To enable beyond-visual-line-of-sight (BVLOS) UAV flights for agricultural support in mountainous areas with steep terrain, it is necessary to estimate the flight range of a UAV in which stable communication can be maintained with a control terminal on the ground, allowing for the transmission of data such as conditions and observations of people or vehicles along the flight path. In the existing study, radio wave propagation simulation based on ray tracing methods which can account for terrain structure is utilized to estimate wireless communication quality centered around the operator terminal in mountainous areas. However, when incorporating complex, non-geometric structures such as vegetation, the computational complexity of the ray tracing method increases exponentially. As a result, it becomes difficult to perform simulations frequently, especially in agricultural areas where terrain structures change continuously. Therefore, in this study, we propose a new estimation method of wireless communication quality by integrating lightweight radio wave propagation simulation and machine learning technology. The proposed method roughly estimates the communication quality by utilizing the simulation considering only coarse terrain structures such as elevation at first and then improve the accuracy by utilizing the machine learning model that is trained so as to output the accurate communication quality based on the structure information including complex terrain features.