A machine learning approach to rapidly project climate responses under a multitude of net-zero emission pathways

Abstract

Navigating a path toward net-zero, requires the assessment of physical climate risks for a broad range of future economic scenarios, and their associated carbon concentration pathways. Climate models typically simulate a limited number of possible pathways, providing a small fraction of the data needed to quantify the physical risk. Here machine learning techniques are employed to rapidly and cheaply generate output mimicking these climate simulations. We refer to this approach as QuickClim, and use it here to reconstruct plausible climates for a multitude of concentration pathways. Higher mean temperatures are confirmed to coincide with higher end-of-century carbon concentrations. The climate variability uncertainty saturates earlier, in the mid-century, during the transition between current and future climates. For pathways converging to the same end-of-century concentration, the climate is sensitive to the choice of trajectory. In net-zero emission type pathways, this sensitivity is of comparable magnitude to the projected changes over the century.

QuickClim, a machine learning technique that mimics outputs of climate model simulations at a fraction of the computational cost, could support rapid and efficient assessment of future climate responses to a wide range of carbon emissions scenarios and decarbonisation trajectories.