Describe quantitatively the amount and quality of ocean data available. Some examples:
Subsample a dataset to estimate the sampling error in a bulk measurement, e.g., global mean temperature.
Compare similar measurements from two datasets (e.g., Reid & Mantyla versus eWOCE, or WOCE shipboard CTD versus WOCE drifter CTD). You can do this in a region of interest, or in a chosen bulk measurement (e.g., global temperature, Mediterranean temperature).
Compare similar measurements but separated in time. Is there signal above the noise? How can you estimate noise (or has someone else done it for you)?
Don't be afraid to piggyback off of other published work. Use google scholar or Web of Science to locate articles by keyword, and then see what they say.
Feel free to use the Matlab references, and refer to similar projects from last year, Ready to Roll and What Do You Mean Mean?.