On the long-term memory characteristic in land surface air temperatures: How well do CMIP6 models perform?

Stemming from the multi-scale interactions of various processes, long-term memory (LTM) has become a well-recognized property in the climate system. Whether a dynamic model can reproduce the observed LTM is a widely used criterion for model evaluation, especially regarding its ability in simulating natural variabilities. While many works have shown poor model skill in simulating the LTM of land surface air temperature (LSAT), it is not yet known whether CMIP6 models offer any improvement. In this study, the performances of 60 CMIP6 models in simulating the LTM characteristics in LSAT were evaluated. Results showed that most models reproduced the LTM in the global-mean LSAT, among which AWI-ESM-1-1-LR and E3SM-1-0 performed best. All 60 models reproduced the variation in LTM with latitude. CNRM-CM6-1 and HadGEM3-GC31-LL performed best in simulating the LTM of LSAT at the global scale. The multi-model mean (MMM) performed better than any single model. The biases of the MMM and CRUTEM5, and among the 60 models, were significant in the equatorial and coastal regions, which may be attributable to the simulation differences of the models in terms of their ocean–atmosphere coupling processes.摘要利用去趋势涨落分析 (DFA) 方法计算序列的长程记忆性 (LTM) , 以CRUTEM5数据集的结果作为观测参照, 评估了60个参与第六次国际耦合模式比较计划 (CMIP6) 的气候模式对地表气温LTM的再现能力. 结果表明: 大部分模式可以再现全球平均地表气温序列的LTM特征, 其中AWI-ESM-1-1-LR和E3SM-1-0的模拟效果最好; 60个模式均能模拟LTM随纬度带的变化; 综合来说, 全球水平上CNRM-CM6-1和HadGEM3-GC31-LL对地表气温LTM的模拟性能最好; 多模式平均相比单一模式模拟性能更好; 多模式平均与观测结果的偏差以及模式之间的模拟差异显著体现在赤道和沿海区域, 这种偏差可能源于模式对海气耦合过程的模拟差异.