Interesting thoughts!
Since this is a game and not a professional simulator it's unrealistic to expect a detailed and accurate radar model, especially since radar performance data is as classified as it gets. And I also know that line of sight calculations are complicated stuff that require quite some computing power.
With that said, I have been thinking about what makes this different from the visual representation and if the logic behind this would be applicable to radar as well. I mean, look at an object inside a bunker from above and all you see is the roof of that bunker. Obviously there's logic that masks out the object inside the bunker, so shouldn't it be possible to use that logic to mask out the object for the radar as well?