Abstract. Radiosonde (RS) is widely used to detect the vertical
structures of the planetary boundary layer (PBL), and numerous methods have
been proposed for retrieving PBL height (PBLH) from RS data. However, an
algorithm that is suitable under all atmospheric conditions does not exist.
This study evaluates the performance of four common PBLH algorithms under
different thermodynamic stability conditions based on RS data collected from
nine sites in January–December 2019. The four RS algorithms are the
potential temperature gradient method (GMθ), relative humidity
(RH) gradient method (GMRH), parcel method (PM) and Richardson number
method (RM). Atmospheric conditions are divided into convective boundary
layer (CBL), neutral boundary layer (NBL) and stable boundary layer (SBL) on
the basis of the potential temperature profile. Results indicate that SBL is
dominant at nighttime, whilst CBL dominates at daytime. Under all and SBL
classifications, PBLH retrieved by RM is typically higher than those
retrieved using the other methods. On the contrary, the PBLH result retrieved by
PM is the lowest. Under CBL and NBL classifications, PBLH retrieved by PM is
the highest. PBLH retrieved by GMθ and GMRH is relatively
low under all classifications. Moreover, the uncertainty analysis shows that
the consistency of PBLH retrieved by different algorithms is more than
80 % under CBL and NBL classifications. By contrast, the consistency of
PBLH is less than 60 % under SBL classification. The average profiles and
standard deviations of wind speed and potential temperature under consistent
and inconsistent conditions are also investigated. The results indicate that
consistent cases are typically accompanied by evident atmospheric
stratification, such as a large gradient in the potential temperature
profile or a low-level jet in the wind speed profile. These results indicate
that the reliability of the PBLH results retrieved from RS data is affected
by the structure of the boundary layer. Overall, GMθ and RM are
appropriate for CBL condition. GMθ and PM are recommended for NBL
condition. GMθ and GMRH are robust for SBL condition. This
comprehensive comparison provides a reference for selecting the appropriate
algorithm when retrieving PBLH from RS data.