Analysing effects of meteorological variables on weather codes by logistic regression

Abstract

The qualitative parameters describing present weather are particularly difficult to automate. The weather types which create most of these difficulties are known, but little attention has been given to investigating the reasons for disagreements between the primary reference, the professional observer and an automated instrument. This paper provides a method - multiple logistic regression - to compare the WMO present weather codes detected by a professional observer and an automated system. A new approach is introduced to explain the errors relative to the official weather variables. Many weather periods have been analysed, but here results are presented for a snow period with slight and very slight precipitation. The best predictive variables were the dew point temperature, wind direction, relative humidity and visibility. The catalyst for this study was the need for better quality control and for tools to enable the development and the manufacture of better instruments and systems to detect present weather.