A hybrid parametrisation for precipitation probability of exceedance data

Bureau of Meteorology Research Report


Precipitation probability of exceedance (PoE) curves are graphs that represent the odds that rainfall will exceed a given amount on a given day. They allow users to understand the likelihood of possible rainfall amounts at their location in a coming period. This is useful for a wide range of users including agricultural producers and planning authorities.


These graphs can be constructed based on historical data or based on ensemble forecasts. Ensembles are a set of several forecasts that allow for the odds of different potential outcomes to be investigated, typically the Bureauís ensembles are made up of 99 model outputs. Given issues with sampling error, model uncertainty, measurement error and model error, particularly when sample size are small, the data is often fit to smooth curves. Fitting these curves becomes problematic when there are instances of zero precipitation.


Researchers at the Bureau of Meteorology have demonstrated the benefits of using a method to fit curves even when there are zero values. The method is shown for 556 stations and 23 regions in Australia on timescales from weeks to seasons. The fit failure rate associated with the PoE curves fitted using this method was 1.51% for stations and 2.91% for regions. Across locations, the mean error was 0.91% for forecast data and 1.4% for climate data. This is an accurate method for most sites, has relatively low computational requirements, and avoids overfitting while providing an easily interpreted smoothed curve.