The return periods of extreme rainfall (or temperature) are calculated using the Fisher-Tippett Type I probability distribution. This is a statistical calculation of extreme values based on existing time-series of data. A very simple explanation of this will now be attempted:
In order to calculate the extreme high values of say 24 hour rainfall, the daily rainfall data for a particular station is extracted. The highest 24-hour rainfall for each month, for each year is extracted and ordered in chronological date order. All the data for say January is then fit to a probability curve. Based on this unique curve, a value is determined which will meet the condition of only occurring every 25 years, or 100 years. This value itself may not exist in the initial time series from which the curve was drawn, but is calculated from the time series and uses the trends and characteristics in the time series to come up with a value. An extreme value calculated at a return period of 1:100 years will correspond on the probability curve to a probability of 0.99, 1:25 years to a probability of 0.96 and so on (Probability = 1-1/(Return Period in years)).
In simple terms, a 1:100 rainfall value of say 120mm in January implies that rainfall close to 120 mm in 24 hours can only be expected to occur in January once every 100 years. It is possible, however, for 120mm of rainfall to fall this year and then again in 5 years time, but the likelihood is that this rainfall will then not be experienced again for at least another 150 to 200 years.
The entire process is repeated for each month of the year. The values for the year are calculated by taking the highest 24 hour rainfall for each year and fitting the data to the probability curve. The return values are always higher for the whole year than for any of the individual months because the time series of annual extremes consists of higher values than any of the individual months.