Abstract Surface‐incident solar radiation is essential for studying climate history and forecasting. This study investigates whether the simulated has improved for European Centre for Medium‐Range Weather Forecasts Reanalysis (ERA) v5 (ERA5) compared with its predecessor ERA‐Interim (ERAI) in Japan. Additionally, we employ the Shapley additive explanation (SHAP) to analyze the underlying causes of their discrepancies. Homogenized observations, which primarily eliminate the inconsistent impact caused by instrument replacements and site relocations on the raw observations, were used as reference data. ERA5 shows a smaller bias (0.93 W·m −2 ) than ERAI (7.06 W·m −2 ), especially in northeastern Japan. Observations decrease at −2.0 W·m −2 per decade before 1995 and increase at 1.6 W·m −2 per decade afterward. Both ERA5 and ERAI can capture this long‐term transition, although the magnitudes are underestimate by 25–45%. For 1982–2014, the brightening was well reproduced by ERA5 (0.48 W·m −2 per decade), whereas ERAI decreased (−0.35 W·m −2 per decade). For 2000–2014, both ERA5 (−0.6 W·m −2 per decade) and ERAI (−0.4 W·m −2 per decade) missed the observed brightening of 0.5 W·m −2 per decade. discrepancies in these two reanalyses were mainly quantified by their performances in simulating total cloud cover (TCC) and solar radiation under clear‐sky conditions (). /TCC biases can explain 55%/45% of biases for ERA5 and 58%/42% for ERAI by SHAP. Additionally, /TCC trend biases can explain 73%/27% of trend biases for ERA5 and 75%/25% for ERAI. These findings suggest that the accuracy of in reanalysis has a greater impact on , particularly on trend. It can be inferred that improvements in aerosol procedures are urgently needed, including the assimilation of more related observational data in the next generation of reanalysis.
Gao et al. (Fri,) studied this question.