> Both things are true: The climate system is vastly complex, and we’re certain about what we are doing to it. How can we be so confident in a hundred-year projection when we can’t predict the weather with any reliability more than a week out?

TFA doesn't really seem to address this directly.

But it seems straightforward to me. It's because the hundred-year prediction is a much simpler claim (and also conditional on various models of future human behaviour), and also about a single-dimensional metric that readily allows for noise to be averaged out. When you get into questions of the detailed effects of an average global temperature increase (like sea level rises, loss of biodiversity or even, like, average local temperature increase in specific locations) it seems like the error bars are wider.

It's not any more paradoxical than noting that we can extract a trend in historical global average temperature, even when the slope of that trend is on the order of degrees per century within data that includes variations of tens of degrees per half-year, or even per day.