Spencer Patrick Clark: Pattern scaling in an overshoot world
Author Information
Spencer Clark (1,2), Andrew King (1,2), Zebedee Nicholls (1,3,4) & Malte Meinshausen (1,4)
- School of Geography, Earth and Atmospheric Sciences, The University of Melbourne, Australia
- ARC Centre of Excellence for the Weather of the 21st Century, Clayton, VIC, Australia
- Energy, Climate and Environment Program, International Institute for Applied Systems Analysis (IIASA), 2361 Laxenburg, Austria
- Climate Resource, Melbourne, Victoria, Australia
Introduction and Context
As human-induced warming approaches the 1.5ºC Paris Agreement limit, understanding the risks of overshoot, referring to a temporary exceedance and subsequent decline below this limit, is increasingly important. Systematically assessing such risks, however, remains challenging as Earth System Model (ESM) simulations exist for a select few overshoot pathways.
Methodology and Project Summary
In this work, we assess the validity of employing linear pattern scaling to emulate ESM behaviour for a range of overshoot pathways. Linear pattern scaling rests upon the assumption that local-scale climate changes scale linearly with changes in global average temperature. While this assumption has proven robust under transient warming pathways, its validity under overshoot remains less clear. We benchmark linear pattern scaling against scenario-driven and idealised ESM overshoot simulations and identify regions and climate variables of poor performance. In doing so, we seek to provide a benchmark by which other emulation methods for overshoot can be evaluated.
Results
We show generally poorer pattern scaling performance post-peak warming under overshoot, though in some regions the assumption of linearity performs reasonably well. Poorer performance stems not only from the Earth system’s gradual adjustment towards equilibrium but also from other nonlinear processes emerging under overshoot. We further demonstrate that a multivariate pattern scaling extension incorporating additional predictors, such as ocean heat uptake, consistently reduces emulation error under overshoot.
Audience
Our work is of relevance to researchers in climate model emulation, particularly those developing or evaluating machine learning methods for producing spatially-resolved projections under overshoot.
Key Words
Climate model emulation, Earth System Modelling, climate overshoot