Detection of inhomogeneities in daily climate records to study trends in extreme weather (daily stew project)

全球变化不仅可能影响长期平均气温和降雨率，还可能导致极端天气的变化，即热浪和寒潮、干旱、洪水和风暴可能变得更加常见。测量记录提供了一些证据，表明上个世纪可以找到极端降水的趋势。然而，目前尚不清楚气候记录的当前质量是否足以得出明确的结论。众所周知，长期气候记录包含不均匀现象，即不代表气候变化的变化，而是测量条件的变化(搬迁、避难所和仪器的变化等)。如果从元数据中知道这些不均匀，则可以消除它们，但在大多数情况下，需要通过与邻近站点的比较来识别它们(统计相对均质化)。目前，非均质性通常是按月到年的气候变量平均值进行分析的。我们将争辩说，许多不均匀主要影响每日数据分布的尾部，因此在聚集数据中可能无法检测到。因此，对极端趋势最感兴趣的不均质性很难被检测到。修正通常仅限于改变手段。对大部分温度分布的校正算法是可用的，但需要高度相关的相邻站点。由于变化，例如仪器方面的变化，经常在整个网络中传播，不均匀不仅可能导致趋势估计中更大的不确定性，而且还可能导致偏差。为了对极端趋势提供更可靠的估计，我们将开发一种新的温度和降水不均匀检测算法，该算法针对分布尾部的突变。我们建议不要修正数据，而是开发能够意识到这些突变并明确忽略它们的趋势测试。因此，我们绕过了用不完善的方法校正不均匀数据。这将使我们能够更可靠地估计极端情况下的趋势。有了这些工具，我们将研究德国的气候趋势。

Global change may not only affect the long-term mean temperatures and precipitation rates, but may also lead to changes in weather extremes, i.e. heat and cold waves, droughts, floods and storms may become more common. The measurement record gives some evidence that trends in extreme precipitation can be found for the last century. However, it is not clear whether the current quality of the climate record is sufficient to draw firm conclusions. Long climate records are known to contain inhomogeneities, i.e. changes that do not represent climatic changes, but changes in the measurement conditions (relocations, changes in shelters and instruments, etc.). These inhomogeneities can be removed if they are known from meta data, but in most cases need to be identified by a comparison with neighbouring stations (statistical relative homogenisation). Currently inhomogeneities are typically analysed in monthly to yearly means of climatic variables. We will argue that many inhomogeneities mostly affect the tails of the distribution of the daily data and may thus not be detectable in aggregated data. Consequently, the inhomogeneities that are of most interest for trends in extremes are poorly detected. Corrections are typically limited to changing the means. Correction algorithms for the bulk of the temperature distribution are available, but require highly correlated neighbouring stations. Since changes, e.g. in instrumentation, are often spread through the entire network inhomogeneities may not only lead to larger uncertainties in trend estimates, but may also lead to biases. In order to provide more reliable estimates of trends in extremes, we will develop a new inhomogeneity detection algorithm for temperature and precipitation, which targets breaks in the tails of the distribution. We propose not to correct the data, but rather to develop trend tests that are aware of the breaks and explicitly ignore them. Thus, we circumvent correcting inhomogeneous data with imperfect methods. This will allow us to estimate trends in extremes much more reliably. With these tools, we will study climate trends for Germany.