A Bayesian Attribution Analysis of Extreme Temperature Changes at Global and Regional Scales

Abstract

Abstract Recent studies showed that anthropogenic greenhouse gas (GHG) increase is a major driver of the observed increases in extreme temperatures at global and regional scales using an optimal fingerprint (OF) method, which is a frequentist approach based on linear regression. Here, a Bayesian decision method is employed, which finds the most probable cause of the observed changes by comparing likelihoods of different forcings in view of observations. To quantify individual forcing contributions, a new modified attribution procedure based on Bayesian decision is proposed, i.e., computing the likelihood ratio [Bayes factor (BF)] between different forcings. First, the contribution of anthropogenic forcing (ANT) is measured by BF between anthropogenic-plus-natural forcing (ALL) and natural forcing (NAT) using a threshold for “substantial” evidence (lnBF ≥ 1). Similarly, the NAT contribution is assessed by BF between ALL and ANT. Further, the GHG contribution to the detected ANT is quantified by BF between ANT and anthropogenic aerosols (AA), and the AA contribution is evaluated by BF between ANT and GHG. The devised Bayesian approach is applied to HadEX3 observations and CMIP6 multimodel simulations for extreme temperature intensities (warmest day/night and coldest day/night) for global, continental, and regional domains following previous studies. Bayesian attribution results indicate that the ANT signal is detected in many continental and subregions for all extremes indices. This is generally consistent with OF-based results but with less frequent detection, indicating that the Bayesian method is slightly stricter than the OF method. However, GHG contributions to the detected ANT are identified over more subregions in the Bayesian attribution, suggesting its potential advantage over conventional methods in case of low signal-to-noise ratio and high collinearity.