<p>High quality global ozone profile datasets are necessary to monitor changes in stratospheric ozone. Hence, various methodologies have been used to merge and homogenise different satellite datasets in order to create long-term observation-based ozone profile datasets with minimal data gaps. However, individual satellite instruments use different measurement methods and retrieval algorithms that complicate the merging of these different datasets. Furthermore, although atmospheric chemical models are able to simulate chemically consistent long-term datasets, they are prone to the deficiencies associated with the computationally expensive processes that are generally represented by simplified parameterisations or uncertain parameters.</p><p>Here, we use chemically consistent output from a 3-D Chemical Transport Model (CTM, TOMCAT) and an ensemble of three machine learning (ML) algorithms (Adaboost, GradBoost, Random Forest), to create a 42-year (1979-2020) stratospheric ozone profile dataset. The ML algorithms are primarily trained using the Stratospheric Water and OzOne Satellite Homogenized (SWOOSH) dataset by selecting the UARS-MLS (1992-1998) and AURA-MLS (2005-2019) time periods. This ML-corrected version of monthly mean zonal mean TOMCAT (hereafter ML-TOMCAT) ozone profile data is available at both pressure (1000 hPa - 1 hPa) and geometric height (surface to 50 km) levels at about 2.5 degree horizontal resolution.</p><p>We will present a detailed evaluation of ML-TOMCAT profiles against range of merged satellite datasets (e.g. GOZCARDS, SAGE-CCI-OMPS, and BVertOzone) as well high quality solar occultation observations (e.g. SAGE-II v7.0 (1984-2005), HALOE v19 (1991-2005) and ACE v4.1 (2004-2020). Overall, ML-TOMCAT shows good agreement with the evaluation datasets but with poorer agreement at low latitudes. We also show that, as in different merged satellite data sets, ML-algorithms show larger spread in the tropical middle stratosphere. Finally, we will present a trend analysis from TOMCAT and ML-TOMCAT profiles for the post-1998 ozone recovery phase.</p>
Spatiotemporal continuous estimates of PM2.5 concentrations in China, 2000–2016: A machine learning method with inputs from satellites, chemical transport model, and ground observations
