Testing for trend in variability of climate data: measures and temporal aggregation with applications to Canadian data

Summary It is not clear whether different measures of dispersion of weather attributes could lead to different conclusions on the nature and direction of climatic variability. The range is commonly used as a measure of variability, while the presence of trend is typically studied on seasonal and/or annual basis. In this study, we used daily average temperature values at 15 sites spatially distributed across Canada to test for the presence of trend in variability (measured by the range, the standard deviation and the IQR) using a bootstrap method. The length of the series varied from site to site, ranging from 30 to 151 years. The analysis was undertaken for each month, each season, and the annual data. When calculating the standard deviations, estimates of the annual mean temperatures were used to make the results invariant to the presence of trend in mean. The monthly and seasonal analysis revealed the presence of either increasing or decreasing variability for some months and some seasons. The results for the annual data were not so revealing, especially at sites where some months have increasing while others have decreasing trends. The results across sites did not exhibit a clear geographic pattern. However, there were consistently increasing trends in variability at Toronto and St. Johns during non-summer months, and mostly decreasing trend in Edmonton. The significance of trend in variability using the range and the standard deviation were consistent in less than 30% of the time across sites and across the monthly, seasonal and annual aggregations. There was not much agreement between the standard deviation and the IQR either, highlighting the importance of the choice of a measure of variability.