are most susceptible to change or most likely to provide early warning of impending change. Together, observations and models offer the possibility of differentiating between natural variability and anthropogenic change. They have already provided a sketchy description of the timing and character of natural climate variability, and tentatively identified some explanatory mechanisms. The results highlighted above, which suggest that decade-to-century-scale climate fluctuations over the last few millennia have been as varied and extensive as many of those observed over the last few decades, emphasize the need to distinguish between natural and anthropogenic signals.
Separation of natural climate fluctuation and anthropogenic change will require additional evidence involving a well-chosen combination of modeling studies and observational studies. As models improve, a better data base—one with a longer time span, broader spatial representation, and more climate variables—will permit the verification of the distinct signatures within, or key relationships between the specific components of the climate system that the models reveal. (The absence of such data sets for initialization, diagnosis, and validation already hinders modeling progress in some areas.) Increased collaboration among the designers of observation systems, the data analysts, and the modelers is needed, as well as interaction between modelers working on different scales of space and time.
Sophisticated validation methodologies can bridge the gap between observations and the simulations of statistical and dynamical models. For example, sparse data are typically processed (if only by simple interpolation and gridding) to facilitate comparison with the model output; when model output is subjected to the same processing, one model's results can be compared with another's, or with observational data, within a common statistical framework.
Additional data are needed to supplement and expand the currently sparse and sporadic record of past natural climate variability. Such additional data would ideally reflect an assortment of variables and represent a broad range of collection strategies. In some cases they are already available, but have been under-utilized (for example, the indirect or proxy data described in Chapter 5); in others they need to be obtained through special programs or refinement of existing collection programs. In either case, greater sensitivity to consistent data quality, continuity, and uniform data-management practices will be key.
Proxy Data. A critical source of natural-variability information that can augment current instrumental records and model results is the various proxy indicators of the climate of the past several millennia. Tree rings, corals, ice cores, and ocean and lake sediments (see Chapter 5) are proving invaluable in supplying information over long periods of time at annual or even seasonal resolution. These data are particularly relevant to studies of natural climate variability, because (unlike modern observations) they represent records of climate prior to significant human interference. Because these proxy records' utility to the study of modern climate was discovered relatively recently, we still need to identify new indicators, improve our understanding of existing ones, and hone our skill in collecting them. Increased acquisition, processing, and archiving of such valuable climate data will also be important.
Other proxy indicators that merit more active study and collection are traditional paleoclimate data, geochemical tracer data, and biological data.
— Traditional paleoclimate data (e.g., deep-sea sediment records, ice cores) typically encompass tens of thousands of years or longer; while their resolution is a few millennia to hundreds of years at best, they constitute an excellent data set against which to test a model's ability to simulate climate under conditions significantly different from today's. Successful simulation is critical to establish confidence in the fundamental physics of the models, while eliminating uncertainties related to parameters calibrated against modern conditions.
— Geochemical tracer data provide not only insight into the ocean circulation, but unique information on the rates of gas, heat, and momentum exchange between the ocean and atmosphere.
— Proxy data from modern-day biological indicators—for example, the marine life discussed in Dickson's (Chapter 3) and McGowan's (Chapter 5) papers—often reveal information on distinctive aspects of climate, because of their sensitivity to integrated climate conditions and to nonclimate variables such as species interactions or predation. Despite their complexity, they warrant additional study to see whether their climate signals can be successfully extracted.
Historical Records. Historical records (e.g., travelers' journals, ships' logs, newspapers) offer important information on a variety of climate indicators over hundreds of years. Since much of it remains undiscovered, ''archeological" data searches and associated data administration (e.g., reprocessing and quality control) are required to recover it. Besides providing unique, if sometimes subjective, information on past climate variability, historical data serve as a basis for evaluating records of proxy indicators.
Operational Data. Only operational sources can provide the wide-ranging data sets required to initialize, force, diagnose, and validate models with the long-term consistency and coverage required for climate-change prediction. Some such sets are available from present-day monitoring networks, but they have not always been collected systematically enough to yield the required quality, consistency, and uniformity. For example, the data currently gathered for weather forecasting often lack the