Selecting the best climate model for future projections
Many end users of climate model simulations are often interested in projections of how the climate will change in the coming century. However, there are dozens of climate models from 20 different modeling groups that contributed to CMIP5, and there will be many more options form the 33 modeling groups that will contribute to CMIP6.
The default approach is often to take the multi-model mean projection without any selection or preference given to one model over another. One reason for this is that it is difficult to prove within the relatively short historical period that a given model is better than others at reproducing the aspect of the climate we are interested in and that skill in the historical simulations implies skill in future projections. However, this neglects some of the known biases within certain climate models, especially in regard to regional climate.
In a new paper currently under discussion in Biogeosciences we discuss this problem with respect to the climate conditions that drive the formation of E. Huxleyi blooms in Arctic seas. Previous work by colleagues at NIERSC used a machine learning approach to identify the climate variables which were good predictors of the formation and life cycle of E. Huxleyi blooms. We then assessed the skill of CMIP5 models in simulating the seasonal cycle, interannual variability, climatological trends, and signal-persistency of these variables within the historical period as a basis for selecting models for projection of future climate.