Machine-Learned Cloud Classes From Satellite Data for Process-Oriented Climate Model Evaluation

Abstract

Clouds play a key role in regulating climate change but are difficult to simulate within Earth system models (ESMs). Improving the representation of clouds is one of the key tasks toward more robust climate change projections. This study introduces a new machine-learning-based framework relying on satellite observations to improve understanding of the representation of clouds and their relevant processes in climate models. The proposed method is capable of assigning distributions of established cloud types to coarse data. It facilitates a more objective evaluation of clouds in ESMs and improves the consistency of cloud process analysis. The method is built on satellite data from the Moderate Resolution Imaging Spectroradiometer (MODIS) instrument labeled by deep neural networks with cloud types defined by the World Meteorological Organization (WMO), using cloud-type labels from CloudSat as ground truth. The method is applicable to datasets with information about physical cloud variables comparable to MODIS satellite data and at sufficiently high temporal resolution. We apply the method to alternative satellite data from the Cloud_cci project (ESA Climate Change Initiative), coarse-grained to typical resolutions of climate models. The resulting cloud-type distributions are physically consistent and the horizontal resolutions typical of ESMs are sufficient to apply our method. We recommend outputting crucial variables required by our method for future ESM data evaluation. This will enable the use of labeled satellite data for a more systematic evaluation of clouds in climate models.

Characterizing clouds with the CCClim dataset, a machine learning cloud class climatology

“…The consistency of the derived cloud-type RFOs was validated against their related physical variables and those of the classes in the CC-L dataset. This showed that the cloud types in CCClim are consistent with the classes obtained with the CloudSat algorithm (Kaps et al, 2023a).…”

“…We then apply a Random Forest (RF (Breiman, 2001), which is used as a regression model to predict the RFO of each of the nine classes (see Table 1). In Kaps et al (2023a), the physical consistency of the predicted RFOs was validated by using the independent ESA Cloud_cci dataset as input to the RF, which showed good agreement with the cloud type distribution in CC-L.…”

“…The final product of this framework is the RF model, which can predict relative cloud-type amounts using low-resolution cloud properties as inputs. For details on RF training and performance evaluation see Kaps et al (2023a) or the respective code via zenodo 1 .…”

“…Note that the RF model is not applied to the same grid resolution it was trained on. This is possible since the method depends little on horizontal resolution in this range, as shown in the supplementary material of Kaps et al (2023a).…”

“…The CCClim dataset presented here has been created using a combination of supervised ML methods trained using passive as well as active satellite sensors, using a previously published framework (Kaps et al, 2023a).…”