The researchers describe their methods and findings in this youtube video. "The most important part of our work was testing the skill of each of these approaches in reconstructing unobserved temperatures. To do this we took the observed data and further reduced the coverage by setting aside some of the observations. We then reconstructed the global temperatures using each method in turn. Finally, we compared the reconstructed temperatures to the observed temperatures where they are available... While infilling works well over the oceans, the hybrid model works particularly well at restoring temperatures in the vicinity of the unobserved regions."
The authors note that "While short term trends are generally treated with a suitable level of caution by specialists in the field, they feature significantly in the public discourse on climate change.""