Comparing the skill of precipitation forecasts from high resolution simulations and statistically downscaled products in the Australian Snowy Mountains

Statistically significant improvements to a “Poor Man's Ensemble” (PME) of coarse-resolution numeral precipitation forecast for the Australian Snowy Mountains can be achieved using a clustering algorithm to classify upwind soundings for each day according to one of four classes, and adjusting the precipitation forecasts using a linear regression. This approach is a type of “statistical downscaling”, in that it relies on a historical relationship between observed and forecast precipitation amounts, and is a computationally cheap and fast way to improve forecast skill. For the “wettest” class, the root-mean-square error for the one-day forecast was reduced from 26.98 to 17.08 mm, and for the second “wet” class the improvement was from 8.43 to 5.57 mm. Regressions performed for the two “dry” classes were not shown to significantly improve the forecast. Statistical measures of the probability of precipitation and the quantitative precipitation forecast were evaluated for the whole of the 2011 winter (May-September). With a “hit rate” (fraction of correctly-forecast rain days) of 0.9, and a “false alarm rate” (fraction of forecast rain days which did not occur) of 0.16 the PME forecast performs well in identifying rain days. The precipitation amount is, however systematically under-predicted, with a mean bias of -5.76 mm and RMSE of 12.86 mm for rain days during the 2011 winter.

To compare the statistically downscaled results with the capabilities of a state-of-the-art numerical prediction system, the WRF model was run at 4 km resolution over the Australian Alpine region for the same period, and precipitation forecasts analysed in a similar manner. It had a hit rate of 0.955 and RMSE of 5.16 mm for rain days. The main reason for the improved performance relative to the PME is that the high resolution of the simulations better captures the orographic forcing due to the terrain, and consequently resolves the precipitation processes more realistically, but case studies of individual events also showed that the choice microphysical parameterisation was very important to precipitation amounts. The WRF model is capable of reasonably good forecasts of the sounding “class” for Wagga Wagga, with an accuracy of 80% for the first day and 65% for the third day of the forecast, facilitating the use of the PME downscaling for a number of forecast days instead of only the day of the sounding.

To better understand the dynamical mechanisms that contribute to the enhancement of orographic precipitation across the Snowy Mountains, high resolution (1.33 km) simulations were performed with both the WRF model and the Australian Community Climate Earth-System Simulator (ACCESS) for two case study wintertime storms in the 2011 winter. In addition to assessing the impact on the surface precipitation predictions, the forecasts of synoptic conditions, structures of the orographic clouds, as well as the skill of prediction of supercooled liquid water and isotherm heights were also examined