Identifying and Correcting Urban Bias in Regional Time Series: Surface Temperature in China's Northern Plains

Abstract

A detailed study of urban bias in surface temperatures of China's northern plains is described. Temperatures of climatological surface stations were examined using a statistical rank-score procedure that allows screening of the data without knowledge of the station history information. Time series found to exhibit large potential discontinuities (i.e., those introduced as a result of nonclimatic factors such as observation schedule changes, instrument replacements, and station moves) were excluded from further analysis. In addition to the usual total population statistics, census area classifications and population densities were used to distinguish between 21 urban and 8 rural stations. Location-related biases associated with latitude and longitude positions were first removed from all station data, however, using ordinary least-squares regression techniques. Finally, a systematic sampling strategy was employed to estimate magnitudes and trends of urban bias in annual and seasonal mean temperatures. Results of the study indicate that temperatures for stations located in or near the most highly and densely populated urban centers exhibit the largest biases. For most of these urban stations, magnitudes and trends of the bias are greater during spring or summer than during autumn or winter. Standard errors of the estimated urban biases are large, however. Therefore, only the regionally averaged temperatures were adjusted to remove magnitudes and trends of urban bias. Trends in the original and adjusted temperatures of this study and in gridded temperatures taken from the widely used dataset of Jones et al. were also compared. It is suggested that despite past efforts to remove the effects of the urban beat islands from this and other large-scale, land-surface datasets, large urban warming biases may still remain.