Paper deep dive
Assessing the Added Value of Onboard Earth Observation Processing with the IRIDE HEO Service Segment
Parampuneet Kaur Thind, Charles Mwangi, Giovanni Varetto, Lorenzo Sarti, Andrea Papa, Andrea Taramelli
Intelligence
Status: succeeded | Model: google/gemini-3.1-flash-lite-preview | Prompt: intel-v1 | Confidence: 95%
Last extracted: 4/10/2026, 3:48:40 AM
Summary
This paper evaluates the operational value of onboard Earth Observation (EO) processing within the Italian IRIDE programme, specifically focusing on the Hawk for Earth Observation (HEO) service segment. By comparing traditional ground-only processing pipelines (like Copernicus/EFFIS) with a hybrid onboard-ground architecture, the authors demonstrate that onboard AI inference enables reduced latency, lower bandwidth requirements, and improved detection of small-scale events (e.g., 3-hectare burnt areas). The study positions the IRIDE HEO capability as a complementary, intelligence-supported layer for national-scale emergency and land-management workflows.
Entities (5)
Relation Signals (3)
IRIDE â operates â HEO
confidence 100% · Within this framework, Hawk for Earth Observation (HEO) enables onboard generation of data products
HEO â utilizes â Intel Myriad X
confidence 95% · The PDP is responsible for executing AI inference using the Intel Myriad X Vision Processing Unit
IRIDE â complements â Copernicus
confidence 90% · Rather than replacing existing Copernicus services, the IRIDE HEO capability is positioned as a complementary layer
Cypher Suggestions (0)
No Cypher suggestions yet.
Abstract
Abstract:Current operational Earth Observation (EO) services, including the Copernicus Emergency Management Service (CEMS), the European Forest Fire Information System (EFFIS), and the Copernicus Land Monitoring Service (CLMS), rely primarily on ground-based processing pipelines. While these systems provide mature large-scale information products, they remain constrained by downlink latency, bandwidth limitations, and limited capability for autonomous observation prioritisation. The International Report for an Innovative Defence of Earth (IRIDE) programme is a national Earth observation initiative led by the Italian government to support public authorities through timely, objective information derived from spaceborne data. Rather than a single constellation, IRIDE is designed as a constellation of constellations, integrating heterogeneous sensing technologies within a unified service-oriented architecture. Within this framework, Hawk for Earth Observation (HEO) enables onboard generation of data products, allowing information extraction earlier in the processing chain. This paper examines the limitations of ground-only architectures and evaluates the added value of onboard processing at the operational service level. The IRIDE burnt-area mapping service is used as a representative case study to demonstrate how onboard intelligence can support higher spatial detail (sub-three-metre ground sampling distance), smaller detectable events (minimum mapping unit of three hectares), and improved system responsiveness. Rather than replacing existing Copernicus services, the IRIDE HEO capability is positioned as a complementary layer providing image-driven pre-classification to support downstream emergency and land-management workflows. This work highlights the operational value of onboard intelligence for emerging low-latency EO service architectures.
Tags
Links
- Source: https://arxiv.org/abs/2604.07120v1
- Canonical: https://arxiv.org/abs/2604.07120v1
Trouble viewing inline? Open PDF directly â
Full Text
31,912 characters extracted from source content.
Expand or collapse full text
ASSESSING THE ADDED VALUE OF ONBOARD EARTH OBSERVATION PROCESSING WITH THE IRIDE HEO SERVICE SEGMENT Parampuneet K. Thind ââ ⥠, Charles Mwangi â⥠, Giovanni Varetto âĄâ„ , Lorenzo Sarti â , Andrea Papa â , Andrea Taramelli âĄÂ¶ â Sapienza Universit ` a di Roma, Rome, Italy â Argotec Srl, Turin, Italy ⥠Istituto Universitario di Studi Superiori (IUSS), Pavia, Italy ¶ Institute for Environmental Protection and Research (ISPRA), Rome, Italy â„ Universit ` a di Milano-Bicocca, Milan, Italy â European Space Agency (ESA), ESRIN, Rome, Italy Email: parampuneet.thind@uniroma1.it, charles.mwangi@uniroma1.it, g.varetto@campus.unimib.it, lorenzo.sarti@argotecgroup.com, andrea.papa@esa.int, andrea.taramelli@iusspavia.it AbstractâCurrent operational Earth Observation (EO) ser- vices, including those provided through the Copernicus Emer- gency Management Service (CEMS), the European Forest Fire Information System (EFFIS) (within the CEMS), and the Copernicus Land Monitoring Service (CLMS), rely primarily on remote, ground-based processing pipelines. While these systems deliver mature large-scale information products, they remain constrained by downlink latency, bandwidth availability, and lim- ited capability for autonomous prioritisation of observations. The International Report for an Innovative Defence of Earth (IRIDE) programme is a national Earth observation system launched by the Italian government to support public authorities through timely, objective information derived from spaceborne data. Rather than a single satellite constellation, IRIDE is structured as a constellation of constellations, integrating heterogeneous sensing technologies into a unified service-oriented architecture. Within this framework, Hawk for Earth Observation (HEO) generates onboard data products that enable information extraction earlier in the processing chain. This paper examines the limitations of ground-only service architectures and evaluates the potential added value of onboard processing at the operational service level. The IRIDE burnt-area mapping service is adopted as a representative case study to illustrate how onboard intelligence can be used to target higher spatial detail (sub-three-metre ground sampling distance), smaller detectable event sizes (a min- imum mapping unit of three hectares), and improved overall sys- tem responsiveness. Rather than replacing existing Copernicus- based services, the IRIDE HEO capability is positioned as a complementary layer providing image-driven pre-classification to support downstream emergency and land-management work- flows. By situating the IRIDE HEO service segment within the broader evolution of onboard processing technologies, this work demonstrates the operational value of onboard intelligence for emerging low-latency, intelligence-supported Earth observation service architectures. Index TermsâOnboard Processing, Earth Observation, IRIDE, HEO, Service Value Chains (SVC), Edge Artificial Intelligence (AI), Operational EO Services, High Resolution Mapping, Burnt- Area Mapping I. INTRODUCTION Operational EO services are dominated by remote, ground- based processing pipelines in which full-scene imagery is downlinked before any information extraction occurs [1]. Such architectures traditionally assume that high-level data products must be generated through extensive preprocessing chains executed on the ground [2]. Core European services such as CEMS (including EFFIS), and CLMS will take advantage of this downlink-first paradigm [3, 4]. While this approach has proven effective for continental-scale monitoring [5], it inherently ties product timeliness to downlink availability and ground-segment throughput [6]. As a consequence, such architectures are prone to latency bottlenecks and bandwidth saturation during crisis events [7], motivating hybrid onboardâ ground processing concepts aligned with next-generation hy- brid constellation and NewSpace mission architectures [8â10]. At the same time, the EO domain is experiencing rapid growth in data volume, spatial resolution, and revisit fre- quency, intensifying pressure on ground infrastructures [11]. In response, concepts derived from terrestrial edge processing where computation is relocated closer to the data source to reduce latency and data movement [12] are increasingly extended to space systems as orbital edge computing [13]. In this context, onboard machine inference is emerging as a core architectural element for future EO missions [14]. Over the last decade, multiple missions have demonstrated the technical feasibility of onboard AI for EO [15]. The ΊSat- 1 mission implemented the CloudScout deep neural network for onboard cloud detection [16]. CloudScout operated on hyperspectral imagery acquired by the HyperScout-2 im- ager [17]. Onboard inference was executed on the Intel Movid- ius Myriad-2 vision processing unit, a low-power Commercial arXiv:2604.07120v1 [cs.CV] 8 Apr 2026 Off-The-Shelf (COTS) accelerator used for in-orbit deploy- ment [18]. Building on this line of development, ΊSat-2 extends the concept toward a multi-application onboard-AI platform for operationally relevant tasks [19]. The mission concept has been presented as supporting applications such as cloud detection and image-to-map conversion [20]. More recent descriptions further include capabilities such as vessel awareness, wildfire detection, anomaly detection, and onboard compression [21]. Additional recent missions, including real- time fire detection on the FOREST-2 satellite, further illustrate the increasing maturity of onboard intelligence for time-critical event monitoring [22]. At the system level, roadmapping efforts on spaceborne data handling identify onboard pro- cessing as a key enabling technology for future autonomous missions [23]. These roadmaps further highlight the role of AI in supporting increased onboard autonomy and responsiveness under downlink constraints [24]. Complementary technology- focused assessments emphasise onboard AI as a driver for next-generation spacecraft data systems [10]. Despite these advances, most in-orbit AI applications re- main technology-driven demonstrators focused on algorithm validation and hardware feasibility [25, 26]. As a result, onboard AI is rarely embedded within institutional EO service chains with formally defined product specifications, accuracy targets, and operational end users [27]. The IRIDE programme addresses this gap through a user-driven, service-oriented systems engineering approach for a national, multi-platform EO system [28], in which institutional requirements such as target spatial resolution, revisit time, latency, and thematic accuracy are formalised at service level and translated into constellation architectures and sensing strategies through up- stream, model-based design and optimisation studies spanning multiple thematic domains [29â31]. Within this framework, the HEO microsatellite constellation provides agile multispectral optical sensing at sub-three-metre spatial resolution from low Earth orbit (LEO), operating at an altitude of approximately 500â600 km, together with onboard computing resources dimensioned for AI-based in- orbit processing [32]. Unlike previous experimental platforms, HEO is conceived as an operational component of the IRIDE service portfolio [33]. Burnt-area mapping is a representative service targeting national-scale binary products with a three- hectare minimum mapping unit and> 95% thematic accuracy, supporting institutional users and service evolution pathways identified for Copernicus-based monitoring systems [34, 35]. This paper evaluates the added value of onboard processing at the operational service level using the IRIDE HEO onboard service segment as a representative national-scale case. Specif- ically, we: (i) analyse the structural limitations of remote-only processing in existing operational EO services (i) describe the IRIDE HEO architecture and its role within the institutional SVC and (i) assess the operational implications of onboard intelligence through burnt-area mapping service. The objective is to demonstrate, through an end-to-end onboard processing chain, the tangible service-level benefits enabled by onboard AI within next-generation national EO architectures. I. SERVICE-LEVEL COMPARISON OF OPERATIONAL BURNT-AREA SYSTEMS A. Structural Limitations of Remote-Only Architectures All current large-scale operational burnt-area services rely on a remote-only processing paradigm in which full-resolution imagery is systematically downlinked prior to analysis, as exemplified by the CEMS EFFIS service. While effective at continental and multi-national scale, this architecture intrinsi- cally couples product timeliness to downlink availability and ground-processing capacity. Furthermore, medium-resolution sensing baselines inherently constrain the minimum mapping unit, limiting the detection of small and fragmented burn scars that are often operationally relevant at national and local scale. B. Comparison Between Copernicus Services and the IRIDE National Burnt-Area Service Table I compares the service-level characteristics of selected operational European burnt-area mapping systems with the na- tional burnt-area mapping service developed within the IRIDE programme and enabled by the HEO optical constellation. The comparison focuses explicitly on downstream service characteristics, distinguishing between acquisition strategies and product generation logic, rather than on upstream mission tasking or platform autonomy. In particular, the comparison highlights overall timeliness as a key differentiating factor, showing how the IRIDE HEO service segment reduces the time to first actionable information through onboard process- ing and hybrid product generation. The EFFIS service provides systematic, European-scale burnt-area products derived from medium-resolution satel- lite data, optimised for seasonal reporting and continental monitoring. Product generation follows periodic processing cycles executed entirely on the ground. The CEMS Emergency Mapping component, while not a dedicated burnt-area moni- toring service, delivers on-demand crisis products activated in response to specific events and relies on heterogeneous data sources, including commercial very-high-resolution imagery. The CLMS supports long-term land-cover and land-use moni- toring through systematic acquisitions and multi-annual update cycles, rather than near-real-time fire event characterisation. By contrast, the IRIDE burnt-area service is conceived as a national operational service based on systematic satel- lite acquisitions, in which product generation is dynamically driven by detected fire events. In this context, fire occurrences identified through external monitoring systems and institu- tional information flows trigger dedicated processing chains on newly acquired imagery, enabling rapid generation of preliminary burnt-area information without relying on changes to nominal acquisition plans or autonomous satellite retasking. The service targets three-metre spatial resolution, a Minimum Mapping Unit (MMU) of three hectares, and full territorial coverage under institutional operational priorities. Unlike the other services considered, IRIDE integrates onboard process- ing as part of its operational value chain, enabling early information extraction prior to full data downlink. TABLE I SERVICE-LEVEL COMPARISON OF OPERATIONAL BURNT-AREA MAPPING SYSTEMS, WITH EMPHASIS ON OVERALL TIMELINESS Figure of meritEFFISCEMS EMCLMSIRIDE (HEO) Acquisition mode Systematic On-demandSystematicSystematic Product triggeringPeriodicCrisisPeriodicEvent-driven Processing location GroundGroundGroundHybrid Spatial resolution20 m10â30 m â 10â100 mâŒ3 m Minimum mapping unit âŒ10 haâ âŒ0.5â1 ha3 ha Time to first infoDaysHoursMonthsâyearsMinutesâhours (target) End-to-end latency DaysHoursâdaysYearsHours (target) Downlinked data volume Full scene Full sceneFull sceneReduced (ROI / thematic) Onboard process- ing NoNoNoYes Operational scope EuropeanCrisisEuropeanNational â Sub-metric achievable with commercial VHR imagery. I. THE IRIDE CONSTELLATION AND THE HEO ONBOARD SERVICE SEGMENT A. The IRIDE Programme and National Service Architecture The IRIDE initiative, developed by the European Space Agency (ESA) with the support of Agenzia Spaziale Italiana (ASI), represents Italyâs flagship national EO programme un- der the Piano Nazionale di Ripresa e Resilienza (PNRR) [36]. Designed as a multi-segment, end-to-end system, IRIDE inte- grates: (i) an upstream segment of six satellite constellations (optical, SAR, hyperspectral), (i) a downstream segment consisting of ground infrastructure for satellite operations, data processing, mission planning, and (i) a service segment ded- icated to delivering geospatial products and thematic services to institutional and commercial users [37]. The upstream segment comprises multiple satellites with diverse sensors. The downstream segment hosts critical sys- tems such as the Flight Operation Systems (FOS), Payload Data Ground Segment (PDGS), and Central Mission Planning Module (CMPM). These components interact to orchestrate satellite tasking, receive insights or raw data, and distribute final products to users via the IRIDE marketplace. Figure 1 illustrates the IRIDE system architecture, with the upstream spacecraft and onboard processing elements reflecting the HEO platform developed by Argotec, the downstream service and marketplace components aligned with the official archi- tecture described in the IRIDE programme brochure [37]. B. The HEO Optical Segment: Onboard-Ready Satellites Among IRIDEâs constellations, the HEO optical micro con- stellation carries high-resolution multispectral sensors (sub- 3 m GSD) and has the potential to perform first-order inference directly onboard, enabling real-time responsiveness to events such as wildfires. Each HEO unit is equipped with the Unibap iX5-106 SpaceCloud Âź platform, hosting the Payload Data Pro- cessor (PDP) and the Onboard Computer and Data Handling (OBC&DH) subsystems. The PDP is responsible for executing AI inference using the Intel Âź Myriad X Vision Processing Unit (VPU), while the OBC&DH remains responsible for platform control and system monitoring, and the PDP supports it by executing computationally intensive tasks, including onboard image processing. The onboard architecture supports multiple radio interfaces: the S-band for telemetry and telecommand (TM/TC), and the X-band for payload data transmission. C. Integrated Tasking and Product Delivery Workflow The downstream segment includes multiple FOS instances, each associated with a specific upstream constellation and responsible for managing satellite health, telemetry, and link control, including payload data downlink. Within this archi- tecture, a dedicated FOS is assigned to the HEO optical constellation and is operated directly by Argotec. Each FOS interfaces with the CMPM to receive observation requests and tasking updates, which are then dispatched to the correspond- ing satellites. Following acquisition, raw data or onboard infer- ence products are downlinked to the corresponding FOS and subsequently forwarded to the PDGS for ingestion, processing, and distribution through the IRIDE marketplace. This distributed operational setup enables a hybrid service chain in which inference products may be generated onboard the satellite or derived on ground from raw data, depending on data quality, cloud conditions, and service urgency. The ar- chitectural integration of constellation-specific FOS instances with the CMPM and the downstream product management chain ensures end-to-end traceability from tasking to product delivery, as well as timely responsiveness for high-priority applications such as civil protection. IV. ADDED VALUE OF ONBOARD PROCESSING FOR BURNT-AREA MAPPING Integrating onboard processing in IRIDE enables service level gains beyond incremental ground-segment scaling. A. Latency and Bandwidth Reduction By performing radiometric pre-processing and first-order inference directly onboard, the system can downlink thematic outputs or prioritised regions of interest instead of complete raw image strips. This reduces downlinked data volume and alleviates bottlenecks in large-scale fire scenarios. Critically, the end-to-end latency from acquisition to product generation is reduced to minutes or hours, enabling faster situational awareness for emergency response. B. Improved Detection of Fragmented Fires The spatial resolution of 3 m and MMU of 3 ha significantly enhance the ability to detect small or discontinuous burn scars in heterogeneous landscapes. For national authorities such as ISPRA, SNPA, and the Civil Protection Department, this capability is essential for legal enforcement and ecological damage assessment in wildland-urban interfaces [38]. C. Implications for Onboard Model Design Deploying AI models onboard radiation-constrained, low- power satellite platforms such as the iX5 Myriad-X requires careful optimisation of model size, execution time and robust- ness. Addressing these constraints through hardware-aware S5: Hydro meteo - rological climate S1: Coastal and Marine Monitoring Commercial User Interface IRIDE Data resources (EO, in situ, Cyber Italy models) 3rd party data (In situ, EO, others) Co-registration, Time series, Change detection, Specific algorithms Thematic Services Processing and Analytics Resources S2: Air Quality S3: Ground Motion S4: Land Use/Land Cover S6: Water Management S7: Emergency S8: Security (SERVICE PROVIDER) (ALGORITHMS AND MODEL PROVIDERS) (DATA PROVIDERS) Commercial Supplier Interface User Interface MARKETPLACE STRUCTURE Central mission planning and monitoring (CMPM) CMPM receives all request from marketplace and dispatches it to the designated FOS Flight operations Segment (FOS) Payload data ground segment (PDGS) Request sent from Marketplace FOS receives request from CMPM FOS S-band antenna sends command to designated satellite FOS X-band antenna receives data from satellite PDGS receives data from FOS and processes it Data is provided to Marketplace and IRIDE services CPUGPU On-Board Computer (OBC) S Band Radio Payload App n CPU Onboard Processing Framework App 1 App 2 L0 - L1B Pre-proc essing Pipeline X Band Radio Payload Data Processor (PDP) ... HEO sub-constellation Spacecraft GPU VPU Copernicus Core Services Fig. 1. System-level architecture of IRIDEâs three core segments: upstream (spaceborne sensors including HEO), downstream (FOS, CMPM, PDGS), and service (marketplace-enabled user delivery). Arrows indicate command and data flow paths. model and pipeline co-design enables reliable, real-time in- ference in orbit, ensuring that onboard processing delivers consistent service-level benefits within the operational IRIDE SVC. V. DESIGNING ONBOARD AI MODELS FOR OPERATIONAL EO SERVICES Translating operational service requirements into deployable onboard AI requires a system-aware approach to AI model design. Unlike ground-based processing pipelines, onboard inference must operate under strict constraints on memory footprint, computational throughput, power consumption, and execution latency. As a result, the feasibility of onboard processing for operational EO services is determined not only by algorithmic accuracy, but by the joint optimisation of models and target hardware. A. Deployment Constraints and Requirements The onboard VPU adopted within the HEO constellation, the Intel Âź Myriad X VPU, is representative of a class of low- power accelerators designed for embedded and spaceborne inference. While capable of high-throughput execution, such platforms impose hard limits on model size and require support for reduced-precision arithmetic (e.g. INT8 or FP16). Infer- ence latency must remain compatible with real-time or near- real-time product generation to ensure that onboard processing delivers tangible service-level benefits. Meeting these constraints typically requires a combination of optimisation strategies, including quantisation, pruning and knowledge distillation, and approaches like Neural Architec- ture Search (NAS), enabling automated co-design of model structure and deployment characteristics. The HEO onboard AI stack leverages inference toolchains such as OpenVINO, supporting model conversion, layer fu- sion, and runtime optimisation for deployment on the Myriad X accelerator. These tools provide a practical bridge between algorithm development and operational onboard execution. B. From Feasibility to Service Integration Recent work [39] demonstrated that hardware-aware NAS can generate lightweight segmentation models satisfying the latency and resource constraints of Myriad and Orin-class processors, achieving substantial speedups with respect to manually designed baselines. While that study focused on methodological validation and benchmarking across hardware platforms, it establishes the technical feasibility required for deploying high-resolution segmentation models in orbit. Within the scope of the present paper, this class of hardware- aware design approaches is considered an enabling capability rather than a delivered service component. The focus remains on assessing the architectural and operational conditions under which onboard processing provides added value at the service level, independent of any specific model instantiation. VI. CONCLUSION AND OUTLOOK This paper assessed onboard processing as a service-level enabler within operational EO architectures, using the IRIDE HEO service segment and burnt-area mapping as a case study. Relative to ground-only pipelines, hybrid onboard- ground chains can reduce latency and downlink load while improving continuity under emergency acquisition peaks. The analysis positions IRIDE HEO as complementary to Coper- nicus services, targeting higher spatial detail and smaller event detectability at national scale. Ongoing work focuses on integrating a deployable onboard burnt-area model into the IRIDE service value chain, leveraging hardware-aware design approaches such as NAS demonstrated in [39]. REFERENCES [1] K. Kaku and A. Held, âSentinel asia: A space-based disaster management support system in the asia-pacific region,â Inter- national Journal of Disaster Risk Reduction, vol. 6, p. 1â17, 2013. [2] G. Meoni, R. Del Prete, L. Ancos-Villa, E. Albalate-Prieto, D. Rijlaarsdam, J. L. Espinosa-Aranda, N. Long Ì ep Ì e, M. D. Graziano, and A. Renga, âE2e: Onboard satellite real-time classification of thermal hotspots events on optical raw data,â Astrodynamics, vol. 9, no. 3, p. 447â463, 2025. [3] European Forest Fire Information System, âEFFIS - Rapid Damage Assessment,â https://forest-fire.emergency.copernicus. eu/about-effis/technical-background/rapid-damage-assessment, accessed: 2025-12-04. [4] CopernicusLandMonitoringService,âEuropean imagemosaic,âhttps://land.copernicus.eu/en/products/ european-image-mosaic, accessed: 2025-12-04. [5] P. Barbosa, J. Kucera, P. Strobl, P. Vogt, A. Camia, and J. San-Miguel-Ayanz, âEuropean forest fire information system(effis)ârapiddamageassessment:Appraisalof burnt area maps in southern europe using modis data (2003â2005),â Forest Ecology and Management, vol. 234, p. S218, 2006. [Online]. Available: https://w.sciencedirect. com/science/article/pii/S0378112706007857 [6] Z. Fang, P. Yue, M. Zhang, J. Xie, D. Wu, and L. Jiang, âA service-oriented collaborative approach to disaster decision support by integrating geospatial resources and task chain,â International Journal of Applied Earth Observation and Geoin- formation, vol. 117, p. 103217, 2023. [7] S. Chien, J. Doubleday, D. R. Thompson, K. L. Wagstaff, J. Bellardo, C. Francis, E. Baumgarten, A. Williams, E. Yee, E. Stanton, and J. Piug-Suari, âOnboard autonomy on the intelligent payload experiment cubesat mission,â Journal of Aerospace Information Systems, vol. 14, no. 6, p. 307â315, 2017. [Online]. Available: https://doi.org/10.2514/1.I010386 [8] S. Chien, R. Shenvood, D. Tran, B. Cichy, G. Rabideau, R. Castano, A. Davies, R. Lee, S. Shulman, S. Ungar et al., The EO-1 Autonomous Science Agent Architecture. Pasadena, CA: Jet Propulsion Laboratory, National Aeronautics and Space . . . , 2004. [9] S. Madry, Innovative Design, Manufacturing and Testing of Small Satellites, ser. Springer eBooks. Springer Nature, 2018. [10] M. Pastena, B. Carnicero Dom Ì Ä±guez, P. Mathieu, A. Regan, M. Esposito, S. Conticello, C. Dijk, N. Vercruyssen, P. Manzillo, R. Koelemann, and J. Hefele, âEsa earth observation directorate newspace initiatives,â 08 2019. [11] D. Selva and D. Krejci, âA survey and assessment of the ca- pabilities of cubesats for earth observation,â Acta Astronautica, vol. 74, p. 50â68, 2012. [Online]. Available: https://w. sciencedirect.com/science/article/pii/S0094576511003742 [12] W. Shi, J. Cao, Q. Zhang, Y. Li, and L. Xu, âEdge computing: Vision and challenges,â IEEE Internet of Things Journal, vol. 3, p. 1â1, 10 2016. [13] B. Denby and B. Lucia, âOrbital edge computing: Machine in- ference in space,â IEEE Computer Architecture Letters, vol. 18, no. 1, p. 59â62, 2019. [14] B. Denby, B. Lucia, S. Noghabi, K. Chintalapudi, and R. Chan- dra, âKodan: Addressing the computational bottleneck in space,â in ASPLOS 2023 : Architectural Support for Programming Lan- guages and Operating Systems, March 2023. [Online]. Avail- able:https://w.microsoft.com/en-us/research/publication/ kodan-addressing-the-computational-bottleneck-in-space/ [15] G. Giuffrida, L. Fanucci, G. Meoni, M. Batic, L. Buckley, A. Dunne, C. Dijk, M. Esposito, J. Hefele, N. Vercruyssen, G. Furano, M. Pastena, and J. Aschbacher, âTheΊsat-1 mis- sion: The first on-board deep neural network demonstrator for satellite earth observation,â IEEE Transactions on Geoscience and Remote Sensing, vol. 60, p. 1â1, 01 2021. [16] G. Giuffrida, L. Diana, F. de Gioia, G. Benelli, G. Meoni, M. Donati, and L. Fanucci, âCloudscout: A deep neural network for on-board cloud detection on hyperspectral images,â Remote Sensing, vol. 12, no. 14, 2020. [Online]. Available: https://w.mdpi.com/2072-4292/12/14/2205 [17] M. Esposito and A. Zuccaro Marchi, âIn-orbit demonstration of the first hyperspectral imager for nanosatellites,â 07 2019, p. 71. [18] Intel Corporation, âIntelÂź movidiusâą myriadâą 2 vision processing unit,â 2017, accessed: 2025-05-02. [Online]. Avail- able:https://w.intel.com/content/w/us/en/products/sku/ 122461/intel-movidius-myriad-2-vision-processing-unit-4gb/ specifications.html [19] N. Melega, N. Long Ì ep Ì e, A. Pa Ë skevi Ë ci Ì ut Ì e-Kidron, V. Marchese, O. Aragon, I. Babkina, A. Marin, J. Nalepa, L. Buckley, G. Guerrisi, S. Oliveira, and H. Steyn, âDevelopment and implementation of theΊsat-2 mission,â in Small Satellites Systems and Services Symposium (4S), 03 2025, p. 105. [20] A. Marin, C. Coelho, F. Deconinck, I. Babkina, N. Long Ì ep Ì e, and M. Pastena, âPhi-sat-2: Onboard ai apps for earth observation,â in Proceedings of Space Artificial Intelligence, 09 2021. [21] N. Longepe, âThe ESAΊsat-2 mission: an AI empowered 6U Cubesat for Earth Observation,â in AGU Fall Meeting Abstracts, ser. AGU Fall Meeting Abstracts, vol. 2024, Dec. 2024, p. IN31Dâ2039. [22] F. Sch Ì ottl, A. Spichtinger, J. Franquinet, and M. Langer, âReal- time on-orbit fire detection on forest-2,â in IGARSS 2024 - 2024 IEEE International Geoscience and Remote Sensing Symposium, 07 2024, p. 2360â2364. [23] G. Furano and A. Menicucci, Roadmap for On-Board Process- ing and Data Handling Systems in Space, 08 2018, p. 253â281. [24] G. Furano, G. Meoni, A. Dunne, D. Moloney, V. Ferlet-Cavrois, A. Tavoularis, J. Byrne, L. Buckley, M. Psarakis, K.-O. Voss, and L. Fanucci, âTowards the use of artificial intelligence on the edge in space systems: Challenges and opportunities,â IEEE Aerospace and Electronic Systems Magazine, vol. 35, 12 2020. [25] H. Li, H. Zheng, C. Han, H. Wang, and M. Miao, âOnboard spectral and spatial cloud detection for hyperspectral remote sensing images,â Remote Sensing, vol. 10, no. 1, p. 152, 2018. [26] X. Dai, X. Wu, B. Wang, and L. Zhang, âSemisupervised scene classification for remote sensing images: A method based on convolutional neural networks and ensemble learning,â IEEE Geoscience and Remote Sensing Letters, vol. 16, no. 6, p. 869â 873, 2019. [27] V. R Ì u Ë zi Ë cka, A. Vaughan, D. De Martini, J. Fulton, V. Salvatelli, C. Bridges, G. Mateo-Garcia, and V. Zantedeschi, âRavĂŠn: unsupervised change detection of extreme events using ml on- board satellites,â Scientific reports, vol. 12, no. 1, p. 16939, 2022. [28] G. Costa, T. Rumpf, E. Cadau, L. Sapia, V. Faccin, and A. Papa, âIride: The new italian space programme,â in IEEE Interna- tional Geoscience and Remote Sensing Symposium (IGARSS), 07 2024. [29] F. Conti, M. Eugeni, M. Bove, P. Marzioli, L. Marini, M. Pasquali, E. Schiavon, A. Nguyen Xuan, T. Antonella, S. Geraldini, F. Piergentili, A. Taramelli, and P. Gaudenzi, âA model-based approach for the preliminary design of the sar upstream element for the italian iride eo constellation based on usersâ demand,â CEAS Space Journal, vol. 17, 11 2024. [30] A. Taramelli, E. Schiavon, A. Tornato, E. Valentini, M. Righini, S. Geraldini, and L. Piedelobo, âOptimization of the Earth Observation Upstream Segment through an Analysis of User Requirements at National Level,â in AGU Fall Meeting Ab- stracts, ser. AGU Fall Meeting Abstracts, vol. 2020, Dec. 2020, p. SY009â03. [31] A. Taramelli, A. Tornato, M. L. Magliozzi, S. Mariani, E. Valentini, M. Zavagli, M. Costantini, J. Nieke, J. Adams, and M. Rast, âAn interaction methodology to collect and assess user-driven requirements to define potential opportunities of future hyperspectral imaging sentinel mission,â Remote Sensing, vol. 12, no. 8, 2020. [Online]. Available: https://w.mdpi.com/2072-4292/12/8/1286 [32] C. Molteni, D. Greco, A. Vitiello, G. Gutierrez, D. Riccobono, and F. Miglioretti, âHAWK for Earth observation: a novel Earth observation microsatellite constellation for the IRIDE program,â in Small Satellites Systems and Services Symposium (4S 2024), M. Petrozzi-Ilstad, Ed., vol. 13546, International Society for Optics and Photonics.SPIE, 2025, p. 135460Q. [Online]. Available: https://doi.org/10.1117/12.3061315 [33] E. Filippi and A. Aiello, âReasons, objectives and potential impact of the italian nrrp investment for the IRIDE earth observation programme,â LâIndustria â Rivista di economia e politica industriale, no. 1, p. 113â158, 2024. [34] E. Schiavon, A. Taramelli, A. Tornato, and F. Pierangeli, âMoni- toring environmental and climate goals for european agriculture: user perspectives on the optimization of the copernicus evolu- tion offer,â Journal of Environmental Management, vol. 296, p. 113121, 2021. [35] E. Schiavon et al., âMaximizing societal benefit across mul- tiple hyperspectral earth observation missions: A user needs approach,â Journal of Geophysical Research: Biogeosciences, vol. 128, no. 12, 2023. [36] European Space Agency (ESA), âIRIDE â A National Earth Observation Constellation for Italy,â https://w.esa.int/ Applications/Observing theEarth/IRIDE, accessed: 2025-12- 10. [37] IRIDE Spazio, âIRIDE: official programme brochure,â https: //iridespazio.it/, accessed: 2025-12-10. [38] A. Taramelli, âMirror copernicus per lo sviluppo e la competitivit ` a della space economy italiana,â ENEA Magazine / EAI, 2021. [Online]. Available: https://w.eai.enea.it/?option= com content&view=article&id=1338&Itemid=2066 [39] R. Del Prete, P. K. Thind, A. Mazzeo, M. Whitley, L. Papa, N. Long Ì ep Ì e, and G. Meoni, âOptimizing deep learning models for on-orbit deployment through neural architecture search,â Scientific Reports, vol. 15, no. 1, p. 37783, 2025. [Online]. Available: https://doi.org/10.1038/s41598-025-21467-8