One of the more interesting things the paper tried to tease out is whether centralization of Lightning nodes reduced demand for blockchain space. They suspected that a more decentralized LN would require more channels to be opened/closed as nodes attempt to connect to others and make payments, but their results didn't seem to indicate that. It might be some kind of methodological or data deficiency, but nonetheless it's an interesting finding.

On my reading, I found it strange that they didn't realize they could recognize and quantify channel opens/closes on the timechain, because channel opens/closes are pretty easy to detect for the time being.