- It is not a physically-based climate model
- Quality of ouput depends on quality of input data
- Translating climate projections into time series introduces an additional source of uncertainty
- Potentially large resource requirement
- Validation is crucial
- Outputs are not spatially consistent
- Estimation of some types of extreme events not necessarily possible
- Weather Generator output is not a weather forecast!
- Poor representation of large scale climate processes
The UKCP09 Weather Generator learns about daily climate variable relationships from observed climate data and uses what has been learned to develop statistical relationships. Although these relationships can be interpreted with some physical sense, there is no explicit basis in physics or meteorology within the UKCP09 Weather Generator methodology. This makes the UKCP09 weather generator quite distinct from global climate models, which provide a mathematical representation of the processes that are understood to govern the climate system. This means there is no guarantee that the UKCP09 Weather Generator will always reproduce correct daily or hourly behaviour. What are produced are time series which are indicative possible and plausible realisations of the daily climate (future and baseline periods).
The time series generated by a weather generator can only be as good as information put in and the statistical relationship that are its basis. This means the UKCP09 Weather Generator output is reliant on the quality of both the observed climate information (used to allow the Weather Generator to learn about baseline climate) and the UKCP09 projections (used to perturb the baseline climate to produce plausible future time series). The observed climate information is derived from a network of climatological and meteorological stations [194kb] across the UK. There are uncertainties associated with this data, arising from among other sources the uneven spatial coverage (relatively few stations are located in northern and western parts of the UK) and the lack of availability of some desirable information (relatively few stations [164kb] provide hourly observations. As a result, it is important to validate the Weather Generator output relative to the observed data (comparing statistical nature of the generated time series with that of the observed baseline period) before using the generated outputs from future time periods (see the Weather Generator report for more details).
The UKCP09 Weather Generator, like any method used to translate time-averaged climate model projections to local-scale time series, adds an additional source of uncertainty which is difficult to quantify. The implications of these uncertainties can be explored through uncertainty analyses using the set of time series generated.
The UKCP09 Weather Generator will produce 100 simulations of 30 years of daily time series for the baseline period and a minimum of 100 time series for a future time period. Depending on choices made when configuring the Weather Generator run, the files produced could potentially be up to 1 GB in size (when zipped or in binary format). This would typically take about 6 hours to download with a 1 Mb/sec broadband connection. Users should also consider that, to make efficient and appropriate use of the generated time series, there is potentially a large resource requirement, both in terms of understanding what the outputs provide and in terms of the further computation needed to store, process and analyse these time series within an assessment.
All Weather Generator outputs should be validated (i.e. statistics of time series compared with those for the baseline observed data) to understand how reliable the outputs are. Only by comparing output for the baseline period with observed statistics can users truly evaluate how future simulations are likely to be. Some validation is provided in the Weather Generator report.
The weather generator outputs are produced for a single location (5 km grid square). As a result of the stochastic nature of the underlying methodology, the sequencing of the daily (and hourly) values at neighbouring locations will not necessarily be consistent on a day-by-day or hour-by-hour basis – not spatially consistent.
This lack of spatial consistency in separately generated time series means that if output is generated for neighbouring locations (i.e. adjacent 5 km grid squares), the generated variables within the two generated series will not necessarily be consistent (e.g. at any given Weather Generator day (or hour) it might be raining at one location and not the other). The statistics and other characteristics of the adjacent generated climates may or may not be consistent depending on the homogeneity of the baseline climate and of the projected changes (including whether or not the adjacent locations are using the same 25 km projected climate)
As such, Weather Generator output should not be used where spatially-distributed climate information is required involving multiple locations. In such cases, the 11-member RCM output might be more appropriate.
Since the UKCP09 Weather Generator allows users to produce long time series (e.g. 1,000 years of daily output), it is possible, in principle, to investigate long events (such as the incidence of long heatwaves or a 1 in 100 year rainfall event). In practice, the UKCP09 Weather Generator has been trained using a much shorter record of climate observations (35 years; 1961–1995) which does not necessarily include such long return period events. It is unreasonable to assume that these extreme events can be accurately simulated in future time series, regardless of how many years of output are simulated. Using the example of cars on a motorway, if no yellow cars pass a particular point in a five minute period, it is difficult to predict how many yellow cars would pass the same point in a 30 minute period in subsequent days.
Despite what its name might suggest, the purpose and design of a weather generator is not to provide a weather forecast for the future. A weather forecast gives an indication of what the weather is predicted to be on a particular day. A Weather Generator provides a multiple plausible daily all of which are (statistically) consistent with both the baseline climate (1961–1995) and with the UKCP09 probabilistic projections of future climate change. It allows users to generate many different, but statistically equivalent time series that can be used to evaluate different eventualities and possible response strategies/measures using risk management approaches.
The series generated may be thought of as sequences that closely mimic the characteristic of real daily and hourly climate that could happen but have not or most certainly will not actually be observed.
The UKCP09 Weather Generator output is produced using the statistical relationships of climate observations with time-scales ranging from one day to one month as captured by a 30-year baseline climate period. This means that (natural) regional scale climate fluctuations related to for example, the (NAO), or El Niño Southern Oscillation (ENSO) events, are not explicitly represented. Consequently, there is little or no incorporation of seasonal (such as that produced by changes in sea-surface temperature which can influence warm or cold winters) or annual (such as multi-season dry spells like the 2004 to 2006 drought in southern England) climate variability in the generated time series from these sources. The weather generator is only capable of reflecting seasonal and annual variability consistent with whatever is in the baseline observed climate data set.