A dyadic structure is one with two dimensions, presumably space and time in your application. Let's begin with space. First, I want to divide space into nesting squares. I am not a member of the Flat Earth Society, so I mean this only in an approximate sense. Good for a city, but probably not a State and certainly not a country or a continent. Next, data can come at any location, inside a square, at a vertex of squares, or on a straight line between vertices. We are interested in something that occurs at these data points (e.g., density or property value, or perhaps something else). Let's just call this thing we are interested in our dependent variable. Now I want to estimate surfaces over these squares such that the surfaces are continuous from one square to the next contiguous square. However, these surfaces are not usually differentiable along the lines that connect the vertices. The method you would use to estimate the surface is found in Article A Primer on Piecewise Parabolic Multiple Regression Analysis...

It turns out that the regression you will run will have parameters that are essentially the heights of the surface over the vertices. So now we can begin to consider the time dimension. Of course, you need a time variable. Let's keep it simple for now and say that our time variable is just linear time, month numbers, year numbers, or whatever floats your boat. Finally interact (meaning multiply) your location variables by your time variable.

This should do it. You can now predict the surface of the dependent variable at any point in time.

http://people.soc.cornell.edu/strang/articles/Spatial%20and%20Temporal%20Heterogeneity.pdf

From the technical point.

I you use R :https://cran.r-project.org/web/views/SpatioTemporal.html

But, I guess (I am R user) matlab offers fancier stuff for this specific issue (it is har for me to admit).