Abstract: High-precision measurements of Precipitable Water Vapor (PWV) and Liquid Water Path (LWP) are crucial for enhancing weather forecast accuracy and evaluating climate models. This study utilizes observational data from the G-band Water Vapor Radiometer (GVR) and the Aircraft Integrated Meteorological Measurement System (AIMMS-20) mounted on the Beijing Artificial Weather Modification Center"s King Air aircraft, combined with historical sounding profiles (2009–2023) from the Beijing Sounding Station. Using the passive and active microwave radiation transfer mode (PAMTRA) to generate training and validation datasets, with data from 2021 to 2023 reserved as the validation set, we developed three machine learning algorithms—Backpropagation Neural Network (BP), Extreme Gradient Boosting (XGBoost), and Random Forest (RF)—for high-precision PWV and LWP retrieval. Validated against independent radiosonde profiles, the newly developed machine learning algorithm demonstrates superior performance, with RMSEs of 0.62 mm for precipitable water vapor and 0.04 mm for liquid water path, significantly outperforming the GVR system"s built-in retrieval algorithm. Analysis of GVR-measured brightness temperature data reveals that while all algorithms capture consistent temporal trends in PWV and LWP, the GVR"s built-in algorithm exhibits systematic underestimation, with maximum deviations reaching 3.82 mm for PWV and 0.24 mm for LWP compared to the RF algorithm. Furthermore, in sensitivity tests incorporating ±1 K instrument errors, the built-in algorithm demonstrates the poorest stability, yielding the largest Mean Absolute Error (MAE) (0.19 mm for PWV, 0.02 mm for LWP), whereas the newly developed BP neural network exhibits the strongest noise resistance.