Improved Homogenisation of Observations Shows Steadier and Faster Historical Global Warming

Instrumental climate records are foundational for understanding Earth's climatic history, not only for quantifying historical climate changes but also for benchmarking climate models and refining the accuracy of climate predictions. Despite their crucial role, the task of estimating Earth's surface temperatures since 1850 —through data gathered from an extensive network of terrestrial weather stations and oceanic voyages -— presents substantial challenges. Specifically, early temperature measurements are fraught with biases and errors from diverse sources, complicating efforts to achieve a consistent and homogeneous dataset.

In this talk, I will review the history and recent advances in statistical methods for homogenising ocean and land temperature records, as well as the inter-calibration across these two data archives. These include a physics-informed group-wise inter-comparison method for homogenising sea-surface temperatures, an improved pair-wise homogenisation algorithm for land station temperatures, and a coupled model of air-sea heat exchange for inter-calibrating ocean and air temperatures along coasts. Compared with existing approaches, these methodological advancements improve the accuracy of detecting, estimating, and adjusting data biases. Multiple lines of evidence including metadata, physical knowledge, paleo proxies, and historical documents are used to check the results of these statistical analyses.

Applying these methodological advancements, we first homogenise ocean and land temperature archives individually and then across the two archives to obtain a consistent estimate of global temperature records, which we call the Dynamically Consistent ENsemble of Temperature (DCENT). DCENT is a global, gridded data set that features 200 members that comprehensively represent uncertainty.

DCENT shows a steadier and 10% larger increase in global mean surface temperatures since 1850. The steadier warming in DCENT better aligns with the expected temperature responses to external radiative forcing, implying a reduced role for internal variability in historical warming patterns. We envision DCENT to be a suitable dataset for studying decadal variabilities and their underlying mechanisms. Moreover, the greater warming points to an earlier surpassing of the 1.5°C target set forth in the Paris Agreement, with 20-year-average temperatures expected to exceed this threshold by 2028.