From data to discovery: understanding tropical middle stratospheric ozone variability through causal inference

Ozone (O₃) plays a critical role in the atmosphere by absorbing harmful ultraviolet solar radiation and also shaping the thermal structure and dynamics of the stratosphere. Variability in O₃ levels is driven by a complex interplay of factors, including long-term climate change, the abundance of ozone-depleting substances (ODSs), and non-linear interactions between transport and chemical processes. Changes in tropical stratospheric O₃ are particularly intricate due to a strong altitude dependence (WMO, 2022). In the tropical middle stratosphere, a region characterized by strong O₃ production and loss, during the early 2000s satellite measurements revealed an unexpected decline in O₃. Since then, O₃ levels in this region have increased again, but the underlying mechanisms driving such variability remain insufficiently understood, highlighting the need to investigate further the processes driving O₃ concentrations.

In this study, we show the pivotal role of causal inference in disentangling the complex chemical-dynamical influences on O₃ behavior in the narrower region of the tropical (10°S-10°N) middle (10 hPa) stratosphere. Causal inference can add significant value to traditional statistical methods by inferring causal relationships, distinguishing genuine causal links from spurious correlations, and quantifying their strength. The framework integrates qualitative physical knowledge through a causal graph applied to satellite observations and state-of-the-art 3-D chemical-transport model (CTM) TOMCAT simulations. By leveraging causal inference, we provide robust insights into the drivers of O₃ fluctuations and showcase the method’s potential for uncovering causal relationships in stratospheric chemistry-dynamics interactions. To validate this approach, we first construct a simplified toy model that reproduces major chemical-dynamical interactions in tropical middle stratospheric O₃ that are based on the NO_x (=NO + NO₂) catalytic ozone destruction cycle and stratospheric dynamics via stratospheric residual velocity w*. Using this toy model, we demonstrate that causal discovery reproduces the connections between w*, nitrous oxide (N₂O), nitrogen dioxide (NO₂), and O₃ in the tropical middle stratosphere. This successful application establishes a foundation for extending causal effect estimation to observed and modelled chemical processes, including their time lags. We split the periods 2004-2018 into two subperiods (i.e. 2004-2011 when O₃ concentrations declined, and 2012-2018 when O₃ concentrations increased in the tropical middle stratosphere) to demonstrate differences in the w*-N₂O connection that drives distinct O₃ behaviors. Additionally, a process-oriented analysis of different Quasibiennial oscillation (QBO) regimes, combined with bootstrap aggregation, reveals robust patterns in chemical-dynamical interactions. These results highlight the potential of causal inference as a transformative tool for advancing our understanding of stratospheric O₃ variability and its response to dynamic forcing.

World Meteorological Organization (WMO). Scientific Assessment of Ozone Depletion: 2022, GAW Report No. 278, 509 pp.; WMO: Geneva, 2022.