Vasileios Baousis

Earth Observation (EO) and environmental research increasingly relies on AI methods that require access to large datasets, scalable cloud infrastructures, and high-performance computing (HPC) resources. At the same time, the transition of research outcomes into operational, industry-ready services remains challenging, often demanding substantial re-engineering of data pipelines, execution environments, and deployment models. This separation between research-oriented and industry-oriented infrastructures continues to limit the reuse, scalability, and real-world impact of EO innovations.Addressing this gap, the European AI-on-Demand Platform (AIoD) [1] was recently expanded to support both research and industry within a unified digital infrastructure. The platform brings together research-driven AI assets (such as models, workflows, and datasets) with industry-grade tools and services for the development, training, and operationalisation of AI applications across cloud and HPC infrastructures, in an efficient and responsible manner. As a unified gateway, the AIoD connects previously fragmented resources across the European AI ecosystem, making them accessible, reusable, and adaptable to diverse user needs. In parallel, efforts are underway to explore interoperability with emerging European AI Factory initiatives, including PHAROS (the Greek AI Factory) [2], aiming to support future federated access to specialised AI computing resources.We illustrate this approach through Earth Observation and environmental services and use cases that are jointly accessible to researchers and practitioners, including the mapping of sea surface features and marine pollutants, satellite image enhancement through super-resolution, and AI-based prediction and analysis of extreme weather events, enabling a seamless transition from experimentation and validation to scalable, operational deployment. These developments extend earlier work on European AI and Earth Observation convergence [3].[1] http://aiodp.eu[2] https://www.pharos-aifactory.eu [3] A. Troumpoukis et al., European AI and EO convergence via a novel community-driven framework for data-intensive innovation. Future Gener. Comput. Syst. 160: 505-521 (2024) https://doi.org/10.1016/j.future.2024.06.013This work has received funding from the European Union’s Digital Europe Programme (DIGITAL) under grant agreement No 101146490.

CLIMRES (LEADERSHIP FOR CLIMATE RESILIENT BUILDINGS) is a European project addressing the growing vulnerability of buildings and urban environments to climate change impacts. The project combines climate data, Earth Observation (EO) products, and impact assessment methodologies to identify climate-driven hazards, assess building and urban-scale vulnerabilities, and support decision-making through dedicated tools and measures. These solutions are validated through large-scale pilot demonstrations across several European countries. A key enabling outcome of CLIMRES is the design and deployment of a scalable, cloud-native infrastructure hosting the project’s Federated Data Exchange Platform (FDXP) and digital services. The infrastructure is hosted by ECMWF on a dedicated Kubernetes cluster within the Common Cloud Infrastructure (CCI), part of the European Weather Cloud co-managed by ECMWF and EUMETSAT. The underlying cloud environment is based on OpenStack with Ceph storage, providing elastic compute and scalable object storage capabilities for data-intensive workloads. This infrastructure provides the technical backbone for integrating heterogeneous datasets, executing data-processing workflows, and delivering operational services that underpin climate resilience assessments and decision-support applications. The CLIMRES platform follows cloud-native design principles and adopts containerization and microservice-based architectures to ensure modularity, scalability, and operational robustness. Kubernetes is used as the core orchestration layer, while Rancher provides centralized cluster management, monitoring, and operational visibility. All services, including the FDXP and supporting applications, are deployed consistently across environments using GitOps principles, ensuring reproducibility, traceability, and elimination of configuration drift. Continuous Integration and Continuous Delivery (CI/CD) pipelines automate the full software lifecycle, from source code changes to container image building and deployment. Docker images are built through automated pipelines and deployed via Git-driven workflows, enabling transparent, auditable, and predictable releases. Semantic versioning and changelog generation are fully automated, ensuring consistent release management across services. This contribution presents the CLIMRES cloud infrastructure as a production-ready case study for EO- and climate-driven applications. It demonstrates how cloud-native technologies can effectively support scalable data management platforms and operational services for climate resilience.

MediTwin is a European research initiative aimed at developing digital twin technologies for the Mediterranean region, integrating Earth observation data, numerical modelling and artificial intelligence to support environmental monitoring and decision-making. In this context, the MediTwin Summer School 2025 was organised to provide hands-on training on data-driven workflows, cloud-native tools and AI/ML techniques for Earth system applications. The school targeted early-career researchers, PhD students and technical staff from research institutions, with a total of 20 participants. The School required a scalable, secure and reproducible cloud infrastructure capable of supporting hands-on training activities in Earth system digital twins, data analysis and AI/ML workflows. This contribution presents the design and provisioning of the cloud-native infrastructure deployed to support the school, with a focus on Infrastructure as Code (IaC), Kubernetes-based orchestration and hybrid GPU-enabled workloads. The infrastructure was deployed on the ECMWF on-premises cloud, based on OpenStack and backed by Ceph software-defined storage, providing elastic compute, networking and persistent storage services. The Kubernetes cluster was provisioned in a high-availability configuration using Terraform and Rancher Cluster Manager, following established GitOps best practices. The cluster architecture comprised dedicated control-plane, worker, ingress and GPU nodes, enabling both standard cloud-native services and accelerated AI/ML workloads. Cluster lifecycle management, configuration drift prevention and application delivery were handled through a GitOps approach using Rancher Fleet. GitLab acted as the central orchestration platform for source control, CI/CD pipelines and IaC automation, hosting Terraform modules, Helm charts, Rancher cluster definitions and configuration templates. This ensured full traceability, auditability and reproducibility of both infrastructure and application deployments. Sensitive credentials and API keys were securely managed using HashiCorp Vault and dynamically injected into workloads. To support interactive training activities, a JupyterHub service was deployed on Kubernetes using the official Helm chart, customised for resource management, authentication and storage integration. GPU acceleration was enabled via the NVIDIA GPU Operator, which automated driver installation, device discovery and scheduler integration. In addition, outside the Kubernetes environment, 20 GPU-enabled virtual machines were provisioned directly on OpenStack using an Ansible role executed through AWX, itself deployed on the Kubernetes cluster, to accommodate specific student exercises requiring isolated VM-based access. This experience demonstrates how modern cloud-native and DevSecOps practices can be effectively applied to provision short-lived yet production-grade scientific training infrastructures, ensuring scalability, security and reproducibility for future Earth observation and digital twin education initiatives.

Large-scale foundation models trained on multi-sensor satellite imagery has been driving recent advances in Earth Observation (EO) tasks. Although such models achieve impressive transferability across diverse downstream tasks, their computational and memory demands hinder accessibility, reproducibility, and deployment in resource-constrained environments. This work explores a compact and efficient alternative, introducing a metadata-aware Mixture-of-Experts Masked Autoencoder (MoE-MAE) for EO representation learning (Albughdadi, 2025).The proposed MoE-MAE is a self-supervised transformer-based architecture with only 2.5 million parameters. It combines sparse expert routing and geo-temporal conditioning. The sparse routing allows token specialization while keeping active computation low. The geo-temporal conditioning injects information about latitude, longitude, and cyclic temporal attributes directly into the model. The proposed design enables the algorithm to exploit spatial and temporal regularities inherent in EO data without requiring dense, and computationally costly transformers.The model is pretrained in the BigEarthNet-Landsat (BEN-LS) (Corley et al., 2025) dataset using a masked reconstruction loss function augmented with auxiliary unmasked and load-balancing losses to encourage stable expert utilization. The learned encoder representations are then evaluated using linear probing on two benchmark datasets: (1) BEN-LS, a multi-label land-cover dataset with explicit metadata, and (2) EuroSAT-Landsat (EuroSAT-LS) (Corley et al., 2025), a single-label classification datasets without metadata. Despite the encoder’s small size (~2.3 M parameters), the proposed MoE-MAE achieves competitive results with models’ orders of magnitude larger. On BEN-LS, the frozen encoder achieves a micro mean average precision of 0.767, comparable to SSL4EO-L ViT-S/16 MoCo v2 (0.775) (Stewart et al., 2023). On EuroSAT-LS, the model maintains strong transferability, achieving 84.2% accuracy, even in the absence of geo-temporal metadata.Expert specialization across spatial patterns is revealed through adequate ablation and visualization studies, which show that some experts respond primarily to vegetation, others to water or textured regions. This demonstrates interpretable behaviour and complementary feature learning. Additionally, only about half of the model’s expert feed-forward capacity is activated per token, confirming computational sparsity in practice. These findings suggests that such models can retain strong representational power while substantially reducing training and inference costs.This work presents a first step toward small-scale architectures for EO representation learning by integrating metadata, and leveraging sparse computation to approach the performance of massive transformers. Future work will extend this framework to multi-sensor and multi-temporal datasets to capture dynamic Earth processes efficiently.Albughdadi, M. (2025). Lightweight Metadata-Aware Mixture-of-Experts Masked Autoencoder for Earth Observation.arXiv:2509.10919.Stewart, A. J., Lehmann, N., Corley, I. A., Wang, Y., Chang, Y.-C., Braham, N. A. A., Sehgal, S., Robinson, C., & Banerjee, A. (2023). SSL4EO-L: Datasets and Foundation Models for Landsat Imagery. arXiv:2312.05241.Corley, I., Sharma, L., and Crasto, R. (2025). Landsat-Bench: Datasets and Benchmarks for Landsat Foundation Models.arXiv:2506.08780.

Europe faces a growing frequency of extreme weather events, from heatwaves and floods to wildfires and earthquakes, increasingly threatening urban environments. Unusually warm winters are becoming progressively common, destabilizing ecosystems and altering traditional weather dynamics. Addressing these crucial changes, the CLIMRES project aims to foster a “Leadership for Climate-Resilient Buildings” by identifying and categorizing vulnerabilities within the built environment and assessing their effects within urban systems. This effort integrates diverse data sources, including Copernicus services, IoT networks, and municipal datasets, and considers hazard warnings and weather forecasts. Moreover, a liaison with the Destination Earth initiative enhances the project with the capacity to leverage extreme weather predictions and future climate models. CLIMRES aims to deliver vulnerability assessment and impact evaluation methodologies, along with a “hub of measures” inventory for cost-effective building design and materials against climate risks, as well as decision support tools, to aid building owners, policymakers and stakeholders in planning effective interventions and to address vulnerabilities, targeting three levels of decision making at strategic, tactical and operational levels. The project deploys cloud technologies like OpenStack and Kubernetes to host an interoperable platform for vulnerability analysis, data harmonization, and decision-making. Its solutions will be tested and validated on 3 Large Scale Pilots in Spain, Greece, Italy, and Slovenia, addressing hazards such as heatwaves, flooding, fires, and earthquakes. A multi-hazard replication pilot in France will further evaluate the scalability and versatility of these approaches across diverse contexts. Insights from these pilots will feed into a replication roadmap and a capacity-building program designed to train future leaders in climate-resilient urban development. By fostering co-creation with local stakeholders and communities, CLIMRES ensures its innovative solutions are practical, cost-effective, and replicable, targeting Technology Readiness Levels (TRL) 6-8. CLIMRES aims to bridge innovation with actionable solutions, equipping building owners, policymakers, and communities with the tools needed to enhance urban climate resilience. This presentation highlights the project’s interdisciplinary approach, outputs and technological underpinnings, offering insights into scalable solutions for climate adaptation in urban settings.

The detection of environmental changes caused by natural disasters is critical for rapid response and effective management. In this study, we present a methodology for unsupervised change detection that leverges optical Sentinel-2 [1] and Synthetic Aperture Radar (SAR) Sentinel-1 [2] accessed through public SpatioTemporal Asset Catalogs (STAC) [3] along with Earth Observation (EO) foundation model, namely, Clay [4]. The analysis was conducted independently for each dataset to capitalize on the unique properties of these satellite sensors. Sentinel-1 offers robust surface texture sensitivity with its all-weather, day-and-night imaging capability, while Sentinel-2 provides detailed spectral and spatial information critical for vegetation and land-use analysis.Clay foundation model, a large-scale pretrained Vision Transformer trained on EO data from various missions (Sentinel-1, Sentinel-2, Landsat, Planet, NAIP, LINZ, and MODIS), was used to extract spatially and spectrally rich embedding from Sentinel-1 and Sentinel-2 images. The model takes as an input the satellite imagery along with information about location and time and outputs mathematical representations of a given area at a certain time on Earth’s surface. The images were fed to the model as patches of size 256×256 along with the timestap of the scene, the spatial location and other metadata of the input image to estimate the embeddings that can be rearranged to be of size (1024×32×32). These embeddings were then analyzed using pixel-wise distance metrics to quantify changes between pre- and post-even imagery and the resulting distance image was then spatially interpolated to the size of the input image.The approach was validated on satellite imagery of the Valencia region in Spain, an area significantly impacted by recent flooding on the 29th October 2024. For Sentinel-1, the method effectively highlighted surface water changes and structure affected by the floods in two scenes acquired on the 7th October and the 12th November 2024, while Sentinel-2 data captured variations in vegetation areas that was impacted by the floods using two scenes acquired on the 1st October and the 10 November 2024. By analyzing the datasets independently, this framework demonstrates the complementary insights offered by radar and optical imagery in assessing disaster impacts.This study highlights the potential of leveraging open satellite data available via STAC catalogs and EO foundation models for unsupervised change detection in disaster monitoring, enabling rapid response without relying on specialized models tailored to specific regions. Unlike traditional approaches that require retraining for new areas due to geographical variability, this methodology is both scalable and adaptable, providing a generalizable framework for environmental monitoring, disaster response, and resilience planning. The results emphasize the value of integrating multi-sensor satellite imagery to enhance understanding of disaster impacts, facilitating more informed and timely decision-making.References:[1] https://earth-search.aws.element84.com/v1[2] https://planetarycomputer.microsoft.com/api/stac/v1[3] https://stacspec.org/en[4] https://clay-foundation.github.io/model/index.html

The DeployAI project [1] designs and delivers a fully operational European AI-on-Demand Platform (AIoDP) to empower the European industry with access to cutting-edge AI technology, and to promote trustworthy, ethical, and transparent European AI solutions, with a focus on SMEs and the public sector. To achieve this, the platform enables the development and deployment of AI solutions through the following core solutions: (i) AI Builder [2], which allows the assembling of reusable AI modules into AI pipelines; (ii) seamless access to Cloud and HPC infrastructures (e.g., MeluXina and LUMI); (iii) a marketplace for the listing and distribution of ready-to-use AI products; (iv) an expansive and growing library of diverse AI-driven use cases.As part of its domain-driven solutions, AIoDP seeks to empower Environmental Scientists, AI Engineers, Developers, Researchers, and SMEs via the DeployAI Earth Observation Services. These services will accelerate the development of AI-driven environmental applications, by providing pre-trained models that simplify satellite imagery processing, land usage classification, and image segmentation. Key models available as modules within the DeployAI’s AI Builder include:Leaf Area Index (LAI) Model: Enables precise monitoring of vegetation health and ecological dynamics by calculating leaf area per unit ground [3]. Object Detection Model: Identifies specific objects in high-resolution satellite images, supporting applications such as infrastructure monitoring, pollution tracking, and deforestation assessment [4]. Segment Anything Model (SAM): Simplifies analysis across diverse environmental applications through the capabilities of SAM that allows flexible, prompt-based image segmentation for new datasets, with zero-shot and few-shot learning [5]. These models, along with the broader functionalities of AI Builder, enable users to create custom AI pipelines that address their specific environmental challenges in several environmental areas, including vegetation health monitoring, water balance analysis, climate modeling, urban planning, traffic management, pollution monitoring, and infrastructure maintenance. Users can leverage the visual pipeline editor to easily assemble pipelines from reusable AI modules without needing to write code. Once created, these pipelines can be deployed as AI applications on various execution environments. DeployAI facilitates seamless transitions between these environments by providing connectors to a host of target infrastructures, including Cloud platforms and HPC systems. This empowers users to leverage the most suitable computational resources for their specific needs.By providing a user-friendly platform with access to cutting-edge AI technology and Cloud/HPC resources, DeployAI empowers users to address critical environmental challenges and unlock new possibilities for sustainable development.[1] https://deployaiproject.eu[2] https://gitlab.eclipse.org/eclipse/graphene[3] https://github.com/DeployAI-Environmental-Services/depai-lai[4] https://github.com/DeployAI-Environmental-Services/depai-yolov8-obb[5] https://github.com/DeployAI-Environmental-Services/depai-sam-interactiveThis work has received funding from the European Union’s Digital Europe Programme (DIGITAL) under grant agreement No 101146490.

The Copernicus program has fostered Earth Observation (EO) and Earth Modeling by offering extensive data and services to European Citizens. Sentinel satellites’ data is accessible through platforms like the Copernicus Open Access Hub and the Copernicus Data Space Ecosystem, which provide a wide range of information on land, ocean and atmospheric conditions. Complementing these resources, six specialized Copernicus services deliver data in domains such as the atmosphere, marine environment, land monitoring, climate change, security and emergency response. To streamline access and usability, cloud-based Copernicus Data and Information Access Services (DIAS) offer centralised platforms equipped with cloud infrastructure and processing tools. Building on these efforts, the Copernicus Data Space Ecosystem (https://dataspace.copernicus.eu/) enhances existing DIAS services with advanced functionalities like improved search capabilities, virtualizations and APIs. Meanwhile, the Destination Earth (DestinE) initiative led by ECMWF, EUMETSAT and ESA, aims to develop high-precision digital Earth models - or digital twins - that simulate natural and human activities. These models mainly focus on weather-induced extremes and climate change adaptation, generating valuable Earth Modeling data. Furthermore, European Data Spaces integrate datasets across diverse domains, including agriculture, health, energy, and environmental monitoring, creating opportunities to combine these resources with Copernicus and DestinE data through advanced technologies like artificial intelligence (AI) and machine learning (ML). This integration paves the way for innovative solutions and public-facing products and services. Despite the volume and richness of Copernicus and related EO data, its accessibility remains limited, with most users being experts or scientists. For broader industry adoption and the development of impactful applications that benefit society and the enviroment, significant barriers must be addressed. EO data is often fragmented, complex, and difficult to process, requiring domain expertise for tasks such as data discovery, pre-processing, storage, and conversion into formats suitable for analytics and Geographic Information Systems (GIS). The EO4EU platform (https://www.eo4eu.eu/), showcased in this presentation, introduces a multi-cloud ecosystem designed for holistic management of EO data. Its primary objective is to bridge the gap between domain experts and end users, leveraging technological advancements to broaden the adoption of EO data across diverse markets. By enhancing the accessibility and usability of EO data, EO4EU supports market growth through advanced data modeling, dynamic annotation, and state-of-the-art processing, powered by European cloud infrastructures such as WEkEO/DIAS and CINECA. EO4EU provides a suite of innovative tools and methodologies to assist a wide range of users, from professionals and domain experts to general citizens, in benefiting from EO data. Its key features include: Knowledge Graph-based Decision Making: Facilitates insightful feature extraction from diverse repositories, enabling a more comprehensive understanding of datasets. AI/ML Marketplace: A centralized hub for AI & ML models, algorithms, techniques, and metadata. Big Data Processing Engines: Optimized for cloud environments to efficiently manage large-scale datasets. User-friendly Interfaces: GUI, CLI, APIs, and immersive VR experiences, targeting both technical and non-technical users. Workflow Engine: Simplifies the definition and execution of recurring tasks for EO data retrieval and processing.

The BUILDSPACE project harnesses the transformative potential of cloud computing to evolve urban development and resilience practices. By integrating advanced Earth Observation (EO) data with state-of-the-art satellite and cloud technologies, BUILDSPACE addresses critical urban challenges, including climate adaptation, energy efficiency, and disaster resilience, while contributing to the European Green Deal’s objectives of sustainability and carbon neutrality. Central to BUILDSPACE are five innovative services designed to support urban decision-making. At the building scale, the project facilitates the generation and visualization of detailed digital twins through interactive displays, virtual reality (VR), and augmented reality (AR) interfaces. These digital twins enable precise simulations for energy optimization, operational efficiency, and climate impact assessment. At the city scale, BUILDSPACE provides tools to address climate scenarios, such as urban heat islands and flooding, empowering municipalities and urban planners with actionable insights through interactive, map-based platforms. The project’s technical foundation lies in a robust, cloud-native architecture built on Kubernetes and OpenStack, combined with a DevOps methodology to streamline both infrastructure services and application deployment. Kubernetes orchestrates containerised workloads, enabling efficient automated deployment, scaling and management of applications, while OpenStack provides a flexible infrastructure for managing compute, storage, and networking resources. Through the DevOps approach, we ensure continuous integration and delivery (CI/CD), fostering rapid development cycles and operational agility. By adopting open-source cloud platforms, the project ensures interoperability, reproducibility and automation across diverse environments, driving consistency and efficiency throughout the lifecycle of both infrastructure and applications. The project’s services are being validated across four European cities representing diverse climatic conditions, namely Warsaw, Riga, Piraeus and Ljubljana. These validations focus on two scenarios: construction companies monitoring building processes with advanced digital tools, and municipalities analysing the impacts of climate change on urban infrastructure. By advancing from TRL 5-6 to TRL 7-8, BUILDSPACE aims to deliver market-ready solutions that align with the European GNSS and Copernicus initiatives and to synchronise with the advances, concerning Digital Twin technologies and data federation mechanisms, of the Destination Earth initiative, while paving the way for a broader adoption of cloud technologies in EO-based urban resilience applications.