Also, dT/dz (differentiation of T with respect to z) was ranging from -1.71 to 1.66 in the data that was provided (20 days) so in theory, any model trained wouldn’t be able to guess the case where the stability of air is at Class A (Pasquill Stability Class) which corresponds to dT/dz being < -1.9.

Well first because it’s easy and fast and it’s also applicable as we were trying to find a linear relationship between the weather and the concentration of PM10 at each station. We wanted to create a dense neural network but we couldn’t do that with the short amount of data. So I guess we were limited by the amount of data (20 days).

Can you please check if it’s working?