Estimation of Uncertainty in Temperature Observations Made at Meteorological Stations Using a Probabilistic Spatiotemporal Approach

Abstract

AbstractA probabilistic spatiotemporal approach based on a spatial regression test (SRT-PS) is proposed for the quality control of climate data. It provides a quantitative probability that represents the uncertainty in each temperature observation. The assumption of SRT-PS is that there might be large uncertainty in the station record if there is a large residual difference between the record estimated in the spatial regression test and the true station record. The result of SRT-PS is expressed as a confidence probability ranging from 0 to 1, where a value closer to 1 indicates less uncertainty. The potential of SRT-PS to estimate quantitatively the uncertainty in temperature observations was demonstrated using an annual temperature dataset for China for the period 1971–2000 with seeded errors. SRT-PS was also applied to assess a real dataset, and was compared with two traditional quality control approaches: biweight mean and biweight standard deviation and SRT. The study provides a new approach to assess quantitatively the uncertainty in temperature observations at meteorological stations.