An integrated reconstruction problem of past climates in statistical analysis

Shi, Feng [UCL] Goosse, Hugues [UCL] Yin, Qiuzhen [UCL] Guo, Zhengtang

The natural climate varies on time scales ranging from interannual to millennia or longer. It is impossible to explore these time components using the instrumental climate records of the last 150 years during which anthropogenic impact already occurs. Paleoclimate reconstructions are suitable for this purpose. Multiple types of proxy records have been used to extend climate information to pre-instrumental period, such as tree rings, ice cores, speleothems, corals, and ocean and lake sediments. In view of enormous achievements in the time series reconstruction of temperature and precipitation at a single location based on one type of proxy, synthesized results of previous reconstructions becomes more and more highly developed to obtain large-scale climate variability. This leads to the question about how to obtain an optimal climate reconstruction using multi-proxy records, especially through statistical analysis. This involves two aspects, data and method. Each type of proxy record has its benefits and limitations. Most proxy indicators are controlled by more than one factor. For example, the tree-ring width chronology is widely used to reconstruct the climate variability over the past millennia because of its accurate dating, high resolution, wide distribution, and ease of duplication. However, the tree-ring width chronology is usually co-affected by both temperature and precipitation. In addition, the tree growth trend also restricts to extract low-frequency climate variability in tree-ring width records. Thus, the tree-ring width chronology is considered as a temperature signal with large non-temperature noise. Moreover, in various reconstruction methods, different regression equations are used to calibrate the proxy records during the instrumental period. Different regression coefficients would make different reconstructions agree well on the interdecadal time scales, but not for low frequency variability. The cooperation between climate science and applied statistics has started to exhibit substantial efficacy to solve these problems.