For International Financial Institutions, the challenge is no longer whether Earth Observation can add value, but how satellite data can be integrated into operational analysis in a way that is reliable, scalable and cost-effective. This question is at the centre of the Asian Development Bank’s 2025 Economics Working Paper Beyond Pretty Pictures: Combined Single- and Multi-Image Super-Resolution for Sentinel-2 Images, which demonstrates how advanced super-resolution techniques can significantly improve the performance of freely available satellite data for urban land-cover classification in Hanoi. Supported through the European Space Agency’s Global Development Assistance (GDA) Fast EO Co-Financing Facility (FFF), the work illustrates how EO can move beyond experimentation and into practical use within IFI workflows.
The Hanoi case provides a concrete example of this transition. Air pollution in the city presents a severe and growing health risk, exposing residents to fine particulate matter that penetrates deep into the lungs and cardiovascular system. This exposure contributes to diseases such as stroke, heart disease, lung cancer, chronic obstructive pulmonary disease, and respiratory infections. According to the World Health Organization (2024), major sources include industry, transportation, coal power plants, and household solid fuel use. In 2019, Hanoi was ranked the second most polluted city in Southeast Asia. As pollution levels continue to rise, they increasingly threaten both public health and economic productivity, underscoring the need for evidence-based mitigation strategies.
Understanding how people respond to air pollution information is a key part of this challenge. To address this, the Asian Development Bank (ADB) is conducting a Randomized Control Trial (RCT) to assess how air pollution forecasts influence people’s movement patterns and their exposure to risk factors associated with non-communicable diseases. Within this context, Earth Observation data were used to map Hanoi’s green areas in detail, providing a consistent spatial reference to analyse how participants adjust their daily activities when informed about pollution risks.
This work was supported by ESA’s Global Development Assistance Fast EO Co-Financing Facility. The FFF aims to enhance the impact of IFI-led projects by embedding EO into operational workflows across sectors and along project lifecycles. In Hanoi, EO was not used as a standalone technical exercise, but as an analytical input aligned with an ongoing ADB public health study, demonstrating how satellite-derived information can directly support development finance activities.
Over a six-month period, the ESA GDA-supported project delivered a key service: the development of a Land Cover and Land Use (LCLU) classification algorithm tailored to Hanoi’s dense and heterogeneous urban environment.
Enhancing land cover and land use
High-resolution satellite imagery from 2022 and 2023 was used to create a detailed LCLU map of the city. A deep learning algorithm, adapted from the U-Net architecture, was applied to identify urban patterns and land-use features. Imagery was adjusted in resolution to meet the analytical requirements of the study. Additional datasets, including building footprints from Google Open Buildings and global land cover information from ESA WorldCover, were used to support reference data generation and improve classification accuracy.
Method, results and impact
Using Pléiades imagery adjusted to a 2.5-metre resolution, the LCLU classification model achieved an overall accuracy of 86 percent, based on both pixel-based and point-based validation. The outputs show consistent, high-quality segmentation results (Figure 2), with finer spatial detail illustrated in Figure 3. These results reflect the effectiveness of the workflow, where prediction aggregation, post-processing, and ensemble modelling together deliver smooth and reliable classifications suitable for urban analysis.
In parallel, the ADB team enhanced the workflow by applying super-resolution techniques to Sentinel-2 imagery. Using deep learning, Sentinel-2’s native 10-metre resolution was increased to 2.5 metres. The LCLU classification model was then adapted to process these super-resolved Sentinel-2 images, allowing for a direct comparison between data sources. The results demonstrate a clear improvement: classification accuracy increased from 44 percent using original Sentinel-2 imagery to 75 percent with super-resolvedWorkshops conducted in November 2025 with key stakeholders, including Vietnam’s National Statistics Office, focussed on raising awareness of geospatial technologies and the use of EO-derived information.
Sentinel-2 data, approaching the performance achieved with commercial high-resolution Pléiades imagery.
These findings directly reflect the conclusions of the Beyond Pretty Pictures working paper, which shows that combining single-image and multi-image super-resolution can significantly narrow the performance gap between freely available Sentinel-2 data and costly high-resolution imagery. Supported through the GDA FFF, this work highlights how super-resolution can enhance the analytical value of open satellite data and make its use more viable at scale for IFI operations.
Beyond technical performance, the project strengthened ADB’s operational efficiency by supporting internal capacity, promoting advanced EO-based methods, and demonstrating workflows that can be reused in other urban contexts. The results underscore the value of EO-derived data for sustainable urban planning and show clear potential for replication in other Asian cities.
Stakeholder engagement
Alongside technical delivery, the collaboration placed strong emphasis on capacity building. These sessions supported uptake by enabling Vietnamese institutions to apply advanced EO solutions within their own analytical processes. By linking technical development with local capacity building, the ADB project supports sustainable and scalable use of EO in Hanoi and beyond.
Conclusion and next steps
The project establishes a strong foundation for deeper integration of Earth Observation into air pollution management and public health analysis in Hanoi. At the same time, it demonstrates how EO, when combined with super-resolution techniques, can be embedded into ADB workflows to support development finance objectives. The Hanoi case illustrates the role of the GDA Fast EO Co-Financing Facility in enabling this transition, showing how advanced EO methods can move from research into practical use within IFIs.
