A Comparison of Methods for Filling Gaps in Hourly Near-Surface Air Temperature Data

Near-surface air temperature observations often have periods of missing data, and many applications using these datasets require filling in all missing periods. Multiple methods are available to fill missing data, but the comparative accuracy of these approaches has not been assessed. In this comparative study, five techniques were used to fill in missing temperature data: spatiotemporal correlations in the form of empirical orthogonal functions (EOFs), time series diurnal interpolation, and three variations of lapse rate-based filling. The method validation used sets of hourly surface temperature observations in complex terrain from five regions. The most accurate method for filling missing data depended on the number of available stations and the number of hours of missing data. Spatiotemporal correlations using EOF reconstruction were most accurate provided that at least 16 stations were available. Temporal interpolation was the most accurate method when only one or two stations were available or for 1-h gaps. Lapse rate-based filling was most accurate for intermediate numbers of stations. The accuracy of the lapse rate and EOF methods was found to be sensitive to the vertical separation of stations and the degree of correlation between them, which also explained some of the regional differences in performance. Horizontal distance was less significantly correlated with method performance. From these findings, guidelines are presented for choosing a filling method based on the duration of the missing data and the number of stations.