Stacking Machine Learning Models Empowered High Time-height Resolved Ozone Profiling from Ground to the Stratopause Based on MAX-DOAS Observation

Abstract

Ozone (O3) profiles are crucial for comprehending the intricate interplay among O3 sources, sinks, and transport. However, conventional O3 monitoring approaches often suffer from limitations such as low spatiotemporal resolution, high cost, and cumbersome procedures. Here, we propose a novel approach that combines Multi Axis Differential Optical Absorption Spectroscopy (MAX-DOAS) and machine learning (ML) technology. This approach allows the retrieval of O3 profiles with exceptionally high temporal resolution at the minute level and vertical resolution reaching the hundred meters scale. The ML models are trained using parameters obtained from radiative transfer modeling, MAX-DOAS observations, and reanalysis dataset. To enhance the accuracy of retrieving O3, we employ a stacking approach in constructing ML models. The retrieved MAX-DOAS O3 profiles are compared to data from in-situ instrument, lidar, and satellite observation, demonstrating a high level of consistency. The total error of this approach is estimated to be within 25%. On balance, this study is the first ground-based passive remote sensing of high time-height resolved O3 distribution from ground to the stratopause (0-60 km). It opens up new avenues for enhancing our comprehension of O3 dynamics in atmospheric environments. Moreover, the cost-effective and portable MAX-DOAS combined with this versatile profiling approach enables the potential for stereoscopic observations of various trace gases across multiple platforms. This work has been supported by Sino-German Mobility Program (M-0509), National Natural Science Foundation of China (grant number 42075097, 22176037, 42375089, 22376030).