Dynamical Drivers of Interannual Variability in OMI Stratospheric NO 2 Revealed by Machine Learning

Abstract Stratospheric nitrogen dioxide (NO 2 ) plays an important role in atmospheric chemistry and serves as an indirect indicator of dynamical variability in the stratosphere. However, the drivers of its interannual variability remain incompletely quantified. In this study, we analyze two decades (2005–2024) of monthly anomalies in stratospheric‐dominated total column NO 2 observed by the Ozone Monitoring Instrument (OMI) on board NASA Aura satellite. We first evaluate the stratospheric NO 2 from the NASA Goddard Earth Observing System (GEOS) Multi‐Decadal Nitrogen Dioxide and Derived Products from Satellite (MINDS) simulation with the Global Modeling Initiative (GMI) chemistry and show that it reproduces the observed spatial patterns and interannual variability of OMI stratospheric NO 2 column. This good model‐observation agreement supports the use of GMI‐derived dynamical diagnostics, including tropopause pressure and surface pressure, as physically informed predictors. Using these predictors together with seven major atmospheric and oceanic variability indices, the Quasi‐Biennial Oscillation (QBO) and the El Niño‐Southern Oscillation (ENSO) in particular, we compare traditional multi‐linear regression with a machine learning (ML) approach based on gradient‐boosted decision trees. The ML framework substantially improves the reconstruction of OMI NO 2 interannual variability in the stratosphere, indicating the importance of nonlinear relationships and interactions among dynamical drivers. The interpretable ML analysis further reveals that tropopause pressure, QBO, and ENSO are dominant drivers of stratospheric NO 2 variability, with clear latitudinal dependence and distinct lagged influences across regions. These results provide new quantitative insight into the dynamical controls on observed stratospheric NO 2 variability.