Answer to Roy William's Question 1:
Where are the main institutions that produce/use data for/from climate and weather models?

Click here to e-mail Peter Lyster.

I'll provide here the responses from Peter Lyster and Jay Larson

Peter Lyster's Response

It's probably best to think of three kinds of data.

  1. Actual observations. This includes raw and processed data; in the language of the Earth Observing System Data and Information System (EOSDIS) this is called level 1, and level 2 data.
  2. Reanalysis data. This is a controversial term, but I think of it as data that is gridded and has gone through some processing, such as combining it with other data (or model forecasts) to make the final product more accurate. When we at the Data Assimilation Office (DAO) say "data assimilation" we mean that model forecasts have been combined with observational data to get a best estimate of the state of the atmosphere. In EOSDIS, gridded data (a superset of what we call reanalysis data) is called level 3 data.
  3. Model generated data. This is probably what you think of as "Climate Modeling" at laboratories and universities around the world. They don't necessarily use real data to generate initial conditions (although they often use analyzed data for boundary conditions and whatnot).

I don't know the total volume of observational data there is in the "world". However there are interesting observations (I mean conclusions) one can make in reference to the term "modeling" (since I think that the main thrust of Roy's questions comes from concern for modeler's data). When you flat out grid data you need more data than gridpoints (or else you won't have enough information). However, when you do reanalysis that involves an ongoing time-iterated model such as a General Circulation Model (GCM), then the amount of data you feed into the model per day is often smaller than the number of gridpoints you can output per day to disc. Similarly, if you run a straight (e.g., GCM) model simulation then you have the capacity to produce a lot more output (gridpoints) than you have data (to say, compare the result with).

So let me summarize with some numbers. Staying with Atmospheric GCMs, modern GCMs at 2o (latitude) x 2.5o (longitude) x 20 (levels) have about 106 gridpoints (or modes if you're talking about a spectral model). It's common to output "high-frequency" data every six hours of simulation time, so about 1 gigabyte of data may produced per simulation day. By contrast, the world-wide daily meteorological observational data set that is collected over the Global Telecommunications System (GTS), is about 200 megabytes. As an extreme case, if you run a Climate Modeler's GCM for 100 years of simulation time you could produce 36.5 terabytes of data. More modest amounts of data are produced by reanalyses because the timescale is limited by the extent of the meaningful input dataset. The DAO, as well as meteorological centers such as the European Centre for Medium-Range Weather Forecasts (ECMWF) and the National Centers for Environmental Prediction (NCEP, in Maryland) perform decadal-length reanalyses. For example, a reanalysis of 7 years of observations may produce about 2.5 terabytes of data. The ECMWF archives their data onsite, NCEP and the National Center for Atmospheric Research (NCAR) archives data at one of NOAA's extensive network of the data centers (see pointers later in this document, especially the National Climatic Data Center (NCDC) at Asheville NC), and the DAO archives it's data at NASA/Goddard Space Flight Center as well as at NASA's Distributed Active Archives (DAACs). As I just pointed out, the input dataset for reanalyses is typically about 1/5 of the output (archived) dataset. As I said, GCMs that are run in pure climate forecast mode are capable of producing significantly more output than the more data-bound reanalyses; so nobody (to my knowledge) archives the high-frequency datasets for long-term GCM forecast runs. The Atmospheric Model Intercomparison Project (AMIP) that is supported by the US DOE and run out of the Lawrence Livermore National Laboratory (LLNL) compares GCMs that are run in forecast climate mode with both observations and reanalysis data. The full high-frequency output is not archived, but temporal and spatial averaged data are archived and these may be compared with reanalyses datasets. Typical 7 year runs that are performed under the current AMIP II project (involving about 30 different models from through the world) produce about 30 gigabytes of archived data, that is, about 1/70 of the data volume for a full reanalysis high-frequency archived dataset over the same period of simulation time.

A brief list of institutions where climate models are used to assess, for example, the impact of doubling CO2 et cetera:

Climate models are used in a range of scientific studies at:

Numerical Weather Prediction Models that are used for meteorological analysis and forecasting as well as for reanalyses

Jay Larson's Response

Climate modelers often do long integrations of General Circulation Models (GCMs). Time-based Statistics are calculated from these integrations, and these comprise the climate. Popular climate quantities are time-averages for seasons JJA (June-July-August) and DJF (December-January-February) of Surface Temperature (TS), Sea-Surface Temperature (SST), Mean Sea-Level Pressure (MSLP), surface windfields, upper-air windfields, geopotential height (i.e. how high is the surface of constant pressure at, say, 500 hPa), and specific humidity. Scientists also study statistics for zonal averages (i.e. take of time-averaged data and then average it along a latitude band), and in some cases meridionally averages (along longitude bands) of time-averaged quantities. That gives you a flavor for how scientists model climate. In some cases scientists also study variability by taking variances. In short, it's like the old saying "climate is what you expect, weather is what you get."

Where are GCMs maintained and used?

A typical spectral GCM solves the primitive equations in spectral space, where the modes are the spherical harmonics (by solve, I mean the horizontal advection and time evolution). The "source" or "forcing" terms in the primitive equations are calculated in physical gridpoint space, approximated in terms of mode coefficients, and then applied in spectral space. The truncation of the spectral approximation is intimately related to horizontal grid resolution. For example, the model resolution may be R15 (rhomboidal truncation at 15th wavenumber ~ grid spacing of 4.5 degrees in latitude and about 7 degrees in longitude), or T42 (triangular truncation at wavenumber 42 ~ 2.8 degrees by 2.8 degrees), et cetera. By contrast, some general circulation models solve the primitive equations via a finite-difference scheme.

NASA: The GCM used by the DAO is distinct from much what exists in terms of GCMs because it is a finite-difference (rather than spectral) model. The model is called the Goddard Earth Observing System (GEOS) GCM. The version used by the DAO in its data assimilation system is maintained by Larry Takacs. View the DAO web page at

NCAR: Formerly called "Community Climate Model (CCM)," their freely available model has gone through development tracks CCM1 (1987), CCM2 (1992), and is now called "Climate System Model (CSM)." CSM includes CCM3 (1996) and also includes some ocean model options. Scientists, both in America and abroad, use this model for numerical experiments (e.g., they might modify the model to include some new/improved parameterizations of physical processes, but the main dynamics of the model are left the same). Some contacts at NCAR are are Dave Williamson and Jim Hack. View the home page at http://www.ucar.edu -- follow links around to NCAR divisions and look under the CGD (center for global dynamics) pages.

BMRC: Australia's Met Bureau's Research Centre was the home of the first spectral GCM (The NCAR CCM models were originally developed from this), and they have their own GCM used for climate experiments. Bryant McAvaney a contact there.

CSIRO: Australia's Commonwealth Scientific and Industrial Research Organization (CSIRO) have their own GCM as well. Barry Hunt is a contact in their Division of Atmospheric Research that develops this model.

ECMWF: European Centre for Medium-Range Weather Forecasts. Their GCM is used in weather forecasts and reanalyses.

There are lots of GCMs used worldwide. Others include UK Met Office (UKMO), Canadian Climate Center (CCC), National Center for Environmental Prediction (NCEP - formerly NMC). Larry Gates at Livermore is a good contact person for the AMIP project.

Go back to Primary Document Page.