Houghton Lecture - Dr. Benjamin Santer (UCLA)

Fingerprint research seeks to improve understanding of the nature and causes of climate change. The basic strategy is to search in observed climate records for the patterns of climate change (the “fingerprints”) predicted by a computer model. Fingerprint studies exploit the fact that different factors affecting climate have different characteristic signatures. These unique attributes are clearer in detailed patterns of climate change than in records like the average temperature of Earth’s surface. Fingerprinting is a powerful tool for separating human and natural climate-change signals. Results from this research provide scientific support for findings of a “discernible human influence” on global climate and contributed to work recognized by the 2021 Nobel Prize in Physics.

Twenty-seven years ago, at the time of publication of the “discernible human influence”
finding in the Second Assessment Report of the Intergovernmental Panel on Climate Change (IPCC), most fingerprint studies relied on surface temperature. Critics of this work argued that
a human-caused fingerprint should be identifiable in many different aspects of the climate system - not in surface thermometer records alone. Climate scientists responded to this justifiable criticism by moving beyond early “temperature only” fingerprint studies, interrogating modeled and observed changes in rainfall, water vapor, river runoff, atmospheric circulation, salinity, and many other independently monitored climate variables. As the 2021 Sixth IPCC Assessment report recently concluded, the existence of human-caused fingerprints in the climate system is now “unequivocal”.

My lecture will look back at key scientific milestones in the historical evolution of climate fingerprinting. It also addresses some of the scientific challenges ahead, particularly in terms of communicating the results of fingerprint research to policymakers and the public.

In October 2021, the Sixth Asessment Report of the Intergovernmental Panel on Climate Change (IPCC) concluded that “it is unequivocal that human influence has warmed the atmosphere, ocean, and land”. Despite this definitive finding, there are widely publicized claims that uncertainties in disentangling human and natural influences on climate are far greater than scientists have acknowledged. Under this narrative, findings of “unequivocal” anthropogenic fingerprints are premature.

This “uncertainties are underestimated” narrative has little basis in fact. For nearly three decades, scientists engaged in climate change detection and attribution (“D&A”) routinely explored how their findings are affected by uncertainties in climate models and observational data. D&A analysts also considered whether positive identification of human-caused climate fingerprints was sensitive to the choices made in selecting and applying D&A methods.

Today, state-of-the-art D&A studies are typically conducted with large multi-model and single-model ensembles. D&A research explicitly considers uncertainties in model anthropogenic fingerprints, external forcings, and natural internal variability estimates, as well as in the observations themselves. An array of different D&A methods is employed, including simple pattern correlation approaches, “optimal detection” techniques, machine learning, and anchor regression. Additionally, analysts assess the sensitivity of D&A results to the quality of the models selected, the inclusion or removal of global-mean information, the climate variable considered, and whether the D&A analysis involves single variables or is truly multivariate.

My lecture will give numerous examples of such “stress testing” exercises. It will illustrate that the IPCC’s 2021 finding of “unequivocal” anthropogenic fingerprints is supported by compelling evidence that is remarkably robust to current scientific uncertainties.

This lecture was motivated by a close encounter with Mount St. Helens in April 1980. Since then, I have had a long- standing fascination with volcanic effects on climate.

As in the case of anthropogenic signal identification, identifying volcanic effects on climate is a signal-to-noise (S/N) problem. This S/N problem is complicated by the fact that the surface and tropospheric cooling signals of the two largest recent volcanic eruptions (El Chichón in 1982 and Pinatubo in 1991) were partly obscured by short-term warming induced by two separate El Niño events. Standard multiple linear regression approaches can be affected by this fortuitous “ENSO masking” of volcanic cooling. An iterative regression method partly resolves the problem of collinearity between volcanic signals and ENSO variability, yielding more reliable estimates of the true cooling signals induced by El Chichón and Pinatubo. Such noise removal issues are important to consider when analysts compare smoothed, multi-model average volcanic cooling signals from the Coupled Model Intercomparison Project (CMIP) with volcanic signals estimated from the single noisy realization of observations.

Following the pioneering 2011 Science paper by Solomon et al., considerable scientific attention was focused
on the climatic impact of a succession of “moderate” post-2000 volcanic eruptions. When ENSO variability is removed from observations, signals of these “moderate” eruptions are statistically identifiable in a wide range of different climate variables. The eruptions increased the stratospheric aerosol optical depth (SAOD), leading to negative volcanic radiative forcing over the early 21st century. Models participating in phase 5 of CMIP neglected this negative volcanic forcing, thus introducing a systematic warming bias relative to observations. One important lesson learned from this work is that such forcing errors can have non-negligible impact on inferences regarding consistency between modeled and observed temperature trends. This is evident from numerical experiments in which the same climate model is run with both CMIP5 and CMIP6 volcanic forcing.

Several recent modeling studies rely on simulations with prescribed volcanic emissions of SO2 rather than on simulations with prescribed stratospheric distributions of volcanic aerosols. Such studies allow analysts to consider whether information on total column volcanic aerosol burdens yields more confident volcanic signal detection than use of stratospheric aerosol burdens alone. Use of SO2-driven simulations may be particularly important for detecting the climate signals of high-latitude eruptions and of effusive volcanic eruptions with aerosol burdens that are primarily confined to the troposphere.