If you are going to tweak your GW model variables to curve fit the 'real world' with a temperature increase over time for the last 100 years then the model will no doubt continue to show future GW. But I suspect there may have even had some good fits to the 'real world' but then suddenly nosedived (global cooling) in the future. This can happen when attempting to model closed non linear heterogeneous systems with vast mathematical matrix's. But I bet they get thrown out or dismissed.
I would suggest the following for some GW expert. Might make a nice PhD for someone. Go back to the historic 'real world' data sets (last 100 years) for each mercury thermometer then compensate the temperature data for the local thermometers by build an historic micro climate model to adjust the temperature record data for that thermometer. The 'Real World' temperature curve might suddenly flatten out with everything within +-0.5 degrees Celsius of 14.0C i.e. the measurement error