Neutral density revisited

The possibilities of defining and computing an approximately neutral density variable are reexamined in this paper. There are three desirable properties that a neutral density variable should possess. Firstly, the isosurfaces of this variable should coincide with (approximately) neutral surfaces. This would facilitate the analysis of hydrographic data on the most appropriate mixing and spreading surfaces. Secondly, the horizontal gradients of the neutral density should agree with the gradients of the in situ density, and thirdly the vertical gradient of the neutral density variable should be proportional to the static stability of the water column. A density variable that approximates the latter two properties can be used in ocean circulation models based on layer coordinates, and would reduce substantial errors in present isopycnal models due to the use of a potential density variable. No variable can possess all the three properties simultaneously. The variable gamma(n) introduced by Jackett and McDougall (1997: J. Phys. Oceanogr. 27, 237-263) satisfies the first of the properties exactly but is not designed for the use in models. Based on climatological data in the North Atlantic, an alternative neutral density variable <(nu)over tilde>(S, Theta) is defined, which is shown to approximate the two gradient criteria much better than any potential density. We suggest that this neutral density variable may be useful in isopycnal ocean models as an alternative to potential density, since it could significantly reduce errors in thermal wind relation and vertical stability. (C) 1999 Elsevier Science Ltd. All rights reserved.