Adjustment of Temperature Trends in Land stations After Homogenization (ATTILAH)

气候站的迁移和观测方法的变化导致了观测温度记录的不均匀性。有一些迹象表明，这些虚假的跳跃平均倾向于向下，因此它们在数据中引入了冷却趋势。均匀化算法通常用于去除这些伪趋势。然而，由于原则上的原因，完全修正是不可能的，类似于回归技术中唯一部分可解释的方差。特别是对于低信噪比（SNRs），当噪声方差比非均匀性引入的方差大时，实际必要的趋势校正只是基本的应用。如此低的信噪比普遍存在于世界上数据稀疏的时期和地区，这些地区的台站距离较远。不幸的是，对于全球平均值，这些孤立的站点权重很大，因为它们代表了很大的区域。因此，我们预计，即使在应用均匀化之后，全球温度趋势也会严重不足。利用模拟数据，我们将确定两种常用校正方案的性能。对于不同的实际信噪比，将确定所获得的和必要的校正之间的相关性。然而，均匀化算法的检测部分也会间接影响校正的性能。因此，通过测试八种原型算法来评估检测和校正的综合效果，这些原型算法代表了所有常用的均质方法。最后，这些信息将用于估计和纠正现有的、广泛使用的、已经均匀化的全球数据集的温度趋势偏差，并估计这些无偏趋势估计中的不确定性。

Relocations of climate stations and changes in observation practices induce inhomogeneities into the observed temperature records. There are some indications that these spurious jumps tend to be downward on average so that they introduce a cooling trend into the data. Homogenization algorithms are commonly applied to remove these spurious trends. However, for principle reasons a full correction is impossible, analogous to the only partly explainable variance in regression techniques. Especially for low signal-to-noise ratios (SNRs), when the noise variance is large compared to the variance introduced by the inhomogeneities, the actually necessary trend correction is only rudimentary applied. Such low SNRs prevail in data sparse periods and regions of the world, where stations to compare with are far away. For the global average, these isolated stations get unfortunately a large weight because they represent large areas. Thus, we expect the global temperature trend even after applied homogenization to be heavily undercorrected. Using simulated data we will determine the performance of two commonly-used correction schemes. The correlation between attained and necessary correction will be determined for different realistic SNRs. However, also the detection part of homogenization algorithms may influence indirectly the performance of the correction. Therefore, the combined effect of detection and correction is assessed by testing eight prototype algorithms that represent the full variety of the commonly-used homogenization methods. Finally, this information will be used to estimate and correct temperature trend biases of an existing, widely-used, and already homogenized global dataset and to estimate the uncertainties in these unbiased trend estimates.