EGU General Assembly 2022
© Author(s) 2022. This work is distributed under
the Creative Commons Attribution 4.0 License.

Reconstructing zonal precipitation from sparse historical observations using climate model information and statistical learning

Marius Egli1, Sebastian Sippel1, Angeline Pendergrass2, Iris de Vries1, and Reto Knutti1
Marius Egli et al.
  • 1ETHZ, IAC, Zürich, Switzerland (
  • 2Cornell University, Earth and Atmospheric Sciences, Ithaca, NY, US

Changes in precipitation due to climate change are having and will continue to have substantial societal impact. Although physical process understanding allows insights into some of the model-projected changes, we face many challenges when turning to observations in order to detect these changes, such as large internal variability and limited observational coverage both in time and space.

Here, we aim to address these challenges with a tool from statistical learning, by implementing a regularized linear model to (1) reconstruct historical seasonal full (land+ocean) zonal mean precipitation starting in 1950 and (2) detect anthropogenically forced changes in zonal mean precipitation. The linear model is trained using a climate model large-ensemble archive with its coverage reduced to match gridded station observations on land only. Once trained, the linear model can reconstruct the full zonal mean precipitation from the partial coverage given by observations. The reconstructions (1) are compared against independent satellite observations and other sources of historical precipitation reconstructions. Our approach is successful at recovering a large part of the variability in zonal precipitation. In the Northern hemisphere extra-tropics, with relatively high station coverage, the reconstructions achieve an agreement of R=0.8 (Pearson correlation) or higher with independent satellite precipitation. But correlation values decrease considerably in the Southern hemisphere and parts of the tropics. Next, we estimate trends in the forced response (2) in seasonal zonal-mean precipitation, many of which lie outside the likely range in a preindustrial climate. The detected trends are, in line with the projection of climate models forced with historical greenhouse gas and aerosol emissions but are sensitive to the underlying observational data set.

Our results show that for large scale metrics such as seasonal zonal mean precipitation our reconstruction method can facilitate new insights for the detection and attribution of changes in the hydrological cycle. 

How to cite: Egli, M., Sippel, S., Pendergrass, A., de Vries, I., and Knutti, R.: Reconstructing zonal precipitation from sparse historical observations using climate model information and statistical learning, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-4619,, 2022.