Effects of model averaging for estimating future pavement temperatures under climate change

Anthropogenic climate change drives temperature trends that should be accounted for in mechanistic-empirical pavement design. Accounting for these impacts requires modifications to existing design approaches, including analysis inputs for temperature. One portion of the incorporation of future temperature into mechanistic-empirical pavement design that is without consensus is the selection of an appropriate number of global climate models to mitigate uncertainty and optimise computational efficiency. While average ensemble approaches have been applied, there are questions regarding their applicability due to the attenuation of extreme temperatures. The goal of this study was to investigate average ensemble and individual GCM projected subsurface temperatures for pavement sections in the long-term pavement performance (LTPP) InfoPave database and determine the statistical significance of differences. A conventional average from the literature was compared with the proposed rank, percentile, and Bayesian average approaches. The rank and percentile averages are unreliable because of the attenuation of extremes and statistically significant differences compared to individual GCMs.