Identifying Changes in Australian Temperature Distributions using Quantile Regression — Australian Meteorological and Oceanographic Society

Identifying Changes in Australian Temperature Distributions using Quantile Regression (#68)

Elisa R Jager 1 , George Takacs 1
  1. University of Wollongong, Wollongong, NSW, Australia

An upwards mean trend in Australian and global temperatures has been well established but the question of whether temperature distributions are changing is not yet resolved. Looking at trends in different parts of the distribution using quantile regression can give deeper insight into the nature of these changes and their implications for climate extremes. Lewis and King (2017) have studied quantile trends in CMIP5 simulation results for the 21st century, and there have been studies on other national data, but not much work has been done on actual past trends in Australian data. Daily max and min temperature data from the ACORN-SAT database ranging back to 1910 were analysed with trends found in a full range of quantiles. The usage of raw station data preserves the underlying distribution and allows study down to a local level. To prevent issues with overlapping entries and bias from historical unit conversion the data was jittered and the procedure repeated. Multiple time periods were explored, using successive 30 year windows of data, as well as looking at individual seasons and months. Different methods of trend fitting and new measures of convergence/divergence of the distribution, i.e. narrowing or become wider respectively, were explored. For example, distinct difference of the slopes in very high and very low quantiles (taking their confidence intervals into account) is a measure of broadening/narrowing of the whole distribution. This measure showed no uniform pattern of convergence or divergence across the country and seasons, but definite spatial and temporal patterns on smaller regional scales, with the clearest result being a significant number of locations showing converging behavior in Autumn. This agrees with some previous work done on record breaking temperatures which showed less than predicted breaks in Autumn.

  1. Lewis, S, King, A, 2017, 'Evolution of mean, variance and extremes in 21st century temperatures', Weather and Climate Extremes 15
#AMOS2019