Fiorino95b Fiorino, M., 1995b: PCMDI validation database. Abstracts of the First International AMIP Scientific Conference, Monterey, California, 112.

Comprehensive validation of the AMIP integrations has been severely constrained by the availability of temporally consistent, global observational data during the AMIP decade (1979-1988). The advent of reanalysis at three centers in the U.S. and at ECMWF, however, promises to greatly improve the situation. Thus, reanalysis has become the primary focus of PCMDI validation data base development.

The original PCMDI observational data base has been completely overhauled and most of the major data sets have been updated with newer versions. These major data sets include: 1) the NMC Climate Diagnostic Data Base (CDDB); 2) the Oort upper-air analyses for the period 1979-89; 3) monthly mean ECWMF analysis (initialized and "TOGA"); 4) land surface hydrology from NASA GSFC; 5) NESDIS SSM/I atmospheric hydrology; 6) MSU daily and monthly oceanic precipitation; 7) the University of East Anglia, over-land surface air temperature and precipitation; 8) ERBE/SRB/NOAA surface and TOA radiative fluxes ; 9) ISCCP clouds; and 10) various COADS surface flux climatologies.

More importantly, PCMDI has obtained a comprehensive set, in some cases eight years, of monthly mean and higher frequency data from the three U.S. reanalysis projects. Preliminary intercomparison of the U.S. reanalyses shows good agreement in the analyzed variables (e.g., winds) with no obvious noise in the time series due to model/assimilation system changes, as found in the operational products. However, changes in the observing system (e.g., the introduction of SSM/I oceanic surface wind speeds in July of 1987) can cause disturbances in the analyzed fields, particularly for diagnostic quantities such as precipitation. Nonetheless, reanalysis will be the best source for validating the atmospheric general circulation in the AMIP integrations.

The PCMDI observational data distribution strategy is to derive "products" from the original data for a small number of common grids and to make these sets available on the World Wide Web or by anonymous ftp. The derived products are intended to free the user from dealing with format details (every original obs data set seems to be on a different grid and in a different format) and to the simplify model-obs and obs-obs intercomparison while maintaining fidelity to the original. Data which directly match the standard output from AMIP (I and II) will be produced first and will be referred to as the "AMIP Observational Data Set" (AODS).

Each set in AODS will consist of a compressed (using the freely available GNU utility "gzip") ASCII file containing: 1) basic documentation and references; 2) meta data; and 3) gridded fields in a packed integer format (with gzip, the compression performance is comparable to GRIB). Utilities for data access, decode and validation will be provided (FORTRAN and ANSI C) together with a transform utility (ANSI C) to convert the packed integer fields to a form accessible by two visualization and analysis systems-the Grid Analysis and Display System (GrADS) and PCMDI's VCS. However, the resulting binary format is sufficiently simple so that the data can be read by other packages such as IDL. A sample session for accessing, checking, transforming and displaying AODS data will be demonstrated at the meeting.