With the exception of those models based on the response method, most global ocean tide models comprise only a few constituents. Subsequent tidal prediction, which must account for the whole tidal spectrum, relies on tidal inference to supply the missing minor constituents. Inference involves either interpolation or extrapolation of tidal admittances, with the admittance curves defined by the major constituents.

A network of 150 bottom-pressure stations (Ray, JGR, 2013) is used to test some aspects of tidal inference. Some of these tests can be done only with the CNES FES models, since only those models have included direct solutions for a number of minor constituents. Inference of constituents like J1, OO1, and 2N2 at the edges of the tidal bands require extrapolation of admittances, and this appears to cause large (relative) errors, as might be expected. (Shailen Desai recently came to the same conclusion based on his analysis of tidal polar motion.) In contrast, inference of constituents in the middle of the tidal bands can sometimes be more accurate than the directly solved-for constituents, probably because the solved-for constituents rely on observations that are relatively noisy (owing to their small amplitudes). This is even true for the fairly large constituent P1 (although it does not hold for model TPXO.8). The bottom-pressure data confirm that inference of P1 needs to account for the nearly diurnal free-wobble resonance caused by the earth's fluid core. Implications for GDR tide corrections will be discussed.