regards of crop area and production statistics, Eurostat requires countries to collect data about the area of and production from a wide range of crops. Under the Management of grasslands data collections, countries are required to collect detailed data about grasslands, such as the age of permanent grasslands and temporary grasses and grazings, the management practices taking place on permanent grasslands, as well as the area of permanent grasslands with trees/shrubs and managed agroforestry cover.

In Eurostat, the Earth Observation for Statistics (EO4S) activities were pursued by the mandate of the Warsaw Memorandum. The aim is to integrate Earth Observation workflows into official statistical production. To this end, Agriculture Statistics has been one of the most active topics, and more specifically the estimation of crop area as well as grassland area & age, and the detection of wind turbines on agricultural holdings. Eurostat and the ESS in general not having necessarily the required skills and resources for full Earth Observation pipeline implementation, the strategic decision was made to focus on Earth Observation products and liaise with partners that produce those. Regarding crop and grassland area estimation, the first efforts focused on the EUCROPMAP 2018 and 2022, produced by the JRC and BOKU Vienna. To showcase the immense processing resources required, a limited area was re-processed using the code shared by BOKU, adapted to the CDSE environment. Furthermore, a methodology was created to aggregate raster files in statistical units such as 1km grid, NUTS3, NUTS2, etc. Finally, a method to alleviate classifier bias based on EBLUP was implemented but did not produce acceptable results. Building on this, we then focused on the successor product by EEA, namely the HRL VLCC Cropland and Grassland products. We liaised with EEA and the GAF consortium to acquire intermediate products, i.e. classification probabilities (confidence layer) for various classification stages. This allowed us to propose a new bias alleviation method for crop area estimation, based on unit-level multinomial models, which provided better results. Comparisons between official statistics and various modalities of these products were also made in an extensive way for a specific study area, as well as in a more limited capacity for EEA-38. A set of statistical requirements were proposed to the EEA to be considered in the production of future HRL VLCC iterations. In 2026, extensive work on grassland area and age estimation, using the relevant HRL VLCC products will ensue. As an in-house project, and partly in collaboration with INE Portugal, wind-turbine detection was pursued. Three Earth Observation methodologies were studied, which covered a spectrum of input datasets and model architectures. Out of those, none were of adequate quality. A hybrid geospatial method was proposed to assist this reporting requirement, which consists of combining national renewable energy register data, with cadaster and or agricultural parcel data.

Agricultural farm data at high resolution is an essential input for many applications, evolution of farm structure, productivity projections, ecosystem services, or climate change impacts and adaptation. Various sources of data such as farmers’ declarations, surveys, and Earth Observation are available, with pros and cons. National farmer’s declaration use national crop names, often with spelling errors. Surveys and censuses have their own nomenclature, not always identical. The remote sensing products will often need one-to-many or many-to-many spatial and semantic matching with the other data sources, leading to dubious relationships, in addition to inconsistent definitions of parcel boundaries. Farmers’ declarations are often considered the gold standard, as reporting crop areas at parcel and holding level is mandatory for subsidy applications in EU Member States. However, non-subsidised crops are typically excluded and data access is frequently restricted due to confidentiality and privacy regulations. Agricultural censuses take place regularly, usually conducted every ten years following FAO recommendations, cover all farms regardless of subsidy status. However, the precision might be lower for a census if the questions are complex, and spatial data is often simplified to point data, which may or may not be close to the actual parcels with the reported crops. Remote sensing is a third option and earth observation data from Copernicus are used to detect annual crop rotations. Whereas remote sensing gives high spatial detail at pixel level and regionally consistent data, heavy pre-processing is usually necessary, there are fewer crop categories and they may not be easily distinguished from each other. Here we show how the three sources of data can be combined for better insights into crop area, by overcoming semantic, geospatial, and technical interoperability issues.

The launch of the Copernicus Sentinel-2 mission marked a paradigm shift towards crop-specific agricultural monitoring using high-resolution Earth Observation (EO) data. However, its operational uptake within official crop monitoring and agricultural statistics systems has remained limited so far. Key constraints included the absence of sufficiently long time series, the lack of reliable in-season crop type maps, and the challenge of deploying scalable operational processing chains. As Sentinel-2 enters its second decade of observations, these barriers are progressively being overcome, allowing high-resolution, crop-specific information to meaningfully complement legacy monitoring systems based on coarser-resolution sensors (MODIS, VIIRS, SPOT-VGT, Sentinel-3). This contribution documents the transition from experimental Sentinel-2 analyses to an operational crop-specific monitoring framework within the JRC MARS crop monitoring system, which supports monthly agricultural production assessments for the European Commission. Beginning in 2022, crop-specific monitoring was progressively introduced to support crop analytics. By 2025, it had been operationally implemented in four countries, leveraging remote sensing derived in-season crop type maps (Ukraine, France) and provisional GeoSpatial Application (GSA) datasets (Denmark, Netherlands). Building on this experience, a fully automated processing chain was developed to enable wall-to-wall crop-specific monitoring across Europe for the 2026 season, covering approximately 40 million agricultural polygons and 150 million hectares. The system relies on the Copernicus High Resolution Layer on Vegetated Land Cover Characteristics (HRL-VLCC, 2017–2024) to train an AI-based in-season crop classifier, producing consistent monthly crop masks from May to October at continental scale. Aggregated crop-specific time series at subnational level and on a regular 10-km grid demonstrate the capability of the proposed approach to catch the key-events that marked the observed agricultural seasons. Our results demonstrate that Sentinel-2 has crossed a critical operational threshold, opening a new era of crop monitoring. The presented Crop-specific analysis fundamentally improves the relevance of agricultural indicators, providing a concrete and operational route for embedding EO within institutional crop monitoring and agricultural statistics.

Earth Observation (EO) data in the agricultural domain have significant potential to support policymakers in decision-making and legislative development. EO-based policy-relevant indicators can complement traditional information sources, such as in-situ measurements, farmer declarations, surveys, and censuses. These indicators can support various stages of the policy cycle, including implementation (continuous monitoring), evaluation (ex post assessments), and design (lessons learned and scenario analysis). In 2025, the Copernicus Land Monitoring Service released the High-Resolution Layers on Vegetated Land Cover Characteristics (HRL-VLCC), providing, for the first time, operational EO data on croplands and grasslands. HRL-VLCC offers spatially continuous coverage, regular annual updates since 2017, and a spatiotemporally consistent methodology. While the data are not yet timely enough to support policy implementation, they offer significant opportunities for policy evaluation and design. However, two main barriers remain: quality challenges due to the overall uncertainty of the data, and the lack of in-situ data to validate most layers on agricultural practices, and a gap between the raw EO layers and the indicators needed to evaluate policies. This contribution presents the general methodology used at the Joint Research Centre to develop indicators from the HRL-VLCC, addressing the previous two barriers. To tackle data quality issues, we have evaluated the cross-layer and temporal consistency of the HRLs, reported the issues to Copernicus, and corrected the layers when needed. For instance, we found spatial inconsistencies between the grasslands and croplands layers, and temporal inconsistencies in the crop sequences and cropping seasons. A key message is that regular reprocessing of the data, following the concept of collections in the space segment, is the only way to correct the identified quality issues while preserving temporal consistency. Regarding the gap between HRLs and policy needs, we present a methodology for developing indicators by combining different layers, aggregating data spatially and temporally, and using the resulting indicators to assess spatiotemporal patterns and their drivers. Preliminary results on EO-based indicators are presented for: (1) crop diversity and crop rotation, based on crop type layer, (2) secondary crops and crop phenology dynamics, based on cropping patterns layers, and (3) grassland age and management, based on grassland layers. The results demonstrate the potential of HRL-VLCC data to be integrated as indicators within official policy evaluation frameworks, for example, as a new indicator on cover crops to assess the effectiveness of carbon- and nitrogen-fixing measures under the Common Agricultural Policy.

Crop diversity is a key component of resilient agricultural systems, contributing to soil health, biodiversity, and climate adaptation. Yet, until recently, consistent monitoring of crop diversity across the European Union (EU) was not possible. This study presents the first harmonised, annual assessment of landscape-level crop diversity across the EU-27 using the Copernicus High Resolution Crop Type (HRL CTY) dataset (10 m resolution, 2017–2023), complemented with Copernicus grassland layers and Farm Accountancy Data Network (FADN) data. Using a Shannon diversity index derived from satellite-based crop maps, we quantify spatial and temporal patterns of crop diversity at both regional (NUTS3) and local (municipality/LAU) levels. Over 2017–2023, the mean crop diversity across 1 166 NUTS3 regions was 4.6 crop types (median 4.9). At local scale, 25% of agricultural municipalities are located in highly diverse landscapes (>5.9 equally distributed crop types), while 25% fall in low-diversity areas (

Earth observation enables detailed, area-wide mapping of agricultural areas from regional to national level or beyond. Fueled by the availability of dense time series of the Copernicus and Landsat missions, maps with a high level of spatial and thematic detail are regularly collected, which can e.g., support agricultural census by adding area-wide information and spatial detail to sample-based surveys. Still, there is a lack of historical maps, which hinders to reveal long-term trends. This study aims to close this gap by consistently mapping the dominant agricultural land use types for each year from 1990 and 2023, using Germany as a use case. Therefore, we processed all available Sentinel-2 and Landsat images with a cloud cover below 70% for the years 1990 to 2023 and combined these data, in a first step, with samples a from the German Authoritative Topographic Cartographic Information System (ATKIS) to map six general land-cover types: urban, forest, cropland, grassland, wetlands, and other land. In a second step, we combined the time series data with samples from the Integrated Administrative Control System (IACS), which was available for several German federal states and years between 2006 and 2022, to differentiate the resulting agricultural area into grassland and 13 crop type classes. In this two-step approach, we followed Pham et al. (2024) who proposed an approach, which makes use of Temporal Encoding by assigning all clear-sky observation pixels to their respective acquisition dates in a constant array of 365 days and then applies two data augmentation techniques, namely Random Observations Selection and Random Day Shifting, to learn data sparse situations and phenological variations. The final maps showed good accuracies and derived areas matched reported census data for major crop types in Germany, highlighting the ability of the proposed approach to address data gaps in traditional agricultural statistics.

Advances in digital agriculture remain constrained by fragmented modeling, EO-centric embedding spaces, and the scarcity of high-quality labels needed to capture fine-scale agroecosystem processes. These limitations hinder our ability to represent agroecosystem function across complex landscapes, where soil properties, crop performance, and environmental fluxes are simulated by disparate and often incompatible systems. In distinct agroecosystems, such as those of the Canadian Prairies, micro-scale, process-based Soil-Plant-Atmosphere-Water (SPAW) dynamics are particularly pronounced. These localized processes at both within-field and regional scales limit the transferability of AI-driven models and expose the inadequacy of existing EO-centric featurization representations that try to capture localized biophysical continua and within-field spatiotemporal variabilities. For this reason, we present GAIG-Embeddings (the Geospatial Agroecosystem Inference Generator Data), an open, unified agroecosystem-data embedding that fuses EO time series with varying-scale representations of SPAW processes and the regional STEP-AWBH (Soil, Topography, Ecology, Parent material; Atmosphere, Water, Biotic, Human) continuum. This latent representation isolates stable landscape structure (STEP) from dynamic environmental forcings (AWBH), enabling consistent assimilation of agroecosystem states across space, time, and observational contexts. Using high-dimensional encoders and self-supervised transformer architectures, GAIG-Embeddings generates decision-ready geospatial time-series data at 10 m resolution and serves as the first agroecosystem-oriented embedding space for diverse downstream agronomic tasks, including: (1) land productivity-marginality assessment, (2) multi-year yield forecast, (3) digital soil mapping, (4) N2O emission classification, and (5) in-season crop type discrimination. By integrating agroecosystem-oriented insights directly into a unified-foundation model, GAIG-Embeddings enables an actionable pathway toward a multi-year field-scale agricultural productivity and sustainability inference, which aligns with UN-SGDs. Comparative evaluations demonstrate superior capture of within-field variability and close alignment with ground observations across the Canadian Prairies. Taken together, GAIG-Embeddings establishes a scalable, agroecosystem-informed geospatial foundation model, highlighting opportunities to improve transferability, interpretability, and operational integration in next-generation agricultural AI systems that advance global sustainability objectives.

Statistics Sweden currently relies on time-intensive surveys to produce agri-environmental statistics for national, EU and international environmental and climate reporting. In the EU funded project “Earth Observation Data for Urban Land Use and Agri-Environmental Statistics” (Project 101195601 — 2024-SE-GEOS_SATDES), RISE and Statistics Sweden jointly evaluate whether Earth Observation (EO) data can provide annually updated, high-quality detection of crop-residues at the field level, potentially modernizing statistics production and reducing survey dependency. We developed our approach using annotated field polygons derived from a 2024 sample survey in which Statistics Sweden collected information on crop-residues from approximately 3,000 farms. To ensure reliable ground truth, we included only farms reporting uniform residue management across all fields (either all with or all without residue), excluding those with mixed practices. These fields were divided into training and evaluation sets. Post-harvest EO data from the Sentinel-1 radar and Sentinel-2 optical missions were prepared for each field. For Sentinel-1, the first acquisition after harvest was used. For Sentinel-2, scenes were filtered based on cloud presence, and the first cloud-free observation within one month after harvest was selected for each field. To ensure consistent model inputs across fields of varying size and shape, we applied the pixel-set encoding approach of Garnot et al. (2020). Pixels were randomly sampled from each field and encoded using a shared multilayer perceptron. The resulting embeddings were combined into a binary prediction of crop-residue presence. Models were trained and evaluated separately for radar and optical inputs. Preliminary results are promising, showing that crop-residues can be detected using post-harvest Sentinel-2 data, while detection using Sentinel-1 data remains more challenging. Nonetheless, radar imagery is important when optical data are unavailable due to cloud cover. Overall, the findings demonstrate the potential of EO-based methods to generate annually updated crop-residue statistics, reducing reliance on time-intensive surveys.

A fully automated procedure is presented for modeling crop yields at field level using Sentinel-2 data and machine learning methods. The workflow was developed by the Hesse Statistical Office for all of Germany and is now about to be transferred into national official statistics production. The motivation for developing a new procedure is the increasing occurrence of missing values in regional yield statistics due to a declining number of voluntary respondents. Given the age structure of these respondents, this problem is expected to intensify further. Compared with respondents’ estimates, the advantages include greater objectivity, faster and area-wide results, as well as the possibility of more detailed regionalization down to the municipal level, thereby increasing the relevance of the yield statistics. For these reasons, an EU project with Statistics Netherlands (Centraal Bureau voor de Statistiek) is examining the transferability of the procedure to the Netherlands. The method is currently operated for seven crops: winter wheat, winter barley, winter rapeseed, rye, triticale, spring barley, and oats. In addition to the NDMI (Normalized Difference Moisture Index) and REP (Red Edge Position) indices derived from Sentinel-2, the methodology is largely based on the EU’s LPIS data and in-situ yield measurements from official statistics. An ensemble of different ML algorithms is trained to achieve robust results. The highly promising cross-validation results show, for 2025 and at the field level, deviations from the in-situ measurements between 10.65% (winter wheat) and 17.91% (oats), and an explained variance (R²) between 0.54 (spring barley) and 0.76 (rye, oats).

Healthy soils are essential for sustainable food production and environmental stewardship. Yet, soil management practices remain difficult to quantify in official agricultural statistics. Sowing, harvesting, and tillage timing determine periods of soil exposure that strongly influence erosion and nutrient losses. These dynamics are currently monitored mainly through ground surveys, which are expensive and limited in spatial detail, and often difficult to compare across regions. Earth Observation can derive field-level indicators of soil management automatically and at a large scale. In this research, we first combine the strengths of radar and multispectral satellite data by proposing a novel Hybrid Bare Soil Radar Index (HyBRIS) to retrieve sowing, harvest, and tillage dates at the field-level. Validation across 11 European countries and 40 different crop types demonstrates the generalizability of the approach (R2=0.86, MAE=24.3). Second, we characterize agricultural soil cover by integrating satellite time series with time-stamped, geotagged field photos collected across Europe in 2018 and 2022. Fractional cover of bare soil, green vegetation, and non-photosynthetic vegetation was derived from field photos using deep learning, automatically generating a reference dataset. The estimates were validated using a human-in-the-loop approach (R2=0.9, MAE=0.05), and used to train satellite-based models. We compared recurrent deep learning architectures with emerging foundation-model approaches to assess the potential of satellite data for estimating soil fractional cover in near–real-time. Experiments were carried out using the Land Use and Cover Area Frame Survey (LUCAS), a publicly available field survey dataset coordinated by Eurostat, which requires surveyors to acquire field photos in addition to land cover information. Together, these methods deliver quantitative indicators of soil management dynamics at the field-level, enabling monitoring of conservation practices and the production of spatial statistics on soil exposure. These indicators facilitate comparison of soil management among European farming systems and support the assessment of agri-environmental policies.

Timely and reliable crop yield forecasts play an important role in supporting national and international agricultural and food security policies, stabilizing markets and planning food security interventions in food-insecure countries. EO data, agrometeorological gridded data are sourced from the JRC ASAP Early Warning System and used as predictors for the data-driven yield forecasting at both national and sub-national scales. An operational machine learning pipeline is applied to forecast national level crop yield in about 90 countries distributed globally. Yield of the three main crops per country is estimated during the growing season, typically at 75% of the local growing cycle, using 24 years of ASAP data, additional selected socioeconomic indicators and FAOSTAT yield data as target. Crop calendars are obtained from satellite-derived crop phenology cross-check with global crop calendars. Sub-national yield forecasting uses a different pipeline exploiting the larger sample size available at the subnational level (typically 24 years for several administrative units). Yield data are gathered directly from national institutions or through FEWS NET and NASA HarvestStat Africa initiative. Yield forecast figures are complemented by hindcasting uncertainty statistics and drivers’ explainability figures and support JRC and GIEWS analysts in their Early Warning activities. Preliminary results of the use of ECMWF SEAS5 seasonal forecasts to anticipate the forecasting time are presented.

Grassland areas in the European Union, including Latvia, have been declining for decades due to changes in agricultural practices, resulting in significant biodiversity loss affecting ecosystems, plant communities, meadow birds, and pollinators. In response, the EU provides subsidies to encourage the preservation and restoration of grasslands, creating a growing need for reliable and comparable monitoring data. Satellite remote sensing combined with machine learning offers a unique opportunity to address this need, as it provides consistent land-use information covering the entire territory of Latvia. This presentation introduces a pilot project developed by the Central Statistical Bureau of Latvia in collaboration with the University of Latvia to map grassland extent and estimate grassland age using Copernicus Sentinel satellite data. The project emphasizes a reproducible workflow to ensure long-term comparability of grassland area change and age dynamics. A land cover classification model was built using a structured sampling and training approach, integrating multi-source reference data with Sentinel-1 and Sentinel-2 observations and an XGBoost classifier, achieving high accuracy for grassland detection. Grassland age was derived through an automated time-series analysis that identifies recent grassland renewal events based on indicators of soil exposure and vegetation structure. By combining optical and radar-based indices, the workflow enables consistent estimation of grassland age across large areas. The results demonstrate the potential of satellite-based, machine-learning-driven methods to support agricultural policy monitoring and biodiversity-related reporting at national and EU levels.

Agricultural prices and market opportunities are strongly influenced by global developments in the grain market. Farmers, processors, and traders operate within a globally interconnected system in which international production variability directly affects European price levels and sales prospects. Reliable and objective global yield information is therefore essential. At present, however, yield estimates are produced independently by countries, regions, and statistical offices, resulting in fragmented methodologies, heterogeneous data quality, and limited comparability. This highlights the growing need for homogenization, standardization, and statistically robust integration of Earth Observation (EO) into processes for agricultural statistics, and reflects current discussions around User Needs, Experiences, and Challenges. To address these challenges, YPSGlobe is introduced as a consistent, unbiased, scalable, and methodologically robust global forecasting system that provides harmonized, yield predictions for millions of agricultural parcels. The approach combines high-resolution Copernicus EO data, offering dense spatial information and statistically sound, representative sampling with the physically based crop growth model PROMET, forming a digital twin of global crop production. We demonstrate what distinguishes this approach from existing yield forecasting methods and present validation results. Our upscaling methods enable flexible stratification and customizable spatial aggregation, allowing statistical offices to co-design inputs or tailor stratification schemes to their needs, enhancing trust in EO-derived products and transparency of uncertainty estimates. From 2026 onwards, YPSGlobe will deliver weekly global wheat yield and production, including country-level uncertainty information, with possible extensions to additional crops such as barley, rapeseed, maize, and sugar beet. Results will be disseminated through the web-based platform https://portal.vista-geo-service.de/ and standardized APIs, ensuring accessibility, interoperability, and compatibility with Copernicus Services and data infrastructures. By enabling seamless integration into institutional workflows, YPSGlobe supports capacity building and represents a concrete step toward the operational integration of EO into official agricultural statistics.

Grasslands are crucial for agricultural landscapes and provide key ecosystem services such as biodiversity conservation and climate regulation. One important factor is the grassland age, which is known to increase carbon sequestration and resilience to disturbances. Recent EU legislation, including the Statistics on Agricultural Input and Output (SAIO), as well as international climate reporting obligations under the UNFCCC, therefore increasingly require spatially explicit and consistent data on grassland permanence and age. However, this data is currently often unavailable at a national level. Here, we present an approach to estimate grassland age in temperate regions of intensive agriculture based on the long-term Landsat archive (1986–2023), in combination with recent Sentinel-2 data. In the first step, all clear-sky observations over grassland are classified as bare soil or non-bare soil and normalized to derive the seasonal bare soil frequency. Next, the occurrence of bare soil is used as an indicator of agricultural land-use transitions, and a rule-set is defined that translates temporal patterns of bare soil occurrence into grassland establishment dates and age estimates. We demonstrate that this approach yields robust predictions of grassland establishment in Germany (F-score = 79%), as well as the detection of grassland sites that are persistent throughout all analyzed years (overall accuracy = 99%). The grassland age is estimated with a mean absolute error of 1.3 years, and the predictions remain reliable throughout periods with low observation densities. The observed spatial and temporal patterns correspond with major policy and socio-economic changes, including agricultural restructuring following German reunification and reforms of the European Common Agricultural Policy. Our results show that multidecadal EO time series can deliver consistent grassland age indicators at a national scale to support EU reporting obligations (e.g., SAIO) as well as to improve inputs for soil carbon modelling and data inputs for GHG inventories.

Statistics Portugal (INE) fosters the enrichment of its Spatial Data Infrastructure (SDI) using Artificial Intelligence (AI) technologies to support the upcoming challenges in statistical production. To this end, INE is participating in the EU project ‘AI/ML on Earth Observation Data’, aiming to develop methodological and implementation guidelines for research findings in official statistics. This initiative pursues to facilitate the transition from experimentation to production for projects using AI and Machine Learning (ML), ensuring that Earth Observation-based solutions produce valid and comparable results across different countries and timeframes. As part of this initiative, this study presents a comprehensive error analysis of a crop type model (CTM). The original implementation distinguishes multiple crop types—such as winter crops, spring crops, and rapeseed (the latter being inapplicable to the Portuguese context)—using Sentinel-1 radar time series combined with advanced object-based classification software. This study validates the model’s performance in the Oeste e Vale do Tejo (PT11D) NUT2 region of Portugal, using the 2025 Land Parcel Identification System (LPIS) as the ground-truth reference. The methodology involved cross-referencing data from nearly 84,000 LPIS parcels with the CTM, focusing specifically on 79,600 hectares of arable crops, of which 56% correspond to winter and spring crops. Validation results indicate that while the algorithm identified 91,500 hectares, 74% of this classified area overlaps with LPIS arable crop data, with 93% of that overlapping area specifically identified as winter and spring crops. Within this intersection, the algorithm achieved a 96% match for spring crops and a 68% match for winter crops, resulting in an Overall Accuracy (OA) of 89% for distinguishing between seasonal cycles. Comparative analysis shows that while the algorithm is highly effective at identifying the spatial footprint of spring cycles, it tends to over-identify seasonal cycles in areas not registered as temporary arable land in the LPIS. Furthermore, a closer look at the CTM’s performance regarding specific crops reveals that although the algorithm identifies over 90% of the area dedicated to maize, tomato, and rice, these crops simultaneously account for the most significant portion of the area missed by the classification (omission errors). This research concludes that CTM approaches show high potential for identifying seasonal crop cycles. The project remains in full development, and these findings reflect the current state of the model. The implementation of robust spatial masks and refined segmentation is vital. These steps are essential to ensure that Earth Observation-based solutions meet the rigorous quality standards required to produce official statistics.

Effective air quality regulation and climate change mitigation depend on reducing greenhouse gas emissions and air pollutants. To achieve sustainable green development, this study constructs a spatio-temporal hierarchical model to assess the Air Quality Index (AQI) across Nigerian states and to derive actionable insights for environmental sustainability. Specifically, this study develops a three-level Bayesian hierarchical model based on latent Gaussian likelihoods to capture both random effects and systematic differences among states in Nigeria. The model structure allows for state-level variation, covariate effects, additional unexplained variation and variance decomposition. Air Quality Index data from five geopolitical zones (covering 12 states) were used to validate the developed model. The study revealed the priority level and the actions required for each state, as well as the important contributors to air pollution in Nigeria. All industries must endeavour to reduce emissions and prepare for changes in air quality associated with rising temperature. Stakeholders should mitigate the effects of industry pollution on land biodiversity and preserve natural habitats against the degradation caused by air pollution. The model accuracy is very high, indicating high correlation between the observed and the predicted values. The Bayesian model developed was properly validated, the model’s strong fit supports the analytical approach and affirms the demonstrated ability of identified predictors to explain variations in air quality across Nigerian states. States in the South South region requires emergency response to achieve sustainability. This study gives assurance of applying the model in policy interventions and projections in the future.

For the reporting of green house gasses (GHG) emissions in Wallonia, we’ve utilized remote sensing time series to support reliable statistics on land use/land cover change (LULCC). The base product is a set of 10 yearly 2-m resolution maps averaging 93% of overall accuracy. The three pillars for sound LULCC related GHG statistics are analysed in details: an appropriate legend, a time consistent classification and an efficient accuracy assessment. The legend of a geographic database is sometimes overlooked, but it could be a major source of errors if it does not match the expected concepts. Remote sensing primarily provides information about the surface land cover, ideally supported by robust frameworks such as the EAGLE concept. However, this biophysical information often needs contextual information to become compliant with the definitions of regulatory documents, in this case based on IPCC guidelines. The land cover maps are based on annual 25 cm orthophotos, aerial LIDAR (2012-2012 and 2021-2022), Sentinel-2 seasonal composites and ancillary vector data. As we focus on the change, the consistency of the classifications is more important than the yearly overall accuracy. Geometric inconsistencies are therefore addressed using morphological mathematics, and a multi-sensor approach is used to detect and confirm changes. Multiannual information is also needed for specific classes. A dual sampling strategy assessed the quality of the annual maps and the change detection. For the overall accuracy, the sampling was clustered in order to minimize displacements for double-checking, on the ground, uncertain photo-interpretation. Additional samples were concentrated in detected changed areas to increase the precision of the user accuracy estimators. Map-based areas and sample point frequencies are then combined to predict accurate class areas and their confidence interval.

Grasslands account for over half of Ireland’s land cover and are central to national agricultural productivity, climate mitigation strategies, and greenhouse gas (GHG) inventory reporting. Robust quantification of grassland carbon exchange, alongside reliable identification of management activities, is therefore critical for climate-smart agriculture. Eddy covariance (EC) flux towers provide high-quality measurements of Net Ecosystem Exchange (NEE), yet temporal data gaps constrain their operational utility. This study presents a unified Earth Observation (EO) and machine-learning (ML) framework to address these challenges through integrated satellite-EC analysis. We developed an ML model (Random Forest) integrating satellite-based spectral indices (NDVI, EVI, SAVI, NDII), photosynthetic photon flux density (PPFD), temporal covariates, and light-use efficiency interactions across nine Irish grassland EC sites (2,357 site-days; 2023–2024) spanning intensive dairy, rotational grazing, and permanent grassland systems. The ML modelling supported daily NEE gap-filling. Management signals were identified using NDVI threshold methods and gross primary productivity (GPP) anomaly detection and validated against 204 independently reported farm events. Site-calibrated models achieved strong predictive performance for carbon flux gap-filling (R² = 0.70–0.79), enabling robust annual carbon budget estimates. Management detection exhibited complementary strengths: GPP-based methods identified 61% of events, while NDVI-based detection captured 15% of events with higher precision, particularly for cutting (67%). This integrated EO–ML framework enables continuous monitoring of carbon flux alongside the automated detection of grassland management activities. The results demonstrate how coupled spectral and radiative controls govern both ecosystem carbon dynamics and management-induced signals in Irish grasslands. The approach is directly relevant to Common Agricultural Policy reporting and offers a scalable pathway for implementation within the IPCC Tier 3 agricultural GHG inventories.

High-resolution methane observations from space are now technically mature, but their systematic use in official statistics and regulatory reporting is only beginning. GESat GEN1, a new European mission operated by Absolut Sensing as an emerging Copernicus Contributing Mission, was launched in January 2025 and entered its operational phase in April 2025. Its primary objective – detecting and quantifying methane hotspots at tens-of-metres resolution – has been demonstrated in collaboration with ESA and the Copernicus Rapid Response Desk. This contribution looks ahead: how can such assets evolve into an operational, trusted component of European methane statistics and policy indicators? We outline an end-to-end integration roadmap, from satellite tasking to ingestion by National Statistical Institutes (NSIs) and regulators. On the EO side, we describe the planned evolution from a single demonstration satellite to a small constellation, improving temporal resolution and enabling near-real-time (NRT) detection of large emission events for early warning and rapid assessment. At processing level, we present a modular pipeline that combines physics-based inversion with AI-assisted components (for source attribution, scene classification and anomaly detection) and that is designed to feed digital-twin style representations of facility-level methane budgets. On the user side, we discuss strategies for making GESat-derived products operationally useful to NSIs and other authorities: stable product definitions; harmonised, machine-readable metadata; explicit uncertainty and quality flags; and service-level commitments on latency, coverage and reprocessing. We also address long-term sustainability, including integration with the Copernicus Data Space Ecosystem and Copernicus Contributing Missions framework, and options for ensuring continuity of publicly accessible methane datasets. The paper concludes with a set of concrete steps and interfaces through which high-resolution methane EO can transition from pilot projects to a permanent, policy-relevant component of Europe’s statistical infrastructure.

Jurisdictional REDD+ programs and national GHG inventories increasingly rely on Earth Observation-derived deforestation rates to establish historical baselines against which emission reductions are credited. The accuracy of these baselines depends critically on the duration of reference periods (used to calculate historical averages) and crediting periods (over which reductions are assessed). Current carbon standards and reporting frameworks specify varying requirements—typically 5 or 10 years—but the uncertainty introduced by these methodological choices remains poorly characterized. We evaluated how reference and crediting period durations affect baseline forecast accuracy using a global remote sensing dataset of annual deforestation rates for 53 tropical countries spanning 1991–2020. A sliding window approach compared observed deforestation during hypothetical crediting periods against predictions derived from historical averages over preceding reference periods. Forecast error was quantified across all feasible period combinations, with separate analysis for countries exhibiting statistically significant deforestation trends. Results reveal a fundamental trade-off: short periods amplify interannual variability while long periods fail to capture changing deforestation dynamics. Optimal accuracy occurred with 3–5 year periods for both reference and crediting. Using 5-year durations—common in REDD+ frameworks—yielded 37% average forecast error across countries. Country-specific optimization reduced error to 10%, but standardized crediting frameworks preclude such customization. Stratifying by trend detection provided negligible improvement. These findings have direct implications for GHG reporting under the Paris Agreement and voluntary carbon markets. The 37% average forecast error represents a quantifiable uncertainty that should be incorporated into emission reduction estimates and communicated transparently in statistical reporting. We conclude that current 5-year requirements are reasonable but highlight that historical baselines introduce substantial, often unreported, uncertainty into forest carbon accounting.

Accurate, timely, and spatially granular greenhouse gas (GHG) emissions data are essential for effective climate policy, regulatory compliance, and progress toward the UN Sustainable Development Goals (SDGs). This presentation introduces the Greenhouse Gas Emissions Watch Service, developed under ESA’s InCubed program by GHGSat (UK) Ltd and Terrabotics Ltd, as a case study in the integration of Earth Observation (EO) into national statistical and policy reporting frameworks. The project addresses critical data gaps in existing bottom-up GHG inventories across industrial and waste sector activities. Current approaches rely on infrequent reporting, proxy data, or self-declared estimates. Emissions Watch Service provides an independent, space-based evidence layer that improves confidence in emissions detection, quantification, and verification. The service leverages a unique combination of high-resolution GHGSat satellite data, Copernicus Sentinel-2/5P imagery, and advanced AI-driven facility attribution to deliver empirical, facility-level methane emissions data across the UK. By linking observed emissions directly to responsible parties and integrating these measurements with national inventories, the platform addresses key challenges in uncertainty reduction, regulatory enforcement, and cross-agency data harmonization.The project aims to validate a set of targeted operational use cases. These include the identification and prioritisation of emission hotspots across landfill sites and energy-intensive industrial facilities; the detection of abnormal emissions indicative of unreported operational changes, malfunctioning equipment, or episodic emission events. Unlike existing workflows that often detect issues retrospectively, the project will validate the use of satellite emissions data to proactively identify emission hotspots, abnormal behaviour, and episodic events in near-real time. We will detail the technical workflow, including data ingestion, automated plume detection, facility mapping, and uncertainty quantification. The service’s outputs will support the UK’s National Atmospheric Emissions Inventory (NAEI), regulatory compliance (e.g., UK ETS, Methane Action Plan), and SDG indicator reporting. We will discuss lessons learned so far in the project, aligning EO-derived data with official statistical requirements, ensuring traceability, and building trust with the stakeholder community.

The world is not on track to meet the Sustainable Development Goal (SDG) of ending hunger, food insecurity, and all forms of malnutrition, with climate variability and extremes being among the main drivers. Climate extremes such as droughts and floods have been shown to worsen food security outcomes. The African population is particularly vulnerable, with 20.4 percent facing hunger, a rising prevalence of undernourishment (PoU), and levels of moderate or severe food insecurity almost double the global average. Malawi, located in southeastern Africa, is especially vulnerable to climate change and climate extremes. With climate change, more severe droughts and floods are expected. More than 80 percent of the population in Malawi live in rural areas, with a high share of the households involved in farming. The country frequently experiences droughts and floods, which pose major challenges for households that rely on food from own production or staple crops such as maize. Farmers depend heavily on rain-fed agriculture and heat-sensitive crops, which make them even more vulnerable to a changing climate. However, how these weather extremes affect households’ food access, depending on their severity, spatial extent, and timing, remains unclear. The National Statistical Office of Malawi has taken part in a Food Security Statistics project led by Statistics Norway in partnership with the Common Market of Eastern and Southern Africa (COMESA). As part of this project, food consumption data from Malawi’s fifth Integrated Household Survey (IHS 5) was prepared for food security statistics. In this study, we make use of this data by analysing food security and nutrition indicators in rural households and link it with satellite data at the Enumeration Area (EA) level. The objective is to explore whether extreme weather conditions in the period preceding the survey interviews affect food security indicators such as calories consumed from own production and purchases, as well as selected diet-related indicators. We employ three indices derived from satellite imagery: the Normalized Difference Water Index (NDWI), the Normalized Difference Vegetation Index (NDVI), and the Normalized Difference Moisture Index (NDMI). Various linear regression models will be applied to examine potential relationships between the food security indicators and the three selected indices. Data preparation is ongoing, and preliminary results are not yet available. The analyses will demonstrate the value and potential of combining food security indicators from household consumption and expenditure surveys (HCESs) with earth observation (EO). Such analyses can support national authorities in identifying rural households’ vulnerability to food insecurity in the aftermath of extreme weather events, and simultaneously highlight the opportunities that lies within the use of food consumption data from HCES. Understanding how satellite imagery can help detect vulnerability will be particularly valuable for policymakers, including disaster management agencies, and will lay the foundation for future applications by National Statistical Offices.

This work presents a methodology for producing gridded macroeconomic data—with a primary focus on gridded GDP by economic activity—by combining spatial random-forest–based estimates with official national and subnational statistics. Traditional macroeconomic data are highly aggregated and obscure the spatial patterns needed to understand localized economic activity, exposure to physical risks, and infrastructure gaps. Our hierarchical approach integrates official macroeconomic accounts with Earth observation predictors—such as nighttime lights, built-up areas, land cover, and population grids—to deliver high-resolution, gridded estimates that remain fully consistent with national totals. The methodology complements and enhances official statistics by adding spatial granularity through open geospatial datasets and will be extended to generate gridded capital stock estimates.

The increasing availability of high-resolution remote sensing data opens up new opportunities for supplementing and further developing official statistics. The German Federal Statistical Office (Destatis) is currently conducting feasibility studies to examine the extent to which Earth Observation data can be used for statistical purposes in order to increase the timeliness and spatial granularity of statistical information. The aim of this work is also to systematically evaluate the potential of remote sensing data for supplementing, validating and, in the future, integration into statistical production processes. This contribution provides an overview of ongoing exploratory analyses and methodological approaches that combine satellite-based, and airborne data with official statistics and administrative data sources. The focus is particularly on economically relevant issues, such as land use, construction activity, industrial and infrastructure development, and regional economic observation. Methodologically, spatial analyses and machine learning methods for classification and object recognition are used to derive relevant information from remote sensing data. Correlation analyses are used to evaluate the consistency with classical survey methods used in official statistics. A key component of the feasibility studies is the critical evaluation of data quality, stability of results, reproducibility and compatibility with the requirements of official statistics, particularly with regard to methodological transparency and statistical confidentiality. Finally, this contribution discusses the conditions under which remote sensing data could be transferred from experimental analyses to regular statistical production processes in the future and the potential they offer for timely, small-scale, and evidence-based economic observation.

Over the past three years there has been a decrease in building permits in Germany, despite the government’s target of building 400,000 apartments annually. Partly occurring reporting delays can disturb timely provision of data in construction statistics. To address this, EO4ConStat, a joint project between the Federal Statistical Office of Germany (StBA), the Federal Agency for Cartography and Geodesy (BKG), and the German Aerospace Center (DLR) develops an independent Earth Observation-based framework for detecting, dating and validating construction activity as a data resource for statistical quality assurance.Construction site detection is performed using SAMLoRA, a deep learning model combining the Segment Anything Model (SAM) and the parameter-efficient Low-Rank Adaptation technique (LoRA). The image encoder is frozen, therefore the pretrained weights from SAM can be used while only the LoRA layers are trained during fine-tuning, enabling efficient training in resource-constrained environments. The training dataset was semi-automatically generated and contains over 4000 construction sites in over 450 digital orthophotos (DOP) in North Rhine-Westphalia from 2018 to 2023. On the test dataset, SAMLoRA achieves a precision of 0.75 and an F1-Score of 0.68. A higher precision over recall is preferred due to the need for reliable construction site detection. Including typical false-positive examples in training, such as construction yards, improved model reliability.To date the construction activity, Sentinel-2 time series are evaluated using a conditional random forest classifier on spectral indices to differentiate between construction and non-construction time series. Subsequently, change point detection is applied to the most influential indices or bands (RGB, NIR, SWIR1, SWIR2) to estimate the start and end dates of construction. For 107 out of 235 construction sites, the start is detected within three months.By providing a framework to detect, date and validate construction activity, this project demonstrates the potential of Earth Observation to assess the quality of official statistics.

The Carbon Lab research consortium, Global Pasture Watch, has produced comprehensive global pasture maps at 30-meter resolution covering the period 2000–2022. Generating consistent global information on pasture is statistically challenging due to the need to discriminate among multiple grassland types, management intensities, and closely related land-cover categories. Here, we present the first statistically rigorous global area estimates for the grassland domain, distinguishing six grassland classes: (1) natural and semi-natural permanent grasslands without visible livestock, (2) natural permanent woody vegetation within open woodlands, (3) semi-natural permanent grasslands with visible livestock presence, (4) temporary grasslands linked to other land use, (5) improved permanent grasslands associated with livestock, and (6) improved permanent grasslands associated with other land use, alongside estimates for other land-cover classes. We describe the sampling design used to validate the 2020 static map and to derive unbiased global and continental area estimates across all grassland classes, based on an initial global sample of 50,000 locations. The design incorporates stratification targeted at rare classes, systematic interpretation protocols, and estimation procedures that explicitly propagate interpreter uncertainty into class-specific area estimates and confidence intervals. Additional temporal samples were drawn to quantify grassland dynamics over the periods 2000–2010, 2010–2015, and 2015–2022. Particular emphasis was placed on securing sufficient samples for improved and temporary grassland classes, enabling statistically robust estimates of their global extent and change. Together, these results provide the first consistent quantification of the global distribution and dynamics of grassland systems, supporting applications in land-use modelling, biodiversity assessment, carbon-cycle research, and analyses of management intensity and human influence. They also contribute to ongoing discussions around a potential global “pasture peak,” for which emerging evidence suggests a decline beginning around the year 2000.

Earth observation–derived land use and land cover (LULC) datasets support an increasing range of official statistics, including greenhouse-gas inventories, natural-capital accounting, agricultural monitoring, and LULC change assessments. Operational statistical workflows therefore require long, spatially consistent LULC time series that support area-adjusted estimation and robust map-to-statistics conversion. The ESA Climate Change Initiative (CCI) / EU Copernicus Climate Change Service Medium Resolution Land Cover (MRLC) product provides global annual LULC change maps at 300 m resolution from 1992 onwards, using a categorical legend of 38 classes and validated to CEOS Stage-4 standards. However, the categorical nature and spatial resolution of this long-term global LULC record limit its direct use for quantitative LULC statistics, change accounting, and post-classification inference, despite its temporal depth and strong spatio-temporal consistency. Here we present a globally consistent framework to transform the MRLC categorical map series into a spatially explicit global plant functional type (PFT) fractional dataset that preserves the temporal consistency of the original record while resolving sub-pixel LULC composition. Sub-pixel fractions for 19 PFTs are derived annually by integrating the MRLC time series with multiple independent high-resolution (10–30 m) EO datasets, replacing class-level assumptions with spatially explicit information informed by finer-resolution observations than the native 300 m grid. The resulting product delivers more than 30 years of annual global PFT sub-pixel fractional cover maps at 300 m resolution, covering bare soil, surface water, permanent snow and ice, built-up areas, natural and managed grasses, and woody vegetation resolved by major functional types. It substantially reduces reliance on mosaic classes and exposes intra-class variability relevant for area-adjusted accuracy assessment and map-to-statistics conversion. Key advances include a global rebalancing of shrub and grass fractions, explicit C3/C4 grass partitioning, and refined tree functional-type attribution aligned with large-scale ecological gradients. The PFT dataset is distributed as a companion product to the MRLC categorical map series, ensuring traceability and compatibility with existing reporting frameworks. As the ESA CCI / EU C3S LULCC record extends beyond 2022, the associated PFT map series will evolve consistently, enabling long-term integration into LULC and environmental accounting frameworks.

Conducting national agricultural censuses and formulating rural public policies depend on a territorial infrastructure compatible with official statistical systems. In vast countries like Brazil, rural heterogeneity and the prevalence of unpaved roads pose a structural challenge for road network representation, directly impacting census logistics, accessibility analysis, and statistical quality.Brazil’s current road databases are fragmented across collaborative, commercial, and administrative sources. Individually, these sources exhibit significant limitations such as coverage gaps in remote areas, geometric and topological inconsistencies, scale disparities, and limited compatibility with statistical territorial units, restricting their application in census operations and territorial analysis.This work presents an integrated methodology for geospatial data fusion and validation, developed by the Brazilian Institute of Geography and Statistics (IBGE), to construct a national rural road network. The approach integrates multiple sources from the Earth Observation ecosystem and open road map databases, including OpenStreetMap, Overture Maps, Planet Labs' Road Detection, IBGE’s operational databases, enumerators' GPS tracking data from previous censuses, and state-level records. Each source is standardized before integration into a unified network, with post-processing to remove redundancies while preserving original metadata.The resulting road network is a graph enabling navigation to points of interest and census data collection units. This infrastructure supports field operation planning, allows automatic route optimization, improves census coverage, and enhances accessibility analysis at municipal and intra-municipal scales. Furthermore, this database establishes a standardized territorial framework for all IBGE field operations. Preliminary results indicate increased rural road coverage and reduced operational uncertainty in underrepresented areas. By integrating Earth Observation, vector infrastructure, and official statistics, this initiative highlights the strategic role of geospatial infrastructure in supporting field operations and evidence-based public policy.

In recent years, the availability of EO-based forest biomass information has grown substantially, yet its uptake for reporting national statistics in environmental policy frameworks (including REDD+ and GHG inventories) remains limited. National Forest Inventories (NFIs) continue to be the primary source for forest biomass estimation and emission factors, but many countries face challenges in completing or updating their inventories, affecting the timeliness and quality of reporting. Key barriers to the uptake of EO products include the lack of comprehensive guidance with country-relevant examples. While both the IPCC 2019 Refinement and the 2020 GFOI Methods and Guidance mention the possibility of using EO biomass products for national reporting purposes, operational guidance remained limited. Since then, expert-exchanges across different communities (e.g., country representatives, map developers, research institutions, verification bodies) examined the suitability of EO products for operational use. These exchanges highlighted a shared need for clearer guidance on how biomass maps can be used and integrated with NFI data. In response, a new GFOI Module was recently released to support informed decision-making on the use of space-based forest biomass datasets within multipurpose National Forest Monitoring Systems. This presentation will provide an overview of the new guidelines and the process that led to their development, illustrating how EO can support national forest monitoring and reporting in climate policy frameworks. It will then showcase a case study from Peru, exemplifying how space-based biomass products can be integrated with NFI data to improve the precision of biomass estimates using different inferential approaches. The example compares robust statistical techniques from traditional design-based to more complex model-based approaches, clarifying their assumptions and suitability across geographic scales, and highlights key opportunities and challenges for operational integration. We aim to show how accounting for user needs, having clear guidance, and practical applications can bridge EO advances into policy frameworks.

Earth Observation (EO) based monitoring of forest cover change represents a key element of policy relevant indicators supporting environmental sustainability and climate change mitigation. Reliable and spatially consistent information on the extent and dynamics of tree cover underpins assessments of biodiversity, land degradation, carbon balance, and the impacts of human activity. This contribution presents the use of satellite derived indicators disseminated through the National Satellite Information System (NSIS-https://nsisplatforma.polsa.gov.pl/?lang=en) operated by the Polish Space Agency as an operational framework for integrating EO data into public administration workflows. The platform provides access to a wide portfolio of satellite products covering the territory of Poland, generated on a regular basis and supported by operational annual monitoring schemes that produce consistent time series data suitable for long term environmental analysis and official reporting. Among these products are: national maps of tree covered areas (2024, 2025) and multi-year tree cover change maps (2016-2024) divided into dominant forest type. The satellite products are generated using Sentinel data and automated algorithms calibrated to the specific environmental and land use conditions of Poland, ensuring overall accuracy>90%. The indicators support climate and land use reporting carried out by the Ministry of Climate and Environment, including national obligations under Article 8 of climate governance frameworks, by providing spatially consistent evidence on tree cover and its changes. They also complement national inventories used by the Statistics Poland, enabling the implementation of environmental economic accounts in line with Regulation EU 2024/3024 through harmonised and statistically consistent EO based data. This contribution demonstrates how an operational NSIS platform, designed to be user-friendly and allowing even non-specialist officials to easily access and use it, can bridge the gap between satellite data and official reporting, enabling the systematic use of validated EO-based indicators for environmental statistics, sustainability monitoring, and evidence-based policy making.

Originally designed for the management of forest resources, contemporary Forest Inventories (FI) now address a more complex set of variables that capture the diversity, health, and structural composition of forests. As a result, FI are increasingly used in research addressing climate change and biodiversity conservation. Within the Earth Observation (EO) domain, FI data have proven to be a key source of information for calibrating and validating EO-derived models and supporting spatial and temporal predictions. Ultimately, FI-EO integration enhances estimates of forest biomass, characterizes forest structure and productivity, improves understanding of the role of forests in climate and biodiversity, and increases the coverage and timeliness of forest information for monitoring and official reporting purposes. Research integrating FI and EO data is now widespread across temperate, tropical, and boreal forests at subnational, national, and continental scales. These efforts are particularly relevant in the context of land-related policies that emphasize transparent, timely, and robust reporting. However, despite progress, no consolidated overview exists that synthesizes the diverse purposes, approaches, and methods underpinning FI-EO integration. An overview of current FI-EO integration efforts, is crucial for understanding its full potential and benefit for environmental monitoring. This presentation addresses this need by summarizing findings from our systematic review of over 200 studies on FI-EO integration for environmental applications. We provide a comprehensive overview of current application areas, examine regional and temporal trends in integration methods, and highlight how FI-EO approaches are being leveraged to inform policy-relevant sustainability indicators. By synthesizing methodological advances and emerging opportunities, this work aims to support collaboration between data producers and users, support future inventory and earth observation campaigns, and enhance the utility of integrated FI-EO data for reporting and environmental statistics. In doing so, it outlines pathways to strengthen the contribution of FI–EO integration to sustainability monitoring and climate action.

Tropical countries require activity data derived from the interpretation of satellite imagery to comply with a range of national and international policy requirements. As countries work to establish and operationalise National Forest Monitoring Systems (NFMS), it is critical to develop multi-purpose and interoperable methodological approaches to satisfy multiple policy objectives. The proposed approach relies on the development of a suitable stratification based on, preferably, national land use/forest cover and associated changes or, alternatively, global, map products. The stratification is optimised by first interpreting sample-based observations organised in a systematic sample over the whole country. This sample is then used to assess the quality of the stratification with a view to optimise change detection and ensure that changes are appropriately captured. An intensified optimized stratified systematic sampling approach is then applied, using the estimation of variance of change in each stratum to apply the Neyman algorithm for optimising the allocation of sample units in each stratum. A suitable response design is applied for visual interpretation of contemporary high resolution satellite imagery for each reference year based on a detailed nomenclature not only capturing forest cover and change but also other land cover/use types including cropland providing an exhaustive assessment of relevant carbon pools. This approach was applied in the Republic of Guinea with the visual interpretation of 2170 sample units. The total forest area of Guinea amounts to 9.4 million ha with an uncertainty estimated at 3.4% at 90% confidence interval. The deforestation between 2015 and 2020 represented 0.5 million ha with an uncertainty of 8.5% at 90% confidence interval corresponding to an annual deforestation rate close to 1%. Forest losses. Results for 2020-2025 will soon become available and the approach will also be implemented in other countries in the Upper Guinea basin as part of a World Bank initiative.

Forest ecosystems are increasingly threatened by climate-driven stressors, calling for consistent and repeatable monitoring approaches able to assess their condition and stability over time and space. The development of spatially explicit indicators capturing forest health and status is therefore a priority for climate action frameworks, including the Sustainable Development Goals, the EU Green Deal, and international climate and biodiversity reporting obligations. Within the context of the EO4Resilience project, we present a synergistic multi-modal approach based on simulated data from the future Copernicus Expansion missions, specifically the hyperspectral CHIME (Copernicus Hyperspectral Imaging Mission for the Environment) and L-band SAR ROSE-L (Radar Observing System for Europe in L-band) satellite missions. This approach enables the retrieval of key functional and structural forest traits, here proposed as EO-derived policy relevant biodiversity indicators. A machine learning-based workflow is developed to estimate species traits (e.g. Leaf Chlorophyll, Nitrogen, and Water Content) and ecosystem function properties (e.g., Leaf Area Index, Specific Leaf Area). Complementarily, SAR-based physical models (e.g. Water Cloud Model) and polarimetric decomposition are exploited to retrieve structural parameters (e.g., Above Ground Biomass), as well as the Forest Water Stress (FWS) index to track canopy water content using SAR-derived moisture change indices. The methodology is demonstrated on a "Representative Dataset" simulating CHIME and ROSE-L acquisitions through optical and SAR airborne data (AVIRIS-NG/4, F-SAR) and existing satellite data (PRISMA, EnMAP, SAOCOM) over three diverse European test sites (Switzerland, Italy, France) characterized by a broad ecological gradient. Furthermore, extensive field campaigns provided ground-truth data for calibration and validation of functional biodiversity indicator maps. Results show ecologically coherent spatial patterns and robust transferability across sites, confirming the feasibility of generating operational and scalable indicators of forest condition from future spaceborne missions. These indicators might directly support environmental reporting, policy monitoring and sustainability assessment.

The Food and Agriculture Organization (FAO) complements data collected via a country reporting process through periodical remote sensing surveys. These surveys are based on stratified random sampling and visual interpretation of a large number of samples by local experts to produce robust estimates on forest area, forest area changes, and other key indicators. This contribution presents the latest enhancements of the Remote Sensing Survey (RSS) methodology to improve efficiency of the data collection process and precision of global forest area and forest area change estimates. Building on the FRA 2020 RSS design—which combined 20 Global Ecological Zones with four tree-cover change strata derived from the Global Forest Change dataset (Hansen et al., 2013 ) - FAO designed a new survey with improved sampling efficiency through (i) restratification, updating strata to reflect recent forest change dynamics , and (ii) reallocation, using updated stratum-specific variances to optimize the allocation of the samples. An additional stratum was added to target changes that occurred between 2018 and 2024, and re-interpretation of existing samples was streamlined to increase efficiency. The refined methodology was piloted in Zimbabwe and Bolivia. The pilots demonstrated that sample reallocation resulted in a substantial improvement in standard errors of forest loss estimates, whereas re-stratification's impact was more modest. Together, these changes indicated that comparable precision to FRA 2020 could be achieved with approximately 15 percent less samples. At the global level, these improvements resulted in a reduction of the sample size from around 400,000 to approximately 340,000 plots and reduced the time used to interpret the samples. To further streamline the process and reduce expert interpretation workload, we also tested a vision-capable large language model (VLM) workflow that integrates plot geometries and multi-source satellite imagery to generate standardized land-use interpretations. Our preliminary results show that this approach can further reduce the workload in interpreting the samples as the system reliably screened no-change cases with high precision (minimizing false positives) and flagged candidate classes, supporting expert review. Recognizing that photo-interpreting around 340,000 plots is time‑intensive and subject to human error, these results underscore the promising role of AI to accelerate future RSS cycles—particularly for pre-screening, quality control, and targeted expert allocation—while preserving the statistical robustness central to FRA RSS.

Forests are constantly undergoing changes driven by both natural and human-induced processes, which can significantly alter their extent. Regularly updated information on forest extent is therefore essential for effective monitoring, sustainable management, and evidence-based policy development. At the European level, such information is primarily derived from national forest inventories and, more recently, from remote sensing based products such as those provided by the Copernicus Land Monitoring Service (CLMS), including the High-Resolution Forest Type Layer. However, inconsistencies between forest definitions adopted by remote sensing products and those used in national and international reporting frameworks, such as the FAO Global Forest Resources Assessment, can lead to substantial discrepancies in forest extent and change estimates. While remote sensing products typically capture tree cover as a land cover category, national inventories classify forests based on land use, which includes temporarily unstocked areas resulting from management activities such as clear-cutting or disturbance events. Mapping forest area instead of forest cover can therefore offer deeper insights into forest dynamics. For instance, detecting net forest area loss can serve as an indicator of potentially unsustainable management practices and help identify regions where forest degradation may lead to the conversion of forest area to different land uses. In this study, we present a workflow based on remote sensing that shifts from a land cover to a land use perspective to map forest areas at the European scale. The method combines current land cover information, specifically Copernicus Land Monitoring Service products, with a historical dataset, the European Disturbance Atlas (Viana-Soto and Senf, 2025). This integration introduces a temporal dimension that enables the identification of non-treed areas that were previously forested but still designated for forestry use, while excluding areas where a land-use conversion has occurred (e.g., from the conversion of forest to agriculture). In detail, the current forest cover (2020) is first delineated by integrating the CLMS Forest Type and Tree Cover Density (thresholded at 10%) products. The Disturbance Atlas is used to detect areas that experienced forest disturbances (natural or anthropogenic) that may currently appear as unstocked forest or reflect a land-use change. To verify whether these areas have indeed undergone a land-use transition, CLMS land cover datasets are used to identify zones presently classified under non-forest land-use categories. It is worth noting that, as CLMS has recently begun providing annual updates of several of its land cover products, the proposed product could be regularly as new versions become available. Preliminary results indicate that, compared to the forest area estimates of 159 Mha (FAO) and 160 Mha (Eurostat) for the European Union, the proposed forest area product yields an estimate of 163 Mha, while the current forest cover derived from CLMS amounts to 157 Mha.

The PathFinder project is developing a prototype European Forest Monitoring System that integrates National Forest Inventory (NFI) data with Earth Observation (EO) data to map and improve estimates of key forest structural attributes, including above‑ground biomass change. ICP Forests data are further used to model soil carbon dynamics. For the prototype, more than 500,000 NFI plots from 14 countries have been linked with Sentinel‑2 and Copernicus datasets. This enables robust detection of annual biomass changes on the order of 1.5%. High‑resolution (10 m) maps of biomass, diameter, height, and dominant tree species have been produced and are now used to simulate future forest development under a range of scenarios. Maps, estimates, and scenario results are available through the PathFinder mapping and estimation platform at https://portal.pathfinder-heu.eu/. The European National Forest Inventory Network (ENFIN) plays a key role in the governance and implementation of a European Forest Monitoring System, ensuring harmonization, coordination, and long‑term sustainability of monitoring efforts.

Introduction: Forest management at the national level requires precise information at various spatial scales to meet the needs of multiple bodies and stakeholders involved in land management. National forest inventories (NFIs) can be the primary source of these data if supported by inference techniques that exploit the availability of remotely sensed data, which is usually available on a large scale and with high spatial and temporal resolutions. Small-area estimation (SAE) approaches use field observations and spatially continuous predictors to estimate forest parameters and their associated standard errors (SEs). Objective: This study presents the first national-level SAE of forest mean growing stock volume (GSV) in Italy, conducted at the municipal level, thereby augmenting the NFI's spatial resolution. Methods: We tested two unit-level SAE estimators, the Empirical Best Linear Unbiased Predictor (EBLUP) and the Mixed Effect Random Forest (MERF), for estimating the total GSV and its SE. Small-area estimation relies on a statistical model that links field observations to unit-level auxiliary variables. The underlying models in the two cases are the nested error linear regression model (NER) and the random forest model. Both were trained on the GSV from NFI field data and used to produce pixel-level predictions using a set of 138 remote sensing predictors. These predictions were averaged at the municipality level, and the MSE and SE were estimated via a semi-parametric wild bootstrap procedure for each municipality in Italy. Results: The R² values were 0.39 and 0.46 for the NER and MERF models, respectively. Mean GSV varies from 30 m³ ha-¹ to 320 m³ ha-¹ with SE ranging from 10 m³ ha-¹ to 68 m³ ha-¹ under the EBLUP framework. With the MERF estimator, the mean GSV ranged between 11 and 819 m³ ha-¹ with SE from 9 to 185 m³ ha-¹. The mean SE at the municipality level was 29 and 27 m³ ha-¹ with EBLUP and MERF, respectively. Significance: This is the first national-scale SAE estimation conducted at the municipality level in Italy. Our results offer new insights into the distribution of forest stocks across the country, providing a spatial scale suitable for finer-scale forest planning and management. Moreover, we tested the potential benefits of random forests in the context of SAE. Random forests make the SAE model robust against model failure (e.g., providing insurance against model misspecification) and can fit complex, potentially non-linear interactions between covariates, offering an effective handling of outliers. Moreover, random forests handle high-dimensional datasets, enabling the use of Big Data sources, which are now the standard in remote sensing. All these benefits led to more accurate predictions of the mean GSV and its SE. Keywords: National Forest Inventory, remote sensing, small-area estimation, EBLUP, random forests

Mangroves are a key component of national sustainability statistics, contributing to climate regulation, coastal protection, and biodiversity conservation. This contribution presents a comparative analysis of mangrove extent and change in Cambodia, Ecuador, and Guinea-Bissau, developed under the ESA-funded GDA Marine activity to support International Financial Institutions. It demonstrates how EO-derived information can underpin policy-relevant sustainability indicators and official environmental reporting in data-scarce contexts. In Guinea-Bissau, a national-scale assessment of mangrove extent between 2018 and 2023 reveals a net loss of approximately 9,100 hectares, corresponding to an overall decline of 5.1%. Spatially disaggregated statistics highlight strong regional contrasts, with several coastal zones experiencing losses exceeding 10-20%, alongside limited localised gains. These results provide quantitative indicators of habitat change that can be directly integrated into national forest, coastal ecosystem, and climate-related inventories. In Cambodia, mangrove change statistics for 2017-2021 were analysed together with trends in marine environmental parameters relevant to ecosystem sustainability. Mangrove losses were concentrated in specific coastal zones, while others remained relatively stable. Over the same period, EO-based marine indicators indicate a sea-level rise of approximately 2-4 cm per year, an increase in surface salinity (up to 0.1 psu per year in some areas), and strong seasonal variability in temperature and chlorophyll levels. These quantitative indicators support the interpretation of mangrove change in relation to environmental pressures and are directly relevant for national coastal management and climate adaptation reporting. In Ecuador, a standardised national baseline map of mangrove extent for 2023 was produced to support consistent national statistics and future change assessments, addressing gaps in up-to-date mangrove indicators required for policy and reporting. Overall, the three case studies illustrate how harmonised EO-derived statistics on mangrove extent, change, and environmental conditions can strengthen sustainability indicators, improve temporal comparability, and support official reporting on ecosystems, biodiversity, and climate action.

As environmental pressures intensify, policymakers require rapid and accurate biodiversity estimates to inform their policy decisions. Traditional monitoring methods, however, cannot meet this demand for speed without sacrificing spatial coverage. This study investigates whether high-resolution aerial photography and sparse field data from the Landelijk Meetnet Flora (LMF) can be combined to estimate forest diversity across Dutch forested ecosystems in a manner that complements official monitoring and reporting obligations. To address the limited availability of labeled field observations, semi-supervised learning is employed to generate pseudo-labels from unlabeled aerial imagery, enabling broader spatial coverage. Forest species diversity is quantified using the Shannon diversity index derived from LMF survey data. The methodological approach compares neural networks trained end-to-end against models utilizing pretrained encoders. This comparison is particularly relevant for national statistical agencies, where the computational costs of training models from scratch must be weighed against the reproducibility and consistency requirements of pretrained architectures. The analysis spans multiple years of aerial imagery coverage across forested areas in the Netherlands. Critically, this neural network-based approach enables biodiversity estimates to be generated immediately upon the availability of aerial photography mosaics, reducing turnaround time from the multiple years typically required for traditional mapping methods to near-instantaneous assessment.

Water quality in inland bodies of water in Poland is a growing environmental concern, impacted by urbanization, industrialization, and inadequate wastewater treatment. These issues lead to significant pollution, damaging ecosystems and human health. Traditional water quality monitoring, relying on manual sampling and laboratory analysis, is often costly and time-consuming, particularly in rural and remote regions. To address these challenges, we propose a cost-effective, real-time solution for water quality monitoring using Earth Observation (EO) data from Sentinel-2 satellites and Python-based data analysis. The project seeks to address these challenges by utilizing the capabilities of satellite-based remote sensing. Sentinel-2 satellites, operated by the European Space Agency, provide high-resolution multispectral imagery that is well-suited for monitoring water quality. By analyzing this data, we can derive critical water quality indicators, such as chlorophyll concentration, turbidity, and the presence of harmful algal blooms. This approach not only reduces costs but also allows for the continuous monitoring of large and remote areas, making it an ideal solution for Poland’s diverse landscape. Our methodology involves acquiring satellite imagery through the CREODIAS platform, processing data with Python libraries like Rasterio and GDAL, and storing results in cloud-based services (Amazon S3 and PostgreSQL). The geoportal will visualize this data using Web Map Services (WMS) for easy access and interpretation by both researchers and the public. This system empowers users to track water quality trends and encourages community engagement in water conservation. The project offers a scalable model for monitoring inland water quality and can be adapted to other regions facing similar challenges. By providing real-time, actionable data, the project supports better decision-making in water management, contributing to environmental sustainability and climate action.

The growing need for timely economic indicators, particularly during periods of economic uncertainty, has highlighted the limitations of traditional macroeconomic indicators, which often suffer from substantial reporting delays. Earth Observation (EO) offers promising opportunities to complement official statistics by providing economic proxies in near real-time. In this study, we examine the potential of Sentinel-1 Synthetic Aperture Radar (SAR) data to monitor production activity in Germany's automotive sector, a critical component of the national economy. Leveraging the advantages of Sentinel-1, including weather-independent data acquisition and revisit rates of a few days over Germany, we developed an approach to estimate production levels through continuous monitoring of production parking lot occupancy at 18 domestic automotive production sites. Analysis covering data from October 2014 to June 2024 reveals a relatively strong correlation of 0.74 between derived parking lot occupancy metrics and production figures from the German Association of the Automotive Industry (VDA). The results highlight both the potential and challenges of satellite-derived economic indicators. While the approach successfully captures overall production trends with significantly reduced latency compared to traditional reporting cycles, limitations remain in deriving production levels solely based on production parking lot occupancy. Detecting short-term production disruptions is particularly challenging due to data limitations and site-specific operational variations. Nonetheless, this study demonstrates how EO can enhance the timeliness of key economic indicators, thereby better capturing the fast-paced nature of global economic developments. Future work should focus on methodological refinement and extension to other economic sectors, including retail or tourism, where occupancy patterns of parking lots could serve as reliable proxies for economic output.

EU peatlands cover only 2% of land area yet store 30% of terrestrial carbon, exemplifying the complexity of Europe's environmental monitoring landscape, these ecosystems require monitoring under six different policy frameworks: the Carbon Removal Certification Framework (CRCF), the Common Agricultural Policy (CAP), the Nature Restoration Regulation (NRR), Habitats Directive (HD), LULUCF Regulation, and Soil Monitoring Law (SML). Each of these mandates different reporting obligations, indicator identifications, spatial frameworks, and temporal cycles. For National Statistical Institutes (NSIs) and national reporting entities, this creates substantial administrative burden through duplicative data collection, incompatible datasets, and diverging reporting schedules. To address this, the Knowledge Centre on Earth Observation (KCEO) conducted a comprehensive interoperability assessment to identify how Earth Observation (EO) can streamline reporting for addressing peatland monitoring needs. Our methodology systematically follows three main steps: (1) Policy context and semantic analysis: analysing policy framework, reporting obligations and terminology conflicts; (2) Application needs: Building on the policy context analysis we identify operational needs where EO capabilities could enhance policy implementation. We translate these needs into concrete application requirements, defining necessary indicators, scale, type of analysis and reporting formats; (3) Fitness for purpose: For each parameter identified, we systematically screen all available EO products and services against the defined technical requirements The cross-policy interoperability analysis indicates that while significant potential exists for "measure once, report many times", critical semantic barriers persist, starting with the semantic difference of how indicators are defined. For instance, Soil Organic Carbon (SOC), vital across all frameworks, is reported inconsistently - concentration (g/kg) for soil health versus stock (tC/ha) for climate policy. Furthermore, incompatible soil depth conventions (0-30cm vs 0-100cm), spatial classifications (LPIS parcels vs, soil districts vs Natura 2000 sites), and temporal cycles (annual crop monitoring vs. 6-year conservation status reporting) prevent data integration despite measuring fundamentally similar parameters. For National Statistical Institutes and reporting entities, EO offers a pathway to transform peatland monitoring from periodic field surveys to continuous spatial statistics. Standardized EO-based indicators can serve multiple reporting frameworks simultaneously, reducing data collection burden while improving temporal consistency and spatial coverage. Achieving this transformation requires: (1) harmonized definitions for key monitoring parameters enabling cross-policy comparability; (2) standardized indicator frameworks that translate EO observations into policy-relevant statistics; and (3) capacity building to integrate geospatial products into official statistical workflows and national reporting systems.

The growing availability of EO-derived indicators for SDG and environmental policy reporting has not yet translated into consistent, high-impact use in official decision-making [3][4]. This contribution presents the emerging Space Impact Forum (SIF) Impact Framework and Impact Labs concept as a prospective model for helping National Statistical Institutes (NSIs) and policy bodies translate EO outputs into trusted, decision-ready statistical narratives [1][2]. Drawing on SIF’s design across Food, Energy, Environment, and Urban Labs, the framework connects EO data and processing chains to indicators, policy outcomes, and wider societal impacts using a Theory of Change, Logical Framework, and sector-specific Key Performance Indicators (KPIs) to structure measurement and learning [1][2]. Impact Labs are conceived as structured, outcome-driven collaborations where satellite intelligence, analytics, policy expertise, and multi-stakeholder engagement converge to co-design interventions that are measurable, scalable, and adaptable, making them a natural testbed for EO-based official statistics and policy applications [1]. The presentation demonstrates how this impact-oriented approach can map EO data → indicators → policy outcomes → societal impact in ways that align with statistical standards while remaining usable for policymakers [3][5]. It outlines a methodology that starts from scoping and baseline definition, proceeds through co-design and indicator selection, data collection and monitoring, and culminates in analysis, evaluation, and reporting, all underpinned by systematic monitoring and evaluation and integrated satellite, geospatial, and operational data streams [1][2]. Using illustrative case scenarios rather than completed deployments, the contribution shows how improved narrative clarity around EO-derived indicators—anchored in explicit impact pathways and supported by outputs such as dashboards, briefs, and toolkits—can increase institutional uptake and support SDG and environmental policy reporting [4][6][1]. Particular emphasis is placed on aligning uncertainty, metadata, and caveats with the communication needs of NSIs and policymakers, including guidance on avoiding misinterpretation or misuse of EO proxies, scale mismatches, and overconfident narratives, which are recurrent challenges in EO-for-statistics integration efforts [3][5][7]. By positioning the Space Impact Forum’s Impact Framework and Labs as a pre-operational, replicable model, the work offers statistical authorities a roadmap for mainstreaming EO-derived indicators into future reporting cycles while strengthening transparency, accountability, and evidence-based sustainable development policy [1][2]. Citations:[1] Impact Labs https://spaceimpactforum.com/impact-labs [2] Space Impact Forum https://spaceimpactforum.com [3] EARTH OBSERVATION FOR SDG: Compendium of Earth Observation contributions to the SDG Targets and Indicators (May 2020) https://eo4society.esa.int/wp-content/uploads/2021/01/EO_Compendium-for-SDGs.pdf [4] Perspectives on EO for the SDGs https://eohandbook.com/sdg/part2_2.html [5] Andries, A.; Morse, S.; Murphy, R.J.; Lynch, J.; Woolliams, E.R. Using Data from Earth Observation to Support Sustainable Development Indicators: An Analysis of the Literature and Challenges for the Future. Sustainability 2022, 14, 1191. https://doi.org/10.3390/su14031191 https://eprintspublications.npl.co.uk/9616/1/eid9616.pdf [6] Green Orbit Space Communications and PR resources https://greenorbit.space/resources/ [7] Connors, S., Schneider, R., Nalau, J. et al. Earth observations for climate adaptation: tracking progress towards the Global Goal on Adaptation through satellite-derived indicators. npj Clim Atmos Sci 8, 359 (2025). https://doi.org/10.1038/s41612-025-01251-1

Reliable and harmonized land use information is a core requirement for official statistics supporting spatial planning, infrastructure monitoring, and environmental reporting. In the Netherlands, this role is fulfilled by the Bestand Bodemgebruik (BBG), the official land use product of Statistics Netherlands (CBS). BBG integrates national registries, topographic reference data, and aerial imagery to produce spatially explicit land use information. Despite recent automation efforts, the production workflow remains partially manual and resource-intensive, motivating Earth Observation (EO)- driven methods for land-use mapping. This study presents an operational deep learning framework designed to support future BBG production cycles. Using nationwide 25 cm RGB aerial orthophotos and BBG parcel-level reference labels, two complementary classification strategies targeting 14 built-up land use classes were developed. The framework was evaluated using geographically separated training and validation regions to support robust operational deployment. The first approach applies a hierarchical RGB-only pipeline combining an EfficientNet-based built-up detector with a Swin Transformer for fine-grained classification. The second extends this pipeline through multimodal late fusion of parcel-level registry attributes such as building function and morphological indicators. Model outputs consist of parcel-linked raster predictions aggregated to official statistical reporting units. Results show robust performance across urban land use categories. Although the overall macro-averaged F1-score increased only marginally (64.3% to 65.8%), the fusion model demonstrated clear advantages for transport and infrastructure-related classes, particularly rail, metro/tram, and main roads, where registry-derived contextual information improved classification reliability. The analysis highlights limitations of purely image-based classification for functionally mixed urban categories. Overall, the framework supports a hybrid operational strategy that selectively integrates registry data when visual information alone is insufficient, and it demonstrates the feasibility of incorporating EO-driven deep learning into national land-use production workflows. This work was funded under the GEOS2023-NL programme (Project PR003214, WP2–D2.2), supporting AI-enabled geospatial production for national statistics.

This study investigates the spatio-temporal changes in land cover in the Pisani Mountains of Central Italy following the 2018 wildfire, utilizing a multi-year dataset from Sentinel-2. The analysis encompasses all years up to 2024. A Random Forest (RF) classifier was used to define changes in vegetation land cover, and it proved to be quite effective in distinguishing post-fire trajectories. Before examining the classification in the burned area, the Random Forest (RF) model was applied to the surrounding region, which had not been affected by wildfires since 2000, to assess the stability of the model's predictions in this reference area. After validating the RF model in the reference area, it was implemented in the burned area to investigate the ecological trajectories of the various land cover classes. The model's performance was evaluated during the calibration process and through a field sampling campaign, where new plots were collected to validate the model's predictions for 2023. To enhance our understanding beyond mere quantitative classification changes of vegetation classes, we analyzed the statistical distribution of the Normalized Difference Vegetation Index (NDVI) within each class. Utilizing a bootstrap resampling method, we assessed the variations in vegetative growth among different classes along all the years. The results reveal that while certain classes experienced significant impacts from the wildfire, others exhibited ecological resilience, demonstrated by consistent recovery and positive vegetation responses.

Regulatory and reporting frameworks increasingly rely on spatially explicit information on agricultural commodities to support land-use monitoring, deforestation risk assessment, and supply-chain analysis. In many tropical regions, agricultural practices such as smallholder farming, mixed cropping, and cocoa agroforestry create heterogeneous landscapes with variable canopy cover and fragmented land-use patterns. While cocoa maps derived from freely available 10 m satellite data offer broad coverage and low cost, their performance can vary strongly with canopy cover, cocoa density, and landscape fragmentation, making it essential to understand in which contexts such products are sufficient and where higher-resolution information provides added value. We present a stratified evaluation framework for cocoa mapping in Côte d’Ivoire, enabled by a large, transparent reference dataset constructed from curated cocoa inventories, auxiliary thematic datasets, and systematic visual interpretation of 0.5 m satellite imagery, followed by manual correction and independent field-based validation. Using this reference, we trained a deep learning ensemble on Pléiades imagery to produce 0.5 m cocoa predictions with high precision while substantially reducing false positives through a precision-oriented decision threshold. To compare mapping performance across landscape contexts, we implemented a stratified design based on a regular 256 m tile grid and two interpretable proxies of mapping difficulty: Tree Cover Density as an indicator of canopy closure, and Patch Density as an indicator of fragmentation and edge-driven spectral mixing. We visually interpreted 9,000 reference points balanced across nine Tree Cover Density × Patch Density strata to quantify stratum-specific, product-level accuracy of VHR-derived predictions and publicly available 10 m cocoa maps. Results show that in structurally homogeneous landscapes with moderate canopy closure, 10 m products provide stable and reliable cocoa information. In contrast, VHR mapping provides clear advantages both in low canopy density regimes (e.g., young or sparsely planted cocoa, where crowns occupy only a fraction of a 10 m pixel) and in high canopy density and/or highly fragmented landscapes (e.g., dense agroforestry and smallholder mosaics), where sub-pixel heterogeneity and boundary effects dominate. These findings provide general guidance on resolution-appropriate data selection and uncertainty-aware use of EO products in regulatory and statistical applications.

The integration of Earth Observation (EO) into official statistics is often constrained by the difficulty of validating detected changes and by the cost of collecting ground-truth data required by supervised classification models. This limitation is especially highlighted when National Statistical Offices require timely and reliable environmental indicators, yet lack dense or regularly updated in-situ monitoring networks to support model training and validation. To address this gap, this work proposes a universal bitemporal indicator for change detection. The methodology is index-agnostic: once a standard reference index—such as NDWI for water or NDVI for vegetation—is selected, the algorithm exploits the Bitemporal Spatial Autocorrelation (BSAC) structure to quantify change directly from EO data, without the need for training datasets. By measuring the loss of symmetry in the spatial autocorrelation matrix, the method produces a statistically grounded, objective, and reproducible global "Change Score". The operational value of the approach is demonstrated through a case study of the 2024 hydrological crisis of Lake Fanaco (Italy). This event, representative of the increasing water stress affecting Mediterranean regions and many parts of the Global South, provides a benchmark for monitoring rapid resource depletion. As reported by national monitoring bodies, the lake reached critically low levels during the 2024 drought. Using satellite observations alone, the proposed indicator successfully quantified the lake’s desiccation process. Overall, this unsupervised framework offers a scalable solution for NSOs to improve the timeliness and spatial granularity of environmental reporting, even where reference data are scarce.

The rapid expansion of EO capabilities for agricultural monitoring, driven by open-access satellite missions and growing computational resources, has markedly advanced both research and operational applications. However, progress is increasingly constrained by a persistent imbalance between the abundance of satellite data and the limited availability, quality, and accessibility of in situ agricultural observations required for the calibration and validation of EO models. Agricultural field data are particularly challenging to collect and disseminate due to strong spatial and temporal variability, resource-intensive field campaigns, inconsistent measurement protocols, and complex data governance issues related to privacy, ownership, and economic sensitivity. Beyond these structural constraints, evidence from multiple scientific domains indicates that technical, cultural, and institutional factors (e.g., limited data management support, uncertainty about reuse, lack of incentives, and concerns over misuse or loss of priority) play a major role in shaping data sharing practices. This presentation aims to initiate a community-wide reflection on current norms, perceived barriers, and enabling conditions for sharing in situ agricultural data in the EO community. To this end, we introduce an international survey targeting researchers and practitioners involved in agricultural remote sensing. The survey seeks to capture sharing practices and attitudes around in situ data, as well as views on potential incentives and safeguards. By engaging conference participants in this discussion and survey effort, we aim to build an empirical basis for future recommendations that are grounded in community perspectives and responsive to the practical realities of agricultural EO research.

Monitoring fisheries-related Sustainable Development Goal (SDG) indicators in Small Island Developing States (SIDS) remains a major challenge due to limited statistical capacity, fragmented data sources, and the complexity of assessing fisheries sustainability for a wide range of heterogenous techniques. Within the EO4SURE project, EO-based data processing pipelines are being developed to support SDG reporting for this topic in Cape Verde and São Tomé and Príncipe, using open-source tools and free licensing to promote long-term adoption. Automatic data processing pipelines are to be implemented within Champion Users digital infrastructure to support reporting of SDG indicators 14.4.1 (sustainable fish stocks) and 14.7.1 (economic benefits from sustainable fisheries). These pipelines combine satellite-derived environmental data from CMEMS, vessel tracking data (AIS), and fisheries yield and revenue statistics to extract timeseries covering the Sentinel era. Results will map areas of high productivity, provide estimates of apparent fishing effort, spatial distribution of catches, and economic performance. Studying the evolution of such indicators shall allow for insights on the impact and risk of climate change for the fishery industry, while also providing the baseline for future stock assessment efforts. The approach supports both the computation of national indicators and their geographical disaggregation, enabling more detailed assessments of fishing pressure and productivity. By design novel solutions based on space data, strengthening national monitoring infrastructures and fostering institutional uptake, EO4SURE contributes to improved SDG reporting, enhanced fisheries governance, and more sustainable management of marine resources in SIDS.

Crop mapping is a crucial step for modeling crop water use and estimating yield through remote sensing–based (RS) approaches. In this context, the Food and Agriculture Organization’s (FAO) Water Productivity through Open-access of Remotely sensed derived data (WaPOR) provides evapotranspiration estimates that have been integrated with crop type information in the Chokwe Irrigation Scheme (CIS) to support yield assessment. Reliable crop maps are essential inputs for applying crop-specific parameters and deriving meaningful spatial estimates of agricultural productivity. Further, they enable the identification of agricultural areas, providing a basis for accurately mapping rice, the main crop of economic in CIS, and ensuring consistency in the derived products. This study focuses on land cover and rice mapping within the CIS, one of the largest systems in Mozambique. Located in Gaza Province (Mozambique) along the Limpopo River, it covers approximately 30,000 hectares and supports numerous smallholder farmers cultivating rice as well as maize, sugarcane and vegetables. The CIS plays a vital role in both local and national food security. The resulting crop map will be used in the Irrigation Performance Tool (IPAT) to support the generation of more accurate yield and crop water use indicators. The IPAT is being developed with the National Irrigation Authority (INIR) under Mozambique's Ministry of Agriculture within the framework of the FAO WaPOR project, with the aim to create an operational tool for irrigation management and to support national agricultural monitoring. This study places in direct alignment with SDG 2 (Zero Hunger) and SDG 6 (Clean Water and Sanitation) indicators. A field data collection campaign was conducted between April and May 2025, resulting in 739 georeferenced ground truth points across six land cover classes: forest, grassland, built-up areas, wetlands, bare land, and cropland. For the latter, specific crop types were recorded, including rice, maize, sugarcane, beans, and tomatoes. Sentinel-2 time series imagery from the Copernicus program was used to derive a set of spectral and geometric indices, which served as inputs for the classification process. A Random Forest (RF) classifier was implemented in the Google Earth Engine environment and trained using all the collected points to automatically map land cover and crop types. The resulting land cover classification achieved an overall accuracy of 83%, with user's and producer's accuracy averaging 90% for the cropland class. The subsequent rice map reached an overall accuracy of 65%, while the rice class itself achieved an average accuracy of 87% highlighting the success of the classification process. Future work will include crop mapping for the second season, incorporating additional crops, and further validating the mapped areas using statistics provided by INIR staff working on-site within the scheme. There is potential for scaling the IPAT tool to other irrigation schemes in the country, which would require staff to be trained in this crop mapping workflow. This pilot-to-scale approach highlights the role of RS in supporting cost-effective agricultural statistics and enhancing the understanding of rice production potential within the Chókwè Irrigation Scheme.

Accurate quantification of greenhouse gas (GHG) emissions and removals associated with land‑use and land‑cover change (LULCC) is essential for tropical countries implementing REDD+ and operating under the enhanced transparency framework of the Paris Agreement. In this context, the Government of Cameroon, together with technical partners including the United States Forest Service (USFS), Spatial Informatics Group (SIG), and the Coalition for Rainforest Nations, has developed reliable activity data covering 2000–2023. With approximately 65% forest cover, Cameroon assessed national land‑use dynamics using Collect Earth Online (CEO), a visual interpretation platform integrating multi‑sensor satellite imagery (Landsat, Sentinel‑2, NICFI Planet). A systematic 4 × 4 km sampling grid produced 29,409 observation units, each subdivided into 25 sub‑points to capture spatial heterogeneity. Interpretation followed a three‑phase protocol (harmonization, paired interpretation, individual interpretation) supported by a national methodological guide. Quality assurance included cross‑checks, Google Earth Pro verification, and a comparative accuracy assessment against the 2015 national forest‑cover map, yielding an overall accuracy of 94.41% and a moderate Kappa coefficient (0.231), reflecting challenges in hydromorphic and plantation areas. CEO‑derived activity data were combined with national, regional, and IPCC (2006/2019) emission factors to estimate carbon‑stock changes in biomass, dead organic matter, and mineral soils. Results show an average deforestation rate of 22,633 ha/year over 2016–2020, mainly driven by conversion to agriculture (154,229 ha/year). Fires affected approximately 4,670 ha/year, while wood extraction—estimated at 3.6 million m³/year (roundwood) and 10.5 million m³/year (fuelwood)—was a major source of emissions. Despite these pressures, Cameroon’s forests remain a net carbon sink, supported by the high productivity of dense humid forests and natural regeneration. This assessment provides a transparent and reproducible foundation for Cameroon’s Forest Reference Level and strengthens national MRV capacities for REDD+, GHG inventories, NDCs, and potential Article 6 mechanisms.

Open-cast mining operations constitute a significant component of the Polish economy, providing approximately 40 different mineral resources that are essential for various industries. This extraction is facilitated by over 6,000 mining facilities across the country, collectively yielding minerals and mineral by-products. Open-pit mining has profound and enduring consequences for the landscape and ecosystems in affected regions. The most visible long-term effect is often a drastically altered landscape, characterized by large, excavated pits and extensive waste piles that can persist indefinitely. These waste piles can pose a long-term threat to the environment due to the potential for leaching toxic materials into the soil and water for generations. The project aims to detect and monitor open-cast mining sites using Sentinel-1 and Sentinel-2. Research focused on identifying active and abandoned mining areas, tracking changes in mining activity from year to year, and assessing environmental impacts. The Sentinel-2 satellite, a key instrument in the study, is equipped with advanced multispectral optical sensors capable of monitoring land use and land cover changes with high precision. These sensors include the Normalized Difference Vegetation Index (NDVI), the Normalized Difference Built-up Index (NDBI), and the Modified Normalized Difference Water Index (MNDWI). The study utilised these indices to analyse land cover classification, vegetation health assessment, and changes in the mining site boundary. Additionally, radar data from Sentinel-1 was used to improve surface deformation and excavation monitoring. Sentinel-1, with its synthetic aperture radar (SAR) capabilities, enables effective tracking of land surface changes, subsidence, and excavation progress. SAR interferometric coherence was employed to quantify temporal changes in surface stability, as mining activities typically lead to a rapid loss of coherence due to terrain disturbance. This project can be particularly beneficial for environmental agencies, governmental institutions, and mining companies, providing critical insights for environmental impact assessments and regulatory compliance monitoring.

One of the aims of Eurostat-funded 2024-FI-GEOS-GSFIBU-project is to investigate the utilization of EO data for the identification of construction project start and on the other hand to accelerate the accumulation of project start information in statistics compared to the information obtained through the register. EO data can be used to improve the up-to-datedness of the statistics and supplement start up data, especially for large office or industrial projects, which can have a significant impact on the figures in the statistics. This project is limited to new construction project start of large buildings, and the start time is when the building of the foundation of building starts. It is hoped that the start time would be estimated within one week from real start. For testing purposes, some recent construction projects have been collected from Ryhti-database with their coordinates, month and year of permission and recorded start and end dates of construction. Sentinel-1 (backscatter and coherence) and -2 (reflectance and image indices like NDVI) timeseries have been collected for these sites. One of the difficulties is the weather (clouds, seasons) for which reason there can be quite long gaps of observations when using Sentinel-2 data. Therefore, it would be good to utilize Sentinel-1. One of the findings have been that the Sentinel-1 backscatter increases heavily after the start of building. Coherence has more variation, but its decrease could be useful feature. NDVI could be useful feature if the clearing of vegetation happens very late before start of construction, and cloud coverage has been very limited. We also tested the viability of state-of-the-art geospatial foundation models for this task by comparing the model embeddings from different timesteps and checking whether the largest changes occurred in the areas with building permits.

Air pollution is an invisible yet pervasive threat to public health, affecting populations on a daily basis. In 2022 alone, an estimated 269,000 premature deaths in Europe were attributed to exposure to fine particulate matter (PM₂.₅), and an additional 66,000 to nitrogen dioxide (NO₂), according to the European Environment Agency. Beyond its overall health burden, air pollution disproportionately affects vulnerable populations. The World Health Organization highlights that health risks vary significantly between population groups: people living in low- and middle-income contexts are often exposed to higher pollution levels and exhibit higher prevalence of pollution-sensitive diseases such as asthma. Additional vulnerable groups include populations residing near major roads or highways, or working in pollution-intensive occupations. Despite the clear societal relevance of these disparities, policymakers often lack spatially explicit information to identify where interventions are most urgently needed. Earth Observation (EO) data offers a unique opportunity to address this gap. The Sentinel-5P satellite, launched under the Copernicus programme, carries the TROPOspheric Monitoring Instrument (TROPOMI), which provides consistent and spatially explicit measurements of key air pollutants, including nitrogen dioxide and proxies for fine particulate matter, at continental to urban scales. This contribution explores how Sentinel-5P/TROPOMI air quality data can be integrated with socio-economic and demographic datasets to identify patterns of environmental inequality and potential leverage points for policy intervention. In collaboration with TNO, the Netherlands Organisation for Applied Scientific Research, which plays a leading role in the development and validation of TROPOMI products, we aim to combine EO-derived air pollution indicators with statistical data on population characteristics, health vulnerability and socio-economic conditions. By doing so, we demonstrate how EO can support evidence-based policymaking, enhance spatial targeting of air quality measures, and contribute to more equitable environmental and public health outcomes within official statistical and policy reporting frameworks. This work is funded under the GEOS2023-NL programme (Project PR003214, WP2-D2.2).

Accurate grassland yield prediction is crucial for precision agriculture but challenged by environmental variability and limited ground truth. We used high-resolution Sentinel-2, agrometeorological data, spatial–temporal stratification, and a neural network on 184 Austrian sites (2018–22). The model achieved R²=0.8 and MAE=330 kg ha⁻¹, generalised to unseen years, and identified LAI as a key predictor. This underpins the upcoming SatGrass system for optimising grassland production and national statistics in Austria. Diverse environmental conditions, topographic variations, and sustainable and efficient management practices present significant challenges. These challenges can be overcome by using high-spatial-resolution Sentinel-2 satellite data, agrometeorological information, spatial and temporal stratification, and machine learning algorithms. Nevertheless, generalizing to wider geographic regions and times often remains limited due to small and insufficient ground truth datasets. To address this issue, we collected an extensive dataset from 184 sampling sites in Austria from 2018 to 2022 at multiple times per growth period. We developed a specialized application to guide the sampling process, ensuring efficient and consistent sampling. We used a fully connected feedforward neural network to estimate dry matter yield, achieving an R² of 0.8 and a mean absolute error of 330 kg per hectare on an independent test dataset following a stratified split by sampling site. Testing on unknown years also showed robust model performance, underlining the model's generalizability. An analysis of predictor importance demonstrates the relevance of remote sensing-based predictors, particularly the Leaf Area Index (LAI). This study presents a robust, extensively tested, and applicable grassland yield model that achieves high accuracy on a national scale, including temporally and spatially heterogeneous grassland areas.The study included the estimation of the start of season, the estimation of grassland mowing events and the modelling of yield for each growing period.

The emergence of sovereign sustainability-linked bonds (SSLBs) and debt-for-nature swaps offers developing nations the opportunity to reduce debt costs while protecting critical ecosystems. However, their adoption is constrained by the absence of statistically robust, independently verifiable Earth Observation (EO)-derived forestry Key Performance Indicators (KPIs) that can be credibly embedded into sovereign debt instruments. Without rigorous monitoring at decision relevant timescales, investors and credit agencies cannot assess environmental performance with sufficient confidence, limiting scale and market confidence. Uganda exemplifies both the challenge and opportunity. Having lost over 62% of its forest cover since 2000, the country faces a sustainability dilemma where economic development pressures directly conflict with long-term environmental stability. At the same time, Uganda is actively exploring sustainability-linked financing instruments with its Ministry of Finance and partners like NatureFinance, creating an urgent need for operational, statistically defensible EO monitoring systems capable of supporting bond-linked performance triggers. This study presents an integrated EO framework to generate credit-relevant forestry KPIs for sovereign finance. A transparent forest baseline is first established using open-access data. Sentinel-1 SAR time series are then used to detect deforestation through abrupt changes in backscatter, enabling cloud‑resilient monitoring in equatorial conditions. Annual canopy height maps derived from GEDI LiDAR and TESSERA embeddings (fusing annual Sentinel-1 and Sentinel-2) are subsequently used to assess the detected forest loss at annual intervals. The framework directly addresses the technical barriers that have made forestry KPIs rare in sovereign finance: measurement frequency, data transparency, attribution, and baseline setting. By leveraging open EO data and transparent methods designed to meet International Capital Markets Association standards, we establish a replicable model for incorporating environmental metrics into sovereign debt structures. Preliminary results demonstrate the approach's capacity to inform Uganda's sustainability-linked bond framework and provide a blueprint for scaling nature-linked sovereign finance across developing economies.

Heavy metals like chromium drive land degradation across Europe and globally, impacting 60-70% of EU soils at €50B annual cost. The EU Soil Monitoring and Resilience Directive (Dec 2025) mandates biennial health assessments and restoration tracking.​ Solution layer 1: Heavy Metal Detection – Antarix Space SRL fuses SWIR hyperspectral from CubeSats, Sentinel-2 multispectral, and EnMAP/PRISMA missions to detect Cr(VI) signatures at 2,200 nm via clay-hydroxide absorption proxies at 5m resolution.​ Solution layer 2: Bioremediation Pipeline – Sequential monitoring tracks biochar amendments, bacterial consortia degradation, mycoremediation (fungal networks), and final phytoremediation using PRI/NDVI spectral recovery indices. AI-driven multitemporal analysis with IoT validation achieves

The Northern Jordan Valley (NJV) is an important agricultural region in Jordan known for its high-quality citrus production. However, the NJV has been facing increasing water shortages for irrigation due to its reliance on limited transboundary water resources and growing competition from other sectors. Currently, the Jordan Valley Authority (JVA) manages the region’s irrigation supply allocation based on annual water availability and farm size. This allocation method does not consider the impact of fluctuating water supplies on water productivity. As water shortages worsen, the risk to the region's agricultural output rises, highlighting the need to reevaluate current irrigation management and identify potential solutions to sustain agricultural production. Our study integrates ground observations of irrigation water allocation with consumption and water productivity data from the FAO’s Water Productivity through Open-access of Remotely sensed derived data (WaPOR). WaPOR contains open access data on actual evapotranspiration, transpiration, and net primary production at a 20 m resolution for the north Jordan Valley. We employed a random forest classifier within the Google Earth Engine platform, to create a map of citrus cultivation. Irrigation performance was then assessed at three levels: farm, demand area, and the entire scheme, using hybrid indicators of irrigation uniformity, adequacy, consumed and beneficial fractions, irrigation efficiency, water productivity, and yield gap. Our results show significant differences in irrigation performance among farms and demand areas, with overall irrigation uniformity across the scheme less than 45%. Our study reveals that besides limited water availability, the low irrigation performance is influenced by inefficiencies in water conveyance, distribution, and on-farm irrigation management practices. The study concludes that improving irrigation management in the NJV requires a combination of efficiency measures across scales (farm, distribution across demand areas, and conveyance). Collaborative decision-making between the JVA and farmers is crucial, focusing on enhancing operational efficiency through canal rehabilitation and improving on-farm irrigation efficiency through advisory services and technological interventions. Our findings highlight the challenges posed by water scarcity and aggravated by irrigation inefficiencies from scheme to farm level. The study offers a data-driven framework for equitable water allocation and improved water productivity in this important scheme.

Floods are among the most damaging climate-related hazards, with growing relevance for financial markets as climate change intensifies hydrological extremes. This study examines whether and how equity markets price firms’ exposure to flood risk using high-resolution, facility-level data. We link satellite-based flood detection from Synthetic Aperture Radar imagery and hydrological exceedance measurements to publicly listed companies owning large industrial facilities in Europe, North America, and Australia. Flood risk is measured along multiple dimensions, including spatial inundation, probabilistic flood likelihood, realized flood severity and duration, and proximity to flooded areas. Using an event-study framework, we estimate monthly abnormal and cumulative abnormal returns around flood events. The results show that investors price both expected and realized flood risk. Higher flood likelihood is associated with systematically lower cumulative abnormal returns, indicating that forward-looking physical climate risk is capitalized into equity prices. Realized flood severity leads to significant valuation losses, particularly when severe flooding occurs close to firm facilities, highlighting the importance of spatial proximity. In contrast, the spatial extent of inundation alone is less informative once likelihood and severity are considered. Flood duration is associated with partial price recoveries, suggesting dynamic adjustment as information unfolds. Overall, the findings demonstrate that floods are not perceived as purely transitory shocks and underscore the value of integrating Earth observation data intofinancial risk assessment.

In the Brazilian Amazon, where land-use change remains a major driver of biodiversity loss and greenhouse gas emissions, Land Use Land Cover (LUCC) products are key inputs for official reporting on land, agriculture, and climate indicators. By combining EO-derived products with complementary official statistics, decision-makers can move beyond retrospective monitoring and obtain forward-looking evidence to anticipate hotspots and prioritize interventions, leveraging multiple data formats. Going beyond traditional simulation models, in this study, we aim to present a deep learning framework for annual LULC change forecasting in the Altamira micro-region (Pará, Brazil), which contains one of the world’s largest protected-area mosaics - currently under increasing pressure by the expansion of pasture and soy crop activities. The approach combined three different layers: Convolutional Neural Network (CNN) to learn spatial context from static EO-derived data, Multilayer Perceptron (MLP) to capture dynamics from LULC processes, and Gated Recurrent Unit (GRU) to account for yearly economic indicators. A late-fusion strategy was adopted to integrate these layers while preserving their informative spatial and temporal features. Model training integrated annual LULC maps and economic drivers for the period 2019-2023, including land prices and cattle and soybean commodity prices, and spatially explicit EO-derived drivers (i.e., elevation, slope, soil granulometry, and network-based travel-time distances to market infrastructure). Model performance was assessed against the most recent validated LULC reference year (2024), and annual forecasts were produced from 2025 onwards. To ensure relevance for registry-linked reporting and policy targeting, pixel-level predictions were converted into annual class-area and transition indicators aligned with administrative reporting units (municipalities) and over property-registry units (Rural Environmental Registry). Model performance was assessed using accuracy, precision, sensitivity (recall), and F1-score. This study is expected to complement EO-based monitoring cycles by providing an early-warning component for LULC dynamics and supporting anticipatory policy planning and risk-based reporting.

Reliable data on the actual extension of irrigated areas and the corresponding amounts of water abstractions from distribution networks and groundwater are seldom available, making difficult to assess the actual exploitation of water resources for irrigation. To this extent Sentinel-2 derived data provide a very valuable source of information for mapping irrigated areas and estimating spatially-distributed irrigation water requirements. This presentation illustrates the results achieved using the IRRISAT methodology, developed by the authors and applied to different environmental context, from Italy to Middle East and Australia. The methodology, was initially developed as an Irrigation Advisory Service in support to the Water Framework Directive (WFD) and the Italian Ministry of Agriculture Decree of July 31, 2015- where a specific set of obligations of measurement and estimation of the irrigated areas and irrigation water volumes was defined. Successively, the methodology has evolved in a framework of procedures aiming at: i) identifying actual irrigated areas based on Machine Learning classification of dense temporal series of vegetation indices; ii) calculation of irrigation water requirements and actual abstractions based on evapotranspiration models with canopy parameters derived from the full spectrum of Sentinel-2 observations. In this study how this framework has been tested and applied in different operational contexts, such as irrigation consortia in Southern Italy for identifying illegal abstractions, and in arid contexts such as Jordan and Australia to assess actual irrigation water use.

The use of satellite data has significantly improved crisis management, especially in the context of increasingly frequent and severe natural disasters. Predicting the extent and locations of events such as floods is now crucial, as factors like unregulated riverbeds, urbanization, and deforestation intensify their impact. Floods cause not only direct material and infrastructure losses but also serious indirect effects on emergency services. To address these challenges, we conducted a project analyzing the flood that struck southeastern Poland in September 2024, focusing on its impact on ambulance routes and response times. Using satellite imagery from both SAR (Synthetic Aperture Radar) and optical sensors, we mapped the flood extent in the most affected areas of the Lower Silesian and Opole Voivodeships. We then integrated GPS data from ambulances and overlaid a grid of points on road networks. By adjusting road lengths in flooded areas, we calculated the fastest possible routes to emergency calls. Based on this analysis, we created an ambulance access map showing areas that became cut off from emergency services during the flood. The results highlighted the importance of considering indirect effects of disasters. Besides delaying ambulance response, the flood reduced access to medical and logistical resources, making rescue coordination more difficult. Our findings demonstrate the potential of this approach for future crisis management. The results can support planning of alternative routes, optimization of emergency algorithms, and investment in disaster-resilient infrastructure. The methodology can also be applied to larger regions and other types of disasters. Potential stakeholders include crisis management authorities, government institutions, urban planners, investors, and residents in flood-prone areas who wish to assess the risk of limited access to emergency services.

In a context shaped by climate change and increasing market uncertainty, timely, accurate and reliable information on agricultural production and practices is essential to support effective public policies. Although the potential of satellite-based Earth Observation (EO) for agricultural statistics has been recognized for many years, its operational uptake by National Statistical Offices (NSOs) has remained limited. Today, however, mature EO capacities - particularly those provided by the Copernicus Programme - offer a solid basis for large-scale, policy-oriented applications. The open-source Sentinel for Agricultural Statistics (Sen4Stat) toolbox facilitates the integration of EO data into official agricultural statistics workflows. Spain has been one of the pilot countries where Sen4Stat was developed and tested through close collaboration with the Ministry of Agriculture, Fisheries and Food (MAPA). The national agricultural survey ESYRCE is an integrated list and area frame survey based on square segments subdivided into agricultural plots. By combining ESYRCE survey data with Sentinel-2 time series, Sen4Stat was used to produce a national crop type map and to estimate cereal crop yields. The integration of EO-derived crop maps with ESYRCE data substantially reduced uncertainty in crop area estimates. For instance, barley acreage estimates in Castilla y León exhibit significantly narrower confidence intervals when EO data are combined with survey observations, compared to survey-only estimates. Similar improvements were observed for yield estimation, effectively increasing the effective sample size and the reliability of aggregated statistics. Furthermore, EO data enabled spatial disaggregation of crop area statistics at the municipal level and supported the production of a national irrigation map, contributing to improvements in the survey sampling frame. Finally, our study will discuss next steps for the operational adoption of EO within official statistics, including institutional integration, organizational and IT requirements, processing costs, which are ongoing discussions between MAPA and Eurostat.

Quantifying informal trade is a challenge in Central Asia due to data scarcity, limited survey coverage, and the structural opacity of non-formal economic activities. The aim of the study is to develop an EO–based analytical framework that moves beyond descriptive spatial analysis toward predictive estimation of informal trade flows. Conducted within the ESA’s EO Clinic framework in support of the World Bank Group, the research focuses on two major regional hubs of informal commerce: Dordoi Bazaar in Bishkek, Kyrgyz Republic, and Barakholka Bazaar in Almaty, Kazakhstan.     High-resolution satellite imagery acquired between 2012 and 2020 was first used to extract time-series indicators characterizing bazaar activity, including total bazaar area, roofed-over surfaces, container and fixed buildings, vehicles in parking areas, and construction sites. These EO-derived indicators were then combined with selected ancillary economic and mobility data to support a predictive analysis of informal trade volumes.     Multiple regression and machine learning techniques were evaluated under different data scenarios. Models were trained on historical data (2015–2018), validated against 2019 benchmarks, and used to generate estimates for 2020. Results yield encouraging results but do not support definitive conclusions due to the limited availability and short temporal coverage of input data, which constrains the effectiveness of predictive techniques. Initial modelling, trained on data from 2015 to 2018 and validated against 2019 benchmarks, achieved prediction errors of approximately 10% for Dordoi Bazaar and 5% for Barakholka Bazaar. The inclusion of ancillary economic and mobility data substantially improved model performance for Barakholka, reducing the validation error to around 2–3%.     The findings demonstrate that integrating EO data with advanced analytics can support the estimation of informal trade in data-constrained contexts, offering a scalable and policy-relevant complement to traditional economic statistics despite remaining uncertainties. 

Monitoring land-use change is essential for providing evidence that supports policy design and implementation for sustainable land management and development. However, consistent monitoring remains challenging in highly heterogeneous landscapes, especially where smallholder agriculture produces fine-grained and seasonally dynamic land-use mosaics. This study is an evaluation of national-scale land-use patterns and trends in Ethiopia from 2018 to 2024. The methodology applied in this study involves the integration of Sentinel-1 and Sentinel-2 satellite imageries with a deep learning classification approach. This enables the annual mapping of 21 natural land-use and agricultural crop classes. We quantify class-specific confidence and examine the spatial distribution of uncertainty over time. Preliminary results show that classes characterised by pronounced and distinct physical signals, such as open water and closed-canopy forest, are mapped with a consistently high degree of confidence and demonstrate stable change patterns. In contrast, classes that are mixed, transitional or management-dependent, which are common in smallholder systems, show persistent confusion across space and over time. This uncertainty is concentrated in intensively used agricultural zones and along boundaries between different land uses. This has direct implications for national change estimates and for how land-use dynamics are communicated to decision-makers. It is therefore essential to take into account the observational limits existing if transparent monitoring and the generation of policy-relevant land-use change evidence from earth observation are to be achieved.

Timely and spatially detailed crop statistics are essential for agricultural monitoring, food security assessment, and policy decision-making. Traditional survey systems provide rigorous estimates but are costly and limited in spatial coverage. Earth Observation (EO) offers wall-to-wall spatial data but suffers from classification biases when used on its own. This study presents an operational framework that integrates optical and radar EO with ground survey data to derive statistically consistent crop area estimates with quantified uncertainty. I used Sentinel-2 surface reflectance data and the LUCAS 2018 survey for Germany to train a Random Forest classifier across 12 major crop types, based on vegetation indices (NDVI, EVI) and within-season timing. The baseline classification achieved an overall accuracy of ~35%, with high discrimination for maize and wheat but weaker separation for minor crops. Raw map-based crop areas derived from these predictions were adjusted using a design-based correction informed by a confusion matrix. Uncertainty was quantified through bootstrap resampling, yielding narrow 95% confidence intervals for major crops. Sentinel-1 radar backscatter was also evaluated but provided limited improvement under current acquisition density. Implemented in Google Earth Engine, this framework demonstrates how EO can enhance the timeliness and spatial resolution of crop statistics while remaining consistent with official statistical principles. The results highlight the importance of bias correction and uncertainty quantification when using EO-derived maps for decision-relevant agricultural indicators.

We present a methodology to develop the integrated climate change transition and physical risk assessment of industrial companies in Europe, Northern America and Australia. There is an increasingly important need for effective large-scale climate change risk assessment solutions with more governments aligning their company reporting regulations with the Task Force on Climate-related Financial Disclosures recommendations. In this paper, we measure key aspects of climate change risks of industrial firms on the globe and vice versa. The study provides valuable insights into climate risk exposure for companies, investors, and consumers, offering a pioneering approach by integrating data from major international registers. We analyse data from 70,000 companies and their 170,000 plants, which report to fragmented Pollutant Release and Transfer Registers and Greenhouse Gas Reporting Programs. For our assessment, transition risks are measured in terms of reported greenhouse gas emissions, while physical risks calculated for all company plant locations in terms of historical cooling energy needs, flood exposure and photovoltaic power potential. We show that climate change transition and physical risks are not correlated, therefore climate change risks are variably felt across different factors. The research contributes to the evolving landscape of climate risk management and highlights the need for standardized methodologies in the face of impending regulatory changes.

In Germany, the percentage of people living in urban areas is given as 80.3%. This figure is calculated by classifying the inhabitants of each administrative unit that has more than 5000 inhabitants as ‘urban’. This statistic is then compared with other countries. Surprisingly, however, these countries use completely different definitions of what is ‘urban’ – Denmark, as one example, uses a threshold of 200 inhabitants, while Japan uses 50,000. The spatial statistics are therefore inconsistent and comparisons are to be questioned. In addition, the question naturally arises as to the extent to which the approach is tenable for Germany itself. In this study, we first project various national threshold values for classifying urban populations that are used in different countries around the world onto our study location of Germany. This shows how differently the degree of urbanization in Germany can be assessed. In addition, we use a grid-based approach to calculate different degrees of urbanization at a high spatial resolution. We apply at grid-level thresholds on population figures but also on the variables building density and the proportion of a certain building type to infer the degree of urbanization. By systematically applying thresholds between minimum and maximum per variable, we track the effects on the resulting degree of urbanization. These permutations produce a wide range of results. To avoid using predefined concepts or thresholds, we combine all these possible variants of the administrative approach with thresholds for population figures and the grid-based approach with thresholds for population, building density and the share of a certain building type. This probability-based approach allows a more differentiated statistical picture of Germany to be drawn: According to our results, Germany can be classified as at least 50.0% to possibly 68.1% of the population as urban, which is nowhere near the stated 80.3% in official statistics. We conclude that the results of the generally used approach to quantifying the urban population are not unambiguous and should therefore only be used with great caution for political and social decisions. Our approach shows that a local approach based on population density and built structures comes close to the perception of urbanity.

The transition area between forest and urban interfaces has become critical in the analysis of risk factors and management of the threat of wildfires and one of the mandatory issues from the Catalan Government to achieve surveillance methods to design, define and monitor policy implementation. Currently, a zone or henceforth wild forest urban interface (WUI) is regulated with a width of 25 meters from the urban limits and is expected to have low fuel loads. There is discussion about whether the width of the WUI should be expanded to 50 or 100 meters to reduce the risk of wildfires, in particular new 6th generation ones. The methodology presented below analyzes the morphology of the vegetation in the strip/s, considering the first 25 meters, but adaptable to 50 or 100 m. The data used for the analysis come in part from Earth observation images captured by ICGC, such as annual orthoimage coverage flights as well as from Sentinel2 satellite images. Likewise, in order to obtain forest mass metrics, the normalized digital surface model is used as a by-product of the ICGC photogrammetric flights and the population area database as a base for urban cover. Two products are offered, analysis of the occupation within the bands in annual temporality, based on the digital surface model and ICGC orthoimages and detection of changes in time “near-real” under the temporal resolution capabilities of the Sentinel 2 images. The resilience of landscape (SDG15) to wildfires becomes a key contribution from Earth Observation data to define policies and assure a fine economic-social and environmental trade offs.

We present a secure, cloud-based geospatial platform developed for SAIL Investments, designed to support screening, monitoring, and reporting of sustainability KPIs in green finance. The tool integrates Earth Observation and open-source geospatial datasets to track deforestation, biodiversity, TNFD metrics, net carbon flux, and other environmental impact indicators, with the flexibility to extend to climate risk assessment or other reporting needs. The platform includes project and portfolio management, enabling authorized users to analyze and visualize impact measurements efficiently, both at the project and portfolio level, and generate standardized reports. Building on automated KPI workflows, it streamlines decision-making and reporting for sustainable investments. We are currently exploring its reusability and market potential, demonstrating how geospatial data can drive measurable, nature-positive financial outcomes, and will present the tool and the outcomes during this event.

Reliable population accounting and urban assessment strongly depend on the availability of up-to-date, high-quality spatial data. While traditional demographic approaches based on census data provide accurate representations of social structure, their infrequent acquisition and high institutional cost limit the ability to capture rapid urban and demographic dynamics. In this context, Earth observation serves as an effective complementary data source, enabling continuous and spatially explicit monitoring of urban areas. This presentation will explore a range of approaches for urban area mapping and assessment using Earth observation data, with a particular focus on the Copernicus Data Space Ecosystem (CDSE) as a key platform for data access, processing, and analysis. Different methodological workflows will be presented, ranging from rapid and easily applicable techniques to more advanced analytical solutions integrating multi-source satellite data, including radar and optical imagery, as well as thematic urban products. In addition, the presentation will highlight the growing role of artificial intelligence in Earth observation analysis. A case study for the city of Arequipa, Peru, will demonstrate the application of classification models and embedding-based representations to analyse urban expansion and structural changes over a multi-year period between census iterations. The region represents a particularly relevant example due to its dynamic spatial growth and migration-driven demographic transformations. Overall, the presented approaches will illustrate how combining Earth observation data and modern AI techniques can significantly enhance urban mapping and demographic analysis beyond traditional data sources.

Two EO-based data production pipelines, aimed at supporting the production of statistics on SDG 6 indicators, are to be implemented within the EO4SURE project. Two case studies will be carried out, in Denmark and Portugal, in close collaboration with key users: Statistics Denmark, the Portugues Statistical Office and the Portuguese Environment Agency. Indicator 6.4.2 (level of water stress) will be fed by calculating total freshwater withdrawal (with a focus on irrigated agriculture) and total available freshwater resources, including storage in surface water bodies. This pipeline includes the modelling of actual evapotranspiration, mapping spatial and temporal extents of water bodies, and the use of a hydrological model to estimate water inflows and outflows at a national or regional level. Indicator 6.6.1 (changes in extent of water-related ecosystems) will focus on mapping of wetland ecosystems using classes from Global Ecosystem Typology. The changes in spatial extent of those wetlands will then be monitored at a monthly timestep with 10 m spatial resolution. In Denmark the study will cover the whole country, while in Portugal the demonstration of these solutions will be focused on the Algarve, a region where water availability is most relevant. In addition to supporting SDG statistics reporting, the solutions will contribute to the European Union Water Framework Directive, and can support decision-making and control of water resources, with direct implications for their sustainability. Results should contribute both to the sustainable management of drinking water and to the protection of water-related ecosystems.

Integrating Earth Observation (EO) into official agricultural statistics requires bridging the gap between raw spectral signals and robust, harmonized indicators suitable for policy reporting. Traditional yield estimation methods often lack spatial granularity, while raw satellite data can be noisy and difficult to aggregate into administrative units. This study presents a scalable framework that transforms high-resolution Sentinel-2 time series (2018–2024) into objective "Yield Productivity Zones" (sYPZ), offering a validated pathway for enhancing national crop yield tables. Using a spatiotemporal multi-crop framework applied to the Danubian Lowland (Slovakia), we processed 7 years of vegetation data at a 10 m resolution to identify stable high, low, and unstable yield patterns. Unlike precision agriculture approaches that normalize data per field, we employed a regional normalization strategy. This is crucial for statistical reporting, as it captures broad environmental gradients driven by climate and soil rather than just local management variability. To demonstrate the validity of this approach for official statistics, we upscaled these granular sub-field metrics to administrative levels. While the core study focuses on pixel-level precision, extended validation reveals a significant correlation when aggregated stability metrics are compared against official NUTS3 yield statistics. This confirms that pixel-level EO stability patterns effectively capture regional productivity trends reported in traditional inventories. Furthermore, by applying machine learning (XGBoost) and SHAP analysis, we uncover granular driver patterns, identifying localized risks such as potential soil degradation linked to specific topographical features. This depth of insight enables stakeholders to navigate interventions more effectively, transforming raw data into the explainable, trusted intelligence required by statistical authorities. The results highlight that future operationalization depends on the standardization and inter-calibration of yield thresholds. Acknowledgement: “Funded by the EU NextGenerationEU through the Recovery and Resilience Plan for Slovakia under the project No. 09I05-03-V02-00013.”

Accurate cropland mapping in data-scarce regions remains challenging due to limited field data, strong interannual climatic variability, and heterogeneous cropping systems. This study proposes an NDVI-based training sample migration framework that transfers labeled samples from reference years in irrigated and rainfed agricultural systems to a target year using time-series similarity analysis. Ten similarity metrics representing geometric, temporal alignment, and correlation-based families were systematically evaluated to identify optimal thresholds and robust hybrid combinations for stable cropland transfer. The migrated samples were used to train a Random Forest classifier to generate binary cropland maps for 2021. Independent validation yielded overall accuracies of 86% in Kazakhstan and 95% in Uzbekistan. Comparisons with global cropland products (WorldCereal 2021 and WorldCover 2021) demonstrated improved spatial coherence and reduced misclassification, particularly in semi-arid environments. The proposed framework extends the temporal utility of existing labeled datasets and supports scalable cropland mapping without the need for repeated annual field surveys.

Given the intense pressure in Flanders on the remaining open space, patches of nature elements such as Historic Permanent Grasslands (HPG) and Small Landscape Elements (SLE) are legally protected by a decree of 1997 [1] and associated regulations of 1998 [2]. HPG and SLE are important for various local biodiversity objectives and act as wildlife corridors. The Agency of Nature and Forest are mandated to enforce the regulations. Since the European Sentinel-2 and Sentinel-1 constellations were fully active, they ordered the production (2017) and operations (2021) of a system for monthly detection of HPG condition changes to support such enforcements. The EO-based enforcement system is anchored in regional reference GIS datasets of protected nature elements, derived from decennial airborne LiDAR campaigns and annual airborne photogrammetric surveys. These acquisition campaigns, as well as the production and maintenance of the derived GIS layers, are coordinated by the Agency Digital Flanders. The reference datasets provide the polygons and associated metadata of the protected features subject to enforcement. The monitoring system analyses multi-source Earth observation time series to detect anomalous temporal behavior, including abrupt disturbances and deviations from expected phenological patterns. Multispectral Sentinel-2 observations provide detailed information on vegetation dynamics through indicators such as NDVI, fAPAR, and fCOVER, and enable the identification of surface water conditions, allowing natural flooding events to be distinguished from potentially reportable grassland disturbances. To ensure temporally consistent monitoring under frequent cloud cover, Sentinel-2 data are integrated with Sentinel-1 C-band SAR observations using the CropSAR AI fusion model [3], which produces a consistent, cloud-free time series of vegetation indicators. Overall, the presented approach demonstrates how EO data can efficiently and timely support regulatory enforcement by directing field inspections to locations where intervention is most likely required. References:[1] Vlaamse Regering (1997). Decreet betreffende het natuurbehoud en het natuurlijk milieu. Vlaams Codex. Available online: https://codex.vlaanderen.be/portals/codex/documenten/1005915.html [2] Vlaamse Regering (1998). Besluit tot uitvoering van het decreet betreffende het natuurbehoud en het natuurlijk milieu. Vlaams Codex. Available online: https://codex.vlaanderen.be/portals/codex/documenten/1006515.html [3] Van Tricht, K., Gobin, A., Gilliams, S., & Piccard, I. (2018). Synergistic use of radar Sentinel-1 and optical Sentinel-2 imagery for crop mapping: A case study for Belgium. Remote Sensing, 10(10), 1642. https://doi.org/10.3390/rs10101642

As Europe experiences the fastest warming worldwide, accurate data is essential for evidence-based decisions and progress toward the SDGs[1]. Earth observation (EO) and citizen science (CS) data play a crucial role in supporting environmental-compliance requirements under EU legislation and international agreements[2]. The EU-funded ENFORCE project[3] develops a pan-European collaboration and innovation hub that aligns EO and CS data with official standards, improving data admissibility in court and applicability in environmental-compliance proceedings. Triangulating EO, CS and sensor-based in-situ data enables evidence gathering on illegal activities such as deforestation, fires, waste disposal and water contamination for legislative and regulatory follow-up. ENFORCE introduces a maturity-model framework to assess the fit-for-purpose of CS data, strengthening data credibility, admissibility and scientific validity through criteria toward a data-readiness scale, ensuring correct application of protocols and data requirements supporting public authority environmental compliance activities and implementing EU policies. Within this framework, ENFORCE demonstrates how EO-derived environmental indicators can be assessed against data readiness and admissibility criteria in concrete regulatory contexts, such as the implementation of the Water Framework Directive (WFD). For ENFORCE, EO data supports the WFD implementation by providing Chlorophyll-a (µg/l)[4], mean suspended particulate matter (mg/l), and indirect insights on dissolved inorganic nitrogen (µmol/l)[5] and dissolved oxygen (mg/l)[6]. The project advances SDGs through the MINKA Citizen Science Observatory[7], recognized as a UN SDG Acceleration Action, which delivers essential biodiversity and ocean variables as non-traditional data sources. MINKA supports SDG 14.1 by integrating marine-litter data to assess coastal plastic pollution, and SDG 14.5 by generating open biodiversity datasets (over 643,214 records) for coastal management. Indirectly, it contributes to SDG 13.3 by providing data to Barcelona City Council’s biodiversity atlas and monitoring dune-ecosystem change, and to SDGs 4 and 5 by promoting inclusion, participation, and education through participatory systems. Bibliography: [1] European Environment Agency (EEA). 2025. Europe's environment and climate: knowledge for resilience, prosperity and sustainability. Copenhagen: EEA [2] Owen RP & Parker A J, 2018. Citizen sceicne in environmental protection agencies. In: Hecker, S., Haklay, M., Bowser, A., Makuch, Z., Vogel, J. & Bonn, A. 2018. Citizen Science: Innovation in Open Science, Society and Policy. UCL Press, London. https://doi.org/10.14324 /111.9781787352339 [3] ENFORCE Project Website. Available at: https://join-enforce.eu/ [4] Attila,J, Kauppila,P, Kallio,KY, Alasalmi,H, Keto,V, Bruun,E, Koponen,S. 2018. Applicability of Earth Observation chlorophyll-a data in assessment of water status via MERIS — With implications for the use of OLCI sensors, Remote Sensing of Environment, Vol 212, Pages 273-287, ISSN 0034-4257, https://doi.org/10.1016/j.rse.2018.02.043. [5] Poikane, S., Kelly, M.G., Herrero, F.S., Pitt, J.A., Jarvie, H.P., Claussen, U., Leujak, W., Solheim, A.L., Teixeira, H. and Phillips, G., 2019. Nutrient criteria for surface waters under the European Water Framework Directive: Current state-of-the-art, challenges and future outlook. Science of the total environment, 695, p.133888. [6] Best, M.A., Wither, A.W. and Coates, S., 2007. Dissolved oxygen as a physico-chemical supporting element in the Water Framework Directive. Marine pollution bulletin, 55(1-6), pp.53-64. [7] MINKA SDG Platform. Accessed at: https://minka-sdg.org/

Unlock the full potential of large-scale Earth observation data without the burden of downloading terabytes of imagery. This hands-on workshop introduces the Sentinel Hub Statistical API, a powerful component of the Copernicus Data Space Ecosystem (CDSE) designed for scalable, cloud-based geospatial analysis. The API allows users to bypass the pixels completely, shifting intensive processing to the cloud and returning ready-to-use statistical summaries in a clean JSON format. This approach dramatically optimizes workflows, enabling rapid time-series analysis without needing to download the underlying satellite imagery. The instructor will: - Introduce the API's capabilities and how participants may already be familiar with using it in Copernicus Browser. - Guide participants through how to build your Statistical API request focusing on how the Time Range and Aggregation Intervals work with the API. - Explain how evalscripts work with the API. - In the CDSE Jupyter Lab, we will then run through some pre-prepared examples together utilizing a wide range of datasets from the Copernicus program including Sentinel-2 L2A imagery and Copernicus Land Cover Monitoring Service (CLMS) datasets. - Finally, a small task will be set testing the participants' learning from the workshop and while they work on this task, there will be an opportunity to asks questions. By the end of the course, participants will hold the knowledge and working code to perform statistical analysis in their own projects and work.

Small Woody features such as hedgerows, shrubs, and linear or patchy clusters of trees provide undoubted ecological benefits supporting habitat connectivity, biodiversity, and carbon sequestration. Despite their relevance are often overlooked in large-scale land monitoring programs as they are small in size, difficult to detect, and scattered across diverse ecosystems in Europe. The Copernicus HRL SLF (High-Resolution Layer - Small Landscape Features) largely solves this problem, providing a high-resolution monitoring framework (nominal 5m resolution) available for users from 2015 with a 3-year revisit cycle. This workshop demonstrates how National Statistical Offices and environmental agencies can integrate SLF data to refine land-use statistics and support policy frameworks like the EU Nature Restoration Law and the CAP. Workshop Format (90 Minutes): Expert Presentation (15 min): Overview of HRL SLF products (SWF, WVL) and the 2015–2021 time-series. Live Demonstration (25 min): A Jupyter Notebook walkthrough showing how to spatially aggregate SLF data (e.g., by NUTS or parcel) and calculate woody feature shares. Interactive Breakouts (35 min): Participants will use pre-configured Jupyter Notebooks (via Binder/Cloud) split into three thematic streams: Group A (Agriculture): Quantifying features within parcels for CAP "non-productive area" reporting (e.g., LPIS cases in Spain/Luxembourg). Group B (Biodiversity): Aggregating woody biomass proxies at NUTS 3 level for Natural Capital accounting. Group C (Urban): Assessing "Green Infrastructure" within Functional Urban Areas (FUA) for climate resilience statistics. Final Recap (15 min): Groups present challenges and opportunities for operationalizing these workflows. Outcome: The workshop will produce actionable recommendations on product selection, multi-scale statistical aggregation, and reporting mechanisms for institutional use.

The Agro-Informatics platform is an FAO Digital Public Good (DPG), the main geospatial infrastructure for storing, searching, sharing and analyzing geospatial data on food and agriculture. To manage the discovery of these vast repository, the platform implements the SpatioTemporal Asset Catalog (STAC) specification. By standardizing metadata across heterogeneous sources, STAC allows for efficient indexing and querying of imagery based on time and location. For statistical applications, it enables the rapid identification and filtering of analysis-ready data before any heavy processing begins, ensuring that statistical aggregations are based on the most relevant and accurate assets available. Building on this discoverability, the platform leverages openEO to standardize the analytical processing layer. openEO provides a unified interface that abstracts the complexity of underlying cloud backends, allowing users to define statistical workflows, without managing the infrastructure. This approach brings the algorithms to the data, minimizing latency and enabling scalable, on-the-fly computation of agricultural indicators directly from the archive. The repository relies on robust data engineering pipelines running on Google Cloud to precompute key statistics. The pipelines run on Cloud Composer- managed Airflow- to construct computational graphs, compute statistics, then write the results into a centralized analytical database. To bring together processed statistics, the GAUL 2025, a global sub-national boundaries dataset has been produced and endorsed as DPG. It provides up-to-date province and district boundaries of the world, compliant to United Nations. These statistics are then accessible through several applications, including the Food Security Risk Intelligence and Early Warning Room[1], the Livestock Intelligence Hub[2], the FAO Remote Sensing portal[3] and the FAO Agri-Guard Application. [1] https://riskmonitor.fao.org/ [2] https://data.apps.fao.org/livestock/ [3] https://data.apps.fao.org/remote-sensing-portal

EarthCODE (https://earthcode.esa.int) is ESA’s strategic initiative to support Open Science, helping users discover, reuse, and build upon EO research to fill scientific knowledge gaps and address pressing societal challenges - a process referred to as “Earth Action”. EarthCODE hosts a repository of FAIR data and reproducible workflows on cloud environments, crucial for transparent statistical reporting and Earth Observation data analysis. In this 3-hour workshop, we introduce EarthCODE’s capabilities and available data. Participants will use the osc-library to search for and access Analysis Ready Data (ARD) from ESA’s Open Science Catalog (https://opensciencedata.esa.int) such as seasonal fires analysis datasets (https://opensciencedata.esa.int/products/seasfire-cube/collection) or data for estimates of Gross Primary Productivity (https://opensciencedata.esa.int/products/gpp-sen4gpp/collection). The hands-on technical content introduces the modern open-source Python ecosystem (Zarr, Xarray, Dask) within Jupyter Notebooks. Participants will learn about: EarthCODE data and how it can be used in your research and applications Using the osc library to search for and access ESA EO research data Working with common EO tools locally and on EarthCODE platforms Publishing research outputs using EarthCODE platforms and services By the end, participants will learn how to access and analyze high-value ESA research data suitable for policy and statistical applications. We will conclude with a feedback session on optimizing EarthCODE for the statistics community.

The constant improvement in Earth Observation (EO) capabilities is enabling us to build better and faster tools to monitor our planet. Crop monitoring is one of the most crucial tasks to support food security as well as national statistics at local and regional scales. The ESA-funded WorldCereal project provides an open and flexible cloud-based processing system that allows the production of local to global 10-meter resolution seasonal cropland and crop type maps that can be used for crop monitoring at different scales. To generate products with high accuracy the classification algorithms require high-quality reference data on land cover and crop types. For this purpose, several tools and initiatives have been undertaken to support the collection of these data sets. One of the tools available for in-situ reference data collection is Geo-Quest, a free mobile phone app that allows users to join targeted tasks or "quests". These quests can be opportunistic, e.g. registering data whenever possible, or directed, e.g. following a sampling design like a stratified random sample. Currently in the app, the Crop Capture quest is available for in-situ crop data collection, allowing users to easily geolocate, delineate and capture crop information, including crop type, irrigation, phenology and diseases. This information can be exported as geo-parquet files, which can be directly ingested into GIS systems but also into the WorldCereal Reference Data Module (RDM), where is used to train and validate the WorldCereal algorithms and products. The RDM allows users to share their data and supports them with AI-assisted legend mapping. If the data is made public, the system supports them also with thorough quality control and data licensing. This workshop will showcase the Geo-Quest application and demonstrate how to ingest the data into the WorldCereal RDM. Participants will acquire expertise in both tools and can start exploring them immediately.

In December 2022, the Kunming-Montreal Global Biodiversity Framework (GBF) was adopted by countries around the world to rebalance our relationship with nature. The value of the System of Environmental-Economic Accounting Ecosystem Accounting (SEEA EA) in mainstreaming biodiversity is explicitly recognized in the GBF monitoring framework. This includes compiling Ecosystem Services Accounts for reporting on headline indicator B.1 - Services provided by ecosystems. Compiling Ecosystem Accounts requires data on ecosystems and their services that are consistent and comprehensive over space and time. Earth Observation (EO) data is highly suited for these needs, deriving input data for modeling ecosystem extent, condition, land cover, change over time, and downscaling statistical data. However, identifying, harmonizing and integrating EO data for modeling ecosystem services requires substantial technical expertise and is resource intensive. The capacity required to make use of these large and complex datasets is a significant barrier to their use for ecosystem service modelling. Consequently, many countries do not have established systems for mapping the provision of ecosystem services to inform ecosystem accounting. To lower these barriers to compiling ecosystem accounts, the Basque Centre for Climate Change (BC3) has developed the ARIES for SEEA tool. The tool is free to access and allows users to produce rapid, standardized ecosystem accounts consistent with the SEEA EA framework. Additional ecosystem service models for global climate regulation, grazed biomass, nature-based tourism and coastal protection have been integrated into the tool under the SEEA-Related Indicators for the GBF project, funded by the European Union and implemented by the United Nations Statistics Division. This workshop will introduce ARIES for SEEA and provide initial training on using the tool and its recent developments. It will guide users through the steps to generate ecosystem service accounts for reporting on headline indicator B.1 and opportunities to customize models for national contexts.

This 1h30 session will showcase the SDGs-EYES platform, how it moves from data and models to usable, trusted services for SDG monitoring. It will combine introduction, live interaction, and expert discussion to demonstrate both the technical capabilities and uptake pathways. A key emphasis will be placed on how the platform ensures: traceability and validation at platform level (data integrity and metadata continuity), validation and traceability within pilots (indicator generation and comparison with ground truth) and importantly, reproducibility, enabling users to access, run, and adapt the underlying workflows. Setting the scene and aligning the audience (Monica Miguel-Lago, EARSC) ● Role of Earth Observation for SDG monitoring ● Key challenges: trust, usability, integration into workflows ● Intro about from data availability to operational use Introduction to the SDGs-EYES Platform (Manuela Balzarolo, CMCC) ● What is the SDGs-EYES platform (vision, scope, services) ● Brief overview of pilots and platform components (from capabilities, gaps to solution) Hands-on Demo / Interactive Walkthrough (Stefano Natali, SISTEMA) and (Alessandro D'Anca & Jishnu Jeevan, CMCC) ● Navigation of the platform, architecture / show concretely how the platform works ● Navigation of the dashboards, maps & tools ● Example use case (GHG emissions (FIRE-TRACE), climate risk (Heat Health Risk Assessment) ● How users interact with services o Platform level (i) Traceability: maintaining the link to original EO products (metadata, source information) (ii) Validation: ensuring data integrity (no corruption from source to platform) o Pilot level (i) Traceability: tracking how EO data is transformed into indicators (ii) Validation (pilot level): comparison of indicators with ground truth / reference datasets o Reproducibility (core feature):Users can access the code behind the servicesRun workflows as they areModify them and test results: This enables transparency, trust, and adaptability to new contexts 30 min Round Table Discussion (EARSC, CMCC, SISTEMA, ISTAT) From research to operations: advancing SDGs-EYES services panel will move from demo to reflection and uptake

For many years, the potential of satellite-based Earth Observation (EO) for agricultural statistics has been widely acknowledged, yet its adoption by National Statistical Offices (NSOs) has remained limited. The open-source Sentinels for Agricultural Statistics (Sen4Stat) toolbox aims at facilitating the uptake of sentinel EO-derived information within official NSO workflows. It consists of an open source EO data analytics system that ingests Sentinel-1 and Sentinel-2 time series and processes them into statistically-sound crop area estimates and yield forecast and estimation. Designed to run at national scale, the EO processing component is linked with (i) a module for in situ datasets quality control, (ii) a visualization tool and (iii) a set of tools for higher-level statistical analyses. Its open-source nature allows users to deploy the system on their own premises or on a cloud infrastructure and to operationally generate products tailored to their specific needs. Following a brief introduction to the Sen4Stat approach, the proposed workshop will combine two complementary types of interaction: • a live demonstration of the toolbox, illustrating the EO processing workflow, system configuration, and crop statistics production, and demonstrating how NSO statistical survey data are integrated and leveraged for the EO data exploitation to enhance the crop statistics production; • a panel discussion focused on experience sharing and lessons learned, providing an opportunity to address the technical and institutional challenges associated with integrating EO data into operational workflows, as well as the strategies adopted to overcome them. The Sen4Stat system has already demonstrated its ability to deliver reliable, robust, and timely information to support agricultural statistics and strengthen food security in various countries. It was presented during the Earth Observation Training for Official Statistics organized in January by the Stakeholder Engagement Facility team, where it attracted strong interest. This workshop provides an opportunity to build on these exchanges, deepen practical engagement, and further advance the adoption of EO technologies within the official statistics community.

The increasing demand for scalable, reproducible, and operational Earth Observation (EO) processing requires moving from monolithic deployments towards flexible cloud-based service architectures. Building on the experience of the Sen4CAP and Sen4Stat systems, this tutorial presents the ongoing cloudification of Sen4CAP and a direction for the future evolution of Sen4Stat.The proposed approach decouples the EO processing components from the original Sen4CAP/Sen4Stat systems and transforms them into independently deployable, containerized services available in the cloud. These processors, covering Sentinel-1 pre-processing, time-series analysis, crop classification, biophysical indicators and vegetation indices computation, are exposed as on-demand services that can be executed independently of the original platform environment. This service-oriented architecture enables scalability, interoperability, and flexible integration into statistical production workflows.The tutorial will demonstrate:● The principles behind the cloudification process (containerization, workflow orchestration, API exposure, and infrastructure abstraction);● How the Sen4CAP/Sen4Stat processors are transformed into standalone cloud services;● How to programmatically integrate Sen4CAP/Sen4Stat services into existing statistical systems;● The roadmap towards the cloud-native evolution of Sen4Stat processors.Participants will gain practical insight into how EO-based agricultural monitoring can transition from locally deployed processing systems to operational, service-based infrastructures aligned with modern cloud and open science paradigms.This tutorial is particularly relevant for National Statistical Offices, agricultural ministries, research institutions, and service providers seeking to operationalize EO workflows in a cloud-based, scalable, and sustainable manner

Trust in EO information products is a fundamental condition for their uptake in official statistics. While EO offers unprecedented opportunities (global coverage, consistency, and timeliness), EO also introduces new challenges related to transparency, uncertainty, and alignment with established statistical frameworks. The workshop will explore how transparency, methodological robustness, and uncertainty quantification can strengthen confidence and trust in EO-derived information. What factors build or erode trust in EO data when used in official statistics? - How can we ensure transparency and reproducibility of EO workflows? - How should uncertainty be quantified and communicated? - What are the best practices for metadata and documentation ? - Which frameworks for data quality assessment are needed? Identify practical approaches to ensure that EO information products are not only scientifically robust, but also trusted, accepted, and usable within official statistical frameworks

At the end of the workshop, we would like to have insights into the resources and the gaps related to EO for statistics among the participants. We expect to plot and match i) the gaps (problems), ii) the existing resources to fill these gaps (solutions), and iii) detect our blank spots. Another outcome is how to organize working together and how to overcome the blank spots. Organiser Details: Paulussen, R.R. (Remco) CBS NL MARTINS Carla (ESTAT) KANDYLAKIS Zacharias (ESTAT-EXT) HOFER, Nina (Statistics Austria) Mugnoli, Stefano (ISTAT, IT) Garioud, Anatol

Reliable statistics are foundational for evidence-based policy and for tracking progress toward national and global agri-food systems. Agricultural statistics inform decisions on productivity, resilience, and sustainability. But conventional sources, such as farm surveys, censuses, and administrative records are costly, infrequent, and uneven in coverage, creating gaps in timeliness, spatial detail, and comparability. These gaps are especially acute for emerging priorities such as land-use change, greenhouse gas (GHG) emissions from agriculture and land use, and the intensity and risks associated with fertilizer and pesticide applications. Earth observations (EO) provide consistent, frequent, and scalable measures to help close these gaps. Optical and radar satellites reveal crop types, phenology, management intensity, and land conversion, and ancillary datasets support inference on emissions drivers and environmental risks. Integrating EO with in situ monitoring and standardized methodologies enables more consistent, granular, and timely indicators that support agri-food systems, climate and biodiversity commitments, and SDG reporting. Major scope of the workshop is to help participants operate EO for official-quality statistics. We will feature 5 to 7 invited talks on the topics above and facilitate brainstorming to develop concrete proposals on: (1) EO requirements for agri-food statistics; (2) harmonized definitions for agri-food system statistics across land cover/use categories; (3) QA/QC and intercomparison of remote-sensing data products; and (4) data infrastructure for delivering remote-sensing products.

The workshop will delve into current needs and opportunities for integrating forest biomass information from forest inventories, national statistics and Earth Observation to strengthen the monitoring and reporting of forest biomass for environmental assessments and climate action. We will begin by reviewing the latest recommendations on using EO-based biomass products within MRV processes and international frameworks, as well as briefly touch upon ESA’s Biomass mission advances. Building upon recent discussions led by the Global Forest Observations Initiative (GFOI), we will discuss three different pathways leading towards the integration of these datasets, namely (1) key considerations informing the design of new ground-based campaigns to ensure both compatibility with and added value from EO datasets; (2) lessons learned from experiences combining and harmonizing different existing in-situ data (e.g., National Forest Inventories, among others) for their integration with EO datasets; and (3) assessments of inferential strategies for the integration of EO-based biomass datasets with in-situ data to enhance the precision of biomass estimates at different geographical scales. These pathways aim to support a broad range of end users in forest biomass estimation for monitoring and reporting purposes. The workshop will open with short presentations highlighting success stories and state-of-the-art examples on these themes, followed by a short round of questions. Afterwards, three parallel round-table discussion groups, focusing respectively on the three pathways for conceiving the integration of in-situ data, EO-based biomass datasets and national statistics. The session will conclude with summary presentations from the three discussion groups, followed by closing remarks and next steps. The workshop will ideally last 2.5 hours, including one short break. We aim to map existing efforts, key considerations, best practices and research needs under these themes, bridging efforts in tropical and temperate regions, and ultimately summarizing these findings in a perspective paper or policy brief.