Do Inhomogeneities Shift the Global Observed Temperature Trend? (DISGOT Trend)

点击翻译按钮获取中文摘要

Temperature trends in climate data are affected by inhomogeneities, which are typically caused by relocation of the stations or changes in observation methods. It is undisputed that these breaks have the potential to severely falsify the individual trend of a climate station. However, only if the introduced jumps are biased, i.e. being non-zero on average, they induce a mean trend bias into the data and not only some scatter that largely cancels out by averaging. Consequently, such biased inhomogeneities are most important, but they are also hardest to detect with relative homogenization methods. Thus, the main goals of the proposed project are to verify or reject the existence of a global mean trend bias due to inhomogeneities and, if applicable, to estimate its magnitude and uncertainty.In the previous project, we tested prototypes of commonly used homogenization algorithms that consist of a combination of those components, which we expect from the literature to be crucial. These are the applied detection and correction modules or the method to build difference time series. The generally good performance of the modern homogenization algorithms, found by earlier Benchmark studies, could be confirmed. However, in the past the focus was often on break detection scores and the improvement of station series, not on the network-mean trend bias. Our preliminary work concentrated just on this specific issue and found mixed results. One of the two principal algorithm types seems to have general problems with the mean trend bias, the other works satisfactory only if dense station networks are available. Thus, there are strong indications that the mean trend bias, at least on global scale, is difficult to detect by current methods, because the station density might be too low in large areas worldwide. However, a full coverage is indispensable for the determination of the global climate trend.The proposed project is structured in four major Work Packages. In WP 1, we will analyze the preliminary indications that the mean trend bias is hardly detectable by current methods and identify the responsible shortcomings of each prototype. WP 2 derives the signal-to-noise ratio (SNR) for different regions of the world, which is the key parameter for the reasonable functioning of any homogenization algorithm. WP 3 tests the real algorithms whether they have similar problems as their prototype counterparts. WP 4 proposes five alternative approaches to determine the mean trend bias and tests their ability to overcome the problems.