Locally weighted scatter‐plot smoothing for analysing temperature changes and patterns in Australia

ABSTRACT The mean maximum monthly temperature data were recorded at 112 stations in Australia. The data (1990–2015) were downloaded from the Australian Bureau of Meteorology (BOM) website. Missing values were imputed using regression models based on information from the nearest stations, as well as the time periods. The data were deseasonalized to remove seasonal variations and then cluster analysis techniques were used to group the stations into six clusters. For each cluster, locally weighted scatter‐plot smoothing (LOESS) and double exponential smoothing (DES) were used to analyse temperature changes and patterns. The results showed that LOESS produced better fits as well as smoother curves compared with the DES. The trends in temperature were increasing in all clusters, whereas the patterns showed periodicity of the temperatures. The positions of 88 stations included in this study with the percentage missing data for each station.