Hartmut Grassl

Coupled general circulation models (CGCMs) integrate our knowledge about atmospheric and oceanic circulation. Different versions of CGCMs are used to provide a better understanding of natural climate variability on interannual and decadal time scales, for extended weather forecasting, and for making seasonal climate scenario projections. They also help to reconstruct past climates, especially abrupt climate change processes. Model intercomparisons, new test data (mainly from satellites), more powerful computers, and parameterizations of atmospheric and oceanic processes have improved CGCM performance to such a degree that the model results are now used by many decision-makers, including governments. They are also fundamental for the detection and attribution of climate change.

Max-Planck-Institute for Meteorology, Bundesstrasse 55, D-20146 Hamburg, Germany. E-mail: grassl@dkrz.de

Historical Development

The weather forecasting models based on the barotropic vorticity equation that emerged in the early 1950s (1) neglected the ocean and all diabatic processes in the atmosphere that drive changes in weather. Nevertheless, they were able to give useful forecasts up to 48 hours in advance solely by analyzing the often rather realistic shift of decaying high- and low-pressure systems that were given as input from a mainly surface-based, synoptic, in situ meteorological network. In the 1960s, larger computers and a growing understanding of baroclinicity (2) allowed a breakthrough to the forecasting of newly developing mid-latitude atmospheric disturbances on time scales of up to about 3 days. As in the 1950s, only the dynamics of the atmosphere were modeled and diabatic processes were largely neglected, but in addition to the surface network, the models used the soundings of the troposphere and lowest stratosphere produced by the nearly global radiosonde network, as coordinated by the then-new World Weather Watch Programme of the World Meteorological Organization (WMO). From the radio soundings, differential advection of temperature and vorticity could be derived that determined the development of new low-pressure systems.

The continuous improvement of weather forecasting in the past two decades, which now includes all major diabatic atmospheric processes and thus also air-sea fluxes of radiation, heat, and momentum, has led to useful forecasts of about 8 days in winter. It has also had major consequences for climate modeling. Those improvements include the following:

1) Coupled atmosphere-ocean models such as the ECHAM model of the Max-Planck-Institute for Meteorology in Hamburg, Germany now often use a meteorological forecast center's model dynamical core and some parameterizations of subgrid-scale processes.

2) Through their reanalyses (3, 4), the weather forecasting centers provide the most consistent validation data sets for coupled climate models.

3) The breakthrough to predictions of seasonal climate anomalies that is already operational at several meteorological centers (5) [especially for areas affected by El Niño-Southern Oscillation (ENSO)] has to a large extent bridged the gap between weather forecast and climate models, because these forecasts combine the integration of nonlinear prognostic differential equations using initial values from observations (as used for weather forecasting) with probabilistic estimates of anomalous weather statistics (climate anomalies).

4) The application of seasonal forecasts in the developing and developed countries will boost the build-up of a global upper ocean observing system and thus will provide a long-needed data set for further improvements of climate models and of ocean GCMs (OGCMs), which will have many more applications than just being part of a climate model.

Evaluation of CGCMs

If CGCMs had the following four capabilities, there would be greater confidence in the use of CGCMs for the projection of future climates: (i) Adequate representation of the present climate; (ii) reproduction (within typical interannual and decadal time-scale climate variability) of the changes since the start of the instrumental record for a given history of external forcings; (iii) reproduction of a different climate episode in the past as derived from paleo climate records for given estimates of the history of external forcings; and (iv) successful simulation of the gross features of an abrupt climate change event from the past.

If a CGCM reproduces the present climate [for example, does not show large systematic errors in sea surface temperature (SST)], especially the seasonal cycle, and does not need flux adjustments, it has successfully passed step one of the above evaluation but must still not be able to project future climate realistically for a given forcing. At present, many CGCMs, both flux-adjusted and non-flux-adjusted, pass step one; that is, they simulate mean climate and the annual cycle correctly on large scales and approach observed variability on time scales up to interannual.

Some models, when forced by scenarios of external parameters for the 20th century, reproduce the climate variability of recent decades, including the impact of volcanoes such as Pinatubo. Step (ii) of the evaluation is then passed within typical decadal time-scale variability. But the history of solar forcing and processes stimulated by solar forcing are still insufficiently known to justify more model studies.

The third step in evaluating CGCMs lies in simulating a past climate state, preferably one rather different from the present one. Such a test needs many paleo data of high quality, which are available for only a few time slices, such as the last glacial maximum (18,000 to 21,000 years ago) or the warmer period in the Holocene (roughly 6000 years ago). Although paleo data for model evaluation are more abundant for the Little Ice Age period of the Northern Hemisphere than for any other period in the past, this period is not so useful because its climate state differed from the present one far less than that projected in scenarios until the end of the 21st century. In the Paleo Climate Model Intercomparison for AGCMs (7), mid-Holocene (6000 years ago) simulations of 18 AGCMs at prescribed SSTs captured the northward extension and intensification of the African Monsoon in the Northern Hemisphere summer. The warmer-than-present conditions in high northern latitudes were also reproduced, but the paleo climate data do not support the modeled drier interior of Northern Eurasia and Northern America, and CGCM runs for the mid-Holocene (8) tend to intensify the African monsoon further than is actually seen.

The hardest test for a climate model is the simulation of an abrupt climate change. With the advent of quantitative paleo climate data, mainly the high temporal isotopic compositions of ice cores and sediments (9), this test came within reach. Because integrations of high-resolution CGCMs with grid sizes of about 100 km consume too much computer time, only models of intermediate complexity have been used for such tests until now. These models recently partially passed such a test. In addition, their components have to be tested by high-resolution component modules. In the ideal case, deposits with a yearly time resolution such as tree rings, lake varves, coral reefs, and ice cores would constitute the validation database for the evaluation of CGCMs under steps (iii) and (iv) above because isotope ratios in these deposits might even give us patterns that reveal circulation anomalies such as El Niño or the North Atlantic Oscillation. Therefore, a small group of scientists from both the climatological and the isotope community has started to enhance the Global Precipitation Network for Isotopes in Precipitation (GNIP), existing since 1961, to find locations that are especially suited to detect circulation anomalies and to help to better transfer isotope information into climate parameters and vice versa (10).

Two climate processes have been considered in particular that lead to abrupt climate change: A major rearrangement of the global thermohaline ocean circulation and the transformation of tropical and subtropical dry savannahs into deserts or even a hyperarid zone, now called the Sahara.

Improved Understanding of Climate Variability or Change

Thermohaline ocean circulation. At present, a major part of the water in the ocean interior had its last contact with the atmosphere up to hundreds of years ago in the Greenland-Iceland-Norwegian seas or the Labrador Sea. The high salinity of North Atlantic water and the cooling near the edge of the sea ice in winter and early spring lead to deep subsidence of dense surface waters. These waters then form North Atlantic Deep Water (NADW), a major portion of the global ocean. NADW reaches all ocean basins as part of the global ocean conveyor belt. In climate history, several events in which this deep convection was stopped abruptly (as revealed from ice cores and deep sea sediments) are known (12), and the strong climate shifts associated with it are documented for the North Atlantic region and beyond. Up to now, modeling of these events has generally been performed with coupled models of intermediate complexity (13, 14) that have been calibrated in their system component modules by higher-resolution AGCMs and OGCMs. The harder test of a higher-resolution fully coupled model (a CGCM) is still not available. Whether current models of intermediate complexity can model abrupt climate change is answered by (15) with a partial yes: "The necessary physics are in these models and allow for thresholds and switches of the thermo-haline circulation. However, their location on the hysteresis now and in the past and the likely evolution in the future are unknown because we do not know whether there are additional stabilizing or destabilizing processes that we must take into account."

The strong interest in thermohaline circulation changes in the past arose with the observation in CGCM runs that deep water formation in the high-latitude North Atlantic would shrink or even stop if there were an enhanced greenhouse effect in the atmosphere (16). However, a mechanism not included in these models may dampen the entire discussion (17). Because most CGCMs used so far for such long-term integrations do not reproduce ENSOs as well as (17), mainly because of higher spatial resolution (0.5° latitude) in the tropics, the earlier studies underestimate increased evaporation in the tropical Atlantic for more El Niño-like events in the transient climate change runs, and thus also underestimate surface salinity in the Atlantic. Increased computing power may help solve this climate change research problem.

Positive vegetation feedback. Vegetation strongly modifies surface energy fluxes as compared to bare soil. Thus, it has the potential to strongly influence regional and global climate. Models of intermediate complexity have recently been used (13) in which vegetation is interactively modifying local, regional, and global climate.

One of the main results is a general enhancement of the monsoons during the warmer part of the Holocene about 6000 years ago, underlining the positive feedback of vegetation and explaining, for example, the dry savannah area in what is now the Sahara desert simply as the result of radiative forcing due to higher insolation in the Northern Hemisphere summer as compared to present conditions. Also, the interaction between earth orbital parameters, ocean, and vegetation can explain strong high northern latitude warming in the Eemian interglacial about 125,000 years ago (13). The main reason for the positive feedback of vegetation is the drastic surface albedo change of up to 50% during the snow-covered period of the year when tundra is replaced by taiga, as well as snow cover lasting into high-insolation springtime at these latitudes.

Projections of Global Climate Change

The Coupled Model Intercomparison Project (CMIP) has helped to assess the performance of about 20 coupled models (18), giving a more reliable range of answers than was known for IPCC's second assessment report (6).

Many CGCMs now show ENSO events; that is, the irregular, interannual climate variability originating in the tropical Pacific, the more so if run with higher latitudinal resolution in the tropics (19). This lends more credibility to models used for projections of climate change because model variability approaches observed climate variability on seasonal-to-decadal time scales (20) that is mainly due to ocean-air interaction. However, because nearly all the models run without variable solar and volcanic forcing, they should not yet fully reach observed variability on time scales of up to decades. On the other hand, low model variability would give high probabilities for the detection of anthropogenic climate change too early.

The sensitivity of model equilibrium to an external forcing cannot be derived from a century time scale CGCM run because the full adaptation of the global ocean to such a forcing takes up to several millennia. Therefore, sensitivity is still derived from so-called equilibrium mixed layer models, in which an atmospheric GCM is reacting to doubled CO₂ concentration and only an ocean mixed layer model fully reacting over decades is coupled to the model atmosphere. In 1999, Le Treut and McAvaney (21) reconsidered 10 such model combinations and found an average warming of 3.3 ± 0.8 K and a precipitation increase of 6.3 ± 3.6% for a doubled concentration of CO₂ (2 × CO₂). Compared to IPCC's second assessment report (6), mean temperature sensitivity decreased slightly from 3.8 K (for 17 models) to 3.3 K (for 10 models). Therefore, it is unlikely that this estimate will change greatly in the upcoming third assessment report of IPCC. The large range can only be reduced substantially if two questions are answered: How strong is the water vapor feedback on average? Will clouds, that cool on average now, give up part of this cooling and thus amplify the warming or will they damp? The answer is subject to better observations of the 3D distribution of liquid water, ice, and water vapor that could come only from new satellite sensors (22). Support for the large mean temperature reaction of about 3 K to 2 × CO₂ in model projections of future climate comes from another type of model experiment. Mixed layer ocean models coupled to atmospheric models for the last glacial maximum need a 2 × CO₂ sensitivity of about 3 K to bring ocean surface temperatures near the ones derived from paleo information (7).

There is, as already mentioned, agreement on increased mean global precipitation if the surface of the water planet Earth warms. However, of greater importance are the questions of where precipitation falls, at what rate, and when. An increased precipitation rate over many land areas is a general CGCM result, as is more precipitation in high northern latitudes and the inner tropics. An increased precipitation intensity and longer time periods without precipitation are of major consequence for many rural societies not only because they depend on rain-fed agriculture but also because the infrastructure protecting them against flash floods is often weak. In one model (23), the return period (the time needed on average for another extreme event of the same magnitude) for the present-day 20-year extreme of daily precipitation would shrink nearly everywhere and could reach 10 years over North America in a 2 × CO₂ climate.

If climate change were reducing the variability of precipitation and temperature, a mean global warming at the surface as the consequence of an enhanced greenhouse effect would not intimidate many people. Because variability changes are more important than shifts in mean values, the width and the shape of probability distribution functions must be assessed in climate change scenarios. As Fig. 1 shows, even a stable shape of the distribution function must cause new extremes on one side when the mean value shifts. Because our infrastructure is normally not adapted to these new extremes, dikes and dams could break more often. However, if the distribution function broadens--that is, the standard deviation grows--even more new investment in better infrastructure would be needed, and so-called natural disasters such as flooding and drought would more often be human-made. For precipitation, most models (6, 21, 24) and observations show a broadening of the distribution function, leading to more frequent major precipitation events. Because most impact studies (for example, those treating agricultural yield) do not include a changed precipitation rate distribution that is shifted to more extreme single events with nearly none or a moderate increase in total amount, they need to be repeated.

We also still have to wait for answers to the following questions on weather extremes in a changed climate: Will northern mid-latitude storms intensify, and how will their main tracks shift? Will tornadoes and thunderstorms become less frequent but more violent or vice versa? Will tropical cyclones be less frequent but more intense if the ocean surface warms in the tropics?

Regional Modeling

An impressive example of the possibilities that regionalization of GCM results opens up was recently given (26). Statistical downscaling was used to convert large-scale circulation parameters from the global European Centre for Medium Range Weather Forecasting (ECMWF) reanalysis (4) into local meteorological variables in a Scandinavian mountain area. The resulting parameter values compared well with local observations. Then these local variables were derived again, but from a 10,000-year run of an AGCM coupled to a mixed layer ocean, and were fed into a glacier model simulating the mass balance and length of several glaciers over thousands of years. The conclusion of this study is, as demonstrated in Fig. 2, that the length variations of the Nigaardsbreen and Rhone-Gletscher glaciers were outside the internal variability range only for the recent major retreat since 1850. But all fluctuations, including those during the Little Ice Age in Europe, as partly recorded for the two glaciers mentioned, were not significantly different from mere natural internal climate variability.

Detection of Change and Attribution to Causes

Has the situation changed? The answer is yes because improved data and methods can be reported now:

1) CGCMs have been run with forcings by solar and volcanic variability (27), different greenhouse gases, and sulfate aerosols. All results point to an anthropogenic contribution, especially in the latter part of the 20th century.

2) Both paleoclimatic reconstructions of the past millennium and new estimates of climate variability in CGCMs show that the observed average warming in the 20th century is significantly above natural variability (20, 28).

3) Several studies applying fingerprinting with both fixed and time-dependent multiple signal patterns (29) to CGCM runs with natural plus anthropogenic forcings conclude that only the combined action of greenhouse gases and tropospheric sulfate aerosols can explain the observed record, especially during the recent decades.

Obviously, the uncertainties surrounding the detection of climate change and the attribution of the observed change to certain forcings have shrunk since 1995.

Remaining Uncertainties

For both abrupt climate change processes we have at least a qualitative understanding, [that is, we can model the principal features of the events (31) or processes], but concerning NADW formation, we do not know how near we are to such an event or whether we are driving toward it (32). Another major uncertainty lies in the broad range still given for climate sensitivity to long-wave radiative forcing by increasing greenhouse gas concentrations in the atmosphere. I see no indication that the next IPCC assessment will strongly reduce the range given in 1996 (6): that the full reaction of the climate system to a doubling of CO₂ (from 300 to 600 ppmv) will lie between a 1.5 and 4.5 K increase in global mean near-surface air temperature. Unless we have measurements of the vertical profiles of liquid water, ice, and aerosols in the atmosphere, we will not be able to improve cloud parameterizations used in CGCMs to such a degree that a significantly smaller range would emerge. We may see better CGCM simulations due to strongly improved cloud parameterizations in about 5 years when new satellite sensors will allow profiling in the atmosphere (22). However, if one looks at the response of CGCMs to a transient forcing, differences between models are smaller.

A further major uncertainty remains the substantiation of the indirect aerosol effect on clouds in models (33); that is, the change in optical and precipitation properties of clouds caused by a changed spectrum of cloud condensation nuclei. This indirect effect is mainly due to the emission of precursor gases such as sulphur dioxide (SO₂), nitrogen oxides (NO + NO₂), ammonia (NH₃), and hydrocarbons, which are transformed chemically into small aerosol particles that grow by coagulation into the cloud condensation nucleus range. But directly emitted black carbon (soot) particles and other carbonaceous particles that absorb solar radiation are also important (34). Typical shifts from maritime to continental water clouds (35), which mimic a pollution effect (Fig. 3), could reduce cloud albedo for clouds exceeding a vertical extent of about a kilometer, counteracting the cloud albedo increase for less thick clouds.

Looking at global climate evolution from a very long-term perspective, it is surprising that despite major glaciations, an Earth mostly without continental ice sheets, a sun with increasing luminosity, and a 10-fold variation in atmospheric CO₂ content mean surface temperature has remained in comparably narrow bounds of about ±5 K as compared to the present mean. We need to understand the negative feedback that stabilizes climate and thus keeps Earth a living planet.

Outlook

The improvement process for climate models that is needed for such applications will continue as it rests on pillars needed for other purposes. These pillars, roughly ordered according to their strength, are:

1) More and as well as more precise global observations of the composition, thermodynamic structure, and dynamics of the atmosphere as well as of ocean and land surface parameters through satellite remote sensing, partly offsetting the often shrinking in situ network. These data sets (36) are ideal for model evaluation and will soon also allow the testing of climate model performance in an event-based mode, not only for time averages.

2) Improved parameterizations of physical and chemical processes in the atmosphere and at the global surface, especially for clouds and vegetation. Examples of where we need strong improvements are in determining mean cloud albedo as a function of the amount of liquid water or ice in a model grid volume, depending on aerosol type and loading, and determining evapotranspiration from a mixture of surface types in complex terrain.

3) The growth of computing power by about an order of magnitude every 6 years, if current trends continue. More oceanic and atmospheric processes will thus become resolved and need no longer be parameterized.

4) Assimilation of all, including asynoptic, observations into forecast or climate models that not only assign values to grid points without nearby observations and discard observations exceeding prescribed error limits but also create a physically consistent starting field for a forecast or a validation data set for a climate model. This holds for land surfaces, the atmosphere, the upper ocean, and the cryosphere.

5) More sophisticated numerical techniques that need less computer time despite improved descriptions of advection and diffusion, thereby not requiring 16 times more computing power if the grid size is halved, but only about 10 times more.

CGCMs will become a primary tool delivering policy-relevant information to many types of decision-makers, including governments. We scientists should create networks of climate research centers across national borders and intensify cooperation with operational weather and climate forecasting centers in order to accelerate progress in understanding the functioning of the Earth system and to better exploit the possibilities for disaster prevention and management. The existing Global Change Research Programmes (35) need not only to cooperate, as they do already to a large extent, but they need the infrastructure to do so effectively. This cooperation and networking would facilitate worldwide dissemination of information and foster further progress.

REFERENCES AND NOTES

1. J. Charney, Fjörtoft, J. Neumann, Tellus 2, 237 (1950) .
2. N. Philipps, Quart. J. R. Meteorol. Soc. 82, 123 (1956) .
3. The longest reanalysis over 40 years stems from the National Center for Environmental Predictions as described in E. Kalnay, et al., Bull. Am. Meterol. Soc. 77, 437 (1996) [ISI].
4. A 15-year reanalysis by ECMWF [J. K. Gibson et al., ECMWF Reanalysis Project Report Series 1, (ECMWF, Reading, UK, 1997)] is something of a forerunner to a 40-year reanalysis from the same center to be available in 2002.
5. An interesting example is global seasonal rainfall forecasting with a CGCM depending on new ocean observations across the tropical Pacific by the so-called TOGA-TAO array of moored buoys implemented during the Tropical Ocean/Global Atmosphere (TOGA) project of WCRP [see, for example T. N. Stockdale, D. L. T. Anderson, J. O. S. Alves, M. A. Balmaseda, Nature 392, 370 (1998) [ISI] ].
6. IPCC, Climate Change 1995: The Science of Climate Change (Cambridge Univ. Press, Cambridge, 1996).
7. S. Joussaume, et al., Geophys. Res. Lett. 26, 859 (1999) [ISI].
8. J. E. Kutzbach and Z. Liu, Science 278, 440 (1997) [ISI] [Abstract/Full Text]; C. D. Hewitt and J. F. B. Mitchell, Geophys. Res. Lett. 25, 361 (1998) [ISI].
9. W. S. Broecker, Science 278, 1582 (1997) [ISI] [Abstract/Full Text].
10. The implementation plan of the Climate Variability and Predictability project (Technical Document 869, WMO, Geneva, Switzerland, 1998) refers to the need to extend quantitative climate information with high temporal and spatial resolution into the preinstrumental period to answer, for example, the following question: Is the increased intensity and frequency of El Niños in recent decades part of a natural long-term fluctuation or human-made?
11. R. T. Weatherald and S. Manabe [Clim. Change 43, 495 (1999)] give an assessment of the development of soil moisture parameterizations in climate models.
12. W. Dansgaard, et al., Nature 364, 218 (1993) [ISI] .
13. V. Petoukhov et al. [Clim. Dyn., in press] describe their model CLIMBER, with which multiple vegetation cover states, depending on initial conditions (wetter or drier), were found and which has also been used by their group [ V. Brovkin, A. Ganopolski, M. Claussen, C. Kubatzki, V. Petoukhov, Global Ecol. Biogeogr. 8, 509 (2000) ] to assess historical global climate change driven by changes in land use. Pioneering work on the quantification of biosphere-atmosphere feedbacks in climate change is also contained in I. C. Prentice, et al., J. Biogeogr. 19, 117 (1992) [ISI] and Texier et al., Clim. Dyn. 13, 865 (1997).
14. T. F. Stocker and D. G. Wright, Nature 351, 729 (1991) [ISI] .
15. A perspective on the understanding of these events, together with a bibliography and attempts to model it, are given by T. F. Stocker and O. Marchal in Proc. Natl. Acad. Sci. 97, 1362 (2000).
16. See, for example, S. Manabe and R. J. Stouffer, Nature 378, 165 (1995) [ISI] .
17. M. Latif and E. Roeckner, J. Clim., in press.
18. G. A. Meehl, G. J. Boer, C. Covey, M. Latif, R. J. Stouffer, Bull. Am. Meteorol. Soc., in press.
19. M. Latif, J. Clim. 11, 602 (1998) [ISI]. A recent study with the same model (17) showed that the more frequent El Niño-like events that occur with enhanced greenhouse forcing, leading to less evaporation in the tropical Atlantic, could give negative feedback, halting the weakening of the thermohaline Atlantic circulation that is a common feature of CGCMs run in transient mode with rapidly increasing greenhouse forcing, but not resolving ENSO events adequately.
20. R. J. Stouffer, G. Hegerl, S. Tett, J. Clim., in press.
21. H. Le Treut and Mc Avaney, personal communication (2000).
22. On behalf of the climate research community, represented by the Joint Scientific Committee of the World Climate Research Programme, I wrote in June 1996 to all major space agencies recommending that attempts be made to measure cloud water and ice profiles; upper soil wetness; tropospheric wind profiles; the fine structure of the gravity field of Earth; and, if possible, sea ice thickness. The very positive reaction of NASA, the European Space Agency, the National Aeronautics and Space Development Agency of Japan, and the Centre Nationale d'Etudes Spatiales of France will provide such new observation capabilities within the next 5 years.
23. F. W. Zwiers and V. V. Kharin, J. Clim. 11, 2200 (1998) [ISI].
24. G. A. Meehl, et al., Bull. Am. Meteorol. Soc. 81, 4277 (2000) .
25. F. Giorgi and L. O. Mearns, J. Geophys. Res. 104, 6335 (1999) [ISI]; F. Giorgi and L. O. Mearns, Rev. Geophys. 29, 191 (1999) .
26. This information is based on two recent publications: B. K. Reichert, L. Bengtsson, O. Åkesson, J. Geophys. Res. 104, 19071 (1999) [ISI]; J. Oerlemans and B. K. Reichert, J. Glaciol., in press.
27. See U. Cubasch, G. C. Hegerl, R. Voss, J. Waszkewitz, T. J. Crowley, Clim. Dyn. 13, 757 (1997) [ISI] for solar forcing; see also (37).
28. M. E. Mann, R. S. Bradley, M. K. Hughes, Geophys. Res. Lett. 26, 759 (1999) [ISI].
29. Examples are S. F. B. Tett, P. A. Stott, M. A. Allen, W. J. Ingram, J.-F. B. Mitchell, Nature 399, 569 (1999) [ISI] ; G. C. Hegerl et al., Clim. Dyn., in press; T. P. Barnett et al., Bull. Am. Meteorol. Soc., in press.
30. A. Berger, Rev. Geophys. 26, 624 (1988) [ISI].
31. WMO and the United Nations Environment Programme assess ozone depletion regularly. The most recent assessment is published as Report No. 44 in the Global Ozone Research and Monitoring series of WMO, Geneva, Switzerland (1998).
32. See (15) and references (11, 12, 14, 15, 17, 19) in that review for the thermohaline circulation hysteresis.
33. E. Roeckner, L. Bengtsson, H. Feichter, J. Lelieveld, R. Rodhe, J. Clim., in press.
34. The original publication in German [H. Grassl, Strahlung in getrübten Atmosphären und in Wolken, Hamburger Geophysikalische Einzelschriften, Serie B, No. 74 (1978)] has in part been repeated in The Changing Atmosphere, F. S. Rowland and I. S. A. Isaksen, Eds. (Chichester, London, 1988), chap. 11.
35. WCRP, the International Geosphere/Biosphere Programme (IGBP), and the International Human Dimensions Programme (IHDP) on Global Environmental Change already jointly sponsor the Global Change System for Analysis, Research and Training, which establishes research networks in large regions (such as Southeast Asia, northern Africa, and southern Africa) and promotes projects such as Climate Predictions for Agriculture. However, the infrastructure (an international secretariat and international project offices) is weak for IHDP, and IGBP is also weak in terms of project offices. A fourth program, called DIVERSITAS and devoted to biodiversity, does not yet have what one can call a functioning infrastructure.
36. Global satellite data sets are especially useful for establishing parameters for which no in situ equivalent exists, for example, monthly mean precipitation (also over oceans) as derived by the global Precipitation Climatology Project of the Global Energy and Water Cycle Experiment [ G. Huffman, et al., Bull. Am. Meteorol. Soc. 78, 5 (1997) [ISI]]. Ocean surface heat fluxes also belong to this category.
37. The theoretical framework for the optimal fingerprint method was laid by K. Hasselmann, J. Clim. 6, 1957 (1993) [ISI]; K. Hasselmann, Clim. Dyn. 13, 601 (1997) [ISI]; and K. Hasselmann, Quart. J. R. Meteorol. Soc. 124, 2541 (1998) [ISI].
38. L. Bengtsson, E. Roeckner, M. Stendel, J. Geophys. Res. 104, 3864 (1999) .

Abstract of this Article Reprint (PDF) Version of this Article Similar articles found in: SCIENCE Online ISI Web of Science PubMed PubMed Citation Search Medline for articles by: Grassl, H. Alert me when: new articles cite this article Download to Citation Manager

Collections under which this article appears: Atmospheric Science

Volume 288, Number 5473, Issue of 16 Jun 2000, pp. 1991-1997.
Copyright © 2000 by The American Association for the Advancement of Science.