Both netCDF variables and attributes are named, typed containers for values. This leads to the question, when is it appropriate to use a netCDF attribute instead of a netCDF variable to hold a value?

Characteristics of variables:

• intended for data
• can hold arrays too large for memory
• may be multidimensional
• support partial access (only a subset of values)
• values may be changed, more data may be appended
• may have attributes
• shape specified with netCDF dimensions
• variables are read when user asks for data

Characteristics of attributes:

• intended for metadata, information about entire variable or dataset
• used for small units of information that fit in memory
• for single values, strings, or small 1-D arrays
• atomic access, must be written or read all at once
• values typically don't change after creation
• an attribute may not have attributes
• length specified when created
• attributes are read when file is opened

2010 Unidata NetCDF Workshop > The Two NetCDF Data Models