Existing Planetary Boundary Layer (PBL) observations, particularly during the pre-convective environment, remain sparse and underutilized, thereby constraining the predictive skill of mesoscale Numerical Weather Prediction (NWP) systems. This study presents a Kalman Filter One-Dimensional Variational (KF1D-Var) data assimilation framework designed to retrieve atmospheric thermodynamic profiles from water vapor and nitrogen channel observations of Raman lidar within the China Meteorological Administration ground-based vertical profiling network. Leveraging Global Forecast System (GFS) forecasts as the background field, the framework generates high-resolution atmospheric profiles at 30-min temporal and 15-m vertical resolutions. The retrievals are evaluated by using nighttime observations collected at Sheyang station, Jiangsu Province during 1–31 August 2024. The results indicate that the KF1D-Var assimilation reduces GFS biases in temperature, humidity, and pressure within the PBL. Water vapor corrections reach magnitudes of −5% (drying above 300 m) and +1.5% (moistening below 300 m during precipitation events). A month-long sensitivity experiment using a Weather Research and Forecasting (WRF) model-based observation-nudging NWP system shows improved forecast skill at 1–10 h lead times, relative to the Control run. The forecasts exhibit enhanced relative humidity between the surface and 650 hPa, as well as improved horizontal wind and precipitation predictions. These results demonstrate the potential of assimilating Raman lidar thermodynamic profiles to refine NWP accuracy, particularly in representing boundary layer moisture and the processes critical to convective-scale forecasting.

National-scale aviation turbulence in China remains poorly understood, largely due to the scarcity of measurements. In the present study, we investigate aviation turbulence across China’s air traffic control zones from 2017 to 2023 by leveraging the combination of pilot reports (PIREPS) and in-situ in-flight turbulence observations. Our analysis reveals a 35% increase in turbulence incidents in the study period, a growth rate that significantly outpaces air traffic throughput. The methods used to diagnose turbulence include single-index, ensemble, and Random Forest (RF) machine learning models. The RF model demonstrates superior diagnostic accuracy, achieving a nationwide area under the curve (AUC) of 0.87, significantly outperforming traditional ensemble and single-index methods. Regionally, the model’s performance was particularly effective in challenging western regions like Northwest China and Xinjiang. Furthermore, notable regional variation of turbulence is revealed. Turbulence in the eastern China is predominantly driven by dynamic factors, while that in the western regions is primarily influenced by thermodynamic processes and complex topography. These findings underscore the potential of machine learning to advance turbulence forecasting and enhance aviation weather services in China.