Accurate forecasts of rainfall are essential for partner organisations, such as the Environment Agency for flood warnings and the Highways Agency for road safety.
Unlike temperature, rainfall might or might not happen. Rain can be highly variable, especially when falling as showers; one side of a town might get rain while the other remains dry. So for the purpose of verification, forecasts of rain are expressed in terms of the 'chance' (probability) of happening. This also means that probabilistic forecasts can never be completely wrong or right. Two rainfall totals are considered; broadly speaking they represent the chance of 'light' and 'moderate' daily accumulations, for two days ahead.
We use a verification method that:
- Compares the probability of the rainfall occurring to what actually happens.
- Compares this to how difficult or easy it was by simply forecasting a constant long-term frequency of such events.
High scores are achieved when:
- A high probability of rain is forecast and rain subsequently occurs.
- A low probability of rain is forecast and it subsequently remains dry.
Conversely, low scores are achieved when:
- A high probability of rain is forecast and it subsequently remains dry.
- A low probability of rain is forecast and rain subsequently occurs.
The best results are achieved when the forecast probability is significantly different from the long-term frequency of events, and the forecast is correct.
The BPM (which is a combined Brier Skill Score over many UK sites) enables us to track forecast improvements over time. A positive Brier Skill Score indicates that the forecasts are, on average, better than simply forecasting a constant long-term probability for each site based on determining the frequency of such events over time. A completely accurate forecast would have a score of 1. Clearly forecasting rainfall at specific locations is extremely challenging.
Last updated: May 12, 2011 12:43 PM