Outliers in Mixed Models for Monthly Average Temperatures

Abstract

Long-term series of monthly average temperatures taken at 28 sites in Valle del Cauca, Colombia, are studied. Mixed models are applied to cater for the within- and between-site variation. Outliers are inevitable in such studies, due to faulty equipment, slip-ups in the recording process, or unusual weather patterns. We apply a simulation-based approach to the assessment of the outlier status of suspected observations. It is a method based on graphical comparisons of user-defined features, related to large residuals, in the real and simulated data sets. Robustness in the identification of the outliers is achieved by applying the procedure with several alternative models. The impact of the identified outliers is assessed. Two meteorological stations, Zaragoza and Monteloro, are identified as having many outliers, so that all the data from them should be discarded.