Models and Observations for Two CMIP5 Ensemble Means

I found csv file of the CMIP5 ensemble means for RCP4.5 and 8.5 on the Climate Reanalyzer website, so I downloaded both so I can do my own comparisons. I plotted multiple GMST datasets including 2 reanalyses with both RCP ensemble means from CMIP5. RCP4.5 over the last 15 years has been trending slightly warmer than observations. I don't have the 95% uncertainty envelopes, but observations here are clearly going to be well within the envelope.  

But I decided to plot 30-year trends for these datasets with the expected 30 year trends in the two model ensemble means. It shows that while the models have expected a higher trends in warming over the last 15 years or so, RCP4.5 expects a steep drop off in warming rates right about now. This suggests to me that these scenarios are expecting a deceleration of our emissions in pretty substantially in ways that simply isn't happening. If current trends continue, we're likely to surpass RCP4.5 over the next decade or two, while RCP8.5 may still be relevant to future warming.

So I decided to go back to the IPCC's AR5 report and look at what the CO2 pathways for the various CMIP5 scenarios. It shows RCP4.5 scenario trending with more emissions than RCP6.0 currently, but this does show a deceleration in emission rates, which plateaus before 2050 and after which, emission rates decrease. This seems to me to explain why it is that RCP4.5 expects slightly more warming than observations now while we are almost certainly going to surpass RCP4.5 by 2100 if we continue as we are.  After all, the pathway is named after the destination of radiative forcing increase by 2100 - it's a pathway that leads to an increase in radiative forcing to 4.5 W/m^2. But there are several ways in which we can get there, and it looks like RCP4.5 expects more emissions currently but much fewer emissions by 2030. This leads to decelerating temperature increases through 2100.