Geographical Distribution of Thermometers Gives the Appearance of Lower Historical Global Warming

Gaps with missing data in the observational temperature record are responsible for an underestimation of the global warming between 1881–1910 and 1986–2015 by 0.1 °C. We found that missing data in the historical observations introduce a warm bias in the early part of the record and a cold bias toward the end. The effect of the nonuniform sampling was explored by comparing the global mean temperature estimated from gridded observations, climate model simulations, and reanalysis. Output from global simulations was subsampled by masking the grid boxes corresponding to those with missing data in the observations to mimic the geographical availability of temperature measurements. A combination of variance depending on region and a varying geographical data sampling over time explains the bias in the global mean. We propose a methodology for estimating the global mean temperature that reduces the effect of the nonuniform variance. Key Points The incomplete sampling of the surface temperature of Earth's surface affects the estimation of trends in the global mean The changing availability of temperature observations over time influences the variance of the data sample An inhomogeneity in the data in terms of changing variance affects the trend estimate for the global mean temperature