"Sizzling 2005 takes fourth place in record books" announced the Herald headline. The national daily temperature was on average 13.1C, the report said, making last year the fourth-hottest year nationally since climate records began in the 1860s. It was topped only by 1971, 1998 and 1999.

But the fine print shows that the averages for these four hottest years differ by a maximum of two-tenths of a degree, and 1971 was only one-tenth of a degree warmer than last year. These amounts are well within the error margin for the data.

The statistics could be re-interpreted to show that the hottest year was 1998, with an average temperature of 13.3C, so you could say that since 1998 there has been cooling. This approach to interpreting climate statistics is what climatologists call data-picking, as in cherry-picking.

Climate is about statistics and the opportunity for new records to be set increases with time because the instrument record of climate is seldom more than about 100 years. In identifying patterns of climate change and variability through statistical averages, return periods and probabilities of occurrence, this is a minute fragment.

Climate scientist and statistician Douglas Hoyt points out that because the probability of establishing a new weather record never drops to zero, every year some region will establish a new rainfall, storm, temperature or other climate record.

Even in the warmest years some locations will set a new monthly mean low-temperature record. Even in the coolest years some locations will have periods of record warmth.

For example, even though last year was one of the warmest globally, Somalia experienced its "first ever" snowfall in May and China entered its coldest winter in 20 years.

It is important to keep in mind is that climate statistics are mathematical constructs.

They are no more than human inventions, the meaning of which is often misunderstood.

For example, it is false to assume that a 100-year heatwave, flood or storm will occur or be exceeded every 100 years. In fact, there is a high probability (63 per cent) of a 100-year event occurring more than once in a given 100-year period.

Moreover, there is a small but significant likelihood that two or more floods that are more serious than a 100-year flood will occur in a given 20-year period.

An example is the two major floods in Greymouth during 1988, where a 13-year and a 36-year flood occurred in the same year.

There is the worry that whatever the statistical norm for climate may be, this will change as climate changes.

Scientists advising the Government's Climate Change Project say most regions are likely to face more varied rainfall patterns through the century and floods will become four times more frequent because rainfall is likely to become more intense.

On the other hand, a great deal of research which, taken together, suggests that extreme climate events may become less frequent and less severe.

American researchers Karl and Easterling report the results of climate models that suggest temperatures in the future will be confined to a tighter range, on average.

The researchers found that most of the increase in global temperatures so far has been occurring during the winter and at night.

If these forecasts are correct, variability in the data will shrink more and reduce the occurrence of extreme hot and cold spells, but only on average.

* Dr Chris de Freitas is an Associate Professor in the School of Geography and Environmental Science at the University of Auckland.