New Study Underscores the Importance of Comparing Global Temperature Records
If there was a single thermometer that captured the Earth’s temperature, you could imagine the red liquid inside slowly creeping upwards, as global temperatures break record after record. But measuring average warming over the entire Earth’s surface is not nearly this simple. Estimates of global warming are constructed from complex sets of observational data gathered around the world. A new study published today in Nature Climate Change and led by Sofia Menemenlis, a Ph.D. student in the Atmospheric and Oceanic Sciences program at Princeton University, demonstrates that, while all datasets of global sea surface temperatures (SSTs) show significant warming over the satellite era, they lead to different conclusions about the precise rate of global warming.
SSTs are measured by a wide range of instruments: ships, buoys, marine observation stations — collectively referred to as in situ observations — and satellites that use infrared or microwave sensors. Several research centers around the world produce publicly available global SST datasets that are indispensable to scientists, climate modelers, and policy makers. The most prominent four are the UK’s Met Office Hadley Centre (HadISST), the U.S. National Oceanic and Atmospheric Administration (which publishes both ERSSTv5 and OISSTv2.1), and the Japan Meteorological Agency (which produces COBE).
These datasets use much of the same sea surface data but differ on a number of counts: how measurement biases are corrected, how SSTs are measured in areas with sea ice, and how incomplete data are estimated. There are some data input differences too. ERSSTv5, for instance, uses only in situ data.
Menemenlis and her co-authors show that these methodological differences can lead to differences in the magnitude of warming estimates. The four datasets show respective SST warming trends for the area between 60°S and 60°N (excluding the polar regions, where different datasets treat temperatures in areas of sea ice differently) during 1982–2024 of 0.108, 0.138, 0.154, and 0.184 °C per decade. While all datasets clearly show significant warming, the differences between them raise questions about how we interpret records of global warming.
Consider the record-breaking temperatures seen during 2023 and 2024, years that coincided with a strong El Niño event. To what extent was this temperature spike a historical anomaly — higher than prior years with El Niño? According to the HadISST dataset, the temperature spike during this period was far above previous El Niño years. But according to OISSTv2.1, the margin of record warmth was roughly on par with that of the 2015–2016 El Niño.
Surprisingly, these discrepancies mostly vanish when looking at global temperature datasets that include both ocean and land surface temperatures. But that’s not because the land data are so dominant — land makes up only about 26% of the planet’s surface between 60°S and 60°N. Rather, as the researchers found, it’s because these datasets rely on only two of the SST datasets — ERSST and HadSST — thereby omitting much of the variation seen across the full collection of SST datasets.
Every one of the SST and global temperature datasets — each a massive undertaking involving millions of individual data points and the cumulative efforts of thousands of scientists at institutions around the world — contributes to our estimates for total global warming. The discrepancies between them emphasize that they’re all necessary to the gradual process of zeroing in on the most accurate temperature estimates and better understanding the plausible range of historical global temperature trends. When even one agency loses funding, staff, or political support, the quality and continuity of all our climate data can suffer.
Menemenlis’s own research team had experience the need for this data firsthand. “We began by trying to understand the drivers of record-breaking warmth in 2023, which coincided with devastating weather extremes around the world,” she explains. “But we came up against these more fundamental questions about uncertainties in the observational global temperature record. The paper describes a straightforward but careful analysis that we hope will be useful to the climate science community.”
Menemenlis and her co-authors emphasize the need for collaboration in this system. They argue that the code and methods behind SST reconstructions should be made publicly available, so scientists can more easily work toward harmonizing estimates and better understanding uncertainties. Until then, they urge caution in treating any one dataset as definitive. The climate is undeniably warming, but pinning down exactly how fast it is remains an ongoing scientific endeavor that requires input and support from groups around the world.