Project Description

Abstract

This research activity investigates the potential of Quantum Machine Learning (QML) for Earth Observation (EO), with particular emphasis on representing and processing EO data using quantum computing. A central objective will be the systematic study of quantum data encoding and measurement strategies, assessing how different design choices affect the information content and interpretability of the resulting quantum representations. Building on this foundation, the research will explore hybrid quantum-classical learning architectures, including Quantum Neural Architecture Search (QNAS), to identify efficient, task-specific quantum circuit configurations while preserving model interpretability.

The activity will progressively move from simulation-based experiments toward hardware-aware analysis and experimental validation on real quantum computing platforms, investigating the effects of noise and current hardware limitations on QML performance. Particular attention will also be devoted to assessing the practical constraints associated with the potential deployment of quantum computing solutions in space-based EO scenarios. The overall research will contribute to the definition of reproducible methodologies and guidelines for quantum encoding and measurement in EO, supported by benchmark analyses and dissemination through high-quality peer-reviewed scientific publications.

Detailed description

Research Question

How can quantum computing theory, hardware-aware design, and Earth Observation data physics be jointly leveraged to achieve experimentally validated quantum utility for Earth Observation tasks?

Fig. 1 Overview of the project concept and research scope, connecting Earth Observation data with Quantum Machine Learning processing.

State of the Art

The development of Artificial Intelligence (AI) has profoundly impacted Earth Observation (EO) by enabling the extraction of meaningful information from the rapidly growing volume and diversity of EO data [1]. Despite these advances, the integration of big data and AI-based EO workflows remains challenging: large-scale Machine Learning (ML), and especially Deep Learning (DL), can be time-consuming and resource-intensive due to limitations in data management, processing pipelines, and computational resource utilization [2]. These constraints motivate exploring complementary computing paradigms that may improve scalability, robustness, or efficiency in selected EO learning settings.

Quantum Computing (QC) leverages quantum-mechanical principles to process information in fundamentally different ways than classical hardware [3]. Within this landscape, Quantum Machine Learning (QML) has emerged as a research field investigating how quantum information processing can support learning tasks, for instance, through quantum feature maps, kernels, and hybrid quantum–classical architectures [4].

Early theoretical and empirical results suggest that quantum models may provide expressive representations and favorable learning dynamics in specific regimes. However, performance is strongly dependent on circuit design choices, data encoding strategies, and hardware noise characteristics [5]. As a result, the question of whether QML can effectively support EO tasks is increasingly being explored.

Actually, most QML studies relevant to EO have focused on hybrid architectures, reflecting the constraints of current Noisy Intermediate-Scale Quantum (NISQ) devices in terms of qubit count and coherence  [6].

Fig. 2 Overview of current trends in Quantum Machine Learning for Earth Observation, highlighting the prevalence of hybrid approaches, simulation-based studies, and commonly adopted quantum computing frameworks.

A particularly promising family of approaches is quantum-enhanced convolution, including quanvolutional operations that use quantum circuits as feature extractors [7,8]. Recent EO-oriented contributions demonstrate the feasibility of adapting quanvolutional ideas to remote sensing problems (e.g., classification and downstream tasks), while also highlighting that model design is often heuristic and evaluation is frequently limited to simulation-based settings [6, 9-12]. These limitations are widely recognized as key barriers to moving from proof-of-concept demonstrations toward reproducible evidence of practical utility in EO.

QC is increasingly discussed in the broader AI landscape, including in relation to modern AI systems such as chatbots, reinforcing the need for rigorous, reproducible evaluations in application domains like EO.

In parallel to academic progress, institutional and programmatic efforts are emerging to assess QML4EO potential in a structured manner. For example, ESA has initiated dedicated activities to investigate quantum computing for EO, such as the QC4EO study within the EO4Society program, which frames opportunities, limitations, and research directions for quantum methods in EO workflows [13].

Such initiatives, together with growing international academic interest, signal the transition of QML4EO from an exploratory topic toward a more organized research area that increasingly requires shared benchmarks, reproducible evaluation, and clearer links to operational constraints.

Against this background, three complementary research directions are increasingly identified as necessary to advance QML4EO beyond simulation-driven experiments: (i) the development of physics-informed quantum algorithms that explicitly account for EO data characteristics and observation models; (ii) hardware-aware benchmarking protocols that quantify the impact of noise, uncertainty, and device variability, enabling reproducible comparisons across platforms; and (iii) modular hybrid architectures that integrate quantum building blocks with classical/HPC resources and, in the longer term, can be assessed under edge/on-board constraints [14]. Together, these directions define a pathway from purely simulated studies toward experimentally validated, hardware-level evidence for EO-relevant tasks.

Despite promising results, current QML4EO research still lacks (i) physics-grounded principles to map EO measurements into quantum states and to design measurement strategies, (ii) reproducible, hardware-aware benchmarking across backends that explicitly quantifies noise/uncertainty and resource/energy-related proxies, and (iii) modular design patterns that enable transfer across tasks and paradigms beyond classification. This project addresses these gaps through three contributions:

• Physics-aware encoding & measurement principles for EO, moving beyond ad-hoc encodings by preserving the physical meaning of EO observables when mapped to quantum states, building also on recent domain-informed representations for polarimetric SAR [15].
• Hardware-aware, reproducible cross-backend benchmarking, quantifying accuracy–robustness–uncertainty trade-offs on real quantum devices and simulators, using standardized tasks, metrics, and baselines, and reporting device variability and noise sensitivity comparably.
• Modular quantum building blocks for EO, including QNAS-ready quanvolutional components and encoding recipes that can be reused across EO tasks and extended to quantum generative paradigms, providing a structured basis for future feasibility assessments under edge/on-board constraints.

Tasks description - Research Plan

The fellowship will be structured around four main Work Packages (WPs), which are interconnected: WP1 provides the theoretical basis for WP2’s hardware experiments; WP2 produces benchmarking evidence that feeds into WP3’s feasibility analysis; WP4 ensures integration, dissemination, and coordination with ESA. An overview of the Research Plan, including objectives, main activities, deliverables, and timelines for each WP, is summarised in the attached table.

This research plan will be carried out with a strong focus on consolidating existing collaborations and establishing new ones. Furthermore, synergies with ESA initiatives will strengthen cross-domain innovation and accelerate the transition from simulation-based proof-of-concept studies toward experimentally validated evidence.

To enable transparent and community-driven comparison, the benchmark will be released via open repositories and accompanied by documentation and scripts that support repeatable evaluation across simulators and hardware backends.

The activity starts from simulation-validated QML components (baseline TRL ~2–3) and systematically progresses toward experimental verification on real quantum devices (target TRL ~3–4 for selected building blocks and benchmarks). The project’s validation strategy is explicitly hardware-aware, focusing on noise and stability characterization and on cross-backend reproducibility, ensuring that results are not only theoretically grounded but also experimentally credible.

All software artifacts and benchmarking scripts will be released through public repositories, together with a final report consolidating guidelines, hardware characterization, and feasibility outcomes.

References

[1] Tuia, D., Schindler, K., Demir, B., Zhu, X. X., Kochupillai, M., Džeroski, S., ... & Camps-Valls, G. (2024). Artificial Intelligence to Advance Earth Observation: A review of models, recent trends, and pathways forward. IEEE Geoscience and Remote Sensing Magazine.

[2] Wang, M., Fu, W., He, X., Hao, S., & Wu, X. (2020). A survey on large-scale machine learning. IEEE Transactions on Knowledge and Data Engineering, 34(6), 2574-2594.

[3] Nielsen, M. A., & Chuang, I. L. (2010). Quantum computation and quantum information. Cambridge University Press.

[4] Schuld, M., & Killoran, N. (2019). Quantum machine learning in feature Hilbert spaces. Physical Review Letters, 122(4), 040504.

[5] Abbas, A., Sutter, D., Zoufal, C., Lucchi, A., Figalli, A., & Woerner, S. (2021). The power of quantum neural networks. Nature Computational Science, 1(6), 403-409.

[6] Sebastianelli, A., Mauro, F., Delilbasic, A., Di Stasio, P., Fan, F., Meoni, G., ... & Ullo, S. (2025). Quantum Machine Learning for Earth Observation: A Review and Future Prospects. Authorea Preprints.

[7] Henderson, M., Shakya, S., Pradhan, S., & Cook, T. (2020). Quanvolutional neural networks: powering image recognition with quantum circuits. Quantum Machine Intelligence, 2(1), 2.

[8]  Fan, F., Shi, Y., Guggemos, T., & Zhu, X. X. (2023). Hybrid quantum-classical convolutional neural network model for image classification. IEEE transactions on neural networks and learning systems.

[9] Sebastianelli, A., Mauro, F., Ciabatti, G., Spiller, D., Le Saux, B., Gamba, P., & Ullo, S. (2025). Quanv4eo: empowering earth observation by means of quanvolutional neural networks. IEEE Transactions on Geoscience and Remote Sensing.

[10] Russo, L., Mauro, F., Memar, B., Sebastianelli, A., Ullo, S. L., & Gamba, P. (2025, May). A Quantum-assisted Attention U-Net for Building Segmentation over Tunis using Sentinel-1 Data. In 2025 Joint Urban Remote Sensing Event (JURSE) (pp. 1-4). IEEE.

[11] Mauro, F., Sebastianelli, A., Del Rosso, M. P., Gamba, P., & Ullo, S. L. (2024, July). QSpeckleFilter: a quantum machine learning approach for SAR speckle filtering. In IGARSS 2024-2024 IEEE International Geoscience and Remote Sensing Symposium (pp. 450-454). IEEE.

[12] Mauro, F., Razzano, F., Di Stasio, P., Sebastianelli, A., Meoni, G., Schirinzi, G., ... & Ullo, S. L. (2025). Quantum-Enhanced Water Quality Monitoring: Exploiting ΦSat-2 Data with Quanvolution. IEEE Geoscience and Remote Sensing Letters.

[13] European Space Agency (ESA). QC4EO Study – Quantum Computing for Earth Observation (EO4Society). https://eo4society.esa.int/projects/qc4eo-study/.

[14] Ralser, B., & Herzog, S. (2025, June 11). Quantum computer from Vienna for outer space. Rudolphina—University of Vienna. https://rudolphina.univie.ac.at/en/quantum-computer-from-vienna-for-outer-space.

[15] Bhattacharya, A., & Verma, A. (2025). Bloch Sphere Representation of Polarimetric SAR Targets. IEEE Geoscience and Remote Sensing Letters.

Project Outcomes

Publications

F. Mauro, L. Russo, A. Miroszewski, A. Bhattacharya, P. Gamba, and S. Ullo, “Co-Design of Quantum Encoding and Measurement for PolSAR Land-Cover Classification,” IEEE Transactions on Geoscience and Remote Sensing, manuscript submitted for publication, 2026.