- To investigate more local scale climate impacts
- To explore some types of extreme events
- To explore a sub-sample of the selected probabilistic projections’ distribution
- To investigate the risk of threshold being exceeded
- To obtain a representative daily climate sequence
- To obtain estimates of hourly time series
Impact assessments of climate change often require more detailed temporal and spatial detail than is provided by the UKCP09 probabilistic climate projections (which provides projections for monthly, seasonal and annual average changes, at 25 km resolution). Outputs from the weather generator are better suited to assess impacts at these finer resolutions, given that they are provided at daily and hourly temporal resolutions and at a spatial resolution of 5 km (or if an area is selected, for an area up to 1,000 km2).
In order to fully explore the range of uncertainty in the , each weather generator simulation is constrained to sample a minimum of 100 model variants of the 10,000 that make up the probabilistic projections. This results in the production of a minimum of 100 weather generator output files for each request made in the User Interface. User should not average the outputs from the generated time series to produce a single time series. Averaging of these time series is not mathematically or statistically sound. The same considerations apply when dealing with the 100 baseline time series.
The UKCP09 weather generator can be used to generate long time series (user-specified of up to 1000 years) that are statistically consistent with a particular average (or stationary – i.e. not varying through time) climate. The weather generator must be run for both the weather generator baseline climate (1961–1990) and for the future climate that is consistent with (and using the same 30-year time periods as) the UKCP09 probabilistic climate projections (sampled data) themselves. This capability can be useful where the recurrence interval of particular events is of interest. For example, the weather generator output can be analysed to examine the recurrence frequency of certain daily rainfall totals.
Whilst there is sufficient generated data in longer time series (100 to 1,000 years) to estimate long and these series are stationary, users should be wary of producing and interpreting these return periods. Extreme statistics for return periods longer than 10 years based on the daily time series should be used with caution and users are advised to carry out uncertainty analysis using the Weather Generated series generated. Hourly extreme statistics are subject to even more uncertainty, and return periods beyond 5 years should be used with caution.
The UKCP09 Weather Generator can be used to explore a specific portion of the selected UKCP09 probabilistic projections. This functionality is possible as the probabilistic projections are available as variants within the that maintain the linkages between variable that were determined together (maintaining internal consistency). In addition, an ID has been associated with each model variant. As such, users have a number of options that can be used to sub-sample the selected probabilistic projections:
- Random sample (default): Selects a user defined set (between 100–1000) of model variant IDs at random
- Sample by model variant ID: Selects a user defined set (between 100–1000) of model variants by user selected IDs for each run. Users must identify a set of model variant IDs (text file) and there must be between 100 and 1000 IDs.
- Sample by variable/probability level/season: Sample a particular part of the distribution that may be of particular interest (e.g. exceedance of a threshold).
The user will not be able to use all of the 10,000 model variants as the maximum number of runs that can be generated in one request within the User Interface is 1000.
Identification of thresholds can potentially provide a useful route into using UKCP09. Where these relate to daily conditions, the weather generator is the most appropriate product to use to analyse the frequency of exceedance both for the baseline period and for the future. For example, the weather generator output can be analysed to examine the risk of daily maximum temperature exceeding a specific temperature (e.g. 30°C). Although the weather generator itself can be used to investigate thresholds, this process is made considerably easier by using the .
Some applications require a typical daily (or hourly) climate sequence. A weather generator is well suited to providing this type of information, though some thought is required about how to handle the weather generator output (typically 100 sequences each containing at least 30 years of daily time series)
Users can also request the Weather Generator provide hourly time series consistent with the generated daily time series and thereby, consistent with the 25 km resolution probabilistic projections. See the assumptions section for more details about how the hourly data is derived.