Using Empirical Mode Decomposition to Filter Out Non-turbulent Contributions to Air–Sea Fluxes

Abstract

A methodology based on Empirical mode decomposition (EMD) was used to filter out non-turbulent motions from measurements of atmospheric turbulence over the sea, aimed at reducing their contribution to eddy-covariance (EC) estimates of turbulent fluxes. The proposed methodology has two main objectives: (1) to provide more robust estimates of the fluxes of momentum, heat and CO$_2$; and (2) to reduce the number of flux intervals rejected due to non-stationarity criteria when using traditional EC data processing techniques. The method was applied to measurements from a 28-day cruise (HALOCAST 2010) in the Eastern Pacific region. Empirical mode decomposition was applied to 4-h long time series data and used to determine the cospectral gap time scale, $T_\mathrm{{gap}}$. Intrinsic modes of oscillation with characteristic periods longer than the gap scale due to non-turbulent motions were assumed and filtered out. Turbulent fluxes were then calculated for sub-intervals of length $T_\mathrm{{gap}}$ from the filtered 4-h time series. In the HALOCAST data, the gap scale was successfully identified in 89% of the 4-h periods and had a mean of 37 s. The EMD approach resulted in the rejection of 11% of the flux intervals, which was much less than the 68% rejected when using standard filtering methods based on data non-stationarity. For momentum and sensible heat fluxes, the averaged difference in flux magnitude between the traditional and EMD approaches was small (3 and 1%, respectively). For the CO$_2$ flux, the magnitude of EMD flux estimates was on average 16% less than fluxes estimated from linear detrended 10-min time series. These results provide evidence that the EMD method can be used to reduce the effects of non-turbulent correlations from flux estimates.