Consider a manifold $M$, say representing the Earth. It seems to me that it doesn't make sense to talk about units of points in $M$, and that it is the charts that have units. That is, if $\phi$ is a chart that gives coordinates $(x, y)$, then $x$ and $y$ map from points of $M$ to numbers with some unit of length.
Then, somewhat counterintuitively, the basis vector fields $\partial/\partial x$ have units of inverse length. If one then had a vector field $V$ measuring, say, wind speed, the components $V_i$ would have units of length over time, and so $V$ itself would have units of inverse time.
The basis one-forms $dx$ also have units of length. So if $h$ is a manifold function with some set of units, its differential $dh$ has the same set of units, but its components would pick up a unit of inverse length.
Am I on the right track? Do any of you know of any sources talking about this? I had a hard time finding some.