资源环境科技发展态势分析平台

Climate impact assessment and decision making in the light of projected future climate change require accurate and robust climate scenarios at the local scale. The latter are targeted by statistical bias correction (BC) and downscaling of climate model output. A nowadays well-established technique is quantile mapping (QM). Here, we apply several different implementations of empirical QM, a parametric (and computationally more expensive) QM method for daily precipitation, and a mean-BC to an ensemble of regional climate scenarios over the topographically structured terrain of Switzerland. The performance of these methods in the current climate and their long-term stability are analysed with respect to distributional and temporal statistics as well as climate impact indices for daily temperature and precipitation. We select an optimal QM implementation, study its effect on the inter-variable consistency, and compare its climate change signal (CCS) to that of a delta-change (DC) method. The results demonstrate that QM effectively reduces raw model biases. The most important improvements are corrected magnitudes and, for precipitation, also wet-day frequencies. The quantile-dependent bias removal is superior to the mean-BC with respect to distribution-tail statistics. Temporal statistics and climate impact indices are also improved. There is no performance benefit from the parametric QM method. The selected empirical QM implementation substantially improves the joint temperature-precipitation distribution and maintains the temperature-precipitation cross-correlation function as represented by raw climate model data. The CCS analysis reveals the superiority of QM over the DC method with respect to distribution-tail characteristics and temporal statistics. This work reveals empirical QM as a reliable and stable method for BC and downscaling from state-of-the-art regional climate models to local weather stations over alpine terrain, confirming and expanding previous studies.