PALAEOCLIMATIC data -- geological evidence of climatic change contained in fossils, rocks, sediments and ice -- play a crucial role in studies of atmospheric changes and their effect on climate because it is the only empirical evidence of what happened in the past.

Such data serve as a yardstick to check the predictions on the basis of climate models made by scientists. Such models are immensely complex because of the large number of parameters involved and require the use of supercomputers to evaluate climatic changes.

One example of how geological data can provide an insight into climatic processes comes from a study of what are called Milankovitch cycles -- the periodic variations in Earth's orbit and orientation with respect to the Sun. Variation in the cycles can change the intensity of seasons in each hemisphere and are considered an important factor in climate evolution.

As a result of these cycles, the Northern Hemisphere has warmer winters and cooler summers now than about 11,000 years ago. This picture was tested by an interdisciplinary research group, COHMAP, by looking at the predictions of climate models for orbital parameters corresponding to the period 6,000-12,000 years ago. The expected pattern of cooler winters and warmer summers was observed. In addition, the model predicted strong summer monsoons that would cause higher lake levels in the region that is now the Sahara desert. All these effects are indeed present in the geological record, providing striking confirmation of the model.

The geological record can also alert scientists to the existence of unusual climatic phenomena unaccounted for in their existing models. One such phenomenon, which occurred about 10,000 years ago, is the Younger Dryas, an abrupt cooling that led to a steep temperature fall in the north Atlantic Ocean, Europe and Greenland. The Younger Dryas is regarded as a sign of instability in the climate system, and was discovered only by study of the geological record.

The geological record sometimes contradicts the predictions of models, thereby pointing to their limitations. For example, most climatic models predict that tropical oceans should share in the global changes in climate, warming up during hot periods and cooling down when temperatures are lower worldwide. However, these conclusions are not borne out by palaeoclimatic data.

Studies of the diversity and abundance of marine organisms called foraminifera indicate that during the Last Glacial Maximum, 18,000 years ago, when large parts of the Earth were covered with ice, the oceans were only about 2 degrees cooler than they are today. Studies of isotopes show that during warm periods like the middle of the Cretaceous era, 100 million years ago, the tropical oceans were only a few degrees warmer than now.

This inconsistency is puzzling because most models predict far higher ocean temperatures in this epoch. "It's a testable prediction and we're not seeing it," says Thomas J Crowley of the Applied Research Corporation in College Station, Texas. "If we hadn't looked in the geological record, we wouldn't even think this prediction could be wrong."

Many scientists believe such paradoxes are the most important contributions of the palaeoclimatic data because they challenge assumptions and lead to important modifications of theory.