Search site

Limitations of Weather Generator 2.0

  • Reliability of hourly data for future convective storms

There are limitations to the rainfall extremes that can be simulated by the UKCP09 Weather Generator, particularly in the case of hourly rainfall extremes. The UKCP09 Weather Generator underestimates summer hourly rainfall because of a simplified equation used to derive hourly rainfall.

Daily information is used to drive the weather generator. In order for the weather generator to generate hourly information it uses a fixed relationship between daily and hourly rainfall statistics. This relationship comes from the observed climate statistics. Using this relationship in the future projections is limiting because it does not allow for a simulation in the increase of convective rainfall events.  Convective rainfall is driven by thermal heating, causing evaporation; therefore, one would expect hotter summers to lead to more convective rainfall. However, in the weather generator, hourly rainfall is obtained from the daily values so it is difficult to project changes in the hourly intensity of rainfall.

The projections for summer show overall decreases in daily rainfall therefore these will be associated with decreases in summer hourly rainfall. This simplification is clearly not consistent with the possibility of an increase in infrequent, but intense convective events. Therefore, in locations such as south east England where summer convection dominates the extreme rainfall, a low confidence should be placed in estimates of future hourly extremes. 

  • Limitations to the representation of extreme rainfall

Although the UKCP09 weather generator provides insight into future projections, caution should be taken particularly when examining future extremes. This is because the UKCP09 Weather Generator uses a record of climate observations (35 years; 1961-1995) as the baseline climatology to “train” the weather generator. Therefore, it is unreasonable to assume that  extreme events, outside of the range of those previously observed, can be accurately simulated in future time series, regardless of how many years of output are simulated.