Estimating exposure biases in early instrumental land surface temperature data

Exposure biases are non-climatic changes in the surface air temperature record which were introduced due to changes in the way thermometers were protected from solar radiation. The possible presence of the exposure bias in early instrumental temperature datasets is a well-known issue, and the impact of changing thermometer exposures, particularly the transition between historic exposures and the Stevenson screen, has been explored by previous studies. However, despite this, very few adjustments have been made to account for the bias, with the exception of a handful of localised studies. As a result, the exposure bias still accounts for significant uncertainty in global surface air temperature compilations, such as HadCRUT5.

In this work we report an attempt to address the exposure bias for extratropical weather stations in a version of CRUTEM5 that has been extended back in time to 1781 (CRUTEM5_ext). We developed statistical models to predict the bias introduced by transitions from four main categories of historic exposure – open, wall-mounted, intermediate and closed – to the Stevenson screen. The models are based on an empirical analysis of the characteristics of the exposure bias observed in 20 parallel measurement studies, together with the temperature and radiation variables that were a priori expected to influence the magnitude of the bias in mean temperatures on a monthly timescale. Separately, we have compiled a database detailing the historic exposures in use at stations and the timing (or approximate timing when a precise time is not known) of the transition to the Stevenson screen. The statistical models, where robust, are then applied to the individual stations within CRUTEM5_ext to make adjustments for the exposure changes according to the database of historic exposures.

This presentation will outline the model development, give a brief overview of the evolution of thermometer exposures in use in the early instrumental period for extratropical stations, and will illustrate the impact the exposure bias adjustments have on the CRUTEM5_ext data. This work forms part of the NERC-funded GloSAT project (https://www.glosat.org/) which is developing a global surface air temperature dataset starting in 1781. Where appropriate, stations used to create the GloSAT dataset will be adjusted for the exposure bias using the models presented here.