Applying machine learning techniques to compute vertical refractivity profiles in maritime environments

Within the Marine Surface Boundary Layer, evaporation ducting greatly affects the radar performances. The vertical refractivity profiles are classically computed by bulk models and these profiles can be fitted, with a very good accuracy, by analytical expressions involving four parameters to be determined. With machine learning techniques and particularly a decision tree regressor, we directly obtained these parameters from the physical bulk inputs. The resulting refractivity profiles are generally very close to the original ones and in terms of propagation results, we obtained mean errors around 1 dB by computing the propagation factor for classical radar scenarios.