Quoting the Environmental Resilience Institute: "To investigate the apparent cooling in coastal California, IU postdoctoral research fellow Alexander Charn and his colleagues took 32 downscaled simulations of the California coast and compared them to six different observational datasets of temperature trends in the same area. They found that of the four datasets that showed cooling, all had failed to take into account artificial jumps in temperature caused by historical changes in the instruments, such as weather stations and satellites, recording the data. The two datasets that did account for these changes showed coastal warming trends.
Though this research focused on California, Charn said that the lesson can be applied anywhere.
“Observational records anywhere, not just in California, have to be careful when calculating trends,” he said. “They have to account for artificial jumps in station records, caused, for example, by changes in station location. Datasets that do not take into account how the data is collected can ultimately distort the analysis of downscaled climate models.”
Charn and his colleagues also investigated how best to ensure that downscaled models predicted trends accurately. While models can be tuned to match historical observations of weather, Charn said it usually isn’t that simple.
“You can have a model that does really well in the past, but that doesn’t mean it will perform well in the future,” he said."