Doppler spectra observations are particularly useful for interpretating hydrometeor populations and tracing their changes in the air. This work aims to combine previously developed tools, PEAKO and peakTree, to characterize hydrometeor populations and to demonstrate their applicability. However, I am not convinced that there is a real need for combining these two algorithms. This manuscript is way too tedious, and I do not see much new contributions to the community. Please see my comments below.

Major comments.

1. The core of spectral analysis is the peak identification which facilitates separations of different hydrometeor populations. Actually, the essential contributions of PEAKO and peakTree algorithms are overlapped. PEAKO uses ML to identify spectral peaks, while peakTree uses a binary tree structure. The authors claimed that the two algorithms are combined. However, one may simply use PEAKO to do hydrometeor population separation without using peakTree. Therefore, I do not see any additional benefits of using peakTree. I do not really see the novelty or the value of publishment of this point.

To me, this work is simply the evaluation of PEAKO. However, this group has already developed and validated some ML methods for spectral peak identification in many papers, do you really need an additional paper demonstrating the validation?

2. In spite of my criticism on the novelty, I do see the need of assessing the impact of spectral average on peak identification. Unfortunately, this part is very poorly addressed. I list some technical questions as below. If they are well addressed upon major revisions, I would like to see its publication.

The number of averaged spectra in Table 1 is not clearly defined, how is it different from the number of coherent integrations? The PAMTRA simulation results in Section 3 shows that this parameter will affect the of spectral peak detection, and smoothing the spectrum in a desirable interval could help to suppress noise peaks. However, the recommended average number is based on the simulation for fixed PAMTRA parameters, in other words, the particle size distribution and atmospheric dynamics do not change. Do the simulation only represent the effect of Gaussian noise in the simulated Doppler spectra?  These parameters can vary greatly over time in actual observations, in that case, additional averaging of Doppler spectra may instead lose useful signals. How should we measure whether to average the spectrum under different cloud environments? Also, the spectral resolution will affect the effect of averaging. How different spectral resolutions will affect the results? 

Spectral averaging is entangled with the effect of turbulence. How do you separate their roles in changing the spectral shape respectively in real observations?

Fig.7(g-i) gives the PAMTRA simulated spectra based on the size distribution measured by HoloBallon, dose the simulations take into account the effect of horizontal wind on the spectra? In the simulations of bimodal distribution, the relative spectral reflectivity of ice peak and liquid peak is different from that observed by cloud radar. The responsible factors should be clearly discussed.

3. The misinterpretation of ice columns to SIP. You may see the presence of ice columns, however, they are not necessarily produced by SIP. If you wish to identify a SIP event, you need to compare N_column with N_INP, as did in (Li et al., ACP, 2021; Wieder et al., ACP, 2022; Billault-Roux et al., ACP, 2023).