As mentioned in your answer, terrian conditions such as
presence of trees and shrubs have great effect on the wind speed. I wonder if all the data are collected in the same site or similar terrians?
Could you confirm if there is any other bad data in the training sets? It seems that PairTrng6 involves fault by visually looking at the plotted mean wind speed of the two anemometers. Could you just let us know if this data set is health or not?