The capability of 20th‐century reanalyses to reproduce the spatiotemporal variation in surface incident solar radiation over Japan for 1931–2010

Abstract Understanding the spatiotemporal variations in surface incident solar radiation ( R s ), especially over longer time spans, is crucial for climate modeling and environmental assessment. This study, referring to the century‐long homogenized R s measurements, comprehensively quantifies the ability of five 20th‐century reanalyses phenomena to estimate R s over Japan for 1931–2010, and furthermore assesses the contributions of total cloud cover and vapor pressure to the biases in R s trends. Dominant features revealed that R s signals boost from the north to the south, which could be captured but is overestimated by all reanalyses. The dominant time evolution of R s was well captured by CERA20C, followed by ERA20C and 20CRv3, corresponding to their relatively better performance cloud cover. Unfortunately, ERA20CM totally fails to reproduce the spatiotemporal variation in R s as its procedure does not include an assimilation step. 20CRv2c fails in estimating R s for 1931–1960, but improves much for 1961–2010 as the quality of assimilated observational variables becomes better. A dimming of −2.76 W·m −2 ·decade −1 for 1931–1960 was detected in the observed R s , which could be reproduced by CERA20C and ERA20C. Afterwards, a brightening of 0.88 W·m −2 ·decade −1 for 1961–2010 was shown in observations, and all reanalyses fail. The trend bias of R s in CERA20C attributes more to total cloud cover before 1960, and to aerosols or other factors after 1960. Cloud cover contributed more than vapor pressure to the trend bias of R s for 1931–1960; vapor pressure became more important in regulating R s for 1961–2010 than for 1931–1960. The combined effects of both factors on R s are insignificant for 20CRv3 during the entire period, for 20CRv2c for 1931–1960 and for CERA20C for 1961–2010, suggesting other factors, such as aerosols, cloud types, cloud optical depth and their interactions may impact R s simulations more and this calls for further investigation.