The best way is to project the elevation data using an appropriate projection. Then interpolation should be no problem with interp2. Choosing a projection will depend on where the data lie on the globe (near the equator?, near the poles?) and how large an area the data spans. If your bathymetry data spans only a few hundred kilometers, perhaps a UTM projection is appropriate. There's a pretty good explanation of projections here. If you have the mapping toolbox, you can project and unproject with projfwd and projinv.
Another option, for a small area of the globe, could be to use interpm2 in the lat dimension to get an array of lats, use it again for the lon dimension, and call that your new grid. But again, this would only be appropriate for a small area of the globe where spacing between meridians does not change very much over the north-south extents of the data.
