served global-mean surface air temperature, from interannual to interdecadal time scales. However, it fails to reproduce the warming trend of centennial time scale (i.e., ~0.5ºC per century) that has been observed since the end of the last century. If the model is assumed to be realisticin spite of its failure to reproduce the quasi-periodic components of the natural variabilitythis result suggests that the observed centennial-scale warming trend is not generated within the climate system by nonlinear interaction among the atmosphere, ocean, and continental surface. Instead, the trend must be caused by a sustained trend in natural and/or anthropogenic thermal forcing, such as changes in solar irradiance, greenhouse gases, and aerosol loading in the atmosphere. Essentially similar results have also been obtained from the long-term integration of a coupled model at the U.K. Meteorological Office (see, e.g., Mitchell et al., 1995).
Identifying predominant patterns associated with the natural, internally generated climate variability would aid in the detection of patterns of anthropogenic change (see, for example, Barnett and Schlesinger, 1987; Hasselmann, 1993; and Santer et al., 1995). If the effect of sulfate aerosols is considered together with the effect of greenhouse gases in GCMs, the spatial distribution of the model-generated change of atmospheric temperature over the decadal time scale appears to become more realistic (Santer et al., 1996). These and other recent results (see, e.g., IPCC, 1996a, for an overview) are leading to a more reliable estimate of the anthropogenically induced climate change, as well as of the natural variability caused by mechanisms internal to the climate system.
In the future, major effort will need to be devoted to observational and modeling studies of internally generated climate variability, so that this variation can be distinguished from anthropogenic climate change. Records from past observations of both ocean and the atmosphere should be compiled and analyzed for variables such as concentration of greenhouse gases in ice cores, sea-level pressure, surface and subsurface temperature and salinity in oceans, air temperature and humidity at the surfaces, and temperature and geopotential height at selected pressure levels in the atmosphere. It is also essential to improve model parameterizations of various feedback processes, in particular those involving cloud, snow, and sea-ice cover, all of which substantially affect incoming solar and/or outgoing terrestrial radiation at the top of the atmosphere. Other factors of critical importance are cumulus convection and land-surface heat and water budgets. Greater use of data from remote sensing and in situ measurements of radiative emissions and river runoff facilitate evaluating and improving the parameterizations of the important processes identified above.
As was noted in Chapter 4, there are practical reasons for dividing the study of climate variations by the time scales on which they occur. The climate system clearly evolves over a continuum of time scales, however, and no "spectral gap" in nature justifies such a separation. To advance our understanding of overall climate change and variability most efficiently, it is important that we explicitly recognize those processes that cannot easily be categorized by scale, and that we particularly emphasize those mechanisms which affect climate variability and change over a range of time scales.
A few specific examples of climate-variability patterns and their possible causal mechanisms that appear on more than one time scale are (1) the interdecadal variability of ENSO, in amplitude, periodicities, and warm or cold anomaly distribution; (2) the North Atlantic Oscillation (NAO) and its purely atmospheric, purely oceanic, or coupled mechanisms; and (3) changes in the carbon cycle, over land, ocean, and the tropical or mid-latitude or polar regions. The modes of variability that cross two or more time scales can arise either from the intrinsically broad-band behavior in time of a specific spatial mode, or from the nonlinear coupling between narrow-band spatio-temporal modes that share certain regional characteristics. Which one of these overall types of behavior is at the root of a given dec-cen climate phenomenon has important implications for its predictability.
Currently, the national and international organizations devoted to the study of physical climate are structured to address separately the high-frequency variability (GEWEX), seasonal-to-interannual variability (GOALS), decade-to-century-scale variability (CLIVAR DecCen), and millennial and longer-scale variability (e.g., PAGES). Each of these groups has identified a suite of high-priority issues that must be addressed. Many of the detailed processes involved in these issues are common to all four units. For example, improved understanding of air-sea exchanges is of fundamental importance to the study of climate, regardless of time scale. Similarly, the patterns of climate variability and the coupled modes are of equal importance to all groups, because their regional manifestations occur on a broad range of time scales. Issues related to these common processes and patterns warrant particular attention, and a dec-cen program that is highly coordinated with GOALS and GEWEX would enable them to be studied most effectively. Furthermore, the physically based studies of climate must be fully integrated with those investigating the chemical-biological aspects, which are currently being addressed by elements of the International Geosphere-Biosphere Programme.