Preparing Array Analytics for the Data Tsunami - Geographical Information Systems: Trends and Technologies - page 6

have emerged, such as PostGIS Raster, SciQL, SciDB, and EXTASCID, these
still lack a rigorous semantics defi nition so that they have suffi ciently been
treated in the previous section.
As it turns out, AML, AQL and RAM can be mapped to Array Algebra ,
therefore this common representation eases the comparison by highlighting
the common functionality and differences between the four models
(Baumann and Holsten 2012).
Array Algebra
An algebraic approach for array modeling is being offered by the Array
Algebra (Baumann 1994; 1999). Originally, it has been developed for
studying image processing and computer graphics such as (Carson and
McGinnis 1986; ISO 1994; Felger et al. 1990). The Image Algebra of the Air
Force Armament Laboratory (AFATL) (Ritter et al. 1994) has proven to be
a valuable starting point for database-centric domain-independent array
processing. The traditional application domains of Array Algebra in the
past consisted of sensor, image, simulation, and statistics data services. The
latest emphasis is on large-scale Earth Science (Gutierrez and Baumann
2007) and Life Science (Roland et al. 2001) data.
Array Algebra is implemented by the rasdaman Array Database
management system and its query language rasql (Baumann 2013). The Web
Coverage Processing Service (WCPS), (Baumann 2008) geo raster query
language standard issued by the Open Geospatial Consortium (OGC) in
2008 is based on the concepts introduced by Array Algebra.
The core model specifies an array to be a mapping from integer
coordinates—the spatial domain —to some array cell data type. The model
consists of three core operators: array constructor, condense operator, and
array sorter . The condense operator reduces an array to a scalar value
by using some aggregation function. The array sorter proceeds along a
given dimension by reordering the corresponding hyperslices. All other
operations, such as common statistics and image processing operators like
matrix multiplication, image convolution, and Fast Fourier Transform, can
be expressed as combinations of these operations.
AML
The Array Manipulation Language ( AML ) is another algebra-based, high-level
language which allows querying data and defi ning new arrays in terms
of existing ones (Marathe and Salem 1997; 1999). The targeted application
domain of the model was image databases especially for remote sensing,
although the authors point out that their model allows customization to
multiple application domains.