A Dynamic Method for Gap Filling in Daily Temperature Datasets

Abstract

regression-based approach for temperature data reconstruction has been used to fill the gaps in the series of automatic temperature records obtained from the meteorological network of Veneto Region (northeastern Italy). The method presented is characterized by a dynamic selection of the reconstructing stations and of the coupling period that can precede or follow the missing data. Each gap is considered as a specific case, identifying the best set of stations and the period that minimizes the estimated reconstruction error for the gap, thus permitting a potentially better adaptation to time-dependent factors affecting the relationships between stations. The best sampling size is determined through an inference procedure, permitting a highly specific selection of the parameters used to fill each gap in the time series. With a proper selection of the parameters, the average errors of reconstruction are close to 0 and those corresponding to the 95th percentile are typically around 0.1°C. In comparison with similar regression-based approaches, the errors are lower, particularly for minimum temperatures, and the method limits inversions between the minimum, mean, and maximum temperatures.