DSS presents a global set of surface and upper air reports operationally collected by the National Centers for Environmental Prediction (NCEP). These include land surface, marine surface, radiosonde, pibal and aircraft reports from the Global Telecommunications System (GTS), profiler and US radar derived winds, SSM/I oceanic winds and TCW retrievals, and satellite wind data from the National Environmental Satellite, Data, and Information Service (NESDIS). The reports can include pressure, geopotential height, temperature, dewpoint depression, wind direction and speed. This data is the primary input to the Global Data Assimilation System (GDAS), which is used to make forecasts and the Final Analyses (FNL).

The "PREPBUFR" processing is the final step in preparing the majority of conventional observational data for assimilation into the the Global Forecast System and Global Data Assimilation System unified grid-point statistical interpolation analysis (GSI) (the "GFS" and "GDAS" networks). This step involves the execution of series of programs designed to assemble observations dumped from a number of on-line decoder databases, encode information about the observational error for each data type as well the background (first guess) interpolated to each data location, perform both rudimentary multi-platform quality control and more complex platform-specific quality control, and store the output in a monolithic BUFR file, known as PREPBUFR. The background guess information is used by certain quality control programs while the observation error is used by the analysis to weigh the observations. The structure of the BUFR file is such that each PREPBUFR processing step which changes a datum (either the observation itself, or its quality marker) records the change as an "event" with a program code and a reason code. Each time an event is stored, the previous events for the datum are "pushed down" in the stack. In this way, the PREPBUFR file contains a complete history of changes to the data throughout all of the PREPBUFR processing. The most recent changes are always at the top of the stack and are thus read first by any subsequent data decoder routine. It is expected that the data at the top of the stack are of the highest quality.