Data Model

Bufrpy data model matches that of the BUFR messages fairly closely, though not as closely as libbufr or the Python libbufr wrappers. Typically BUFR messages are stored in a file so that the file contains a single BUFR message. Some applications (e.g. SAFNWC HWR) store multiple messages in a single file, requiring some tricks to read them.

The top-level data structure in bufrpy is Message that corresponds to one BUFR message. A BUFR message contains 6 sections, some of which contain metadata and some the actual data. The most important sections are numbers 3 and 4. They contain data descriptors and the data itself respectively. Logically BUFR files contain at least one subset of data. Each subset follows the same structure, defined by the data descriptors in section 3. In bufrpy, sections are attributes of Message. Section 4 contains the data subsets as a list of BufrSubset objects.

Logically each subset is a linear sequence of data values. All metadata about the data is stored in the descriptors. However, there are certain looping constructs (replication descriptors) that allow repetition of a sequence of data items. This leads to hierarchical logical structure. Physically the data is still stored in a linear sequence (or in interleaved fashion if a message has multiple data subsets). In bufrpy the data of each dataset is represented by a list of BufrValue objects. A replicated sequence in the data is represented by a list of lists of BufrValue objects so that the outer list contains one list for each repetition, and the inner lists are equivalent to values of a BufrSubset. Hierarchical data is thus naturally represented as nested lists.