Regionalization of the summertime planetary boundary layer height in comparison with various reanalysis datasets over China

Abstract

Knowledge about the regionalization of the planetary boundary layer (PBL) height (PBLH) remains unclear for such a huge country like China, largely to the limited observations. By using K-means clustering of summer radiosonde-derived PBLHs, combined with land surface related factors, we classified China into six regions, including Northeast China, North China, South China, Inner Mongolia, Tibet Plateau, and Northwest China. Overall, the PBLH in China generally reaches the minimum at 0800 Beijing time (BJT), as opposed to the peak at 1400 BJT when the convective PBL is well developed. Concerning the spatial distribution, the PBLH at 14 BJT exhibits a “northwest high-southeast low” pattern with the highest (2259 m) occurring in Inner Mongolia. Furthermore, the performances of several commonly used reanalysis products (ERA5, FNL, and CERA) are comprehensively compared against PBLH. Among all the three reanalysis datasets, The PBLHs from ERA5 present the strongest relationship with those from radiosounding observations across China, suggesting its robustness and suitability in estimating PBLHs. In terms of the performance of the reanalysis dataset in characterizing three PBL regimes such as convective, neutral and stable PBL, the PBLHs from ERA5 show significant positive correlations (R > 0.7) with the observed counterparts for both convective and neutral PBL regimes in most areas. However, all three reanalysis datasets tend to overestimate the PBLH under stable SBL. Our findings indicate that PBLH biases in reanalysis products are partly attributed to the misrepresentation of PBL thermal regimes, which is probably owing to the unsatisfactory simulations of cloud effects in weather models. Overall, this study presents the first regionalization of high-resolution radiosonde-derived PBLHs in China and quantitatively evaluates three popular reanalysis products, which have great implications for better application of datasets and improve the model parameterization of PBL.