Abstract Statistical downscaling methods have been actively developed to improve the applicability of global climate models’ (GCMs) outputs to impact studies at a local scale. This study presents a multivariate bias correction (MBC) method that preserves GCM-driven climate change signals and the interdependence between hydro-climate variables by combining the quantile delta mapping (QDM) method with distribution-free shuffle approach, refer to as MBCDS. Since MBCDS also employs the Gaussian rank correlation directly, it does not require an iterative process to improve the Gaussianity as used in existing MBC methods. This study evaluated the effects of interdependence on univariate- and multivariate-distribution criteria and hydrologic indicators during a historical baseline period. Furthermore, this study examined potential changes under selected CMIP5 climate projections for the three future time windows. Application of the downscaling outputs to hydrologic simulations showed that the univariate method underestimated snowfall considerably during the snowfall period, which leads to less snow accumulation and subsequently resulting in less spring and summer flows. On the contrary, the multivariate methods improved all of multivariate-distribution criteria, hydrologic indicators, and the ability in reproducing snowfall depending on multiple hydro-climate variables (i.e., precipitation and temperature). Moreover, the univariate method projected less increase in winter snowfall compared to those of multivariate methods, indicating that the univariate method may result in biased future conditions. In particular, MBCDS showed better and comparative performance compared to the existing methods with regard to climatic and hydrologic performance measures and considerably improved the computational time, with up to 53% less computing time required than those of other existing methods which include an iterative process. Therefore, this study proved that MBCDS can be a good alternative to provide high-resolution climate projections for impact studies at local scales. Further, the study also demonstrated that consideration of the interdependence between hydro-climate variables using multivariate methods is essential to avoid erroneous assessment of climate change impacts for snow-dominated watersheds such as those in Alberta, Canada.

中文翻译：

---

单变量和多变量统计降尺度方法对加拿大艾伯塔省气候和水文指标的影响

摘要 统计降尺度方法已被积极开发，以提高全球气候模型 (GCM) 输出对局部尺度影响研究的适用性。本研究提出了一种多变量偏差校正 (MBC) 方法，通过将分位数增量映射 (QDM) 方法与无分布混洗方法相结合，保留 GCM 驱动的气候变化信号和水文气候变量之间的相互依赖性，称为 MBCDS。由于 MBCDS 也直接采用高斯秩相关，因此不需要像现有 MBC 方法中使用的那样通过迭代过程来提高高斯性。本研究评估了历史基线期间相互依赖对单变量和多变量分布标准和水文指标的影响。此外，本研究在选定的 CMIP5 气候预测下检查了未来三个时间窗口的潜在变化。将降尺度输出应用于水文模拟表明，单变量方法大大低估了降雪期间的降雪量，导致积雪减少，从而导致春季和夏季流量减少。相反，多元方法改进了所有多元分布标准、水文指标以及根据多个水文气候变量（即降水和温度）再现降雪的能力。此外，与多变量方法相比，单变量方法预测冬季降雪量增加较少，表明单变量方法可能导致未来条件有偏差。特别是，MBCDS 与现有方法相比，在气候和水文性能测量方面表现出更好的比较性能，并显着改善了计算时间，与包含迭代过程的其他现有方法相比，所需的计算时间减少了 53%。因此，本研究证明 MBCDS 可以成为为局部尺度的影响研究提供高分辨率气候预测的良好替代方案。此外，该研究还表明，使用多元方法考虑水文气候变量之间的相互依存关系对于避免错误评​​估气候变化对加拿大艾伯塔省等以雪为主的流域的影响至关重要。