Over the past 2.58 million years — a period scientists call the Quaternary — Earth’s climate has swung repeatedly between cold glacial periods and warmer interglacials. A new study in Nature Communications uses a GCM (General Circulation Model) paired with a faster statistical model called an emulator to investigate what drives these shifts. Because running a GCM continuously over millions of years would take decades of supercomputing time, the researchers instead used one to generate 182 training simulations, then used those results to calibrate the emulator.
The emulator learns the relationships between five key inputs — three orbital parameters (eccentricity, obliquity, and precession), atmospheric CO₂ concentration, and global sea level as a proxy for ice sheet size — and the climate outputs the GCM would produce. Once trained, it can generate a full global map of temperature and precipitation for any combination of those inputs in minutes rather than years. This allowed the team to simulate 2.58 million years of climate history, plus 31 additional sensitivity experiments, at a fraction of the cost of running a conventional climate model.
To validate the results, the researchers compared the emulator’s output against real-world climate records known as proxy data, including ice core temperature reconstructions from Antarctica’s Dome C site and ocean sediment records from across the Pacific. The emulator matched these records reasonably well, particularly over the last 800,000 years, capturing the timing and general magnitude of glacial-interglacial cycles. The strongest agreement was found at Dome C, which achieved a statistical skill score of 620 out of 1,000. Some discrepancies appeared, especially during certain interglacial periods and further back in time, likely due to uncertainties in both the proxy data and the CO₂ records used to drive the model.
The study’s most significant finding came from sensitivity experiments designed to isolate which factor contributes most to global temperature change. CO₂ accounted for roughly 54% of the overall temperature signal, while ice sheets contributed about 35%. The three orbital parameters — despite being the pacemakers of ice age cycles — combined for less than 12%, with each individually below 5%. This means Earth’s orbital geometry does not drive ice ages primarily through direct heating or cooling effects, but rather by indirectly triggering changes in CO₂ and ice sheets, which then amplify the signal. The direct radiative effect of orbital shifts on annual mean temperature is, on its own, negligible. Orbital forcing does, however, play a more meaningful role in seasonal patterns, particularly monsoon precipitation, which the GCM-trained emulator captured to some degree.
Regionally, CO₂’s influence was strongest over tropical landmasses such as South America and southern Africa, while ice sheets had the greatest impact over polar regions including Antarctica and Greenland. Together, these two feedbacks dominate the long-term climate record. The findings reinforce and extend previous work using simpler models, offering a clearer picture: over the Quaternary, greenhouse gas concentration and ice sheet extent are the primary drivers of global temperature change — not the orbital cycles that set them in motion.