Forced Component Estimation Statistical Methods Intercomparison Project (ForceSMIP)

Anthropogenic climate change is unfolding rapidly, yet its regional manifestation is often obscured by atmosphere-ocean internal variability. A primary goal of climate science is to identify the forced response, i.e., spatiotemporal changes in climate in response to greenhouse gases, anthropogenic aerosols, and other external forcing, amongst the noise of internal climate variability. Separating the forced response from internal variability can be addressed in climate models by using a large ensemble to average over different possible realizations of internal variability. However, with only one realization of the real world, it is a major challenge to isolate the forced response in observations, as is needed for attribution of historical climate changes, for characterizing and understanding observed internal variability, and for climate model evaluation.

In the Forced Component Estimation Intercomparison Project (ForceSMIP), contributors used existing and newly developed statistical and machine learning methods to estimate the forced response during the historical period within individual ensemble members and observations, across eight key climate variables (SST, surface air temperature, precipitation, SLP, zonal-mean atmospheric temperature, monthly max. and min. temperature, and monthly max. precipitation). Participants could use five CMIP6 large ensembles to train their methods, but they then had to apply their methods to individual evaluation members, the identity of which was hidden. Participants used methods including regression methods, convolutional neural networks, linear inverse models, fingerprinting methods, and low-frequency component analysis. Here we show how the different methods performed on climate models and what they determined to the be the forced response in observations. Our results show that many different types of methods are skillful for estimating the forced response and that the most skillful method depends highly on which variable and metric is evaluated. Furthermore, methods that show comparable skill can give very different estimates of the forced response in observations, illustrating the epistemic uncertainty in estimating the forced climate response from observations. ForceSMIP gives new insights into the forced response in observations across multiple key variables, but also the remaining uncertainty in its estimation.