Organizing temporally vague Raster Data in Cloud Environments for machine-learning Applications

Abstract. Although geospatial time series derived from historical remote sensing and topographic maps provide critical insights into land-cover evolution, their heterogeneous structures and temporal vagueness complicate interoperability for machine-learning applications. The performance of geospatial access in cloud environments depends particularly on the formats and services used. Furthermore, raster datasets are characterized by the large amount of data, which requires efficient access for reading and writing. This paper outlines common approaches to organizing spatio-temporal raster data for use in cloud environments. It also proposes a concept for modelling temporal vagueness based on object storage. This concept provides the basis for interfacing with traditional ISO 8601 requirements.