A Comprehensive Survey on High-Performance and Efficient Computing for Earth Observation
Remote sensing (RS) for Earth observation (EO) is undergoing an unprecedented data explosion driven by high-resolution satellite missions, new trends such as the development of digital twins (DTs), open-access policies, and cloud-native data ecosystems. At the same time, the rapid evolution of deep learning (DL) and geospatial foundation models (GFMs), a class of large-scale DL models, is transforming the way EO data are represented, transferred, and exploited across tasks and domains. These developments impose demanding computational requirements across the full EO processing workflow, from data acquisition and transmission to storage, analysis, model training, and deployment, all of which increasingly rely on high-performance computing (HPC) capabilities. This survey reviews the intersection of HPC and EO, providing a structured overview of computing paradigms, infrastructures, and hardware accelerators that support modern EO workflows. Rather than proposing a prescriptive rule for selecting a specific computational solution, the survey relates the main stages of a typical EO workflow to the infrastructures commonly used to support them, including edge computing, ground-segment computing, cloud computing, and supercomputing environments. This workflow-oriented perspective highlights how the suitability of each infrastructure depends not only on computational performance but also on data locality, latency, scalability, interoperability, energy constraints, economic cost, and deployment requirements. The survey also analyzes the role of hardware accelerators, including graphics processing units (GPUs), field-programmable gate arrays (FPGAs), tensor processing units (TPUs), and other artificial intelligence (AI) accelerators, as well as emerging paradigms such as quantum and neuromorphic computing. Special attention is devoted to distributed training frameworks, model-sharding libraries, cloud-optimized data formats, workflow orchestration tools, and hardware-aware optimization strategies needed to support large-scale EO processing and GFM development. Finally, the survey identifies open challenges and future research directions toward scalable, sustainable, interoperable, and reproducible EO processing at the convergence of RS, DL, and HPC.
keywords: High-Performance computing, Earth observation (EO), supercomputing, cloud computing, quantum computing, Edge computing
Publication: Article
1789027387349
September 10, 2026
/research/publications/a-comprehensive-survey-on-high-performance-and-efficient-computing-for-earth-observation
Remote sensing (RS) for Earth observation (EO) is undergoing an unprecedented data explosion driven by high-resolution satellite missions, new trends such as the development of digital twins (DTs), open-access policies, and cloud-native data ecosystems. At the same time, the rapid evolution of deep learning (DL) and geospatial foundation models (GFMs), a class of large-scale DL models, is transforming the way EO data are represented, transferred, and exploited across tasks and domains. These developments impose demanding computational requirements across the full EO processing workflow, from data acquisition and transmission to storage, analysis, model training, and deployment, all of which increasingly rely on high-performance computing (HPC) capabilities. This survey reviews the intersection of HPC and EO, providing a structured overview of computing paradigms, infrastructures, and hardware accelerators that support modern EO workflows. Rather than proposing a prescriptive rule for selecting a specific computational solution, the survey relates the main stages of a typical EO workflow to the infrastructures commonly used to support them, including edge computing, ground-segment computing, cloud computing, and supercomputing environments. This workflow-oriented perspective highlights how the suitability of each infrastructure depends not only on computational performance but also on data locality, latency, scalability, interoperability, energy constraints, economic cost, and deployment requirements. The survey also analyzes the role of hardware accelerators, including graphics processing units (GPUs), field-programmable gate arrays (FPGAs), tensor processing units (TPUs), and other artificial intelligence (AI) accelerators, as well as emerging paradigms such as quantum and neuromorphic computing. Special attention is devoted to distributed training frameworks, model-sharding libraries, cloud-optimized data formats, workflow orchestration tools, and hardware-aware optimization strategies needed to support large-scale EO processing and GFM development. Finally, the survey identifies open challenges and future research directions toward scalable, sustainable, interoperable, and reproducible EO processing at the convergence of RS, DL, and HPC. - Dora B. Heras, Rocco Sedona, Álvaro Ordóñez, Gabriele Cavallaro, Sebastián López, Manil Maskey, Gabriele Meoni, Artur Miroszewski, Catherine D. Schuman, Jón Atli Benediktsson - 10.1109/JPROC.2026.3721381
publications_en