Project Description

Abstract

The 3CS4EO mission proposes a technology demonstrator for a new generation of Earth Observation (EO) systems in which sensing, computation, autonomous coordination, and communications are integrated directly into the space segment. Its objective is to demonstrate how advanced on-board processing, Artificial Intelligence, reconfigurable computing, and inter-satellite connectivity can reduce dependence on ground-based processing by converting large EO datasets into concise information products while still in orbit. This approach is intended to shorten the time between observation and user exploitation, improve responsiveness to transient phenomena, and limit the amount of data that must be transferred to the ground.

The Initial Operational Capability (IOC) is based on two spacecraft flying on the same Low Earth Orbit plane and operating according to a Leader–Follower concept. The Leader performs broad-area surveillance using EO sensors suited to rapid scouting and hosts the main high-performance computing and Platform-as-a-Service environment. Its on-board algorithms identify relevant phenomena, derive geolocated event information, and determine whether further observation is required. The Follower complements this capability by acquiring more detailed imagery of selected targets with a narrower field of view and a better spatial resolution.

The two spacecraft implement an autonomous Tip-and-Cue operational chain. Events detected by the Leader can generate targeting information for the Follower, whose mission-management functions evaluate observation opportunities and spacecraft resources before carrying out a refined acquisition. At the same time, compact notifications can be delivered to users via low-latency Direct-to-User (D2U) links, while larger datasets remain compatible with conventional ground distribution. Optical inter-satellite connectivity also enables future integration with external communication infrastructures and other EO systems.

The demonstrator prioritizes applications for which rapid interpretation and reaction offer significant operational benefits within the primary AoI at IOC, including maritime surveillance, wildfire monitoring, forestry, AFOLU, and intelligent cloud screening.

Beyond IOC, the concept is designed to evolve into a multi-node, multi-plane architecture capable of federating heterogeneous EO platforms, communication networks, and distributed computing resources. In this perspective, 3CS4EO serves as a scalable precursor to space-based cloud infrastructures in which data interpretation, tasking, and information generation progressively migrate toward autonomous and collaborative orbital assets

Detailed description

The 3CS4EO concept is built around an integrated in-orbit sensing, processing, and decision-making chain, in which EO spacecraft are assigned a more active role in generating and exploiting information. Rather than limiting the space segment to data acquisition and transmission, the mission architecture introduces on-board capabilities for event recognition, information extraction, autonomous coordination, and responsive tasking, allowing selected operational decisions to be taken closer to the point of observation.

This shift from a predominantly data-delivery-oriented approach to a more cognitive operational model is intended to address key limitations identified by stakeholders, including data-delivery latency, limited temporal and spatial responsiveness, and the burden of processing and downlinking large volumes of EO data. The mission shall address multiple use cases by exploiting the complementarity of heterogeneous EO payloads, with priority given to Maritime Security, particularly vessel detection; Wildfire Detection and Monitoring; AFOLU and forest monitoring; and Cloud Detection and cloud-aware filtering.

At Initial Operational Capability (IOC), the vision is implemented through four core capabilities:

• on-board AI and processing for the conversion of EO measurements into event-oriented information products;
• cooperative multi-satellite sensing between heterogeneous spacecraft;
• low-latency/Direct-to-User (D2U) communications supporting tactical services in which an initial alert can be made available within minutes of acquisition or event detection;
• a Platform-as-a-Service (PaaS) capability providing a modular and reconfigurable on-board computing environment for the controlled deployment and execution of mission-defined and user-provided processing applications.

In the longer-term Extended Capability (EC), the architecture is intended to evolve toward greater autonomy and more payload heterogeneity, with a larger number of 3CS nodes on multiple orbital planes, interoperability and federation with external EO constellations and transport layers, and progressively distributed processing across space assets. This provides a path from the initial cognitive EO demonstrator toward an operational building block for future space-cloud services, while preserving scalability and interoperability as architectural principles.

The IOC system concept shall be based on a two-satellite Leader-Follower configuration in Low Earth Orbit (LEO). The satellites shall be placed in the same orbital plane, with the Follower phased behind the Leader, with an along-track separation sufficient for inter-satellite communication. The Leader is a platform intended to host one or more EO payloads primarily dedicated to wide-area scouting of the Area of Interest (AoI). Accordingly, its sensing suite is expected to favor larger swath coverage over very high spatial resolution, enabling the detection of relevant events or features for further investigation. The Leader shall also host a high-performance on-board computing capability, such as an HPC, to perform advanced data processing, AI-based inference, and autonomous event detection. The Follower is instead intended to provide a more detailed follow-up observation characterized by a narrower swath and higher spatial resolution. It shall be equipped with a processing unit capable of executing onboard processing functions that are compatible with the computational, power, and resource constraints.

The operational logic is centered on an in-orbit sensing-processing-coordination loop. During a Leader acquisition, on-board processing can detect an event or anomaly and generate a compact event-intelligence product, typically including information such as event type, acquisition time, location, confidence level, and, where applicable, a recommended follow-up action. A Tip message can then be exchanged with the Follower through the intra-constellation link; if the requested acquisition is feasible, the Follower performs a Cue observation over the same AoI using its complementary, higher-resolution payload. On-board mission management and intelligent tasking functions may assess acquisition feasibility, priorities, and available spacecraft resources, supporting local replanning and autonomous coordination of the follow-up observation. In parallel, compact alerts may be transmitted directly to the user via the IoT-based D2U capability. At the same time, higher-volume imagery and products are subsequently delivered through the nominal ground data chain. At least the Leader shall have optical inter-satellite communication capability to support connectivity with an external transport layer and, in the EC evolution, federation with external EO systems.

A key element of the 3CS4EO concept is the adoption of a PaaS paradigm, intended to make the on-board computing capability available to the user as a modular and reconfigurable execution environment. At IOC, the Leader shall provide the primary PaaS capability, enabling authorized users to deploy and execute validated mission-specific or experimental software, including AI/ML algorithms and EO processing applications.

Due to the reduced number of space nodes at IOC, a primary AoI shall be identified, within which the target temporal and geometrical performance shall be achieved. Broader and more systematic global service is associated with EC, where additional nodes, orbital planes, and external federation can improve coverage, revisit, and latency. With only two satellites in the same orbital plane, the classical revisit time does not, by itself, represent the distinctive value of the IOC mission. The concept therefore emphasizes E2E responsiveness, rapid event extraction, coordinated follow-up, and efficient information delivery, with performance guarantees focused on the Primary AoI at IOC and progressively extended toward global coverage at EC.

At IOC, the mission shall focus on mission-level performance targets, aiming for a System Response Time below 30 minutes within the primary service area without exploiting an external transport layer. Where available, an external transport layer may provide additional latency reduction; at IOC, its primary role is to improve temporal performance for acquisitions outside the Primary AoI, with a target SRT below 90 minutes. A first actionable alert shall be delivered within 10 minutes from acquisition of the Leader, while coordinated follow-up observations shall be triggered and performed within a few minutes, enabling a significant spatial refinement of the initial detection, with the follow-up observation targeting sub-10 m spatial resolution. For standard monitoring at IOC, the mission shall aim for a maximum revisit time of 6-8 days or better within the Primary AoI, depending on the observation capability and use case. These figures should be considered mission-level performance objectives for defining a future system concept, rather than constraints tied to a specific payload or platform implementation.

Taken as a whole, 3CS4EO provides a scalable reference architecture for Earth Observation in which sensing, processing, autonomous coordination, and communication are treated as a single E2E service chain. The compact IOC configuration demonstrates the core mechanisms required for a future network of cognitive EO assets - on-board AI, cooperative Tip & Cue, heterogeneous sensing, direct user alerting, a reconfigurable PaaS environment for in-orbit software deployment and experimentation, and interoperability with external communication and observation infrastructures - while the EC roadmap extends them toward federated and distributed space-based computing, with the objective of improving revisit, latency, resilience and the range of EO information generated closer to the point of acquisition.

Project Outcomes

Development Outputs

• Vessel Detection: end-to-end dockerized pipeline for Unibap iX10-102/Hailo-8 package and NVIDIA Jetson/PyTorch package. Preprocessing, inference, and postprocessing for vessel detection on optical scenes. Includes a trained model checkpoint and a sample dataset with annotations. The output is intended to support the activation of an alert or tip-and-cue strategy. It consists of a message including metadata and information about the detection, and may also include an image chip of the detected target.
• Wildfire Detection – Unibap iX10-102/Hailo-8 package: end-to-end Dockerized pipeline (preprocessing, inference, postprocessing) for pixel-level wildfire segmentation from thermal bands (11.50 – 12.51 µm), compiled for the Unibap iX10-102/Hailo-8 target platform. Includes the compiled model and labeled sample dataset for validation. The output is intended to support the activation of an alert or tip-and-cue strategy. It consists of a message including metadata and information about the detection, and may also include an image chip of the detected target.
• Participation in the ESA Φ-Innovation Summit 2026 to present the outcome of the 3CS4EO technical feasibility study.

Development Tools

• Vessel Detection Model (YOLOv8n-P2, single Sentinel-2 B04 band, 640×640 input): AI Model prototype for on-board vessel detection in optical imagery, delivered as a PyTorch checkpoint for NVIDIA/Jetson platforms and as a compiled Hailo-8 HEF for the Unibap iX10-102.
• Wildfire Detection Model (U-Net, 16 filters, thermal-only input from Landsat-8 bands 10/11): AI Model prototype for on-board segmentation (pixel-level) of active-fire using only thermal bands, delivered as a compiled Hailo-8 HEF for the Unibap iX10-102, trained on a small sample of expert-labeled fire masks.

Future Development

• Cloud Detection: E2E dockerized pipeline for cloud detection in optical multispectral imagery. The onboard AI model generates a cloud probability mask, together with a message containing relevant metadata and detection information. The output is intended to support the tip-and-cue strategy and decision-making on the retention or rejection of cloud-contaminated acquisitions. The application will be provided for both the Unibap iX10-102/Hailo-8 and NVIDIA Jetson/PyTorch platforms.
• AFOLU: End-to-end dockerized pipeline for forest change detection in optical multispectral imagery. The onboard AI model identifies and localizes relevant changes in forest cover, potentially associated with deforestation, illegal logging, or the development of new access roads, and generates a change-detection mask along with a message containing relevant metadata and detection information. The output is intended to support the tip-and-cue strategy, enabling the prioritization of areas requiring further observation and higher-resolution follow-up acquisitions. The application will be provided for both the Unibap iX10-102/Hailo-8 and NVIDIA Jetson/PyTorch platforms.