Is it possible that the inversion layer creates a structure for sound to reflect/refract back down towards the sensor, when it in fact the original source was on the ground? You might not detect it laterally if there were a bunch of trees in the way.