Sensitivity of the ACCESS forecast model statistical rainfall properties to resolution

Abstract

The representation of the statistical properties of rainfall over Darwin is assessed in four versions of the Australian Community Climate and Earth‐System Simulator (ACCESS) forecast model against the dual‐polarization C‐band polarimetric (CPOL) radar observation for one rainy season (November 2014 to April 2015). This is an extended analysis of a companion study assessing the earlier version of the regional ACCESS‐12 km rainfall properties with the same CPOL observations. By comparing four different horizontal resolutions (40, 12, 4 and 1.5 km) of ACCESS, it is shown that increasing resolution results in improved total domain rainfall but does not change the compensation effect of underestimated intensity and overestimated frequency of occurrence. In all versions where the convective parametrization is still turned on, a strong and spurious land–sea contrast with too much rain over ocean not exporting to land is found. This problem is only solved by switching off the convective parametrization in the model (1.5 km resolution). However, in this explicit version, high rain rates are this time largely overestimated, resulting in a similar order of magnitude of the mean daily rainfall bias as when convection is parametrized, albeit with a different spatial distribution. The diurnal peak of hourly rain rates is also found to occur too early (around noon at the maximum insolation) when convection is parametrized, but is correct (around 1500 local time) when convection is explicit. More work needs to be done to assess if these errors are common to most cloud‐resolving models, and to solve these two main issues of excessive rainfall in the explicit version of the model and artificial land–sea contrast in the parametrized version of the model.