Causal Discovery for Stratospheric Ozone Variability in the Reanalysis and Climate Model Simulations

Kurzfassung

Stratospheric ozone (O3) is crucial in protecting the Earth from harmful solar radiation and impacting climate dynamics. The tropical (20°S - 20°N) lower stratosphere (about 70 hPa) has the highest variability and continual decrease with time in O3 concentrations during the last 20 years. Therefore, evaluating O3 variability in this region, which is affected by atmospheric dynamics and chemical processes, is important for climate science. The purpose of this study is to analyze the connections between the main drivers contributing to O3 variability in the tropical lower stratosphere using causal discovery, which is a novel method based on graphical models known as causal graphs. First, the monthly time series from the European Center for Medium-Range Weather Forecasts (ECMWF) Copernicus Atmosphere Monitoring Service (CAMS) Atmospheric Composition Reanalysis 4 (EAC4) between 2003 and 2019 were analyzed to better understand O3 variability in the tropical lower stratosphere. Then, the monthly time series from Coupled Model Intercomparison Project Phase 6 (CMIP6) and Chemistry-Climate Model Validation Activity 2 (CCMVal-2) historical and future simulations were used to compare causal graphs with those of the reanalysis. The variables were selected based on the literature review and expert knowledge, considering the chemical and atmospheric dynamics factors influencing O3 variability in this region. The focus was set on the following variables: O3, hydroxyl radical (OH), water vapor (H2O), vertical velocity, El Nino-Southern Oscillation (ENSO) and Quasi-Biennial Oscillation (QBO). Second, a toy model was developed, informed by a comprehensive literature review and expert knowledge, by defining the data frame and graph. The toy model enables controlled testing and provides a foundation for understanding before applying causal discovery methods to real-world data. Following causal discovery with the toy model, all resulting time series of main drivers contributing to the tropical O3 variability in the lower stratosphere were further analyzed applying two Peter Clark Momentary Conditional Independence (PCMCI) algorithms, namely, the extended version of PCMCI (PCMCI+) and Latent PCMCI (LPCMCI) to all analyzed data sources. The causal graphs obtained by these methods showed that for the given length of analyzed data points, namely, 204 months, the PCMCI+ and LPCMCI algorithms detected only few anticipated connections from the time series and had challenges in detecting the direction of the causal links expected from the literature review. Therefore, in the final step, the linear causal effect mediation analysis was conducted across all analyzed data sources, utilizing the causal graph defined by the toy model. This approach aimed to estimate the causal strength of anticipated links by determining the directions of all connections. According to the causal graphs provided by the linear causal effect mediation analysis, the effect of upwelling, modulated by the QBO, on O3 variability was the most notable, while the anticipated negative connection between HOx (=OH+hydroperoxyl radical (HO2)) catalytic cycle and O3 was not as pronounced in most graphs. Overall, the resulting link strengths obtained by both causal discovery methods used and the linear causal effect mediation analysis (more evidently obtained) performed, confirm that atmospheric dynamics factors are dominant compared to the chemical contribution to O3 variability in the tropical lower stratosphere due to longer O3 lifetime. Also, the F1-score results generated by the LPCMCI algorithm showed a higher agreement with the toy model in most data sources (except the UKESM1-0-LL and ULAQ models) than those produced by the PCMCI+ algorithm. For future research, the analyzed region can be extended to the middle and upper stratosphere over a longer time period, and another method (i.e. bootstrapping aggregation) can be employed to provide a more robust causal analysis for understanding stratospheric O3 variability.