Temperatures of nine thermometer screens during a 6-year field experiment in De Bilt (the Netherlands) have been compared. The screens are currently in use throughout the world and comprise the following types: an aspirated Young screen, four naturally ventilated round-shaped multi-plate screens (KNMI, Vaisala, Young, Socrima), a slightly aspirated version of the KNMI screen, a synthetic Stevenson screen (both aspirated and naturally ventilated) and a naturally ventilated wooden Stevenson screen. The multi-plate KNMI screen served as a reference. For the daily minimum, maximum, and mean air temperatures (T(n), T(x), T(mean)), the absolute seasonal mean differences with the reference were almost all <= 0.1 degrees C. An exception is the aspirated Young screen for which differences in mean T(x) in summer (JJA) are most notable and amount to -0.43 degrees C. The differences of the individual T(n), T(x), T(mean) values may be much larger than their seasonal averages. For the aspirated Young and the naturally ventilated Stevenson screens they are most notable, where the Young is generally cooler (T(x) and T(mean)) and the Stevensons warmer than the reference. The absolute temperature differences between the screens and the reference are shown to increase with decreasing cloud cover and windspeed. Furthermore, using the original 15-s samples it is shown that random variations cause fast-responding screens to have more extreme T(n), T(x) values than slow-responding screens. For a supposed transition of the natural ventilated synthetic Stevenson screen to the reference screen, transfer functions are successfully developed for the 10-min temperature data and the daily T(n) and T(x) data. It is argued that the explained variance of the models could have further improved when high-accuracy (especially in the range 0-3 m/s) windspeed measurements were available (at screen level and position). Copyright (c) 2007 Royal Meteorological Society.
