Retrieval of planetary boundary layer height from CALIPSO satellite observations using a machine learning approach

The planetary boundary layer height (PBLH) is a key variable in air quality, climate modeling, and weather prediction. Traditional retrieval methods, such as radiosondes, provide high accuracy but lack spatial coverage. This study presents a Random Forest (RF) model based on Machine Learning (ML) to estimate PBLH from ten years of Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) on board the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO), using radiosonde measurements as a reference. The model achieves an R 2 of 0.67 and an RMSE of 278.02 m with a spatial resolution of ≈ 20 × 20 km 2 in a test set that covers mainly Europe and North America. Unlike previous methods, our approach does not require atmospheric typing and uses minimal data filtering, demonstrating robustness under diverse aerosol and cloud conditions. Although validation is currently limited to mid-latitude regions, the method offers a scalable approach to global monitoring and supports the management of climate and air quality. Future work will extend the validation to other geographic zones and explore deep learning models for further improvements. • PBLH Estimation: Random Forest model achieves R 2 = 0 . 71 and RMSE = 308 m using CALIPSO data. • Global Dataset: 4.3M radiosonde launches from 693 stations worldwide were gathered. • No Data Filtering Needed: Model works in multilayer clouds and dust conditions. • Context Matters: Temporal and geographic features are key to performance. • Urban-Ready: Strong in cities ( R 2 = 0 . 68 ) , able for air quality and climate monitoring.