Designing the Future of Scientific Computing: FAIR, Efficient, and Sustainable

Designing the Future of Scientific Computing: FAIR, Efficient, and Sustainable Assoc. Prof. Dr. Ing. Serkan Girgin MSc1,2 [email protected] https://linkedin.com/in/serkan-girgin/ 1Center of Expertise on Big Geodata Science 2Department of Geo-information Processing Faculty of Geo-information Science and Earth Observation (ITC) 21-22 October 2025, Bali, Indonesia 10th International Conference on Sustainable Information Engineering and Technology Digital transformation is a double-edged sword •Information technologies now underpin almost every domain of research and innovation, from AI and big data analytics to cloud computing and IoT. •While they accelerate discovery and efficiency, they also bring significant environmental, social, and ethical impacts. •The production, operation, and disposal of IT infrastructure consume vast energy and resources. Data centers, HPC, and AI have large carbon footprints. •Achieving sustainability depends on collaboration across disciplines and borders. IT can foster this, but only if designed with openness and accessibility in mind. •Digital technologies shape behavior, decision-making, and access to knowledge. Unsustainable or biased design can reinforce inequalities. Illustration by Storyset.com Environmental Consciousness Achieving sustainability in our current digital age requires rethinking how we design, develop, and use information technologies, as they both enable and challenge sustainable development. Illustration by Storyset.com Ethical Responsibility Social Inclusivity Illustration by Storyset.com FAIR Principles Energy Efficiency •FAIR principles ensure that data and software can be discovered, shared, and reused effectively across disciplines. •They form the foundation for open, reproducible and transparent science. •Energy efficiency focuses on reducing the energy, while optimizing time, and computational resources required to achieve scientific results. •It is critical to minimize the environmental impact of large-scale computation. •Combining FAIR principles with energy-efficient practices creates a research ecosystem that is both responsible and resilient for long-term scientific progress. Illustration by Storyset.com https://doi.org/10.1038/sdata.2016.18 FAIR principles focus on four key aspects Source: https://fosteropenscience.eu/learning/assessing-the-fairness-of-data/ FAIR principles focus on four key aspects •Findable - Resources should be easy to locate with rich metadata and persistent identifiers. Findability ensures other researchers can discover resources efficiently. •Accessible - Data and software should be retrievable using open protocols while respecting access rights. Accessibility reduces duplication of effort and facilitates collaboration. •Interoperable – Resources should be presented in standardized formats and described with formal and broadly applicable vocabularies to ensure datasets and tools to work together. Interoperability lowers barriers to combine data and tools. •Reusable – Resources should be well-documented and licensed to ensure they can be reliably reused in other studies. Reusability extends the value of scientific outputs and enhances reproducibility. FAIR principles are fundamental for Open Science Source: https://fredvbrug.github.io/images/OA.jpg Research data publishing bad practices further limit effective data access and interoperability. As a result, spatial research data often becomes "invisible" significantly reducing its findability and accessibility. Illustration by Storyset.com OpenSTAC aims to create an open STAC catalog of public research datasets published at major research data repositories. The project "OpenSTAC: an open spatiotemporal catalog to make geospatial research data findable and accessible" with file number OSF23.2.111 of the research programme NWO Open Science Fund 2023 is financed by the Dutch Research Council (NWO) Illustration by Storyset.com •Research data repositories are regularly monitored for newly published datasets. •Geospatial datasets are detected by checking for common geospatial file formats. •For each identified dataset, spatiotemporal metadata is extracted from the relevant files and combined with the dataset-level metadata from the repository. •This information is used to create corresponding STAC items and collections, which are published alongside cloud-native version of geodata in an open-access STAC catalog. Illustration by Storyset.com https://doi.org/10.1038/s41597-025-05309-w https://opendatastac.org/10.1038/s41597-025-05309-w OpenSTAC enables researchers easily find and access geospatial data using open-source STAC tools, including visual browsers, CLI tools, and client libraries. Illustration by Storyset.com Several challenges still hinder widespread FAIR adoption •Researchers are often rewarded for new publications, not for making data or software reusable. •Inconsistent metadata and documentation practices make FAIR adoption challenging. •Without rich, structured metadata, even open resources are difficult to find or reuse. •Researchers need guidance and support to implement FAIR principles effectively. •A community-driven approach, backed by institutions and supportive policies, is essential to bridge the gaps and strengthen capacity building. Illustration by Storyset.com OSCs are hubs for learning Open Science practices from peers https://osc-international.com An active, self-motivated community is the first step toward success https://osc-international.com/start-your-own-osc/ Illustration by Storyset.com FAIR Principles Energy Efficiency How familiar are you with your scientific computing workflows? Illustration by Storyset.com A shift in energy-aware computing culture starts with monitoring •Measuring energy-related metrics is essential for researchers to understand the energy footprint of their workflows. •Only through this practice can researchers identify inefficiencies and improve their workflows. •Embedding energy monitoring into daily research practices will ensure sustainable habits become routine. •We need tools that provide accessible methods for tracking energy use and carbon footprints of computational tasks. Illustration by Storyset.com Gaps in energy monitoring for Earth Observation Big Data (EOBD) Data source variation •Various sensors with different sampling rates. •No time synchronization. Cloud abstraction •Lack of transparency in virtualized environments. •No detailed energy cost of each running service. Hidden costs •Energy from cooling, PSUs, and networking is often hidden. Task Attribution •Challenges in attributing energy to individual tasks. •Need for hierarchical attribution in virtualized environment Standard benchmark •Lack of standardized frameworks to compare energy efficiency of various EO tasks Icons by flaticon.com Energy Efficiency in Cloud-Based Big Data Processing for Earth Observation: Gap Analysis and Future Directions A. Bhawiyuga, S. Girgin, R. A. de By, R. Zurita-Milla (2025) https://doi.org/10.48550/arXiv.2510.02882 A cloud-native monitoring framework is being developed for EOBD Adhitya Bhawiyuga PhD Candidate Department of Geoinformation Processing [email protected] https://linkedin.com/in/bhawiyuga/ For more information: Diverse computing infrastructures facilitate energy efficiency •Not all computing infrastructures are equally energy-efficient for a given workload. •Low-power CPUs, ARM-based systems, or specialized accelerators may run slower yet consume far less energy overall. •One can shift workloads in time or space to minimize energy footprint, e.g. use low-power systems during peak grid load periods. •Exposure to diverse computing platforms also enables development of energy-aware algorithms and software. •Having access to a heterogeneous computing ecosystem with machines having different energy characteristics allows researchers to choose the best balance between energy use and performance for their specific tasks. Illustration by Storyset.com GCP enables access to energy-efficient edge units for geocomputing https://platform.crib.utwente.nl NVIDIA Jetson AGX Xavier Cluster 8-core CPU NVIDIA Carmel ARMv8.2, 2.26GHz 512-core GPU Volta architecture with 64 Tensor Cores 32GB unified memory 256-bit LPDDR4x, 2133MHz, 137GB/s 32GB internal storage DL and CV accelerators Gigabit ethernet 5-30 W energy consumption Although general-purpose computing platforms offer great flexibility, achieving efficient optimization for arbitrary computations is difficult. Illustration by Storyset.com Google Earth Engine is a gamechanger for geocomputing https://earthengine.google.com openEO provides an open specification and API for geocomputing https://openeo.org Client 1 Client 2 Client 3 Sustainable science is not about doing less, it is about doing better •Computing platforms for research should be FAIR-aligned and energy-aware by integrating both data and software FAIRness and energy-efficient operation. •Reuse and openness should be rewarded by incentives for sharing and collaboration to encourage sustainable practices. •Life-cycle approach should be followed considering the long-term impact of computing workflows on energy, reproducibility, and research continuity. Illustration by Storyset.com