Abstract
Inhomogeneities in climate data are the main source of uncertainty for secular warming estimates. To reduce the influence of inhomogeneities in station data statistical homogenization compares a candidate station to its neighbours to detect and correct artificial changes in the candidate. Many studies have quantified the performance of statistical break detection tests used in this comparison. Also, full homogenization methods have been studied numerically, but correction methods by themselves have not been studied much. We analyse the ANOVA (analysis of variance) joint correction method, which is expected to be the most accurate published method. We find that, if all breaks are known, this method produces unbiased trend estimates and that in this case the uncertainty in the trend estimates is not determined by the variance of the inhomogeneities, but by the variance of the weather and measurement noise. For low signal‐to‐noise ratios and high numbers of breaks, the correction may also worsen the data by increasing the original random unbiased trend error. Any uncertainty in the break dates leads to a systematic undercorrection of the trend errors and in this more realistic case the variance of the inhomogeneities is also important.
Highlights
The main obstacle to accurate long-term trend estimates is the presence of inhomogeneities that are hidden in the data (Parker 1994, Brohan et al, 2006, Aguilar et al, 2003, Menne et al, 2010, Brunetti et al, 2006, Begert et al, 2005, Auer et al, 2005)
More recent numerical validation studies looked at the performance of both the break detection and complete homogenization algorithms (Domonkos, 2008; Domonkos, 2011; Venema et al, 2012; Williams et al, 2012; Chimani et al, 2018; Killick, 2016)
Statistical homogenization algorithms consist of three parts, where the first is dedicated to detecting the break positions
Summary
The main obstacle to accurate long-term trend estimates is the presence of inhomogeneities that are hidden in the data (Parker 1994, Brohan et al, 2006, Aguilar et al, 2003, Menne et al, 2010, Brunetti et al, 2006, Begert et al, 2005, Auer et al, 2005). Statistical homogenization algorithms aim at detecting and correcting these spurious effects by comparing a candidate station to its neighboring reference stations. More recent numerical validation studies looked at the performance of both the break detection and complete homogenization algorithms (Domonkos, 2008; Domonkos, 2011; Venema et al, 2012; Williams et al, 2012; Chimani et al, 2018; Killick, 2016). The HOME benchmarking study for European climate station networks found that the performance for break detection and trend accuracy were only modestly correlated (Venema et al, 2012). This implies that the other components of homogenization methods, including break correction, are important
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
