An imputation method for the climatic data with strong seasonality and spatial correlation.

Missing data were frequently found in the instrumental climatic records, which hindered the statistical analyses on climate change. A novel imputation method, called Imputation Based on Decomposition of Time Series (IBDTS), was developed in this article for the climatic data with strong seasonality and spatial correlation. It was to decompose the time series into three components first, and then to predict the missing values in each component. The trend component was predicted by regression analysis, the seasonal component was predicted by spectral analysis, and the remainder component was predicted by spatial interpolation. The IBDTS imputation method showed relatively small errors in performance, and kept the real attributes of climatic series, including the amplitude and phase with the cycle period of 12 months, and the linear trend. The sensibility to station distance for the IBDTS method was relatively small. In addition, the IBDTS method had the ability to deal with the data with none of or only a few of complete series, and it was possible to be applied not only in the field of climatology but also in other fields as long as the data had the intrinsic properties of strong seasonality and spatial correlation. [ABSTRACT FROM AUTHOR]