Multi-Dimensional Arrays & Slicing Concepts

NetCDF variables represent multidimensional arrays. Each dimension has a name and integer length. A variable's shape is the ordered sequence of dimension names that define its indexing space (e.g. time height1 y x).

The Coordinate Model

In climate and weather datasets following the Climate and Forecast (CF) metadata conventions:

• Coordinate Variables: 1D variables sharing the same name as a dimension (e.g. time(time), lat(lat), lon(lon)) provide the physical coordinates for each index.
• Auxiliary Coordinate Variables: 2D or 3D variables (e.g. curvilinear grids with lat(y, x) and lon(y, x)) specify coordinates for irregular or projected grids.
• Data Variables: N-dimensional arrays storing physical quantities (e.g. air_temperature(time, height, y, x)).

2D Display Axes vs. Extra Dimensions

Standard computer displays render 2D raster heatmaps. When a variable has 3, 4, or more dimensions:

• Display Axes: Two chosen dimensions (e.g. x and y) form the horizontal and vertical axes of the visual image.
• Extra Dimensions (Complementary Dimensions): The remaining dimensions (e.g. time and height1) require an active slice index (e.g. time=0, height1=2).
• Hyper-Slicing: The underlying UCAR data adapter reads a 2D matrix slice data[t, h, :, :] without loading the entire 4D array into memory.

Data Type Abstraction

Capella Datascience provides the generic DataAdapter<T> interface with specialized implementations:

• FloatAdapter, DoubleAdapter: For floating-point continuous fields.
• IntegerAdapter, ShortAdapter, ByteAdapter: For integer/categorized fields.

The adapter normalizes missing data / NaNs, computes dynamic min/max ranges, and applies linear ratio interpolation to map raw values into 8-bit or 16-bit color palettes.