Do internals of neural networks make sense in the context of hydrology