ML-TOMCAT: machine-learning-based satellite-corrected global stratospheric ozone profile data set from a chemical transport model

• Published in: Earth system science data Vol. 13; no. 12; pp. 5711 - 5729
• Main Authors: Dhomse, Sandip S., Arosio, Carlo, Feng, Wuhu, Rozanov, Alexei, Weber, Mark, Chipperfield, Martyn P.
• Format: Journal Article
• Language: English
• Published: Katlenburg-Lindau Copernicus GmbH 10.12.2021; Copernicus Publications
• Subjects: Aerosols; Air pollution; Algorithms; Analysis; Atmospheric models; Atmospheric ozone; Atmospheric processes; Atmospheric research; Chemical transport; Chemistry; Chlorine; Datasets; Decision trees; Emission measurements; Environmental aspects; Experiments; Fourier transforms; Gases; Instruments; Learning algorithms; Machine learning; Measurement methods; Measurement techniques; Measuring instruments; Ozone; Ozone profiles; Parameter uncertainty; Photochemicals; Photochemistry; Process parameters; SAGE satellite; Satellite data; Satellite instruments; Satellite-borne instruments; Satellites; Simulation; Solar flux; Stratosphere; Three dimensional models; Time series; Transport; Trends; Tropopause; Troposphere; Upper atmosphere; Upper Atmosphere Research Satellite (UARS); Canada; Montreal Quebec Canada
• ISSN: 1866-3516; 1866-3508; 1866-3516
• DOI: 10.5194/essd-13-5711-2021

High-quality stratospheric ozone profile data sets are a key requirement for accurate quantification and attribution of long-term ozone changes. Satellite instruments provide stratospheric ozone profile measurements over typical mission durations of 5–15 years. Various methodologies have then been applied to merge and homogenise the different satellite data in order to create long-term observation-based ozone profile data sets with minimal data gaps. However, individual satellite instruments use different measurement methods, sampling patterns and retrieval algorithms which complicate the merging of these different data sets. In contrast, atmospheric chemical models can produce chemically consistent long-term ozone simulations based on specified changes in external forcings, but they are subject to the deficiencies associated with incomplete understanding of complex atmospheric processes and uncertain photochemical parameters. Here, we use chemically self-consistent output from the TOMCAT 3-D chemical transport model (CTM) and a random-forest (RF) ensemble learning method to create a merged 42-year (1979–2020) stratospheric ozone profile data set (ML-TOMCAT V1.0). The underlying CTM simulation was forced by meteorological reanalyses, specified trends in long-lived source gases, solar flux and aerosol variations. The RF is trained using the Stratospheric Water and OzOne Satellite Homogenized (SWOOSH) data set over the time periods of the Microwave Limb Sounder (MLS) from the Upper Atmosphere Research Satellite (UARS) (1991–1998) and Aura (2005–2016) missions. We find that ML-TOMCAT shows excellent agreement with available independent satellite-based data sets which use pressure as a vertical coordinate (e.g. GOZCARDS, SWOOSH for non-MLS periods) but weaker agreement with the data sets which are altitude-based (e.g. SAGE-CCI-OMPS, SCIAMACHY-OMPS). We find that at almost all stratospheric levels ML-TOMCAT ozone concentrations are well within uncertainties of the observational data sets. The ML-TOMCAT (V1.0) data set is ideally suited for the evaluation of chemical model ozone profiles from the tropopause to 0.1 hPa and is freely available via https://doi.org/10.5281/zenodo.5651194 (Dhomse et al., 2021).