Towards automated and global homogenization procedures for early instrumental temperature data.

Homogenization in climate research means the removal of non-climatic changes. This presentation describes the homogenization of the early instrumental dataset (HCLIM; https://doi.pangaea.de/10.1594/PANGAEA.940724) of monthly mean temperature time series. New homogenization algorithm validation methodology is assessed here on early instrumental data, and its use to assess the skill of three different algorithms, when applied to early instrumental data.This study addresses the challenge of homogenizing early instrumental meteorological observations spanning 365 years to accurately reconstruct historical climate change. Manual homogenization methods are time-consuming and prone to errors, necessitating an exploration of automated alternatives. We assess three current automatic homogenization algorithms—CLIMATOL, PHA, and BART—utilizing the global early instrumental dataset HCLIM, covering discontinuous monthly temperature records from 1658 to 2021, with record lengths varying from 15 to 260 years. Our analysis reveals significant differences in break frequency among algorithms, with BART detecting eight times more breaks than CLIMATOL and PHA. The ratio of homogeneous series also varies notably: PHA at 85%, CLIMATOL at 70%, and BART at 31%. Additionally, we evaluate algorithm performance by comparing detected records with collocated reference records from the 20CRv3 reanalysis product. Moreover, we identify certain data points unsuitable for homogenization due to missing neighbouring stations or apparent outliers, while highlighting their potential for identifying extreme climatic events. This research underscores the importance of rigorous assessment and validation of automated homogenization methods in historical meteorological records analysis. Furthermore, we find that both homogenization and the use of 20CRv3 are indispensable for refining results. By comparing the homogenized datasets with a reanalysis dataset, the results reveal that some data points necessitate homogenization, while others do not require this correction based on the reanalysis alone. Furthermore, the study demonstrates that utilizing three homogenization tools in conjunction with the reanalysis dataset yields optimal results for the early instrumental dataset Hclim.