Fig. 1.

Major components of the generic architecture and the basic flow of data between them. (1,4) Inputs: Data are freely available from several sensors, user provides request details such as types and number of classes, the area of interest, temporal information, and spatial resolution. (2,3) Preprocessing: Data are processed into a consistent, normalized, pixel-based format (ARD – CARD4L) so all pixels are comparable; these can be stored in a “data cube” environment to facilitate efficient access to time series and eliminate the data processing burden. The (5a) product generation then requires (5b) reference data for training and validation and is supplied either by the user or a shared reference database. (5c) A variety of algorithms can be made available from an open source “algorithm warehouse”. (6) Validation: The product is assessed for accuracy by comparing with independent validation reference data. (7) Products: the approach can support a wide variety of other products by using different algorithms.