Method1

the first thing I try is using polynomial approximation of 2D azimuth and elevation space to ray hit length

pros : most general, inclusive

cons : warping of un-hitted region for polyfit / computational time(no problem confirmed for simulation)

---

I need to do some gaussian mixture model to this problem

method2

using sampled raycasting result, perform integer-mixed methodlogy

pros: don't need approximation

cons: limited proposal / no way to solve... your situation is too poor