ERA5 reanalysis data is used to evaluate the performance of two seasonal forecast model systems, the SEAS5 and MF7, in simulating the variability of the Saharan Heat low (SHL) at various timescales over the 1993 – 2016 period.  Strengths and weaknesses of each forecast model are discussed, and statistical bias correction methods are applied to improve the forecast of the SHL in the forecast models.  While the methods applied in this paper seem appropriate for the analysis conducted, I do have a serious concern (see below) that needs to be addressed.

My concern is that the entire validity of the results presented rely on the assumption that the ERA5 reanalysis is providing reliably accurate information since it is being used here as the “target” for comparison and the bias correction being applied.  In my opinion this may not be a sound assumption to make over a relatively remote region of the Sahara where there are far fewer observations constraining the ERA5 reanalysis.  This means that there certainly is some degree of uncertainty in the ERA5 reanalysis data, but the authors do not explicitly address this uncertainty anywhere in the manuscript.  What is needed is for the analysis to be expanded such that not just ERA5 is used as a “target”, but also other atmospheric reanalyses (e.g., such as JRA-55 and/or MERRA2, to name a couple) are also evaluated.  In doing so, results from multiple reanalyses can be compared and explicitly discussed to address this uncertainty and provide a means to talk about the greater robustness of the findings in general. 

Without the inclusion of this expanded analysis the findings only have limited value because they are not placed in a broader context.  Unfortunately, what I am suggesting above will likely result in a large reworking of the entire manuscript and will take some time to complete.  Thus, for this reason I have recommended to reject and resubmit for this manuscript.  I encourage the authors to do so because there is good potential to advance our weather prediction capabilities from a study such as this one.  Below are some additional comments I had for the authors as they update their manuscript accordingly.

• Lines 21 – 23: sentence is worded awkwardly and can be misinterpreted.  How it is currently written implies the only reason the SHL emerged as a key component of the WAM system is because of AMMA, which of course is not true.  Suggest the authors update the text to better clarify that the AMMA project significantly highlighted the importance of the SHL in influencing variability of the WAM system.
• Line 34: Typo: Thorncroft and M. 1999  Need to fix this as you appear to be missing the second author’s last name both here and in the references.
• Line 76 – 77: “very hot temperature” Can you provide a range of temperatures here to show what you mean comparable to what you did for RH later in the same sentence?
• Line 87 – 91: “… detected the SHL with occurrence of more than 70% during the boreal summer, ….” Using what data? Daily?  Hourly?  Can you provide more information here about what you mean from all these prior studies that you presumably are taking the same regions?  Likewise, you discuss detecting the SHL here, but you have not yet mentioned exactly how you plan to detect the SHL.  What metric(s) are you using?  I presume this is coming a little later, but maybe it should come first.
• Line 95: Is the daily temperature just for a specific level/levels?  If so, which?  Again – this is related to my other comment earlier that it may be better to explain how you intend to detect the SHL before the discussion in 2.1 and 2.2.
• Lines 132 – 135: What data was the Lavaysse (2015) using (certainly not ERA5), and have you confirmed that it is indeed valid for ERA5 and the MF7 and SEAS5? It would be helpful to convey this explicitly to build confidence in your methodology here.
• Lines 222 – 226: In Figure 1 and other figures with shading (Figs. 4, 6, 7) there is not enough contrast between the different color hues making it hard to visually interpret values from the figure. Thus, it is hard to evaluate how well SEAS5 and MF7 are doing compared to ERA5.  Recommend the authors improve the figures by increasing the contrast between the color values used and possibly add line contours to label interval levels.
• Line 223 – 225: I don’t understand what the authors mean by “coherent climatologies of the SHL over the Sahara”.  I think they mean that the SEAS5 and MF7 reasonably replicate the climatology of ERA5, but I am not certain.  Please clarify.
• Line 225 – 226: It is unclear what is meant by “A progressive decrease in the intensity of the SHL is also observed over the North of Libya”. MF7 does not appear to yield the relatively cooler temperatures over northeastern Africa that are shown in ERA5.  Is this what is meant?  In any case the authors need to clarify this comment better.
• Line 237 – 239: “ … to get a robust selection of events at different periods.” Can you explain more explicitly what is meant by robust selection here?  Also – it would be helpful if the authors would explicitly mention with a sentence or two in the manuscript how the distributions change when the arbitrary threshold changes from 0.5 to 10.
• Line 240: “ …. in all our products ….”  By products, do you mean seasonal model forecasts?  Suggest clarifying to appeal more to readers less familiar with the seasonal forecasting lingo.
• Lines 277 – 279: This seems speculative. Given that you have all the output you could nail down whether or not this is what is happening.

Seasonal Forecasts of the Saharan Heat Low characteristics: A multi-model assessment by Cedric G. Ngoungue Langue et al.

General comments:

The Saharan Heat Low (SHL) is one of the important drivers of summer-time precipitation over the West African Monsoon. The authors discussed the performance of two state-of-the-art seasonal forecasting systems SEAS5 (ECMWF) and MF7 (MeteoFrance) in forecasting the Saharan Heat Low (SHL) against ERA5 reanalysis data for the period 1993-2016. SEAS5 and MF7 show opposite biases and both models under-estimate the interannual variability of the SHL. Statistical bias correction methods reduce biases; however, they do not add much in terms of skill beyond 1 month.  A lot has been done, but there are several critical issues that hamper the suitability of the current MS for its publication in WCD. Below are some points both major and minor that Authors should consider during their revision.

Some points are (major and minor):

• Introduction can be organized better. Line 19: poor precipitation forecast skill is for what seasonal? Subseasonal? The reference provided showcased 1-5 day precipitation skills. Line 37: Adding a demographic map will be good mentioning names of countries etc. Line 41 to 50: the references mentioned are from CMIP5 models, these may be fine, but not much relevant in the current research – this paper is concerning with initialized models aka seasonal forecasting system, so suitable references are those that utilized such models for SHL analysis. Line 51: Mentioning the “Climate Models”, in “above studies”, these are IPCC models, and Author mentioned seasonal forecasting models, so valid reference should be provided, if no reference available, just relying on OBS studies might be enough. Authors should take this concern, as it could avoid a lot of confusion present in the current MS. Line 56-60: rewrite, these are confusing and not clear. Line 60: remove “some”. Line 61: Climate models or seasonal forecasting models? Line 63: Prediction skill is for SF models. Line 62-63: Rewrite. Line 64: Need more details about this, by looking at this one can observe the low skill in SHL in current state-of-the-art SF models, how SHL then can be used as a predictor for the rainfall? “Seasonal models” must be seasonal forecasting models.
• Line 75: Add topography figure as mentioned above.
• Line 80-85: 30E must be 30 N, here and others as well.
• Boreal summer is JJA. A study considering June to September, and some places June to November. Be consistent
• Related to above, Figures 2 and 3, captions say the computation for SHL is performed for June to November period. While text section 2.3 (Line 113) says June to September period. Which one is right?
• Line 95: Be specific, Is this 2m temperature?
• Native SEAS5/MF7 is 36/37 KM, and ERA5 is 0.25, then why remap at 1x1.
• Line 108: SEAS5 forecast is for 0.5 to 5.5. What you mean by “for a period of 6-12 months for SEAS5”
• Line 115 to 120: Organize intro well, otherwise readers will remain in the state of constant confusion. Provide a reference/references that use either NWP or SF models.
• Line 134: Since T850 hPa is used, I would recommend using this term instead of simply sating temperature.
• Line 135: If no detection is performed, better to use a different heading for 2.4.1.
• Line 154: The period of the analysis is also the same for OBS. Make it clear. Are you focusing June to Sep or all 6-months?
• Line 158: “reservoirs” may be better replaced by being components.
• Section 2.4.3: Bias correction depends upon the data. You can consider other OBS datasets as well, matching your analysis window, and may add this OBS sensitivity as supplementary material.
• Line 211: Which “previous analysis”???
• Line215: Provide the definition of CRPS, that it is a quadratic measure of the difference between the forecast CDF and OBS CDF.
• Line 225: This agreement is in terms of OBS and models? Or just in terms of observation. Better to rewrite this sentence.
• Line 229: SEAS5 should be ERA5.
• Line 240: “similar behavior” in SEAS5 and MF7? Rephrase this sentence.
• Line 260: Move this to the Methods section.
• Line 265: The analysis period is from June to September when SHL is active. Why did you add here other months? Is there any reason? And you used Lead-0 in this case. This is strange to see a contrasting behavior in models at Lead-0.
• Line 273: Hot may be replaced by warm.
• Line 275: What is “observed bias”. Rewrite for clarity.
• “hotter trend”, What does this mean? Confusing? Rewrite.
• Line 270 to 280: Why two models show completely opposite bias? Authors must provide some plausible reason for these opposite behaviors in these models? Please also use other observations for a comparison. Maybe adding RMSE is also valuable here.
• Line 301: Avoid “trend”, when explaining bias analysis. This is confusing.
• Line 330-335: Move this to the Method section.
• Line 335: Not clear.
• Line 364: AEJ is defined already. Please check for this and others.
• What is DACCIW?
• Line 406: All datasets mean both SEAS5 and MF7?