tl;dr: If ocean modeling and earth science questions were to swamp out the questions by the current user community, you would probably be better served with your own site. If the questions are mainly about things we already cover a lot on Computational Science Stack Exchange (numerical methods, general-purpose computational science software, computational fluid dynamics), it would probably be a good fit. I personally skew towards inclusiveness, because I think good communities try to be inclusive, and I think that if a topic isn't a good fit, we will see it over time.
What I've seen so far is that application communities (or general niche communities) that start here generally drift away after reaching a certain mass:
We used to get a number of computational chemistry questions; now these are answered more frequently on Chemistry Stack Exchange. Computational chemistry questions are still on-topic here, but if they don't get answered here, they'll probably get migrated.
We got the odd Mathematica question here early on in the beta; now these questions are answered on Mathematica Stack Exchange, and we tend to migrate Mathematica questions there
For a while, we tried being a forum for FEniCS. Now, FEniCS questions tend to get closed, and users are told to ask questions on the FEniCS Q&A site.
Physics tends to migrate computational questions to us. If a question is more about the physics of an application, it's a better fit for Physics Stack Exchange. If a user has a physics problem already set up with all of the equations, and wants to know how best to solve it using numerical methods and software, it's a better fit for this site. We occasionally migrate questions to Physics.
We'll get the occasional statistics or computer vision question; if it gets to be too specialized, we migrate it to the appropriate site.
I think one key is to try and get people who already visit the site to engage. Questions about general computational methods (what numerical methods should I use, what general-purpose computational science software packages should I use, etc.) tend to get answered. Questions about computational fluid dynamics get answered fairly frequently, and would also be a good fit. I think questions about application-specific models for ocean modeling and earth science would probably also be on-topic, if explained in terms accessible to general computational scientists, but they would be less likely to get answered by our current user base.
