Effect of data gaps: comparison of different spectral analysis methods

Abstract. In this paper we investigate quantitatively the effect of data gaps for four methods of estimating the amplitude spectrum of a time series: fast Fourier transform (FFT), discrete Fourier transform (DFT), Z transform (ZTR) and the Lomb–Scargle algorithm (LST). We devise two tests: the single-large-gap test, which can probe the effect of a single data gap of varying size and the multiple-small-gaps test, used to study the effect of numerous small gaps of variable size distributed within the time series. The tests are applied on two data sets: a synthetic data set composed of a superposition of four sinusoidal modes, and one component of the magnetic field measured by the Venus Express (VEX) spacecraft in orbit around the planet Venus. For single data gaps, FFT and DFT give an amplitude monotonically decreasing with gap size. However, the shape of their amplitude spectrum remains unmodified even for a large data gap. On the other hand, ZTR and LST preserve the absolute level of amplitude but lead to greatly increased spectral noise for increasing gap size. For multiple small data gaps, DFT, ZTR and LST can, unlike FFT, find the correct amplitude of sinusoidal modes even for large data gap percentage. However, for in-situ data collected in a turbulent plasma environment, these three methods overestimate the high frequency part of the amplitude spectrum above a threshold depending on the maximum gap size, while FFT slightly underestimates it.