Current and Future Applications of Model Assimilation Systems
In this chapter current applications of model-assimilated data sets in the areas of global atmospheric circulations, mesoscale atmospheric circulations, physical oceanography, the global hydrological cycle, and atmospheric chemistry are reviewed. Development is quite advanced in some of these areas but is only beginning in others, so that the range of achievement to date is quite wide. The best developed data assimilation theory and practice are found in the area of assimilation for global atmospheric circulations and are discussed first. The importance of future applications for the Earth Observing System (EOS) and other programs is briefly discussed.
GLOBAL ATMOSPHERIC CIRCULATION
As discussed in the previous chapter, current operational and research data assimilation systems at the larger global circulation assimilation centers fall into two main families, intermittent insertion and continuous insertion systems. Intermittent insertion is used in the meteorological centers of Canada; France; Japan; Australia; in the United States at the National Oceanic and Atmospheric Administration's (NOAA) National Meteorological Center (NMC), the U.S. Navy's Fleet Numerical Oceanography Center (FNOC), and the National Aeronautics and Space Administration's (NASA) Goddard Laboratory for Atmospheres (GLA); and at the European Centre for Medium Range Weather Forecasts (ECMWF). The continuous insertion sys-
tem is used at the United Kingdom Meteorological Office (UKMO) and the Princeton Geophysical Fluid Dynamics Laboratory of NOAA. Within atmospheric science, both methods have been under development for 15 years or more and are now mature applications in the atmospheric sciences. The theories of optimal interpolation and normal-mode initialization have been extensively developed and applied in both strands of development.
These developments have resulted in important increases in both analysis and forecast accuracy in midlatitudes. Understanding of the problems of tropical dynamics and physics has also advanced, with consequent improvements in tropical data assimilation. For example, operational global models now show levels of skill for tropical cyclone track forecasts that are comparable with the best specialized models for track forecasts.
Important new developments in data assimilation methods are likely to be implemented at operational centers over the next 5 years. These developments are expected to offer more accurate interpretations of remotely sensed data and a more consistent description of the time evolution of the atmosphere. When used along with new observing technologies in global assimilation systems of very high resolution (50 km), these methods will offer the prospect of substantial improvement in accuracy for both analysis and prediction.
For purposes of climate assimilation efforts, the most pressing defects of current model-assimilated data sets arise mainly because of insufficient and relatively poor accuracy of observations of the divergent wind and humidity fields, particularly in the tropics. Because of these defects, descriptions in assimilated data of these fields are largely determined by the predictive component of the assimilation model from other fields (nondivergent wind, temperature, pressure) and so are dependent on the assimilation system used to make the inference.
A second limitation of the operationally produced data sets is the difficulty encountered in studies of the climatic variability of parameters that are sensitive to details of the assimilation systems. Because operational systems have been undergoing constant development, studies of phenomena such as the interannual variability of diabatic heating (based on operational data sets) may have superimposed a nonphysical component of variability caused by changes in the assimilation system itself, in the observing network, or in other nonphysical components. During the proposed reanalysis of operationally produced data sets, the state-of-the-art assimilation system employed can be used to identify these sources of variability.
Further limitations on data sets are imposed by the truncation limit of the assimilation system and by the fact that quality control decisions can never be perfect. An example illustrates the relation between these issues: Suppose a ship in a Norwegian fjord reports a strong local wind. The wind may bear little direct relation to the synoptic-scale flow describable on an as-
similation grid of 200 km. A data assimilation system may correctly reject the (good) observation, because the observation describes a phenomenon below the resolution of the analysis. Problems of quality control and resolution are always encountered in the vicinity of intense phenomena (e.g., the core of a hurricane) that are below the resolution of the assimilation system.
Quality control algorithms apply many checks to observations: Position checks, internal consistency checks, time consistency checks, climatological checks, checks against adjacent data, and checks against the background field. Quality control algorithms are tuned empirically to maximize the likelihood of accepting good, and rejecting corrupt, observations. Because of the uncertainties inherent in such a probabilistic exercise, bad observations are sometimes accepted and good observations are sometimes rejected. The consequences of erroneous decisions for the resulting forecast can occasionally be serious. Nevertheless, the overriding difficulty in data-sparse regions is paucity of data rather than quality control of data (Meteorological Office, 1987).
MESOSCALE1 ATMOSPHERIC CIRCULATIONS
Limited-area operational models in the United States, Europe, and Japan that utilize data assimilation systems routinely produce background states for weather prediction. With some exceptions, these background states are determined by intermittent updating of the meteorological fields (Hoke et al., 1989; Golding, 1989), even though plans generally call for next-generation systems to employ continuous assimilation procedures. The three-dimensional variable fields produced are, at best, only marginally adequate for the spatial resolution of many mesoscale meteorological processes within the domain of the limited-area models. The finest-scale operational model available in the United States—the Regional Analysis and Forecast System (RAFS) of the National Weather Service (NWS)—has a grid increment of 80 km on its highest-resolution "C" grid, which spans the North American continent (Hoke et al., 1989). Therefore, the RAFS will resolve best those phenomena with wavelengths longer than about ~800 km. Its suitability for production of assimilated data sets is limited to the synoptic scale and the higher end of the meso-alpha scale, which ranges from 200 to 2000 km. The RAFS analysis does not satisfy the needs of scientists who wish to
study phenomena with scales smaller than ~800 km. Thus, model-assimilated data sets that resolve all meso-alpha and meso-beta scales of atmospheric phenomena need to be created by special-purpose research models, as do data sets for geographic regions not covered by operational limited-area models. Such models are currently being developed, for example, at the NOAA Forecast Systems Laboratory (FSL) and the National Center for Atmospheric Research (NCAR) in Boulder, Colorado, and at several universities (Colorado State University, Pennsylvania State University, University of Oklahoma, University of Wisconsin at Madison, Florida State University, University of Illinois, and Drexel University).
Research models generally employ physical process parameterizations that are 2 to 3 years more advanced than operational models. These parameterizations can be specifically designed for particular applications. Research models with variable horizontal and vertical resolutions are generally relocatable (not tied to a specific geographical grid) and can be optimized for the phenomenon being studied. They can be used to generate research-quality data sets on all desired scales. On the other hand, the use of limited-area operational models to produce data sets has the advantage that special modeling efforts are not required, with the result that a much larger number of meteorological cases can be made available for research study.
In many studies of mesoscale physical processes, model-based data sets have been generated without the assimilation of any synoptic-scale data during the forecast cycle (except for lateral boundary conditions) because the use of such data through reinitialization would result in the loss of the mesoscale structures generated by the simulation. This is feasible because of the importance of lateral boundary information as determined from observations reflecting the interaction between the mesoscale and larger scales and the relatively good forecasts of atmospheric structure and processes that are often possible for 12-to 72-hour time periods.
The implementation of new technologies in data acquisition, such as the atmospheric profiler and the WSR-88D Doppler radar networks in the NWS Modernization Program, provides the opportunity for the development of four-dimensional data assimilation methods on the mesoscale. For example, NOAA's FSL has begun the development of a mesoscale assimilation model, the Mesoscale Analysis and Prediction System (MAPS). Other examples involving community efforts are the Penn State/NCAR mesoscale model and Colorado State University's Local Analysis and Prediction System (LAPS), which assimilates Doppler radar, satellite soundings, and mesoscale surface observations every hour in real time. In the area of mesoclimatology, the Penn State/NCAR mesoscale model has been used with a four-dimensional data assimilation routine that employs Newtonian relaxation for production of about a 1-year mesoclimatology of flow over North America for the Environmental Protection Agency (EPA) (Seaman. 1989).
The state of numerical modeling and model assimilation in physical oceanography is far less advanced than that of its atmospheric counterparts for a simple reason: the amount of ocean data collected routinely is orders of magnitude less than that collected for the atmosphere. This is primarily due to the fact that society has not required oceanic forecasts to the same extent that it has weather forecasts. The net result is that the global oceans are poorly sampled (especially at depth) and consequently are poorly modeled and poorly understood.
The lack of data is partially compensated for by the slowness of most ocean changes. If, as was often assumed until recently, the ocean were in a steady equilibrium at depths below the level of seasonal variability, then all data, no matter when collected, would be useful in defining this idealized mean state. However, in the last two decades, relatively rapid (i.e., time scale of months) disruptions in oceanic circulation in near-equatorial regions and slower changes (decades) at intermediate depth in high-latitude regions have been detected. Changes at greater depth on still longer time scales (hundreds and thousands of years) are inferred from the paleoclimatic record and predicted by coupled atmosphere-ocean simulations of green-house warming.
At the present time, for practical reasons, data assimilation experiments in the ocean are concentrated on the shorter time scales of weeks to years. Interest and activity in the field have been growing rapidly. An overview of the subject is provided by a special double issue of Dynamics of Atmospheres and Oceans (Vol. 13, Nos. 3–4).
There are currently three distinct types of activities in the area of oceanographic data assimilation and production of model-assimilated data sets. The first is an oceanographic mesoscale prediction activity (Robinson and Leslie, 1985; Robinson et al., 1986, 1989). This activity involves predicting the state of the ocean over a relatively small (100 x 100 km), open region using boundary data as continuous input. In a certain perspective this prediction converts data along the boundary into model-assimilated data sets throughout the region by means of a dynamically consistent interpolation in both space and time using the equations of motion.
The second activity first simulates "data" by running ocean model controls and then assimilates this simulated data into model runs starting from different initial conditions (Philander et al., 1987; Moore et al., 1987; Anderson and Moore, 1989). The use of simulated, rather than actual, data allows for studies of model-based data assimilation in situations where real data are either nonexistent or too sparse to be useful. While the results of these studies are generally optimistic compared to studies using real data, some very useful general principles can be obtained. For example, methods of
data assimilation can be examined (Long and Thacker, 1989a,b; Miller and Cane, 1989), the amount and types of data needed for useful initialization of models can be estimated (Philander et al., 1987), and fixes for various technological problems can be tested (Moore, 1989).
As an example of the usefulness of this simulated data assimilation, Philander et al. (1987) found that a few meridional sections of upper-ocean thermal data alone are adequate to initialize the equatorial upper ocean. Under these circumstances the equatorial undercurrent is generated to the correct amplitude by the model without the use of direct-current observations. On the other hand, Anderson and Moore (1989) showed that both oceanic thermal and current data are necessary to accurately initialize an equatorial Kelvin wave.
It should be pointed out that these studies assume that the simulated data are perfect. It is not at all obvious that actual measured thermal data, say with expendable bathythermograph (XBT) data, would yield the same results.
The third type of activity is the assimilation of actual data into numerical ocean models to produce model-assimilated data sets. Leetmaa and Ji (1989) have used an experimental ocean circulation model to assimilate XBT observations taken in the tropical Pacific and have produced monthly fields of currents and thermal structure. The assimilation of data has successfully corrected model-induced diffusive drifts in the vertical thermal structure and has created model-assimilated data fields downstream of where actual data were introduced. The results suffer from a basic problem characteristic of this type of large-scale data assimilation, namely, that the system is forced by imperfectly known winds, resulting in an oceanic response that is sometimes incompatible with the observed data. The monthly data assimilations therefore sometimes oscillate between states inferred by previous upstream data and states forced by new in situ observations. These problems did not arise, however, in the work of Miller and Cane (1989), who used a simpler model of the tropical Pacific but a more sophisticated assimilation method.
A similar assimilation of real XBT data has been performed for the Atlantic (Morlire et al., 1989) for a 1-year period, with data being introduced gradually throughout each month. The results showed a similar improvement in the model thermal structure and a notable improvement in currents for the simulated year. While improvements were significant, discrepancies still occurred between the observed and modeled data, presumably due to inaccuracies in the imposed surface forcing.
Another example is the work of Derber and Rosati (1989); they used a global ocean general circulation model and a continuous injection technique for the assimilation. The results using the Comprehensive Ocean-Atmosphere Data Set (COADS) and the Master Ocean Observations Data Set
(MOODS) over a 13-month period are encouraging, although further improvements are required.
The dilemma in assimilating data in the presence of imperfect surface forcing can be treated in two separate ways. The first is to make the winds part of the fields that can be corrected by the internal ocean data. Thus, through mutual adjustment, accurate ocean data can be used to correct both the ocean and the wind fields and to provide the most consistent combination of forcing and response fields; this is the approach of Thacker and Long (1988). The second approach, not yet tried, would be to assimilate both atmospheric and oceanic data into a coupled atmosphere-ocean model. The dynamics of such a model would adjust the surface fluxes internally, and, in principle, an optimal state of the coupled atmosphere-ocean system should emerge.
As a final note, not a single oceanographic field program as yet has taken the value-added approach of presenting the wide diversity of field observations in the form of model-assimilated data set output of a numerical ocean model, even though such an output is the best means of ensuring that the various and diverse observations are as dynamically consistent with each other as possible.
The hydrological and energy cycles of the earth system are intimately linked through transports and radiative effects of water vapor in the atmosphere and by transformations between liquid, solid, and vapor states. Rather than dealing with the entire system, the disciplines of meteorology, hydrology, and oceanography have traditionally focused on separate subsystems of the global hydrological cycle.
In the development of the data assimilation/analysis process of atmospheric forecast models, the accurate specification of initial dynamical and thermal fields has, until recently, received higher priority than the specification of hydrological fields. However, the hydrological fields are now receiving increased attention as forecast models become more sophisticated (Tiedtke et al., 1988).
Atmospheric hydrology is the ''fast component" of the global hydrological cycle and plays a fundamental role in weather processes and prediction. Consequently, the atmospheric subsystem is the only component of the hydrological cycle for which global model-assimilated data sets are currently being generated as part of operational numerical weather prediction.
When viewed from an atmospheric perspective, the thermal and hydro-
logical conditions at the interface with the ocean and land surface constitute atmospheric boundary conditions. For weather forecasts up to a few days in advance, it is important to have a reasonably accurate description of the initial boundary conditions. However, evolution of the boundary conditions is typically slow relative to the time scales of day-to-day synoptic variability; therefore, accurate incorporation of time-varying boundary conditions in the models has up to now received less attention in operational short-range weather prediction. In contrast, changing boundary conditions are crucial for resolving seasonal and interannual climate variability (Shukla, 1985). As integrated earth system models are developed under the impetus of the climate and global change programs, development of data assimilation will need to address the entire hydrological cycle (see Large-Scale Field Experiments and Biospheric Studies section).
Land Surface Subsystem
There are important differences in the nature of the current operational models and the character of assimilated data sets between meteorology and hydrology. First, there is a fundamental difference in the most suitable coordinate systems for description and prediction. It is convenient to specify global atmospheric fields on a regular grid or in terms of spectral components. In contrast, the natural "grid element" for surface hydrological operations (e.g., river forecasting and soil moisture accounting computations) is the irregularly shaped drainage basin defined with respect to a stream gauging station at the outflow point.
Second, land surface hydrology has traditionally been focused on the water balance of drainage basins whose area is small relative to synoptic spatial scales. However, it should be noted that this spatial-scale mismatch continues to narrow as the resolution of operational atmospheric prediction models increases. The mismatch is reflected in the sampling by the surface-based observations transmitted by the Global Telecommunications System (GTS) of the World Weather Watch (WWW). The WWW surface-based system was designed primarily to resolve synoptic-scale variability. Even over land, where meteorological observations may be relatively dense, a mismatch remains between the synoptic-scale sampling and the much smaller scales of hydrological variability associated with terrain, diurnal variability, planetary boundary layer processes, and the intermittent subsynoptic character of precipitation systems.
The focus of surface hydrology on individual drainage basins, together with the treatment of precipitation as a measured or specified input to the system, has muted any requirement for global data sets. Thus, no system of international data exchange comparable to that for the atmosphere has been developed for traditional surface hydrological data (e.g., high-resolution
precipitation data and model-generated evapotranspiration and soil moisture estimates). Consequently, a great deal of surface hydrological data exist for many regions that are not generally made available to the operational meteorological centers or the scientific community for inclusion in model-assimilated data sets.
Ocean Surface Subsystem
The ocean represents the inexhaustible reservoir for the global hydrological cycle. Its primary linkage with the other components is through surface exchanges, that is, precipitation versus evaporation and inflow of fresh water from the land margins. Resultant variations in surface salinity affect the stratification and vertical exchange processes and contribute to large-scale current systems. Along with ice cover, they play an important role in determining the surface heat fluxes. In addition, the hydrological cycle over the ocean plays a very important role in determining the buoyancy flux at higher latitudes and hence determines to a certain extent the intensity of the thermohaline circulation of the ocean.
The measurement deficiencies of the hydrological cycle are severe. Many parameters have never been measured except on a local experimental basis; they have to be inferred from empirical relationships or dynamical balance considerations. New satellite-based observations, combined with more effective use of data from existing satellite systems and improvements in the data assimilation and analysis process of global forecast models, offer the only realistic prospects for significant improvement in the description of the global hydrological cycle.
Precipitation, perhaps the most fundamental hydrological parameter, serves as an important example. Precipitation is adequately measured only over a few well-instrumented land areas of the earth. Furthermore, the transmission of daily values is frequently haphazard, and their use in computing monthly averages is questionable at best. Generally, one must rely on the delayed transmission of monthly averages. Precipitation is also an output of present-day forecast models, but the surface observational coverage is inadequate in most regions for verifying the forecast values.
Satellite-based observations, coupled with an adequate "ground truth" program for calibration and analog development, provide the only realistic prospect of ultimately obtaining global precipitation measurements. There is optimism that the means now exist for significant improvement in the estimation from satellite data of precipitation for periods as short as a few days, particularly in the tropics.
No single space-based instrument is currently capable of providing all the information needed to generate precipitation estimates. The possible measurement techniques include using visible and infrared (IR) imagery together to infer precipitation from cloud height and brightness, scattering-mode passive microwave techniques to infer precipitation intensity from the scattering of ice particles, absorption-mode passive microwave techniques to infer precipitation intensity from the brightness temperature of a rain column of known depth, and active microwave techniques to infer precipitation intensity and condensation profiles from scattering and attenuation from raindrops.
The ability to use observations from the established network of geostationary satellites to obtain space-and time-averaged precipitation estimates over the global tropics has led to the establishment of the Global Precipitation Climatology Program (GPCP), sponsored by the World Climate Research Program (WCRP), principally in support of the Tropical Ocean and Global Atmosphere (TOGA) program (World Climate Research Program, 1986; Arkin and Ardanuy, 1989). The GPCP now produces estimates of 5-day precipitation averaged over areas of 2.51° latitude by 2.5° longitude for the tropics and subtropics using an IR thresholding technique. The project is planned to continue for the 10-year period 1987–1996.
LARGE-SCALE FIELD EXPERIMENTS AND BIOSPHERIC STUDIES
There is growing recognition of the importance of viewing the atmosphere, oceans, land masses, and biosphere as interacting components of a comprehensive earth system (National Research Council, 1990). If the complete system is to be understood, each component must be observed and modeled in the context of its impacts on and control by the other components.
Global studies of the physical climate system and the biogeochemical cycles will require the combination of large-scale simulation models, operating in both the predictive and assimilation modes, and a wide range of data types, ranging from standard meteorological observations to specialized ground-truth efforts for validation of the satellite algorithms.
In the area of land-surface-atmosphere interactions, there are a number of significant problems that must be overcome before global modeling studies can be conducted in a realistic, quantitative way (P. Sellers, University of Maryland, personal communication, 1989). Two major problems are:
Model design: Most land surface models currently in use in general circulation models (GCMs) or other large-scale models are based on one-dimensional models of the soil-plant-atmosphere system developed by biologists and agronomists to understand local plant-scale controls on radiation, interception, evapotranspiration, and photosynthesis. For the most
part these models have been used without change to describe area-averaged processes operating over scales of several hundred kilometers. Little work has been done to check the validity of this usage or to develop more sophisticated spatial integration techniques.
Initialization and validation: Even if ''perfect" biosphere-atmosphere models existed, they would be of little use in the absence of effective methods of initializing them and validating their performance. Clearly, satellite remote sensing offers the greatest promise for providing global fields of surface type, photosynthetic capacity of vegetation, albedo, and soil moisture, but the links between the satellite observations and these biophysical quantities either have not been developed or have not been tested with convincing rigor.
These two problems have been and will continue to be addressed by a series of large-scale field experiments. For example, the following experiments are some that have taken place during recent years:
Hydrological-Atmospheric Pilot Experiment (HAPEX) (Andre et al., 1986). This 1986 experiment was conducted in southwestern France over a 100-km2 area. Surface and airborne heat flux measurements were combined with meteorological and hydrological observations to validate a mesoscale modeling effort. The experiment provided data that allowed an evaluation of the contribution of different land cover types to the regional heat flux field and the resultant influence on local circulation.
First International Satellite Land Surface Climatology Project (ISLSCP) Field Experiment (FIFE) (Sellers et al., 1988). FIFE is an intensive effort designed to specifically address the modeling scale issue and the remote-sensing inversion task. The field phase of the experiment was executed from 1987 to 1989 over a 15-km2 area in Kansas that was heavily instrumented with automatic meteorological stations and surface flux rigs. To date, analysis of the data has shown that there are strong relationships between satellite observations in the short-wave part of the spectrum and the biophysical parameters of photosynthetic capacity and canopy water content. An effort is being made to combine all the field observations with a one-dimensional surface-atmosphere column model to calculate surface states and fluxes.
Another French-led experiment, HAPEX-II/Sahel, is planned for execution in 1992 in Niger. It will combine elements of the FIFE and HAPEX experiments.
Concentrations of trace gases in the atmosphere have not been systematically measured on a global scale over long periods of time. Thus, very
limited data assimilation has been performed by the atmospheric chemistry community. An important exception is the case of ozone and related species, such as chlorofluorocarbons and methane, and a few other species such as water vapor, nitric acid vapor, and nitrogen dioxide, which have been measured by instruments such as the Total Ozone Mapping Spectrometer (TOMS), the Solar Backscatter Ultraviolet (SBUV) spectrometer, the Limb Infrared Monitor of the Stratosphere (LIMS), and the Stratospheric and Mesospheric Sounder (SAMS) on board the Nimbus 7 satellite, as well as by the Solar Mesosphere Explorer. These data have been used to verify the consistency of the presently accepted chemical scheme for the middle atmosphere and to derive the distribution of fast-reacting radicals produced from chemical transformations of the measured trace gases.
Studies of specific chemical events in the troposphere over limited geographical domains, usually in connection with observational campaigns, are performed by coupling a rather detailed chemical kinetic code with a transport model. The winds used to simulate transport during these events require model-assimilated data that are provided by mesoscale meteorological models. Difficulties in these simulations include accounting for possible rapid vertical transport and chemical conversion in convectively active clouds and correctly specifing the conditions at the boundaries of the limited domain involved. This type of approach is currently used in connection with acid rain studies and pollution episodes, such as those during which high concentrations of tropospheric ozone are observed over continental areas.
Data assimilation techniques have also been used in a limited number of cases to understand the transport and chemical transformations of trace gases in the stratosphere. Assimilation algorithms have, for example, been used to reproduce the behavior of nitric acid and ozone during a stratospheric warming event. Again, in this type of study, only the dynamical fields (temperature, geopotential height, and winds) area assimilated and used to predict transport of the species.
Finally, a limited number of attempts to assimilate the chemical data themselves have been made by replacing, in a chemical transport model, the calculated distribution of a chemically active trace species by the corresponding observed distributions. An estimation of the effect of this substitution on the calculated concentrations of other species that are present provides a test to validate the chemical scheme adopted in the model.
Data assimilation using models with predictive capability for chemical species will probably be used more extensively when planned future satellites, such as the Upper Atmosphere Research Satellite (UARS) and the polar platforms of the Earth Observing System, begin to provide continuous observations of trace constituents on the global scale over a period of several years.
EARTH OBSERVING SYSTEM
The Earth Observing System (EOS) is a program being planned by NASA in collaboration with the European Space Agency (ESA) and the National Space Development Agency (NASDA) of Japan to utilize satellite measurements to study climatic and global changes of the earth system. The program calls for the deployment of a large variety of earth-sensing instruments on platforms in low-earth polar orbit. Plans for EOS also include a number of earth-observing instruments as attached payloads on the planned low-declination NASA Space Station. The full system is scheduled to operate for a 15-year period, presumably long enough to obtain a useful time series of earth observations for monitoring global change.
From the point of view of producing global atmospheric data sets for climate research, a global satellite observational capability is essential. Two of the most important EOS instruments will be NASA's Atmospheric Infrared Sounder (AIRS) and the Laser Atmospheric Wind Sounder (LAWS). AIRS will measure radiances in more than 115 spectral bands with the goal of obtaining temperature retrievals of 1°C accuracy and 1-km vertical resolution throughout the depth of the troposphere. It will be a prototype for the next-generation operational IR sounder. LAWS is a Doppler lidar system for active tropospheric wind measurement in regions without opaque clouds. The expectation is that an accuracy of 1 to 5 m/s with a 1-km vertical resolution and an average horizontal spacing of 100 km can be achieved.
NASA is planning an EOS Data and Information System (EOSDIS), which will provide for the reception, processing, storage, and distribution of EOS data. An important component of EOSDIS will be a four-dimensional data assimilation system, which will use both EOS and conventional data in delayed mode to produce research-quality data sets for the earth-atmosphere-ocean-land system. The Global Modeling and Simulation Branch of the Goddard Laboratory for Atmospheres (GLA), with cooperating teams from the National Meteorological Center (NMC) and the Australian Bureau of Meteorology Research Center, has been selected by NASA to develop a high-resolution four-dimensional atmosphere-ocean-land data assimilation system for EOS. The panel's recommendation for a nationally focused geophysical data assimilation program calls for close coordination with the EOSDIS program.
USE OF MODEL-ASSIMILATED DATA SETS FOR RESEARCH
Data sets from assimilation of observations and from simulations employing assimilated data as initial state information are widely used by the meteorological community. For example. more than 200 institutes in over
50 countries on six continents have acquired model-assimilated data sets from ECMWF. Most data requests are for research purposes. The heaviest users are in the developed countries, but there are also many requirements for the data sets in the developing world. The advantage of these data sets is that they systematically combine information from many different satellite and ground-based sources, a task that is beyond the resources of small university or national research groups.
The use of model-assimilated data sets for the study of large-scale atmospheric flow was pioneered in the middle to late 1970s by Wallace and Lau at the University of Washington and by Blackmon at the National Center for Atmospheric Research. Their studies of the diree-dimensional structure of midlatitude atmospheric circulation led to a considerable advance in our understanding of low-frequency variability in the atmosphere.
The Global Weather Experiment (GWE) provided, for the first time, a detailed global view of atmospheric structure. The availability of global data accelerated the development of techniques of global data assimilation. The resulting model-assimilated data sets provided good quality descriptions of the mass and wind fields and formed the basis for a wide range of research studies. Vital information on large-scale tropical phenomena such as El Niño and the 30-to 60-day oscillations has been determined through the use of these data sets. In recent years model-assimilated fields have also been invaluable for study of the synoptic environment of tropical cyclones.
Many examples of the utility of model-assimilated data sets in research can be cited. Most of the published work on diabatic forcing of the atmosphere has been accomplished by groups at the universities of Wisconsin, Utah, Reading, and Helsinki using model-assimilated data sets. Most of the diagnostic work on blocking, multiple flow regimes, and low-frequency variability undertaken at NOAA/NMC's Climate Analysis Center (CAC); the Massachusetts Institute of Technology; the universities of Yale, Washington, Stockholm, and Bologna; and the ECMWF have used model-assimilated data sets. Much of the diagnostic work on the structure of the 30-to 60-day oscillations at the universities of Honolulu, Wisconsin at Madison, California at Los Angeles, and Tsukuba and at the NCAR and CAC has been based on model-assimilated data sets, frequently in conjunction with satellite measurements of outgoing long-wave radiation (OLR). The group at the University of Reading, in a recent climatological atlas based on ECMWF model-assimilated data sets, cited 32 papers "published from or inspired by" their climate diagnostics project over the last decade. Groups at ECMWF and the universities of Cologne, Kiel, Wisconsin, and Helsinki have extensively documented the atmospheric energy cycle using model-assimilated data sets. Finally, model-assimilated data sets have proved invaluable for oceanographers concerned with forcing of the oceanic circulation. The data