Constructing the 3D Spatial Distribution of the HCHO/NO2 Ratio via Satellite Observation and Machine Learning Model

Abstract

The satellite-based tropospheric column ratio of HCHO to NO₂ (FNR) is widely used to diagnose ozone formation sensitivity; however, its representation of surface conditions remains controversial. In this study, an approach to construct the 3D spatial distribution of the FNR in the lower troposphere was proposed. Based on satellite and multiaxes-differential Optical Absorption Spectroscopy (MAX-DOAS) data, the horizontal and vertical distributions of the FNR have been respectively obtained. To further enhance the generalizability of this approach, we also reproduced the vertical profiles of the FNR using a machine learning model (Bagged trees) and feature variables. Here, using the three-dimensional distribution of the FNR during the summer of 2019 as an example, a fourth-order polynomial relationship was found between the reconstruction factors (f_{col_i}) and altitudes, demonstrating a correlation coefficient of 0.98. Utilizing this established relationship, a significant difference was found between the reconstructed surface FNR and the satellite column FNR, with the former decreasing by 56.9%. Moreover, the reconstructed 3D spatial distribution of the FNR for the summers from 2018 to 2022 revealed a trend over the five years in Shanghai of the ozone formation control regimes gradually shifting toward the transition and NO_x-limited regimes. Through this newly established approach, not only can the accuracy of identifying surface ozone sensitivity be enhanced from the spaced observation, but also it helps in gaining a comprehensive understanding of the ozone photochemical formation mechanisms in the vertical direction.