Good, bad and the ugly: Seasonal filtering and autoregressive moving average (ARMA) models for detecting and replacing spikes in velocimetric profile data