Bayesian assessment of CMIP6 surface net radiation predictions for Köppen–Geiger climate zones

Abstract Köppen–Geiger climate zones were originally constructed to designate the manifold climates to reflect the mean spatial climate characteristics, which includes the Tropical Zone, Dry Zone, Temperate Zone, Continental Zone and Polar Zone. In this study, 20 CMIP6 models are applied to evaluate historical net radiation ( R n ) climatology during 1950–2014 for the five Köppen–Geiger climate zones, respectively. Additionally, Bayesian model averaging (BMA) approach is applied to obtain the multimodel weighted average predictions, which are further used to assess the seasonal variation of net radiation and decadal trends for far historical period (1951–1970), middle historical period (1971–1990) and near historical period (1991–2010), respectively. Moreover, the Bayesian‐Gaussian mixture PDFs are generated for global land surface as well as five climate zones to quantify the uncertainty associated with the BMA predictions. Results indicate that the global land surface R n is overestimated in general with significant diversity in different climate zones. In addition, R n prediction from BMA model is more accurate and reliable than individual models. The highest R n occurs in JJA for Polar Zone and Continental Zone. However, as for zones in lower latitudes, the peaks of R n reach slightly earlier. The R n has decreasing trends for Tropical Zone, Continental Zone and Temperate Zones during far historical period; however, it becomes to increasing trends after 1970s, which is strongly related with anthropogenic global warming. In addition, compared to other zones, there exists smallest prediction uncertainty associated with BMA prediction for Dry Zone.