INTERNATIONAL JOURNAL OF CLIMATOLOGY Int. J. Climatol. 23: 1495–1521 (2003) Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/joc.945 WERE SOUTHERN SWEDISH SUMMER TEMPERATURES BEFORE 1860 AS WARM AS MEASURED? ANDERS MOBERG, a, * HANS ALEXANDERSSON, b HANS BERGSTR ¨ OM c and PHILIP D. JONES a a Climatic Research Unit, University of East Anglia, Norwich, UK b Swedish Meteorological and Hydrological Institute, Norrk¨ oping, Sweden c Department of Earth Sciences, Meteorology, Uppsala University, Uppsala, Sweden Received 13 September 2002 Revised 10 June 2003 Accepted 10 June 2003 ABSTRACT Temperature series from Stockholm and Uppsala in southern Sweden indicate that summers from the mid-18th century until around 1860 were, on average, warmer than the 1961–90 mean. The station histories suggest that the early observations could have been positively biased, for example because of insufficient radiation protection. We investigate if independent support for warm summers in the early period can be obtained from other climate variables. Using stepwise multiple regression analysis we investigate nine potential predictor variables: six air circulation indices, precipitation, air pressure and cloud amount. Three of these variables — cloud amount (the most important one), meridional geostrophic wind, and air pressure — together explain 65% of the June–August temperature variance in the calibration period 1873–2000. Application of the regression relationship back to 1780 shows that the model is equally successful in predicting year-to- year temperature variability before 1873 as it is in the calibration period, whereas the low-frequency component is poorly reconstructed in the early period. This reduced skill is primarily due to poorer data quality of the predictor variables in the early period, in particular the cloud amount series. The observed decadal mean temperatures during 1780–1860 are found to be above the upper limit of a 95% confidence interval that accounts for uncertainties both in the regression relationship and in the cloud amount series. We conclude that the observed temperatures before around 1860 are, therefore, most likely positively biased. The size of this bias cannot be accurately determined from the evidence used here, but seems to be about 0.7–0.8 ° C for both stations. A comparison with long instrumental temperature series from central Europe suggests a slightly smaller bias (0.5–0.6 ° C). For more accurate assessment of the Stockholm and Uppsala temperatures, we recommend that extensive homogeneity testing of other long northern European temperature series are undertaken. Copyright  2003 Royal Meteorological Society. KEY WORDS: Sweden; multiple regression; standard normal homogeneity test; early instrumental data; temperature; air circulation; cloud amount 1. INTRODUCTION It is well established that the global mean temperature has warmed by about 0.6 ° C since the late 19th century and that at least a significant part of this warming is likely due to anthropogenic emissions of greenhouse gases, although natural factors may also have contributed (Houghton et al., 2001). In the search for further understanding of the underlying causes there is a need to reconstruct climate before the onset of any anthropogenic effect. Only Europe has a large number of instrumental records beginning before the mid-19th century. Prior to using these early data in climate change studies, it is essential to investigate their quality and limitations thoroughly. Although there has been a renewed interest in the early European records in the last decade (e.g. Jones et al., 1999; B¨ ohm et al., 2001; Camuffo and Jones, 2002), there still remain * Correspondence to: Anders Moberg, Climatic Research Institute, University of East Anglia, Norwich NR4 7TJ, UK; e-mail: a.moberg@uea.ac.uk Copyright  2003 Royal Meteorological Society