Ground coupling is something I've delved into since the mid-80's and developed several modeling methods, all utilizing a 2-D finite-difference conduction program by Lester Shen of the (now defunct) Univ. of Minnesota Underground Space Center.

The first and simplest method has been adopted as the F-Factors in ASHRAE 90.1 and is the same as described in the DOE-2 article that Neal mentioned. However, I feel a later refinement done for the California Energy Commission is much improved, more flexible and accurate, and equally easy to use, although it does require calculating the average air temperature over the past 3 weeks as an intermediate boundary condition.

The main thing to note about the F-factors is that they're conductances per lineal feet of perimeter for temperature difference between the inside and the OUTSIDE AIR, not the ground temperature (which would be different than described at the site mentioned in the question). There are actually several ways to implement this model: (1) model a monolithic foundation as an exterior wall and adjust the R-value of the layer so that U*A = F*PL (Perimeter length) -- this is the method illustrated in the doe-2 article, (2) decompose the foundation into two regions - a perimeter strip 1-2 ft wide and a core for the remainder; model the perimeter as an exterior wall, adjusting its R-value accordingly, but model the core as an adiabatic layer. In all cases, add as much soil to the foundation layer as allowed by the program ( ~ 2.5' in DOE-2) to dampen the hourly fluctuations in the outdoor air temperatures.

The main weaknesses of the F-factor method are (1) it "sees" only the outside air temperature, (2) the foundation is treated as a single surface. While there's a temptation to model the foundation as a underground layer tied to the ground temperature, one shouldn't then use F-factors because the boundary temperatures would be wrong.

In the late 90's, I worked with Fred Winkelmann and Vladimir Bazjanac for the Calif. Energy Commission to develop what I think is a much better model that has since been adopted in the Title-24 calculations. The approach is the same as before, i.e., calculate effective conductances, but instead of a single F-Factor, there are six Foundation Conductances for two domains (perimeter and core) and three heat flow paths ("quick" to the outdoor air temperature the past 3 weeks, "slow" to the monthly ground temperature, and "constant" to the deep ground temperature). The model is not only more accurate, but more flexible for different building conditions. If anyone's interested in the report, it's available here.

Fred and I implemented this model in a custom version of DOE-2.1E with new keywords, temperatures,etc., but it never got incorporated into the public version, which by that time was basically dormant. However, I would be happy to share the source code and executable with anyone interested.