资源环境科技发展态势分析平台

Convective permitting simulations are increasingly pursued for providing physically more credible climate projections of rainfall. Their value is likely to be greater for regions where increased resolution not only resolves physical processes better, but also the topographic features of the target domain. Here, we assess the skill of convective permitting simulations to simulate rainfall for water resource assessment work in a climate change context for southeast Australia. Output on 2 and 10 km grid-length resolution from a 5-year regional climate model simulation is assessed for skill in simulating mean seasonal climatologies for days with low or high observed rainfall intensities. Comparison is conducted on spatial grids and for 25 catchments across the study region. No significant difference in skill was found in the loss differential when using absolute error for spatial fields of mean climatologies. Measures focusing on spatial similarity and accuracy in position of high rainfall areas indicate somewhat better skill in the 2 km simulation with regard to positioning (in autumn and winter), and with regard to spatial variability (in autumn and spring). Significant difference in skill was shown when comparing the simulated data sets on a catchment basis; seasonally 5-7 catchments in favor of the 10 km output and somewhat less for the 2 km output (3-6 catchments). When using correlation skill as the test measure, results are overwhelmingly in favor of the 2 km output. We cautiously suggest that results may be overly pessimistic for the 2 km simulation because of inadequate representation of rainfall in high altitude areas in observations.