Earth Observation (EO) satellites have the collective capacity to continuously monitor the entire globe. However, despite the availability of huge (often freely available) volumes of data, less than 5% is currently being analysed. Part of the problem is that this data often requires considerable preparation since it is obtained using numerous types of instruments operating over a range of spatial and temporal resolutions, while the offered products employ a selection of geographical projections. Such preparation is even more challenging considering the massive volumes involved. The required preprocessing and subsequent analysis may, therefore, require technical and scientific expertise not normally held by users for whom the data would be of greatest value.
The CENTURION project aimed to help remedy this by improving upon the analysis and management capabilities employed for terabyte to petabyte-scale volumes of data. This was to be achieved by combining two technologies: datacubes and artificial intelligence as a service, for scalable, flexible, and easy-to-use analytics and data fusion with ensured consistency.
Datacubes are built from large assets of homogenised Earth data leading to Analysis Ready Data (ARD). The employed datacube technology is rasdaman, which includes features such as a standards-based datacube query language, a highly optimized “green computing” engine, the integration of arbitrary external code, and location-transparent federation. The AI processing is provided with functional modular applications, collectively termed AI Knowledge Packs (AI KPs). The AI KPs would allow those with greater technical and scientific knowledge to augment and expand upon them by including their own analysis methods, leading to a “library” of pre-packaged exploitable modular AI applications, while still being accessible to non-experts.
The expected benefits would include a set of readily exploitable AI KPs covering a wide spectrum of real-life use cases. These use cases involved ARD generation, agricultural weather indices, the characterization of maritime shipping dynamics, near-real-time monitoring of forest dynamics, and near-real time flood zone mapping. The platform would be exploitable by current EOD users, while attracting new users who may not have previously employed such data because of a perceived or actual lack of expertise.
