Ocean acoustic tomography was proposed by Munk and Wunsch (1979) as a method for making measurements of ocean variability over large areas. After the successful demonstration of the feasibility of the idea in the 1981 three-dimensional mesoscale experiment (Ocean Tomography Group, 1982) the tomography group has proposed a new experiment to be carried out in 1986 in the eastern Pacific Ocean on ranges as long as the subtropical gyre scale.

In this paper the gyre-scale experiment is simulated in the model ocean, using Holland's eddy-resolving general circulation quasi-geostrophic model. The paper addresses the following issues: (1) measurement of the heat content vertical profile horizontally averaged along the tomographic section: (2) adequacy of the linearized inverse over very long ranges and the need for its improvement: (3) possible improvements in the specification of the field statistics to obtain more accurate estimates and to measure properties like average pycnocline trends; (4) relationship of possible range-dependent information from the inversion to the assigned noise level.

The results of the modeling simulation can be summarized as follows: (1) The linearized stochastic inversion needs to be improved for gyre-scale ranges providing estimates of the average heat content that have warm or cold biases. Iteration is used and shown to provide estimates of the averge heat content. (2) A smaller number of iterations is necessary if the initial estimate is improved. This can be done by including a spatial mean in the horizontal covariance function for regions of the ocean where the energy level in the mean and in the long length scales may be even more important than the mesoscale energy peak. (3) General trends like average pycnocline slopes can be estimated very well by including an inhomogeneous covariance in the inversion. (4) The estimates of the mean heat content values and of the average slopes are rather insensitive to the specified noise level. (5) The type and number of eigenrays used is not a critical facctor in the inversion estimates as long as the eigenrays provide a good sampling of the interior water mass. Increasing the number of eigenrays does not significantly improve the estimate of the average values. (6) Range-dependent information is primarily a function of the assigned noise. For too high a noise the basic quantities that can be measured are the mean value and the average trend.