Cloud-height-dependent cloud–aerosol state classification based on MAX-DOAS observations in a coastal region and validation using lidar measurements

Abstract. Ground-based MAX-DOAS retrievals are strongly influenced by clouds and aerosols. However, existing MAX-DOAS-based cloud classification methods mainly focus on identifying whether observations are affected by clouds, while less attention has been paid to cloud relative height and its different effects on retrievals. We develop a cloud-height-dependent cloud–aerosol state classification method based on MAX-DOAS observations in Guangzhou, a major coastal city in southern China. The method is evaluated using independent 355 nm lidar observations about 3 km from the MAX-DOAS site. It combines the measured zenith signal and its temporal variation, the O₄ differential slant column density (dSCD) at 30° elevation, and the retrieved aerosol optical depth (AOD) to distinguish eight observation states.

Radiative transfer simulations show that low clouds can substantially shorten the effective light path and reduce the O₄ dSCD at 30° elevation, whereas high clouds have a weaker effect because MAX-DOAS measurements are primarily sensitive to the near-surface atmosphere. This behaviour provides the physical basis for distinguishing low clouds from high clouds relative to a site-specific clear-sky reference sensitivity height of about 2 km. In addition, under the clear-sky assumption, cloud-induced light-path changes can be partly misattributed to aerosol extinction, so that low-cloud conditions can lead to overestimated AOD.

Non-clear-sky conditions account for a large fraction of the observations in Guangzhou, with low clouds, cloud holes, and cloud-free high-aerosol conditions occurring frequently. Compared with the lidar-based classification, the proposed method reproduces the main evolution of weather processes reasonably well. After grouping the results into three classes, namely cloud-free, high-cloud, and low-cloud conditions, the overall agreement is 0.658, with fractions of 0.703, 0.666, and 0.524 of the lidar-classified cloud-free, low-cloud, and high-cloud samples, respectively, identified as the same class by MAX-DOAS. Sensitivity analyses further show that the classification framework is generally stable with respect to perturbations in the key thresholds. These results provide information on cloud–aerosol observation states for assessing retrieval applicability under complex cloud conditions.