Stratospheric ozone trends and attribution over 1984-2020 based on satellite data and model simulations with a regularised regression method

Accurate quantification of long-term trends in stratospheric ozone can be challenging due to their sensitivity to natural variability, the quality of the observational datasets, non-linear changes in forcing processes as well as the statistical methodologies. Multivariate linear regression (MLR) is the most commonly used tool for ozone trend analysis. However, the complex coupling in most atmospheric processes makes it prone to the over-fitting or multi-collinearity-related issue when using the conventional Ordinary Least Squares (OLS) setting. To overcome the multi-collinearity, we adopt a regularised (Ridge) regression method to quantify ozone trends and the influence of individual processes. Our MLR model setup is similar to the one used in Dhomse et al., (2022). Here, we use the Stratospheric Water and OzOne Satellite Homogenized (SWOOSH, Davis et al., 2016) merged data set (v2.7) to derive stratospheric ozone profile trends for the period 1984-2020. Beside SWOOSH, we also analyse a machine-learning-based satellite-corrected gap-free global stratospheric ozone profile dataset from a chemical transport model (ML-TOMCAT) (Dhomse et al., 2021), and output from two chemical transport model (TOMCAT) simulations forced with ECMWF reanalyses ERA-Interim and ERA5 (Li et al., 2022).

With Ridge regression, the stratospheric ozone profile trends from SWOOSH data show smaller declines during 1984-1997 compared to OLS with the largest differences in the lowermost stratosphere (>4 % per decade at 100 hPa). Upper stratospheric ozone has increased since 1998 with maximum (~2 % per decade near 2 hPa) in local winter for mid-latitudes. Negative trends with large uncertainties are observed in the lower stratosphere and are most pronounced in the tropics. The largest difference in post-1998 trend estimates between OLS and Ridge regression methods also appears in the tropical lower stratosphere (about ~7 % per decade difference at 100 hPa). Ozone variations associated with natural processes such as solar variability, ENSO, AO and AAO also indicate that Ridge regression coefficients are somewhat smaller and less variable compared to the OLS-based estimates. The ML-TOMCAT data set shows similar results to those using SWOOSH data while model simulations show larger inconsistencies especially in the lower stratosphere. The considerable differences between the satellite data and model simulations indicate that there are still large uncertainties in ozone trend estimates especially in the lower stratosphere where dynamical processes dominate, and caution is needed when discussing results if explanatory variables used are correlated.

References:

 Davis, S. M., et al., The Stratospheric Water and Ozone Satellite Homogenized (SWOOSH) database: a long-term database for climate studies, Earth Syst. Sci. Data, 8, 461–490, https://doi.org/10.5194/essd-8-461-2016, 2016.

Dhomse, S. S., et al., ML-TOMCAT: machine-learning-based satellite-corrected global stratospheric ozone profile data set from a chemical transport model, Earth Syst. Sci. Data, 13, 5711–5729, https://doi.org/10.5194/essd-13-5711-2021, 2021.

Dhomse, S. S., et al., A single-peak-structured solar cycle signal in stratospheric ozone based on Microwave Limb Sounder observations and model simulations, Atmos. Chem. Phys., 22, 903–916, https://doi.org/10.5194/acp-22-903-2022, 2022.

Li, Y., et al., Effects of reanalysis forcing fields on ozone trends and age of air from a chemical transport model, Atmos. Chem. Phys., 22, 10635–10656, https://doi.org/10.5194/acp-22-10635-2022,2022.