Forecasting estimates of surface air temperature changes by the wavelet transformation method

Abstract

Analysis of a series of average annual surface air temperature (SAT) shows that they consist of three major components: a long-term trend, a set of harmonic components, and anomalies whose characters are close to a random process. We propose to use wavelet transformation of the source series to set aside quasi-periodic fluctuations. In this case, the distribution of the transformation coefficients enables one to set aside fluctuations of various scales, both the ones that are almost harmonic and those characterized as nonstationary fluctuation process. Then, forward extrapolation is performed by wavelet transformation coefficients for the set-aside scales in view of their temporal dynamics. The series’ fluctuation component is restored by inverse wavelet transformation. The proposed approach is demonstrated by the example of average annual SAT series registered by stations in the towns of Syktyvkar and Tomsk with an instrumental observation period of more than 100 years.