Detection of non-stationarity in precipitation extremes using a max-stable process model

Abstract

Non-stationarity in extreme precipitation at sub-daily and daily timescales is assessed using a spatial extreme value model based on max-stable process theory. This approach, which was developed to simulate spatial fields comprising observations from multiple point locations, significantly increases the precision of a statistical inference compared to standard univariate methods. Applying the technique to a field of annual maxima derived from 30 sub-daily gauges in east Australia from 1965 to 2005, we find a statistically significant increase of 18% for 6-min rainfall over this period, with smaller increases for longer duration events. We also find an increase of 5.6% and 22.5% per degree of Australian land surface temperature and global sea surface temperature at 6-min durations, respectively, again with smaller scaling relationships for longer durations. In contrast, limited change could be observed in daily rainfall at most locations, with the exception of a statistically significant decline of 7.4% per degree land surface temperature in southwest Western Australia. These results suggest both the importance of better understanding changes to precipitation at the sub-daily timescale, as well as the need to more precisely simulate temporal variability by accounting for the spatial nature of precipitation in the statistical model.