What is downscaling?

Example of a value chain from different emission scenarios via global and regional climate modellers to end-users.

Downscaling is needed because all climate models have a minimum skillful scale due to their design, which make them unable to provide very detailed information about local climate characteristics. There is nevertheless a dependency of the local climate upon large-scale conditions that the climate models are able to simulate in a realistic fashion. This dependency is utilised through downscaling to infer the local response to a large-scale change (downscaling differs from bias-adjustment, where the latter does not necessarily involve a dependency between different scales). The Oxford Research Encyclopedia provides a background on downscaling: Downscaling Climate Information. Downscaling can also include artificial intelligence (AI) and empirical-statistical downscaling is a variant of machine learning. 

Online lectures ("Heavy MET talk") on Downscaling Part I & Part II.