This is an interesting way of visualising the error; the red-blue scheme works better than the yellow-brown IMO. However, if the discharge series were clearer, I don't think it would be necessary. Have you tried reducing the line thickness, and making the simulated series solid rather than dashed?
Also, you may want to avoid the red-green combination if possible for people with colour-blindness. So long as the colours chosen are sharply different, this ought to make them quite distinct.
The lower plots are much clearer but the coloured bars highlight the errors very nicely without having to decipher the plots. A good idea - I might try this myself. I also agree you should avoid red/green and probably red/blue for the plots Red blue bars work OK though.
This is a very interesting perspective. Thanks for the good insights. I will be very interested in peeking at some implementation (in R, preferably), would you like to share some bits of it?
Also, I am interested in a slightly different case. Let us say I have two series of outputs from different modelling setups, which I would like to compare to a series of observed values. And I want them superimposed in the same graph. How would you go with this case?