Hybrid statistical-dynamic downscaling based on multi-model ensembles in Southeast Asia

GCMs under CMIP6 have been widely used to investigate climate change impacts and put forward associated adaptation and mitigation strategies. However, the relatively coarse spatial resolutions (usually 100~300km) preclude their direct applications at regional scales, which are exactly where the analysis (e.g., hydrological model simulation) is performed. To bridge this gap, a typical approach is to ‘refine’ the information from GCMs through regional climate downscaling experiments, which can be conducted statistically, dynamically, or a combination thereof. Statistical downscaling establishes relationships between large-scale climate indicators and small-scale climate variables in the reference (historical) period. Subsequently, these relationships are kept unchanged in the future and used to predict the future variables. On the other hand, dynamical downscaling operates based on the physical processes and the associated interactions in the climate systems and thus can produce a full set of regional climate simulations (e.g., temperature and precipitation fields) that are dynamically consistent. However, traditional dynamical downscaling contains significant biases that are transferred from GCMs and may be enhanced during the process of downscaling, thus degrading the downscaled results. One promising approach to remove these biases is the hybrid statistical-dynamical downscaling method, where GCMs are firstly bias-corrected, and subsequently used as lower and lateral boundary conditions to drive the regional climate models (RCMs).In this work, we apply a hybrid statistical-dynamical downscaling method, following the approach of Xu et al. 2021. The bias-corrected dataset is adjusted to resemble ERA5-based mean climate and interannual variance, and with a non-linear trend from the ensemble mean of the 14 CMIP6 models. The dataset spans a historical period of 1979–2014 and future scenarios (SSP585) of 2015–2100, with a temporal scale of six-hour.The main contributions of this dataset are twofold. First, we provide the open-source and high-resolution (12.5km: Southeast Asia; 2.5km:Southern Malay Peninsula; 500m: Singapore, as shown in the following Figures) datasets, including precipitation, wind, temperature, radiation, etc. Second, through our experiment, this bias-corrected and downscaled dataset is of exceptional quality compared to that of the existing dynamical scaling work (e.g., CORDEX) in southeast Asia in terms of its ability to reproduce regional climate extremes, spatial patterns, etc. This dataset will be useful for policy-makers and researchers to establish the necessary pathways for resilient planning in order to mitigate the dire impacts of climate change.

在CMIP6框架下，全球气候模型（GCMs）已被广泛用于研究气候变化的影响，并提出了相应的适应与减缓策略。然而，其相对粗糙的空间分辨率（通常为100~300公里）限制了其在区域尺度上的直接应用，而区域尺度正是分析（例如，水文模型模拟）所进行的领域。为了弥合这一差距，一种典型的做法是通过区域气候降尺度实验对GCMs的信息进行‘细化’，这些实验可以是统计性的、动力性的或二者的结合。统计降尺度在参考（历史）期间建立了大尺度气候指标与小型气候变量之间的关系。随后，这些关系在未来保持不变，并被用于预测未来的变量。另一方面，动力降尺度基于气候系统中的物理过程及其相关相互作用，因此可以产生一套完整的区域气候模拟（例如，温度和降水场），这些模拟在动力学上是一致的。然而，传统的动力降尺度包含显著的偏差，这些偏差来自GCMs，并在降尺度过程中可能得到增强，从而降低了降尺度结果的质量。一种有前景的方法是采用混合统计-动力降尺度方法，在此方法中，GCMs首先进行偏差校正，随后作为区域气候模型（RCMs）的下边界和侧边界条件进行驱动。在本研究中，我们采用了Xu等（2021年）提出的方法，对偏差校正后的数据集进行调整，使其与基于ERA5的平均气候和年际变率相似，并具有来自14个CMIP6模型集合平均的非线性趋势。该数据集覆盖了1979–2014年的历史时期和未来情景（SSP585）的2015–2100年，时间尺度为六小时。本数据集的主要贡献有两方面。首先，我们提供了开源的高分辨率（东南亚：12.5公里；马来半岛南部：2.5公里；新加坡：500米，如图所示）数据集，包括降水、风速、温度、辐射等。其次，通过我们的实验，与现有的动力降尺度工作（例如，CORDEX）相比，这个偏差校正和降尺度的数据集在再现区域气候极端事件、空间模式等方面的质量尤为突出。该数据集将对政策制定者和研究人员非常有用，帮助他们建立必要的路径，以实现有弹性的规划，以减轻气候变化带来的严重影响。