You should always (at least visually) check your results - we found
strong contour artefacts in vertical curvature (obviously, this
depends on the parameters to s.surf.tps, but we have yet to find a
set of parameter values that avoids this problem and gives a
satisfactory raster all over)
Martin use lower tension (in relatively flat terrain it should be
much lower than the values suggested in the manual) and use smoothing to
avoid overshoots. Smoothing effect for e.g. sm=0.1 is usually much smaller
than the accuracy of data and allows you to use low tension.
We have a new radial basis function which makes it easier to use lower
tension, however we don't have time and worksforce to implement it now,
so you really have to fiddle with parameters to get it right.
I hope this helps,
Helena