Abstract:
Changes in the hydrological cycle under climate change have significant implications for water resources, agriculture, and ecosystems. The large uncertainty in precipitation projections by Earth system models remains a major challenge for climate change impact assessment and climate action. A key parameter of the hydrological cycle changes is the future hydrological sensitivity (HS), which is the sensitivity of global-mean precipitation to temperature in climate change projections. HS has a large model uncertainty, in both Coupled Model Intercomparison Projects 5 and 6. While previous attempts to constrain HS indicate a possible link between HS under climate change versus natural variability, a robust constraint of future HS has not yet been found. In the first part of this talk, I present an emergent constraint on HS via a decomposition into its dynamical and thermodynamical components. Model uncertainty in HS under both natural variability and climate change arises from the dynamical component, which exhibits similar spatial patterns between the two regimes. Motivated by this inter-model relationship, we investigate the sensitivity of atmospheric circulation anomalies to temperature. This yields an emergent constraint on the weakening of the tropical atmospheric overturning circulation under climate change. This allows us to constrain the spatial pattern of the dynamical contribution to HS. In the second part of the talk, I present a custom machine-learning algorithm, designed to obtain the global spatial patterns relevant to emergent constraints. Application of this framework to HS produces results consistent with the traditional approach. This illustrates that machine learning has the potential to discover new constraints of climate-change metrics by processing complex spatial patterns of relevant variables.