An improved ensemble of land-surface air temperatures since 1880 using revised pair-wise homogenization algorithms accounting for autocorrelation.

An 100-member of homogenized monthly land station temperatures from 1880--2023 (Chan et al., 2024). To build this dataset, we apply two automated algorithms (Chan et al., 2024), which accounts for autocorrelation in temperature series, to raw station temperature records compiled under monthly Global Historical Climate Network version 4 (GHCNmV4). The first algorithm, which constitute the 50 members of the ensemble, uses improved standard homogenisation test for breakpoint detection. And the second algorithm, which contribute to the other 50 members, employs a technique called penalised likelihood. Both algorithms remove discontinuities in temperatures arising from changes in measurement approaches or environments. The spread across ensemble members characterizes uncertainties associated with using different combinations of the algorithmic parameters, as well as errors in the magnitude estimate of individual discontinuities.