This study evaluated the performance of 25 earth system models (ESMs), statistically and dynamically downscaled to a high horizontal resolution (0.25° of latitude/longitude), in simulating extreme climate indices of temperature and precipitation for 1980–2005. Datasets analyzed include 21 statistically downscaled ESMs from the National Aeronautics and Space Administration (NASA) Earth Exchange Global Daily Downscaled Projections (NEX-GDDP) and dynamically downscaled Eta Regional Climate Model simulations driven by 4 ESMs generated by the Brazilian National Institute for Space Research (INPE). Downscaled outputs were evaluated against observational gridded datasets at 0.25° resolution over Brazil, quantifying the skill in simulating the observed spatial patterns and trends of climate extremes. Results show that the downscaled products are generally able to reproduce the observed climate indices, although most of them have poorest performance over the Amazon basin for annual and seasonal indices. We find larger discrepancies in the warm spell duration index for almost all downscaled ESMs. The overall ranking shows that three downscaled models (CNRM-CM5, CCSM4, and MRI-CGCM3) perform distinctively better than others. In general, the ensemble mean of the statistically downscaled models achieves better results than any individual models at the annual and seasonal scales. This work provides the largest and most comprehensive intercomparison of statistically and dynamically downscaled extreme climate indices over Brazil and provides a useful guide for researchers and developers to select the models or downscaling techniques that may be most suitable to their applications of interest over a given region.

这项研究评估了25个地球系统模型（ESM）的性能，这些模型在模拟1980-2005年极端气候指数的温度和降水的过程中，经统计和动态地缩减为高水平分辨率（纬度/经度0.25°）。分析的数据集包括来自美国国家航空航天局（NASA）地球交换全球每日缩减规模投影（NEX-GDDP）的21个统计缩减的ESM，以及由巴西国家空间研究所生成的4个ESM驱动的缩减规模的Eta区域气候模型模拟（ INPE）。在巴西以0.25°分辨率对观测网格数据集进行了缩减规模的输出评估，从而量化了模拟观测到的空间模式和极端气候趋势的技能。结果表明，降级产品通常能够再现观测到的气候指数，尽管大多数产品在亚马逊流域的年度和季节性指数表现最差。我们发现，几乎所有缩减规模的ESM的暖期持续时间指数存在较大差异。总体排名显示，三个缩小模型（CNRM-CM5，CCSM4和MRI-CGCM3）的性能明显优于其他模型。通常，在年度和季节尺度上，统计上缩减的模型的总体均值比任何单个模型都能获得更好的结果。