Posted by Dr. Sebastian Wagner, department Climate Extremes and Impacts
Our colleague and PhD student Zeguo Zhang defended on 20 February, 2023, successfully his thesis Reconstruction of climate fields using machine-learning methods at the University of Hamburg.
A typical problem in current and historical climate research relates to the sparseness of available observations and/or reconstructions providing a spatially resolved picture. Zeguo picked up this challenge and applied novel machine learning (ML) methods in his PhD thesis to reconstruct spatially resolved climate fields and comparing his results with traditional methods.
He applied the methods to different fields in geophysical research: A first application was on sea level height reconstructions in the North Sea using only a very limited number of gauge observations bordering the North Sea coast. With the help of machine learning methods and using results of highly resolved ocean model simulations he was able to successfully reconstruct spatially resolved fields of sea level changes over the entire North Sea, with remarkable skill especially for low-pass filtered gauge data.
A second application was on paleoclimate research in the context of so called pseudo-proxy experiments: Here he used the output fields of current global climate simulations, selecting grid points for temperature where tree rings proxy data are available over the Northern Hemisphere. Similar to the gauge observations he used this sparse network to reconstruct spatially resolved fields and comparing them with the direct spatially resolved output fields of the global climate simulations. It turned out that especially one ML method is capable of improving the reconstruction of temperature fields in contrast to classical methods by better reproducing the amplitude of past temperature variations.
