Review of “Determining the sensitive parameters of WRF model for the prediction of tropical cyclones in the Bay of Bengal using Global Sensitivity Analysis and Machine Learning” by Baki et al.

This study investigated the impacts of 24 tunable parameters in the Weather Research and Forecasting model on the simulations of tropical cyclones over the Bay of Bengal region. Three global sensitivity analysis methods were employed and compared. The parameter sensitivity results were found to be consistent across three methods for all the variables, and 8 out of the 24 parameters contribute 80%−90% to the overall sensitivity scores. Compared to default parameters, applying optimal parameters produced remarkable improvements in the simulated 10m wind speed, surface air temperature, surface air pressure, and precipitation predictions. I think the manuscript is well organized and the presentation is generally good. However, there are some aspects need to be improved before considering of publication. 

Recommendation: Minor revision

Minor comments:

1.    The word “prediction” is used in the title and in the main text extensively. Please note that the meanings of “prediction” and “simulation” are not exactly the same, and improved simulation with a better model does not always translate into increases in prediction skills. One good example was given by Liu et al. (2019), who showed that the parameters’ impacts on simulation and prediction might be different. I understand that the topic of this study is “simulation”, so I suggest replacing the word “prediction” by simulation in the title and in the text.

Liu et al. (2019): Validity of parameter optimization in improving MJO simulation and prediction using the sub-seasonal to seasonal forecast model of Beijing Climate Center, Climate Dynamics, 52:3823–3843

2.    Several literatures that are highly related to the selection of parameters are missing in the manuscript. For example, P6 - multiplier of entrainment mass flux rate, P4 - Von Karman constant, and P3 - scaling related to surface roughness, which are found to be important for tropical cyclone simulations in this study, were primarily identified by Yang et al. (2012) and Yang et al. (2017). These papers should be cited accordingly

Yang et al. (2012): Some issues in uncertainty quantification and parameter tuning: a case study of convective parameterization scheme in the WRF regional climate model, Atmos. Chem. Phys., 12:2409-2427

Yang et al. (2017): Sensitivity of Turbine-Height Wind Speeds to Parameters in Planetary Boundary-Layer and Surface-Layer Schemes in the Weather Research and Forecasting Model, Boundary-Layer Meteorology. 162:117–142

Specific comments:

1.    Line 24, “.Singh et al. (2021a).”?

2.    Line 25, “Singh et al. (2019) showed that present warming climate impacts on the …”, please check the grammar.

3.    Line 29, What does “VSCS” mean?

4.    Line 50, “at once” -> “simultaneously”?

5.    Line 108, “in question to”?

6.    Line 395-402, the definition of P6 and the analyses about its impacts largely follows that of Yang et al. (2012), which should be added here. Meanwhile, it is not clear to me why suppressed convection (i.e. weakened consumption of CAPE or instability) leads to more “stable” stratiform clouds. Have the authors checked the vertical profiles of atmosphere temperature and moisture? One explainable for the changes in stratiform precipitation is the competition for moisture between convective and stratiform processes as indicated by Liu et al. 2018.

Liu et al. (2018): Combined impacts of convection and microphysics parameterizations on the simulations of precipitation and cloud properties over Asia, Atmospheric Research, 212:172-185

The study investigates the sensitivity of 8 meteorological variables to 24 WRF model parameters. Three methods are being used for the sensitivity analysis, with very similar results, indicating that the results are robust. Most variables are only sensitive to a few parameters, and some parameters don’t introduce any obvious sensitivity. The results can help to improve forecasting by finding optimal tuning parameters in numerical models.

The study is well-constructed and generally well written. I don’t see any major flaws that would prohibit publication. In fact, I think this is a nice systematic study that can tell us a lot about how to improve forecast models, and how to find optimum value for the myriad of tunable model parameters. I do have some concerns, but they don’t really apply to the methodology or interpretation of results. Please see below.

Major comments:

1.   I feel there is a contradiction between the title and motivation of this study versus the presented results. The title and motivation of the study explicitly refer to TCs, but most of the results are not TC specific. Rather, the results presented in Figs. 3­–15 seem to be derived from the entire domain 2, of which TCs only cover a small fraction of. So, to me it seems the results are general rather than specific to TCs (note that I don’t think this is a bad thing). Only Figs. 16­–18 specifically refer to TCs. This contradiction could be removed by either focusing the analysis to parts of the domain that include the TC (like the panels in Figs. 16 and 17) or by rewording and restructuring the title and text.
2.   Somewhat related to the first comment, the authors seem to treat all variables with the same importance. Again, I don’t think this is a bad thing, but if this study is about TCs, I’d put TC-specific variables, such as 10-m winds and rainfall (and maybe pressure), in the focus. Again this could be done by restructuring the text. 
3.   How did you decide on the ranges in Table 1?
4.   L87: Unless there’s a technical reason for not doing so, to me it makes more sense to say “The objective of the present study is to assess the sensitivity of meteorological variables such as surface pressure, temperature, wind speed, precipitation, to WRF model parameters…”  instead of “The objective of the present study is to assess the sensitivity of the WRF model parameters to various meteorological variables such as surface pressure, temperature, wind speed, precipitation, …”. I think what we’re interested in is the response of the output to the input, and the latter sounds to me as the opposite.
5.   Fig. 18: I wonder if we’re losing some information by showing boxplot aggregates and the average wind speed of all TCs. Often the average is something non-physical, and often does not tell us much. Would it be possible to show 10 timeseries, one for each TC? Also, how do the box plots relate to the colored lines? Shouldn’t the green or blue line go right through the orange lines in the box plots? 

 Specific comments:

1. L. 21: “which alone contributed to an overall increase in the NIO.” – Increase of what? Activity or destructiveness?
2. L 59: It would be interesting to know which two parameters were found to significantly affect the intensity and structure.
3. L194: “spin-up time” instead of "spin-off time" (also L215)

4.     L444: Taylor statistics are…
5.     L446: can be plotted in one Taylor diagram… 
6.     To me there is a discrepancy between Figs. 14 and 15. In Fig 14, it doesn’t look like the optimal parameters are any better than the default parameters when looking at WS10. But when looking at Figs. 15a and 15b, it looks like the optimal parameters are quite a bit better (smaller bias). How do you explain this discrepancy?