Study WP5.1 Machine Learning to Advance Climate Model Evaluation and Process Understanding

Description

This study is led by Lisa Bock. Additional contributors to this Study are Axel Lauer and Veronika Eyring from DLR.

The main CCI ECVs used in this Study are Cloud, Land Cover, Land Surface Temperature, Sea Surface Temperature, Water Vapour, Soil Moisture, Permafrost, and Snow.

It is estimated that this Study will run from September 2023 until March 2025.

The Study comprises three parts. The first focuses on enhancing observational products for climate model evaluation with machine learning. This involved developing and applying a Machine Learning (ML)-based approach to derive cloud classes from high-resolution satellite data and coarse-resolution climate models; the application of NN to ESA CCI Cloud data leading to timeseries of labelled ESA CCI Cloud data; and the use of this dataset for an evaluation of clouds by cloud classes in climate models (here: ICON-A). The second focuses on casual model evaluation for cloud regimes and land cover types by calculating casual networks from the timeseries of several cloud variables of ESA CCI data in order to analyse and investigate the casual connections among the cloud properties and their controlling factors. Then casual networks are analysed for different cloud regimes and different land cover types. The same method is then applied to output from global climate models (here: ICON-A) and resulting casual networks are then compared to the ones obtained from the observations in order to evaluate the models. Thirdly, the evaluation of CMIP6 models with the ESMValTool will be undertaken. This involves CCI Snow and Permafrost datasets being implemented into ESMValTool and whenever possible, the CCI uncertainty estimates will be used to assess whether differences in the model simulations compared with the observations are significant.


Results and conclusions

Results and conclusions will be provided once the Study is complete.