General comments:

It has been well recognized that the estimate of planetary boundary-layer height (PBLH) from radiosonde varies dramatically by the methods used, especially at the morning and evening transitional period. To reduce the inconsistency between existing methods, this manuscript by Chen et al. proposed an ensemble method to confront this challenge based on one year worth of high-resolution radiosonde measurement at Beijing weather station. This algorithm has solid physical basis. The analysis methods are scientifically sound and the results are reasonalbe from my point of view. The manuscript is well organized and figures and tables are presented in a succint way and easy to follow. Nevertheless, the effectiveness of this method seems elusive to me and thus further clarification is neeed. Therefore, I recomment this manuscript be accepted after a minor revision. The specific comments are as follows:

Major comments:

1. ”the effectiveness of the ensemble method” appears several times through the whole manuscript. But I can not find the exact definition for EFFECTIVENESS. For the benefit of readership, necessary clarification for this noun is required. Therefore, the authors are suggested to give a unambiguous definition for EFFECTIVENESS with some explanation for its implication?
2. Figure 9: I do not understand what does the Y-axis on the right-hand side of panel (a) and (b). The authors can clarify this in the figure caption or in the main text.

Minor comments:

L19: “during afternoon, morning, and evening transition periods, respectively.” can be reprased as “at 0800, 1400 and 2000 Beijing time”

L25: what does CBL stands for? For its first appearance, the acronym is supposed to be given a full name.  

L27: It is better to use an adjective to describe the EFFECTIVENESS of the ensemble method developed in the present study.

L36: “with the free troposphere” can be revised to “between the free troposphere and groun surface”.

L47:  the dash line in “wind-profiler” can be dropped.

L73: “As the routine radiosonde generally operates” -> “As the routine radiosonde measurements are generally taken”

L77: “further understanding of the transition period” -> “further understanding of the PBL structure and evolution during the transition period”.

L238 and 245: “the integrated method” is used instead of “the ensemble method” that appear in the title of this manuscript. Are there any differences between them? If not, I suggest the authors use one term through the whole manuscript.

L334: “an accuracy estimation” -> “a reliable estimation”

L336: “first” can be removed.

L367-368: “and that results in some shortcomings of these methods being retained” can be rephrased.

PBLH derived from radiosonde profile is a commonly used method and also taken as the standard for other method. However, there is usually big difference between the PBLH determined based on different variable profiles. An ensemble method based on high-resolution radiosonde data in Beijing is proposed to derive PBLH. The new method aims to decrease the uncertainty of PBLH estimation, but there are still some questions in the new method, which will increase the uncertainty of the results. The paper is recommended for publication after the following comments been properly addressed.

1. 75% quantile of the annual result from the specific gradient method is taken as the threshold in the 2^nd step of the new method. As we all know, it indicates significant seasonal cycle of PBLH, which is usually highest in summer and lowest in winter. If taken the 75% PBLH of the whole year as the threshold, the number of cases needed to be modified will be highest in summer, which will underestimate the derived PBLH in summer, and vice versa. I would like to suggest using seasonal or monthly threshold instead of the annual PBLH threshold.

2. In the new method, if the initial result is greater than 75% quantile, then the highest level of the top 10 smallest gradients under 75% threshold is taken as the derived PBLH. That means if the derived PBLH higher than 75% threshold, it was a false result. Why cannot the derived PBLH higher than the 75% threshold? Based on your method, all of the final derived PBLH are lower than the initial 75% threshold, which will decrease the derived PBLH artificially. Could you give more explain about the 75% threshold from the perspective of atmospheric physics rather than mathematical statistics.

3. In the final step of the new method, the results are divided into several groups by 50 m, and the result of the group with the largest samples is used for the final PBLH estimation. Although seven variable profiles are used in this study, these variables can be divided into three groups, i.e., vertical temperature information group or humidity group or both of them. As show in Wang and Wang 2014, the vertical humidity observations usually suffer from the existence of clouds and the measurement error of humidity instruments, which is the main reason for the difference of PBLH between temperature profile group and humidity kind group. That means, it shows more consistency between the PBLH in temperature group or humidity group or temperature-humidity group. And the difference between the three groups is larger than those in the individual group. So, the number of temperature and humidity variables in the seven selected variables will influence the sample number of each group in the final step of the new method. If five of the seven variables are only related to humidity information, it will increase the probability of the five derived PBLH grouped together in the final step, which will lead to the missing of temperature information.

4. It shows highly vertical resolution of the raw radiosonde data. Three-point smoothing was conducted to avoid the influence of high resolution. I suggest to make a sensitivity analysis about the effects of the number of smoothing points to the final result of derived PBLH.