Intercomparison of the NCEP/NCAR and the NASA/DAO
In response to the climate community's need for long, consistent atmospheric datasets, the NCEP and NCAR are cooperating in a project to produce a 40-year (1957-1995+) record of global atmospheric analyses. Thorough documentation of this "reanalysis" dataset and the assimilation system used to produce it is found in Kalnay et al. (1996). The reanalysis provides a valuable product to the research community because it is produced with a state-of-the-art fixed, global assimilation system and an unparalleled input database. This is a research quality dataset suitable for study of a wide variety of Earth science problems such as climate variability, atmospheric chemistry and surface processes. Moreover, a continuation into the future through an identical NCEP/Climate Data Assimilation System (CDAS) will allow researchers to compare current anomalies with those during the earlier reanalysis years. All reanalysis products will be available to the research community.
The NCEP/NCAR reanalysis project is one of several efforts (see also Bengtsson and Shukla, 1988; Schubert et al. 1993) to re-analyze historical data with a frozen assimilation system. The primary motivation for these reanalyses is that operational data assimilation systems undergo frequent updates which introduce spurious signals that can easily be misinterpreted as climate anomalies. For example, Fig. 1a shows a longitude-time section of the 850-hPa zonal mean virtual temperature in a tropical band (10°N-10°S) from the NCEP Climate Diagnostics Data Base (CDDB) (the data source is the NCEP Global Data Assimilation System, or GDAS, which is the operational analysis system at NCEP; see section 3). Because the GDAS is part of the NCEP operational forecast suite, it has undergone several major changes in recent years (i.e. horizontal and vertical resolution, physics, topography, etc.), and will continue to change in the pursuit of better weather forecasts. Unfortunately, changes to the assimilation system may (and often do) create discontinuities in time histories of model-sensitive quantities thus introducing spurious climate change signals. Unfortunately, quantities linked to the hydrologic cycle (e.g. precipitation, divergence) are especially sensitive to these changes. These signals also become quite evident at seasonal time scales and most evident at interannual and longer time scales, such that the signature of even the dominant climate events (e.g. ENSO) can be masked. Because the NCEP/NCAR reanalysis is produced with a fixed assimilation system, such artificial changes in the global record are removed (Fig. 1b), which makes detection of these signals much more likely. However, since observational systems change over time, some discontinuities are likely unavoidable.
The NCEP/NCAR reanalysis can also be viewed as a benchmark for further development of the NCEP operational analysis. As Fig. 2 shows, correlations of the virtual temperature field between the CDDB and the NCEP/NCAR reanalysis have exceeded 95% in recent years throughout the Tropics (the reanalysis model is the T62/28-level NCEP global spectral model as implemented in the NCEP operational system in December 1994). We might expect that these correlations will begin to decrease in future years as improvements to the operational system are implemented. Thus, by closely tracking these correlations we have an objective measure of the time at which the reanalysis system should be replaced by the operational one for future reanalyses (i.e. Reanalysis 2). Of course there are many other measures, so it is important that a thorough evaluation of the quality and the usefulness of the data be undertaken. While an important part of that evaluation is carried out at NCEP, potential applications of this dataset are quite broad, so feedback from the Scientific community is vital.
The output archives from the NCEP/NCAR reanalysis provide four-times daily estimates of a huge variety of geophysical parameters at the full resolution of the global model. There is a danger that users will treat all assimilated data as the "truth", when in fact certain fields are less reliable because they are only weakly constrained by the assimilation system. Since all reanalysis products will be available to the research community, users must educate themselves about the quality of the data depending on their particular applications. This also places demands on modelers to improve the quality of their physical parameterizations in the assimilating models for future reanalyses, especially for climate applications.
The purpose of this atlas is to provide an early look at the quality of the NCEP/NCAR and the NASA/Data Assimilation System (DAO) reanalyses for a common period (1985-1993). Intercomparisons and comparisons to earlier operational analyses and to in situ and satellite observations are used to establish the level of uncertainty in various parameters. The main results are shown in terms of mean seasonal values and standard deviations with respect to season. Because the NCEP/NCAR assimilation system is frozen, many of the biases identified here are likely to persist in other years that have yet to be reanalyzed.
Section 2 gives an overview of the NCEP/NCAR reanalysis system and gives a brief discussion of some differences with the NASA/DAO assimilation system that influence the intercomparison. Section 3 gives a description of various observational datasets which will be used for comparison. Selected climate mean quantities are presented in section 4; in general the results shown are in terms of mean seasonal values and standard deviations are with respect to season. The results are organized as follows: Sections 4.1 and 4.2 show various climate mean fields as zonal mean and global maps, respectively. These sections also include difference maps between the reanalyses and comparisons to the NCEP operational analyses. The surface energy balance is presented in section 4.3, including comparisons to COADS. Global and regional precipitation estimates, including comparisons to several observationally based products developed at NCEP, are presented in section 4.4. This section also presents an intercomparison of global surface temperatures from the reanalyses and observations. Section 4.5 gives some general conclusions and includes a list of known problems with the assimilated data. Section 5 provides information about the input observational database for the NCEP/NCAR reanalysis in the form of data coverage maps and data count histograms. Section 6 provides information about NCEP/NCAR reanalysis datasets and data access from NCEP, NCAR, National Climatic Data Center (NCDC) and the Climate Diagnostics Center (CDC), including information on CDROMs and the local (NCEP) server.