It's actually funny because you've noticed the issue but can't seem to understand it.
Let's try and think: if a data set shows as noise, it means one of two things. One, it could be that the data is noise. Two, it could mean that the trend requires a much larger amount of data to determine. Bingo! If your uncertainty is significantly higher than any possible trend, it's pointless to use the data. Instead, you look for more data, and lo and behold, you find that when you include the full range of data all the way back to 1980, you get a nice upward slope. 15 years isn't enough, for a system as complicated as our planet, to plot a trend from. Even 35 years isn't enough, really, which is why you generally try to get other data sources going back farther (which is what climate scientists do, go figure).