Abstract The generation of variance by anomalous advection of a passive tracer in the thermocline is investigated using the example of density-compensated temperature and salinity anomalies, or spiciness. A coupled Markov model is developed in which wind stress curl forces the large-scale baroclinic ocean pressure that in turn controls the anomalous geostrophic advection of spiciness. The “double integration” of white noise atmospheric forcing by this Markov model results in a frequency (ω) spectrum of large-scale spiciness proportional to ω−4, so that spiciness variability is concentrated at low frequencies. An eddy-permitting regional model hindcast of the northeast Pacific (1950–2007) confirms that time series of large-scale spiciness variability are exceptionally smooth, with frequency spectra ∝ ω−4 for frequencies greater than 0.2 cpy. At shorter spatial scales (wavelengths less than ∼500 km), the spiciness frequency spectrum is whitened by mesoscale eddies, but this eddy-forced variability can be filtered out by spatially averaging. Large-scale and long-term measurements are needed to observe the variance of spiciness or any other passive tracer subject to anomalous advection in the thermocline.