| Project Detail |
Project Name
Super-resolution Satellite Imagery as a Regional Public Good for Sustainable Development
Project Number
58307-001
Country / Economy
Regional
Bangladesh
Philippines
China, Peoples Republic of
Thailand
Project Status
Active
TA will generate knowledge to: (i) develop a regional public good comprising an open-source method of deriving high-resolution (super-resolved) satellite imagery, and (ii) demonstrate how these images can be used to assess policy-relevant use-cases such as monitoring the green transition over time or assessing the quality of infrastructure. The TA therefore aims to not just provide policy-relevant insights, but to also provide DMCs with the tools to conduct such analyses themselves. The datasets and algorithms together with capacity building activities can be provided to identify DMCs to equip them in developing data-driven climate policies. This TA is well aligned with Strategy 2030, particularly strengthening institutional capacity in DMCs (OP6) in big data capabilities.
Project Rationale and Linkage to Country/Regional Strategy
Satellite imagery has changed the way we observe and understand economies. It offers invaluable data for applications ranging from environmental monitoring to urban planning. However, the utility of satellite images is significantly influenced by their resolution. High-resolution satellite imagery provides detailed information that can support critical decision-making processes for researchers, governments, and policymakers. Despite its importance, the cost associated with procuring high-resolution images remains prohibitively high for many, creating a barrier to access and limiting the scope of policy design and implementation. For one day of data, high resolution satellite data (i.e. 1 meter resolution) can cost about $4 per square kilometer. This means that for Asia, just one day of data would cost approximately $176 million. This contrasts with lower-resolution images like those from Sentinel-2, which are available free but offer less detail at a 10-meter resolution.
For many policy-relevant use-cases, the need for time-series data makes the high cost of high-resolution imagery a significant barrier (Ayush et al., 2021). This has hindered policy development and research in the past. For example, a World Bank study (Engstrom et al., 2017) found that due to the prohibitive cost of purchasing high-resolution data, only 5% of its country of interest could be analyzed using high-resolution data; while other studies have found the high cost of high-resolution images a barrier in the field of conservation biology (Boyle et al., 2014), vegetation monitoring (Guo et al., 2022), and road infrastructure monitoring (Hu et al., 2021).
An emerging, cost-effective approach to obtaining high-quality satellite images is through super-resolution models. Super-resolution, as applied to remote sensing, involves taking free lower-resolution satellite images (e.g. from Sentinel-2), and applying artificial intelligence (AI) models to increase their resolution. There are several approaches to developing super-resolution models for satellite imagery. These include convolutional neural networks (Mueller et al., 2020); generative adversarial networks (Nguyen et al., 2023), diffusion models (Kowaleczko, et al., 2023), and pixel synthesis models (He et al., 2021). The TA will therefore develop several models and select the best performing approach. This will involve comprehensively assessing the performance across a range of geographies, urban/rural areas, and time periods. The model performance will be assessed using metrics such as the Peak Signal-to-Noise Ratio (PSNR), and the Structural Similarity Index (SSI).
Super-resolved satellite images unlock several use-cases. Therefore, the final part of the TA involves applying super-resolved satellite images to a specific use-case. Such use cases may include the following, subject to data availability and quality of the trained model: (i) monitoring the transition to sustainable energy by measuring solar panel take-up; (ii) updating cadasters; (iii) monitoring the transition to sustainable energy by measuring wind turbine growth; (iv) monitoring infrastructure quality. As such, the super-resolved images will be incorporated into examining a specific policy use-case, which will be determined upon consultation with relevant stakeholders (e.g. other ADB areas, ADB Resident Missions and/or Government stakeholders).
For the use-case applications of super-resolved imagery, possible DMCs include Bangladesh, China P.R., Philippines, and Thailand, contingent on consultations with regional departments, government agencies data availability and model quality. Relevant ADB teams will be consulted as part of a collaborative approach to finalize the DMCs. This approach ensures that the SSTA is closely aligned with the operational priorities of ADB as outlined in the corporate results framework.
Impact
The TA is Aligned with promoting big data and technology advancements in DMCs to improve the availability of high-quality, timely, and reliable data in support of data-driven decision making; and strengthened governance and institutional capacity, particularly in the use of big data, in selected DMCs. |
