PM2.5 and PM10 vertical distribution retrieval methods based on coherent Doppler lidar via machine learning: for haze and dust in Qingdao

Accurately profiling the vertical distribution of particulate matter (PM) remains challenging. This study presents a novel approach for retrieving vertical PM concentrations using a single coherent Doppler lidar (CDL) combined with machine learning (ML) models. The models were trained using in-situ PM2.5 and PM10 data as true values, along with input features including the particle extinction coefficient at 1550 nm, signal-to-noise ratio, wind speed, wind direction from CDL observations, as well as temperature and relative humidity from ERA5 reanalysis data. Observations from Qingdao, China, during winter and spring (2020–2024) were used for model development and evaluation. The models show good performance that the R², RMSE, MAE of the PM2.5 test set comparison are 0.787, 18.11 μg/m³, 11.23 μg/m³, and those of the PM10 test set are 0.803 and 29.98 μg/m³, 18.93 μg/m³, respectively. Case studies of haze and dust events demonstrated the capability of this method in capturing and characterizing vertical PM distribution. From the statistical results, the PM layers of the typical haze and dust events in Qingdao were primarily concentrated within 1 km and 1.2 km in altitude, respectively. With the atmospheric dynamics information including wind speed and vertical velocity provided by CDL, the evidence can be found that the enhancement of wind speed and negative vertical velocity in the later phase of the PM events may result in the dissipation and deposition of the PM. Combining the PM backward trajectory simulation, the results show that the haze event during the Chinese Spring Festival in Qingdao was influenced by both local emissions and regional transport, while the dust event in the spring exhibited multi-phase structures driven by long-range transport. This CDL-based approach provides a promising pathway for three-dimensional PM observation with atmospheric dynamics information, offering the potential to enhance air quality monitoring through expanded CDL networks and ML applications.