An ensemble method for improving the estimation of planetary boundary layer height from radiosonde data

Abstract. The planetary boundary layer (PBL) height (PBLH) is an important parameter for both weather, climate and air quality models. Radiosonde is one of the commonly used instruments for PBLH determination and is generally accepted as a standard for other methods. However, mainstream approaches for the estimation of PBLH from radiosonde present some uncertainties and even show disadvantages under some circumstances and the results need to be visually verified, especially during the transition period of different PBL regimes. To avoid the limitations of individual methods and provide a benchmark estimation of PBLH, we propose an ensemble method based on high-resolution radiosonde data collected in Beijing in 2017. Seven existing methods including four gradient-based methods are combined along with statistical modification. The ensemble method is verified during afternoon, morning, and evening transition periods, respectively. The overestimation of PBLH can be effectively eliminated by setting threshold for gradient-based methods and the inconsistency between individual methods can be reduced by clustering. Based on the statistics of one-year observational analysis, the effectiveness of the ensemble method reaches up to 70.8 %, an increase of 14.7 % ~ 61.2 % compared with the existing methods. Nevertheless, the ensemble method suffers to some extent from uncertainties caused by the removal of truly high PBLH, the profiles with a multi-layer structure, and the intermittent turbulence in the stable boundary layer (SBL). Finally, this method has been applied to characterize the diurnal and seasonal variations of different PBL regimes. Particularly, the average CBL height is found to be the highest in spring and the SBL is lowest in summer with about 200 m. The average PBLH at transition stage lies around 900 m and there is no obvious seasonal variation. The findings imply the effectiveness of the ensemble method.