I'm sometimes frustrated by publicly available precipitation forecasts, such as those offered by the Met Office or Google. If I understand these forecasts correctly, they tell me:

  • The expected volume of rain (e.g. 5ml of rain)
  • The overall likelihood of there being any rain (e.g. 50% chance of rain)

This is generally a sufficient level of data, but sometimes it would be useful to see a more statistically nuanced prediction, which could let me know things like:

  • The 95% confidence interval of the expected volume of rain
  • A probability-density of rainfall volume
  • Maximum/minimum possible volume within a given interval of confidence

Is there any rainfall data of that sort available for the UK?

  • $\begingroup$ Two questions: Does the source of the data matter? And does the form of the data matter? I think I have an answer, but it is a rather raw solution, without much human interpretation (and not necessarily unique to the UK) that will take a bit of programming and may not be in a form that you expect. $\endgroup$ May 21 at 21:38
  • $\begingroup$ Source and form don't matter - I'm interested in anything that's out there. I've got a bit of programming/API savvy so don't want to narrow the scope $\endgroup$ May 21 at 21:49

Well there is ensemble data that you can use. For the UK, you can probably use the ECMWF ensemble prediction system data or GEFS from the US NWS. Both are global models, so they can be extended beyond their respective centers.

These ensembles can be used as samples of the probability density function. Normally (pun unintended), with these types of ensembles, the pdf can be assumed to take the form of a Gaussian distribution per ensemble filters. Obviously, this is problematic, as it implies a probability that there is negative precipitation, which is unrealistic. However, if you want to be clever, you can probably fit a kernel to the data. Or you could choose your own distribution. The distribution that is most widely used is the Gamma distribution, though the Pearson distribution has been shown to perform well too. Perhaps using the Kolmogorov-Smirnov test could help you determine which distribution is best.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.