Changes in the Full Distribution of Daily Temperatures with Implications for Extreme Events

Abstract:

Daily maximum and minimum temperatures measured at the weather-station level tend to be non-normally distributed. Unlike normal distributions, non-normality implies the presence of higher order moments of the distribution that covary with mean and variance. Changes in extreme temperature events therefore cannot be attributed to changes in the mean or variance alone. To more completely characterize the observed changes in temperature distributions, we use quantile regression on long daily temperature records from the Global Historical Climatology Network Database. Quantile regression allows for the quantification of the manner in which distributions of temperature are changing as a function of time at percentiles that span the full distribution. Changes in both the upper and lower tails of the distribution are amplified relative to changes in the median. Spatial patterns of change are also identified using singular value decomposition, and the ability to predict regional variations in relation to the base distribution of temperature change is examined.