Multi-model ensemble forecasts of heavy rain events in Australia

Heavy rainfall associated with tropical cyclones, mesoscale convective systems, and slow moving low pressure systems, is a major source of flooding in Australia. To predict heavy rain events forecasters at the Australian Bureau of Meteorology may make use of gridded quantitative precipitation forecasts (QPFs) from the Bureau's global and regional numerical weather prediction (NWP) models, as well as QPFs from several other operational centers. While the models generally agree on the existence of a large rain system, they rarely agree on its precise location and rainfall intensity.

Several recent studies have shown that an ensemble mean forecast of tropical cyclone position constructed from several independent forecasts is generally more accurate than the individual forecasts in the ensemble. Combination and verification of 28 months of QPFs from seven operational NWP models from Australia (2), the United States, ECMWF, the United Kingdom, Germany, and Japan indicate that their ensemble mean gives a more accurate forecast of the location of Australian heavy rain events than do the individual models. The RMS errors and spatial correlation with the observations are also improved. The averaging process increases the forecast rain area and reduces the maximum rain intensity. However, with a reasonable guess for the area bias a linear transformation of the rain rates can be applied to simultaneously reduce the area bias and increase the maximum rain rate.

Another advantage of the ensemble approach is that is allows the probability of precipitation to be estimated. The multi-model ensemble shows probabilistic skill for the first 24-hour period that is comparable to or slightly better than that of large single-model ensemble prediction systems. The usefulness of the multi-model ensemble for rainfall prediction is limited to the first few days of the forecast period, after which the large dispersion of the models degrades the forecast.