In the previous study by Kwong and Chan (2008), CONN is applied to 2D wind field measured by the Doppler LIDAR at the Hong Kong International Airport (HKIA) to forecast the evolution of the airflow. Due to the large amount of LIDAR data involved in 2D geometry and the constraint of computational power, the training dataset is very limited in size (data of a few hours only) and the forecasting capability is much restricted. The present study adopts an alternative approach, namely, focusing on the winds along the glide paths of the aircraft only (and thus training of CONN and forecasting of the wind in a 1D domain) but considering data of a longer time period (in the order of several days). As shown in case studies, this latter approach turns out to perform better than the 2D approach in capturing the evolution of sea breeze and terrain-disrupted airflow, which are two major causes of low-level windshear to aircraft at HKIA. The formulation of CONN for 1D domain and the wind forecasting in the two cases would be discussed in detail in the paper.