The ability to observe large wildfires with proper time and space resolution is mandatory for risk management. Polarimetric weather radars have a chance to identify pyroclastic clouds. Nevertheless, microwave signatures of pyroCb are not well distinguished from sea clutter or clear air echoes. The authors propose here a novel approach based on the statistical properties of Gray Level Co-occurrence Matrices (GLCM) and a Gaussian Mixture Model (GMM) to classify echo sources by combining radar variables with texture-based fields. The work is scientifically interesting and the analysis is rigorously conducted and clearly exposed. Some minor improvements and some further investigations are needed. Section 2.2 deals with weather radar gridded data: to grid those data are needed to move from polar coordinates to Cartesian coordinates. Smoothing these fields is one of the options, usually due to noisy retrieval (e.g. poor sampling), but is not a consequence of gridding. A re-phrase of lines 158-164 is recommended. In the following lines, the authors mention spatial aliasing: the expression aliasing is commonly referred to wind data from weather radar and not the range of observations. Please consider to re-phrase. Line 201 "and run CPUs" is not clear. Line 449, frequency, and radar characteristics are indicated as factors influencing texture fields. Please, list factors more specifically detailing the causes of this influence.

Finally, although the algorithm performance evaluation can not be conducted with direct observations, it could be evaluated as a relative performance with respect to fuzzy logic classification. It is recommended to investigate and discuss this relevant aspect, referring also to the work of Zrnic et al., 2020 (Zrnic, D.; Zhang, P.; Melnikov, V.; Mirkovic, D. Of Fire and Smoke Plumes, Polarimetric Radar Characteristics. Atmosphere 2020, 11, 363. https://doi.org/10.3390/atmos11040363).

This article does an excellent job of summarizing existing literature related to radar echo classification as it relates to fire-related convection and introduces helpful algorithms for improving the existing methods. This publication should be published after a few minor changes.

Adding more references/justification for not using a cartesian grid would be helpful here (ex. Trapp and Doswell 2000). Including solely Brook et al. 2022 does not provide enough information here. More of a discussion about there the noise comes from in this case (sampling deficiencies) should be discussed. A comparison with a cartesian grid might be a useful exercise here as well.

More of a discussion and explanation should be adding around line 201 where the author mentions using CPUs vs. GPUs, and why this is important for the computations, possibly including some benchmark estimates or references to related literature on this topic.

For the figures (ex. Figures 7 and 8), a legend would help illustrate the different cluster classifications instead of including in the figure description. Also, increasing the quality of Figure 7q.