Group seminar on 25. April, 14:00 CET, Bundesstr. 53, Room 022/023

How can you be sure that your statistical model is correct ? Using climate simulations and pseudo-proxies as a virtual laboratory

Dr. Eduardo Zorita

Statistical models are ubiquitous in climate research, for instance, to interpret large data sets or to set up data-driven prediction schemes. One prominent area in which statistical models are necessary is the reconstruction of past climates based on indirect proxy indicators. However, as we cannot directly observe past climates, testing whether a given statistical method provides reasonable or accurate climate reconstructions is challenging. This has led to well-known controversies on the amplitude of past climate variability, for instance, the Hockey Stick controversy. In the early 2000s, we proposed a new approach to test statistical climate reconstruction methods. This approach uses the virtual climate generated by climate models and the generation of so-called pseudo-proxies. In this virtual reality, we can evaluate almost all competing methods, revealing their strengths and deficiencies and providing ways to improve them. This talk will review the motivation and developments of so-called Pseudo-proxy experiments, mainly focused on the Hockey Stick controversy, but with lessons for many other areas of statistical climatology.