Therefore, I gathered future predictions of FWI from 17 models at a 2.5-degree resolution. I then found the daily mean difference between the past (2001-2018) and the future (e.g. 2082-2099). I superposed these changes to the NARR, 0.5-degree resolution and assigned them to my ignitions. E.g. if the daily mean diff in temperature is 5 Kelvin for one pixel at day 150, and two ignitions occurred within this pixel on that given day, I would add the 5 K to the original temperature measured. Thereby, I have the structure of the NARR but with the mean difference of 17 models.
This would be fairly easy if I would look into the 18 years as a whole. But, due to the major revision in 2012, I have now two datasets split as mentioned before.