Weather generator
The weather generator is the bridge between a climate scenario and something a hazard model can run on.
It takes two kinds of input, historical observations of what the weather actually did, and projections of how the climate is expected to shift, and produces future weather that is statistically consistent with both.
Required input data
Four inputs go in. The first three establish observed behaviour; the fourth supplies the future signal.
Rainfall
| Source | HadUK-Grid, plus rain-gauge data from the Environment Agency (EA), Natural Resources Wales (NRW), the Scottish Environment Protection Agency (SEPA) and the Met Office Integrated Data Archive System (MIDAS) |
| HadUK-Grid provides | Daily rainfall at both 1 km and 5 km spatial resolution across the UK |
| Rain-gauge data provides | Hourly records from the four sources above, quality controlled before use |
The two are complementary. The gridded product gives complete national coverage at daily resolution; the gauges give the hourly detail that matters for short, intense events.
Temperature
| Source | HadUK-Grid |
| Provides | Daily maximum and minimum temperature at 1 km and 5 km spatial resolution across the UK |
Secondary weather variables
| Source | Met Office Integrated Data Archive System (MIDAS) |
| Provides | Windspeed, vapour pressure and sunshine duration, at daily timescales |
Future projections of rainfall and temperature
| Source | The Convection-Permitting Model from the suite of United Kingdom Climate Projections 2018 (UKCP18) |
| Provides | Projections at 2.2 km resolution, also made available on a 5 km grid, suitable for analysing local and hourly climate extremes (Met Office, 2019) |
See UKCP18 Local for more on this dataset.
Generated output data
The generator produces stochastic realisations of future weather for rainfall and temperature.
What a "realisation" means here
The generator does not produce one definitive future. It produces many equally plausible versions of it, each consistent with the same underlying statistics but differing in the details of when and where rain falls.
Running a hazard model across a set of realisations, rather than a single time series, is what lets you talk about likelihood instead of just possibility.
Users can request as many realisations as their study needs, then feed them into the relevant impact assessment work. Typical uses include:
- Extreme rainfall totals for a specific return period, such as the depth of a 1-in-100-year event under a future climate
- Temperature extremes, such as how long heatwaves are expected to last
Where the output goes
These realisations are the meteorological forcing for the next stage of the chain. The hazard models take them and turn them into flood depths and heat exposure mapped across a region over time.