Calculating the Temperature Trend Bias Induced by Inhomogeneities Into Climate Data Without Running a Homogenization Algorithm

Inhomogeneities are known to induce sudden jumps into the time series of climate stations. If these jumps tend to be upwards, a spurious positive trend will be inserted into the data, which would falsify the true climate trend. The classic method to identify such biases is to run a homogenisation algorithm and to assess the statistics of the found jump heights. However, in the past it was shown that already small detection errors lead to a strong systematic underestimation of such trend biases. Therefore, an alternative method is proposed that calculates the trend bias directly from the original data without previous homogenisation. First, the Composite Reference (CR) technique is applied where networks of neighbouring stations are compiled and one station is selected as a candidate from which the average of the others is subtracted. In this way, not only the common climate signal is eliminated, but also the common trend bias. However, we do not use the CR data directly, but consecutive differences of it, thus the change from year-to-year within the CR time series. In this way, large data volumes are available which are of course dominated by noise, but also the effect of a possible trend bias is traceable. Every break occurring in the candidate arises also in the reference of all other stations, where it is attenuated by averaging and with opposite sign, as the reference is subtracted. In this way, every inhomogeneity produces one large jump and many small ones with reversed sign so that the median is shifted. This effect is exploited in the proposed method. An application to temperature data from U.S. climate stations shows that there is no significant trend bias caused by inhomogeneities.