Doppler Light-Detection-And-Ranging (LIDAR) system is an essential tool for real-time wind monitoring for aircraft taking off and landing. Single LIDAR model is preferable in terms of cost and being free from synchronization problem of multiple LIDARs. There are many studies for single LIDAR based velocity estimation. In specifying the recognition for typical air turbulences, such as tornado, microburst or gust front, the parametric approach has been introduced in our previous research. However, this method suffers from a large computational time due to solving multiple dimensional and non-linear optimization problem by particle swarm optimization (PSO). Aiming at real-time monitoring, this paper introduces neural network based optimization approach to determine the turbulence model. The results from numerical simulation demonstrate that the proposed method considerably reduces the calculation cost without sacrificing an estimation accuracy, compared with that obtained by the former PSO based method.