The observation and analysis of the difficulties and differences, as well as possible errors, in the information output by automated stations is a complicated process. In observing the present weather the human observer is, for example, able to code many different codes at the same time. A period of mixed winter precipitation events is recognised as a significant challenge in assessing the reliability of automated Present Weather Sensors (PWS). The period of difficult winter conditions presented in this paper was unique in the whole two-year test period. Earlier studies have mostly been based only on liquid precipitation.

This paper presents two basic ideas: the first is to compare many simultaneous human-observed codes with one PWS code; the second is to use logistic regression to compare the observations. A wide variety of weather variables are used as candidates for ‘methodical error’.