Dear all,
I created a new package called divand (data interpolating variational analysis in n-dimensions). Basically it allows to interpolate observations on a regular grid using norm splines. Under some circumstances it is equivalent to optimal interpolation and kriging. But in variational analysis it is easier to take the topology of the domain into account. Such algorithms are often used in oceanography to analyze measurements because they do not introduce a spurious coupling between water masses which are separated by land.
You can find more information about this method in the recently published paper in Geoscientific Model Development (an open access journal):