The (data)cube model for analysis of time series of earth observation raster data, represents the dataset as a multidimensional array with one or more spatial or temporal dimensions. Scalar values in the cube can be selected (or ‘filtered’) and processed based on dimension labels. This allows analysis algorithms to be thought of as a set of operations on the multidimensional array. Technologies that support this model allow to efficiently implement such algorithms.
Some possible operations on a multidimensional cube include: filtering, ‘reducing’ all values along a dimension, ‘aggregating’ values in a dimension, or transforming all values along a dimension. Generally speaking, these operations require the selection of a subset of the data on which work is to be done. This allows implementing the operations efficiently even on very large datasets.
In comparison to file-based processing, most technologies that support cube-based time series analysis reduce implementation overhead, as the user does not need to read and write individual files, also more complex aspects like distributed computing for parallelization can be hidden in a cube based approach. So a cube based approach can also be thought of as an abstraction layer that effectively reduces the need for specific IT-related skills when analyzing earth observation timeseries.
Multiple initiatives support cube based analysis. Some common features include a programming API, often using the Python programming language. Some tools are only accessible as web services, while others can also run locally (on a small dataset). This diversity is still a drawback, as users would need to familiarize themselves with different systems. Initiatives such as openEO try to address this by providing a common API.