长期的气候观测数据有助于我们了解地球过去的气候变化机制，同时能够对未来的气候变化情景提供预测。由于地表入射太阳辐射（简称太阳辐射，Rs）是地表辐射收支、能量平衡的重要参量，因此长达一个世纪的太阳辐射数据集具有非常重要的研究价值。这样长度的、保存完好的观测序列十分稀少。日本地区的太阳辐射观测数据始于1890年，目前为止具有百年以上的观测记录。围绕该数据集，本研究评估了5个不同的20世纪再分析数据集 (由ECMWF生产的ERA20C、ERA20CM和CERA20C，由NOAA/CIRES/DOE生产的20CRv2c和20CRv3)对Rs的模拟能力，刻画了Rs的时空变化特征；利用集合经验模态分解(EEMD, Ensemble Empirical Mode Decomposition)方法将原始数据分为高频分量 (HFC, high-frequency components)、低频分量 (LFC, low-frequency components)以及趋势项，探究了不同时间尺度下再分析数据再现观测Rs的能力；通过Sum of Tree方法量化总云量偏差和水汽压偏差对Rs偏差的相对贡献率；选择Nino3.4指数（中太平洋地区海表温度的平均异常值，用于衡量El Niño和La Niña的强度）、PDO指数（北太平洋年代际海表温度模式变化的主要指标）和东亚大槽位置指数（衡量东亚大槽纬向位置变化的指数）三项气候指数，比较观测数据与再分析数据Rs对气候指数的响应特征，进一步探究20世纪再分析太阳辐射数据产生差异的原因。总结研究内容得到以下结论：
在百年时间尺度上，日本地区的Rs呈现“上升-下降-上升”的趋势，其空间分布信号从北向南逐渐增强。再分析数据CERA20C能有效捕捉到Rs时间序列和空间分布的变化特征。从观测数据和再分析数据Rs和总云量相关性的空间分布图来看，再分析数据Rs模拟能力的改善可能是由于总云量模拟能力的改善。
在五种再分析数据中，CERA20C与观测Rs在HFC (0.76-0.86)和LFC (0.59-0.95)上具有显著的相关性。尽管CERA20C在一定程度上能反映出观测的Rs在HFC和LFC的信号，但其趋势项的表现与观测Rs趋势相反，导致CERA20C无法准确反映Rs的趋势。相比之下，20CRv2c和20CRv3表现出了较好的趋势项模拟能力。
在HFC中，除20CRv2c外，再分析数据中的总云量偏差能解释71%~95%的Rs偏差。在LFC中，ERA20C和20CRv2c的总云量偏差对Rs偏差的解释率高于水汽压偏差；除了CERA20C，总云量偏差能解释1931-2010年58%~68%的Rs偏差。在原始序列中，总云量偏差能解释1931-1960年和1931-2010年ERA20C、20CRv2c和20CRv3太阳辐射偏差的63%~92%，水汽压偏差能解释ERA20C和CERA20C 太阳辐射偏差的50%~77%。总体而言，Rs对总云量的变化表现出高敏感性。除20CRv3外，再分析数据总云量在1961-2010年低频分量LFC模拟结果较差 (-0.59~0.09)，这很可能是再分析太阳辐射数据在低频分量LFC模拟结果不佳的原因。
日本地区Rs对三项气候指数的响应存在显著的季节性差异。结果表明， Rs对ENSO敏感的季节为夏季和冬季，对PDO指数敏感的季节为春季，对东亚大槽位置指数变化敏感的季节为秋季。在再分析数据中，CERA20C和20CRv3能捕捉到气候指数正负异常时500hPa位势高度场的变化情况，但CERA20C很少能捕捉到气候指数正负异常变化时Rs的变化特征；ERA20CM几乎无法捕捉到气候指数正负异常变化时Rs的情况，可能是因为其未能模拟出相对应的500hPa位势高度场特征。
 

Long-term climate observations can help us understand the mechanisms of past climate change on Earth and provide clues for predicting future climate change scenarios. The century-long dataset of surface solar radiation (Rs) is of great research value because it is an important parameter for surface radiation balance and energy balance. Such a long and well-preserved observation sequence is rare. Solar radiation observation in Japan began in 1890 and has more than a hundred years of observation records to date. In this study, we evaluated the modeling ability of five different 20th-century reanalysis datasets (ERA20C, ERA20CM, and CERA20C from ECMWF, and 20CRv2c and 20CRv3 from NOAA/CIRES/DOE) for Rs, explored the spatiotemporal characteristics of Rs, and used the Ensemble Empirical Mode Decomposition (EEMD) method to decompose the original data into high-frequency components (HFC), low-frequency components (LFC), and trends to investigate the ability of reanalyses to reproduce observed Rs at different time scales. By selecting three climate indices, namely the Nino3.4 index (an average anomaly value of sea surface temperatures in the central Pacific region used to assess the intensity of El Niño and La Niña events), the PDO index (a key indicator of decadal sea surface temperature variability in the North Pacific), and the East Asian trough position index (a metric quantifying the latitudinal displacement of the East Asian trough), we conduct a comparative analysis of the response characteristics of observed and reanalysis data to these climate indices. Our investigation aims to further elucidate the underlying factors contributing to the disparities observed in solar radiation data derived from 20th-century reanalysis efforts. In summary, the research findings can be succinctly summarized as follows:
On a centennial timescale, the Rs in Japan shows an "increase-decrease-increase" trend, with the spatial distribution signal gradually strengthening from north to south. The reanalysis data CERA20C effectively captures the changing characteristics of the Rs time series and spatial distribution. From the spatial distribution map of the correlation between observed data, reanalysis data Rs, and total cloud amount, it seems that the improvement in the reanalysis data Rs modeling ability may be due to the improvement in total cloud amount modeling capability.
Among the five types of reanalysis data, CERA20C has significant correlation with observed Rs in both HFC (0.76-0.86) and LFC (0.59-0.95). The trend difference of the trend item is the main reason why reanalysis data fails to accurately capture the trend change of observed Rs. Although CERA20C can reflect the signal of observed Rs in HFC and LFC to some extent, its trend performance is opposite to the observed Rs trend, leading to CERA20C's inability to accurately reflect the trend of Rs. By contrast, 20CRv2c and 20CRv3 demonstrate a better trend modeling capability.
In the HFC, apart from 20CRv2c, the total cloud amount bias in reanalysis data can explain 71%-95% of the Rs bias. In the LFC, the contribution of total cloud cover bias in ERA20C and 20CRv2c to Rs bias is greater than that of water vapor pressure bias.. Apart from CERA20C, the total cloud amount bias can explain 58%-68% of the Rs bias from 1931-2010. In the original sequence, the total cloud amount bias can explain 63%-92% and 56%-59% of the Rs bias of ERA20C, 20CRv2c, and 20CRv3 from 1931-1960 and 1931-2010, respectively. Overall, Rs is highly sensitive to changes in the total cloud amount. Apart from 20CRv3, the other reanalysis data's LFC simulation results are poor (-0.59~0.09) for 1961-2010, which could be the main reason for the poor performance of the reanalysis solar radiation data in the LFC simulation.
The response of Rs in Japan to climatic indices shows significant seasonal differences. The results indicate that the seasons in which Rs is sensitive to ENSO are summer and winter, to PDO index in spring, and to the change in East Asia trough location index in autumn. In the reanalysis data, CERA20C and 20CRv3 can capture the changes in the 500hPa geopotential height field during positive and negative anomalies of climate indices, but CERA20C seldom captures the change characteristics of Rs during positive and negative anomalies of climate indices; ERA20CM is almost incapable of capturing the situation of Rs during positive and negative anomalies of climate indices, possibly because it fails to simulate the corresponding features of the 500hPa geopotential height field.