These tools use a triangulation of your data. They only are able to interpolate if your data is inside one of the triangles. And since a delaunay triangulation only extends out to the convex hull of your data, they do not extrapolate. Anyway, extrapolation is dangerous business.
Note that scatteredInterpolant can apparently extrapolate linearly based on the gradients at the edges, but that extrapolation will be poor in general. Do not trust it for far outside of your data. If you do not have the current release, then TriScatteredInterp does not support extrapolation at all.
Here is the help for scatteredInterpolant for extrapolation. Note that TriScatteredInterp never claims to extrapolate that I see anywhere in the help.
F = scatteredInterpolant (...
selection of an extrapolation method to be used outside the convex hull.
Where EXTRAPOLATIONMETHOD is one of the following:
'nearest' - Evaluates to the value of the nearest neighbor on the
boundary (default for METHOD ='nearest')
'linear' - Performs linear extrapolation based on boundary gradients
(default for METHOD = 'linear' and METHOD='natural')
'none' - Queries outside the convex hull will return NaN