The present study highlights the importance of various physical processes in smoke dispersion from wildfire biomass burning. The manuscript is clearly structured, it is well written and presents important analysis regarding the comparison between twelve state-of-the-art atmospheric models during a specific case study in US. Overall, it will be a very useful addition to current literature, and I suggest publication with a few minor comments as shown below:

1. It is clear from this work that there is still significant lack of knowledge regarding several processes like the quantification of fire emissions and their diurnal cycles, plume injection heights, calculation of AOD and PM2.5, the spatiotemporal representation of smoke plumes in forecasting models etc... However, this is something more-or-less known due to the complexity of the mechanisms involved in these events. Since none of the models managed to provide a realistic description of the case study, the study remains somehow non-conclusive. It might be useful to elaborate more in the conclusions section and provide more physical interpretation on the reported differences between the models as well as some quantification on which of the analyzed parameterizations are more important for similar studies (e.g. plume rise, FRP emissions, etc.). This could help future studies to focus on improving certain model features and discard those that look problematic.
2. The computation of modeled smoke AOD in Section 3.2.2 should be further explained with regards to the optical properties of smoke used in each model.    
3. At several places (e.g. Lines 89-94, 414-419, 607-610) the authors discuss the importance of including the diurnal variation of smoke emissions inside the forecasting window and the possibility to incorporate data from geostationary satellites. A similar system is available in Europe adopting a modeling strategy of hourly-sequential warm start runs with FLEXPART-WRF, driven by METEOSAT geostationary observations (Solomos et al., 2015, 2019). In this approach, the emissions are updated every hour from the MSG/SEVIRI detections, and each simulation is initialized with the smoke from the previous run (warm start). This provides an efficient way of removing the minor or extinguished fires from the simulation and at the same time to enhance the emissions from the actual burning fires, thus representing the diurnal cycle of biomass burning.

Solomos, S., V. Amiridis, P. Zanis, E. Gerasopoulos, F.I. Sofiou, T. Herekakis, J. Brioude, A. Stohl, R.A. Kahn, C. Kontoes, Smoke dispersion modeling over complex terrain using high resolution meteorological data and satellite observations – The FireHub platform, Atmospheric Environment, Volume 119, October 2015, Pages 348–361, doi:10.1016/j.atmosenv.2015.08.066, 2015

Solomos S., A. Gialitaki, E. Marinou, E. Proestakis, V. Amiridis, Holger Baars, Mika Compula, Albert Ansmann, Modeling and remote sensing of an indirect Pyro-Cb formation and biomass transport from Portugal wildfires towards Europe, Atmospheric Environment, https://doi.org/10.1016/j.atmosenv.2019.03.009 , 2019

4. Line 320 typo “biome maps”

Overview

This paper compares fire emission parameters derived from twelve different fire emission models forecasting a 2019 wildfire event in the United States. Parameters are compared among models and observed data. By doing so, the authors aim to derive meaningful insights into the current progress of fire emission models and hence suggest ways to further improve the efficacy of such models. The parameters compared in this paper include biomass burning organic carbon emissions, smoke AOD (magnitude and spatial coverage), surface PM_2.5, plume rise height and ratio of smoke AOD and PM_2.5; hence covering both physics and chemistry aspect of fire emission modeling. The paper suggests areas which current fire emission models can improve on, which includes methodologies to represent diurnal evolution of fire emissions, improved vertical distribution of emitted pollutants and better representation of plume injection heights.

The models and methodology used are clearly described and the paper is well written. I have only two suggestion and a few minor suggestions/clarifications to make.

Major Suggestions

Line 877: It would be highly insightful to understand how the type of emission injection method (within PBL, intermediate, deep) affect the model skill in predicting plume rise heights. Certain emission injection method may be more useful for a certain kind of fire plume (fresh, aged, fire characteristics: smoldering, raging fire, etc.) and not others. If we can associate a better emission injection method with a corresponding type of fire plume, we can improve fire modelling skill. Indeed, further investigations in this regard is necessary and will definitely be a good follow up work.

Line 981: I may not be proficient enough in this aspect, so this is just some thoughts. There might be inherent problems using surface smoke PM_2.5 to smoke AOD ratio when you have different sAOD filters for different models. For models with small sAOD (denominator), the ratio will tend to be bigger and hence result in larger spread. This is consequentially seen in the large nominal mean bias. For example, ARQI and NAQFC have 0.01 sAOD threshold and consequentially have a very large NMB. HRRR smoke and WISC WRF-Chem have 0.02 sAOD threshold and also have very large NMBs. CAMS have a larger threshold, 0.05, and consequentially have smaller magnitude NMBs. AIRPACT is the exception here. This may affect both the magnitude and spread of the ratio calculated and may lead to unfair comparisons between the models. This may affect the model evaluation.

Minor Suggestions

Line 36: For 2.5 in PM_2.5, suggest to be written in subscript.

Line 85: There is one multi-model comparison done by Li, et. al., 2019. Atmos. Chem. Phys., 19, 12545–12567 for many different fire models, which may be worthy to look at.

Line 157: If the forecast system produces more than 1 cycle per day, how is the data treated? Is the data averaged?

Line 220: The style of writing for Section 2.1.6 seems to be slightly different from the rest of the paper. Consider revising.

Line 378: I would like to clarify if the models were in a spun-up condition when model forecasts were extracted to compare with observed data.

Line 446: AERONET is already defined in line 435.

Line 534: It may be insightful to suggest a reason why FRP-driven models results in higher sAOD compared to hotspot driven models.

Line 655: It may be problematic to compare point derived ground-based measurement station data against model grid predictions of smoke PM_2.5. Perhaps a small discussion about this issue will be helpful.

Line 660 and a few other places: May want to revise the use of ‘it’s’.

Line 730: Consider revising this sentence.

Figure 5: Spelling of AERONET in the figure and caption.