Well, I think that in addition to calculating a cummulative integration you also should/need to extract the limits over which you have non-nan values, something like this:
for i2 = 1:size(SALT,2)
i1 = find(isfinite(SALT(:,i2)));
if ~isempty(i1)
I_OK(:,i2) = [i1(1) i1(end)];
CTS{i2} = cumtrapz(SALT(i1,i2));
end
end
In addition you should only use cumtrapz over contiguous regions with finite values, so my example works for the data you gave. You might(should) make sure that the indices i1 are without gaps/or take care of the cases where there are gaps.
HTH
