How Ominous is the Premonition of Future Global Warming?

Article Type

Research Article

Publication Title

Sankhya B

Abstract

Global warming – the rise in global average temperatures observed in recent decades – has drawn significant attention due to its profound and far-reaching impacts on the climate system. A critical question is whether this warming trend will continue into the future. General circulation models (GCMs) are the primary tools for projecting such future climate scenarios, and nearly all of them forecast an alarming increase in global temperatures. While the reality of current global warming is undeniable, the reliability of GCM forecasts remains open to scrutiny. In this study, we undertake a systematic evaluation of these forecasts using our recently developed Bayesian multiple testing framework for model selection in inverse regression problems. Our central question is: How likely is the current global warming pattern, assuming the future projections of the GCMs are correct? This reframes the typical forecasting paradigm into an inverse regression setting, where the present is treated as unknown and inferred from future outcomes. Our framework coherently combines this inverse formulation with conventional forward modeling to identify the best-fitting models. To model the temporal dynamics of global temperature, we adopt a nonparametric compositional Gaussian process (GP) emulator that treats the climate system as an unknown black-box process. Using data from the Intergovernmental Panel on Climate Change (IPCC), we find that GCMs which perform best under various future scenarios still fail to convincingly account for the observed warming trend – if one assumes only their projected future data are accurate. We further analyze ensemble forecasts from all GCMs under each scenario as multivariate time series governed by multidimensional GPs. The inverse-fit results from this multivariate framework strongly reinforce the conclusions drawn from the univariate analysis: assuming the GCMs’ future predictions are valid, the observed global warming pattern appears highly improbable. This casts serious doubt on the representativeness of IPCC-endorsed GCM projections. Lastly, we offer our own forecasts of future global temperatures based solely on the historical data, using our GP-based model. These forecasts do not support the drastic warming trends predicted by most GCMs. In fact, only the projections under the “Commitment" scenario fall within the high-density regions of our Bayesian forecast distributions.

DOI

10.1007/s13571-025-00390-y

Publication Date

1-1-2025

Link to Full Text

Share

COinS