Integrating Science Metadata into Python Using Visad

The initial design motivation of the VisAD system was to enable visualization of any data defined in a scientific program, without the need for users to custom-code display procedures for new data structures they invented as they wrote their programs. The solution was a data model with schemas that could express any data defined in common programming languages, combined with a display model based on mappings from primitive data types to primitive display types that could be used to derive displays of complex data types (Hibbard, Dyer and Paul, 1992; Hibbard, 1995). The VisAD data schemas were patterned after the two primary data-building features in common programming languages: tuples and arrays. Tuples are fixed length sequences of simpler data types, which may have a variety of different types (the obvious example is the C language struct). Arrays are variable length sequences of a simpler data type, where all array elements have the same type (while this is not true in all programming languages, it is true in commonly used scientific languages such as Fortran, C and C++). In order to adequately support the VisAD display model, the data model had to incorporate certain other information absent in programminglanguage data structure declarations. For one thing, VisAD primitive types have names such as time, latitude, temperature and pressure rather than simply being declared as float or real. These primitive names are used to define mappings to primitive display type names (such as XAxis, RGB, Animation and IsoContour). It is a reasonable assumption that values in scientific programs had such names, but they just aren’t expressed because programming languages do not provide ways to express them. Primitive type names are one form of metadata integrated into the VisAD data model. Arrays in scientific programs generally represent finite samplings of continuous functions, but common programming languages provide no way to express this. For example, images and grids are stored in arrays, with one pixel or grid point per array element. These images and grids are really just finite samplings of continuous fields of radiances, temperatures, etc. The VisAD data model uses finite samplings of functions in place of arrays. Thus for example a simple earth image is represented as a finite sampling of a function from latitude and longitude to radiance. This is expressed by the VisAD schema: