Leading with Consequences: A Petascale Ensemble Experiment to Support the Discovery of High Consequence Multi-sector, Multi-scale Climate Change Scenarios

Future climate change, societal evolution, and their interaction abound with uncertainties that are diverse and deep. Reflecting the complexity of that space, a new scenario matrix framework was developed that encompasses shared socio-economic pathways (SSPs), shared policy assumptions (SPAs), and different climate targets (i.e., representative concentration pathways or RCPs). These elements can be used in combination to generate specific scenarios for use in climate change research. Central to that framework are five canonical SSPs that seek to span a range of assumptions about challenges to mitigation and challenges to adaptation. In this work, we demonstrate that the current approach for selecting reference scenarios often hides important consequences, particularly when considering time-varying regional impacts across key sectors. As an alternative approach, this work discovers consequential scenarios by interrogating a large dataset of 33,750 scenarios generated by sampling broadly across the SSP and SPA space using the Global Change Assessment Model (GCAM). Statistical techniques are used to discover which assumptions are particularly consequential, considering a broad range of time-evolving metrics that encompass multiple spatial scales and sectors. Our results highlight that narrowly sampling five canonical SSPs to perfectly attain RCP targets provides a very narrow view of the consequence space, increasing the risk of tacitly ignoring major impacts. Even optimistic scenarios contain unintended, disproportionate regional impacts or intergenerational transfers of consequence. Formulating consequential scenarios of deeply and broadly uncertain futures requires a better exploration of which quantitative measures of consequences are important, for whom are they important, and when. To this end, we have contributed a large database of climate change futures that can support `backwards' scenario generation techniques, that capture a broader array of consequences than those that emerge from limited sampling of a few reference scenarios.