Method for quantitative reconstruction of mean July air temperatures (Tjuly). The quantitative reconstruction of mean July air temperatures (TJuly) is based on calibration chironomid data sets for lakes from northern Russia (Nazarova et al., 2015, doi:10.1016/j.gloplacha.2014.11.015). Mean July air temperatures were inferred using a North Russian (NR) chironomid-based temperature inference model (WA-PLS, 2 component; r 2 boot = 0.81; RMSEP boot =1.43 °C) based on a modern calibration data set of 193 lakes and 162 taxa from East and West Siberia (61-75°N, 50-140 °E, T July range 1.8 - 18.8 °C). The mean July air temperature of the lakes for the calibration data set was derived from New et al. (2002, doi:10.3354/cr021001). The TJuly NR model was previously applied to palaeoclimatic inferences in Europe, arctic Russia, East and West Siberia, and demonstrated a high reliability of the reconstructed parameters. The chironomid-inferred TJuly were corrected to 0 m a.s.l. using a modern July air temperature lapse rate of 6 oC km-1. Chironomid-based reconstructions were performed in C2 version 1.7. The chironomid data was square-rooted to stabilize species variance. To assess the reliability of the chironomid-inferred TJuly reconstruction, we calculated the percentage abundances of the fossil chironomids that are rare or absent in the modern calibration data set. A taxon is considered to be rare in the modern data when it has a Hill N2 below 5. Optima of the taxa that are rare in modern data are likely to be poorly estimated. Goodness-of-fit statistics derived from a canonical correspondence analysis (CCA) of the modern calibration data and down-core passive samples with TJuly as the sole constraining variables was used to assess the fit of the analyzed down-core assemblages to TJuly. This method shows how unusual the fossil assemblages are in respect to the composition of the training set samples along the temperature gradient. Fossil samples with a residual distance to the first CCA axis larger than the 90th and 95th percentile of the residual distances of all the modern samples were identified as samples with a 'poor fit' and a 'very poor fit' with the reconstructed variable (TJuly). CCA was performed using CANOCO 5. In the evaluation of goodness-of-fit, the CCA scaling focused on inter-sample distances with Hill's scaling selected to optimize inter-sample relationships.