Extreme weather exposure identification for road networks in heterogeneous landscapes

Resilient transport infrastructure is essential to the functioning of society and economy. Ensuring network
functionality is particularly vital in the case of severe weather events and natural disasters, which pose serious
threats to both people’s health and the integrity of infrastructure elements. Thus, providing reliable estimates
about the frequency and intensity of extreme weather impacts on road infrastructure is of major importance for
road maintenance, operation and construction. However, against the background of data scarcity in terms of
area-covering, long-term time series, the assessment of extreme weather events is difficult, especially in areas
with diverse landscape properties.
In order to account for heterogeneous small-scale topographic conditions, a hot-spot approach based on selected
characteristic regions is used in this study. For each region, combinations of different extreme value approaches
and fitting methods are compared with respect to their value for assessing the exposure of transport networks to
extreme precipitation and temperature impacts. Four parameter estimation methods (maximum likelihood
estimation, probability weighted moments, generalized maximum likelihood estimation and Bayesian parameter
estimation) are applied to extreme value series obtained via both the block maxima approach (annual maxima
series, AMS) and the threshold excess approach (partial duration series, PDS). Their relative performances are
compared based on the CRMSE5, i.e. the conditional root mean square error for observations with a return period
exceeding 5 years, which gives much weight to the most extreme events.
The viability of the approach is demonstrated at the example of Austria by analyzing five meteorological
indicators related to temperature and precipitation at 26 meteorological stations. These stations have been
selected to represent diverse meteorological conditions and different topographic regions. Results show the
merits of Bayesian parameter estimation methods as compared to traditional fitting methods. Bayesian
estimation of generalized Pareto (GP) distributions fitted to the PDS yielded the best results in 46% of all cases,
followed by Bayesian estimation of Generalized Extreme Value (GEV) distributions fitted to AMS, which
showed the best performance in 35% of all cases. The study suggests that the concept of meteorological hot spot
areas offers a suitable approach for characterizing extreme weather exposure of road networks in heterogeneous
landscapes. The presented framework may contribute to a comprehensive climate risk assessment of
infrastructure networks.