Hi Cam,

I assume you are referring to the MGET tool called Interpolate No Data Cells. If the tool is completing successfully very quickly (e.g. 1 second) and not reporting any messages, it might not be working at all. Usually when that happens, there is an MGET installation problem and it will happen with all MGET tools. If other MGET tools are working, than it is likely not the problem.

So, assuming it is actually running, just generating very low values, one possibility is that there is a low value somewhere (e.g. on land) that has not been masked (set to NoData already). Cells like that can often exert an extreme influence on the interpolation algorithm. To see if this might be happening, you could try allowing the algorithm to interpolate all through the land cells by leaving the Maximum Size of No Data Regions parameter blank and then not masking land yourself after the tool has run. You can then see how the tool would fill in the large land areas. Usually it will be a smooth gradient. If there is a single very low value in land somewhere, it will be like a black hole, sucking values down as you approach it.

I hope that helps. I might be able to help more if you send me the complete output from the tool and one of the rasters that reproduces the problem. (You might have to send that with a file-sending service, as my email will only accept attachments of 20 MB or less.)

Finally, to be proper I should say that there is no guarantee that this tool will produce realistic values, particularly close to shore. The onus is on you to assess whether you think values are realistic. In the open ocean, when scattered cells are missing, it seems reasonable to interpolate from the surrounding cells. Close to shore, it can be more of an extrapolation problem rather than interpolation. For example, chlorophyll concentration often appears to increase exponentially approaching shore. It is hard to say whether the extrapolative behavior of this tool will produce realistic values. If you just need rough approximations, e.g. for a coarse scale modeling exercise, then it might be ok. But if it is important that the values are reasonable, I would encourage you to test the algorithm, e.g. by taking a very clear image, masking out some cells, and seeing how well the algorithm can guess them.

Best,

Jason