General comments

The manuscript “Assimilation of temperature and relative humidity observations from personal weather stations in AROME-France” analyses the value of assimilating T2M and RH2M observations of PWS in the AROME-France NWP model. To answer the questions of how the data should be pre-processed and what influence the DA scheme has, assimilation cycles of one month and two case studies were evaluated. In the case studies, the impact of assimilation on the prediction of a heat wave and a fog event was analysed. The procedure presented in the manuscript is reasonable and well structured. Therefore I only have some minor questions.

Specific comments

1. I miss the information which observation types are assimilated besides the observations of the PWS. It would be great if you could mention which observations are assimilated in your experiements 3DVar and 3DEnVar.

2. Why do you decide to do experiments with T2M/RH2M only and no experiment with combined assimilation of RH2M and T2M PWS stations?

3. Line 399: For me it is unclear if Fig 11c is with or without OI. The sentence: At ground level, the shape of the ground temperature increments with the OI in 3DEnVar experiment (Fig. 11c) is similar to … suggests that Fig 11c is with OI, but in the Figure caption it reads as it is 3DEnVar only.

Technical corrections

Line77f: Sometimes Section and sometimes Sect.

Figure2: A better contrast between observations and the domains would be great. It is difficult to read the numbers of the domains.

Figure4a: y-axis should be labeled with Mean OmB (°C)

Figure6: It is not mentioned what is represented by the coloured stars.

Figure10: The meaning of circle A and B is missing in the Figure caption.

Figure11: Column titles of the shown temperature level would be useful. Label of colourbar should be temperature increments.

The authors have performed detailed experiments of data assimilation of PWS data into the AROME model, testing various DA systems and bias corrections of the PWS data to improve forecasts under certain circumstances. I personally enjoyed reading this paper: it’s very thorough in its methodology and is critical of the limitations that DA systems and PWS ingestion into models have whilst accounting for those limitations and trying to solve inherent biases and issues.

While the paper is methodologically and scientifically sound, I do struggle to read through it at times, especially at the start and in the methodology section, since there is a very widespread use of jargon and abbreviations, most of which are not used often enough to justify their use. The high amount of abbreviations and numbers being thrown around in the text makes it a challenging read and risks losing the main message in a difficult-to-read section, which would be a shame. Please reconsider the amount of abbreviations, especially for the non-intuitive ones such as OmB. Also some of them are written out multiple times along the text. My advice would be to only keep those that come back constantly (PWS) or are essentially common in usage (QC). This will make the text much more readable.

Other than that I think the paper is quite good already, I have listed some more significant comments below that came to mind during my reading. Some of which are more questions borne out of curiosity, and some are points I had some issue with understanding properly. If the writing style is cleaned up and the issues below are addressed, I think the paper is well worth of publishing.

• Squall lines are mentioned, but rainfall observations are not included in the DA? Is this by choice or because the DA system for rainfall typically uses reflectivity and not on-ground data? I think it would be good to reflect on this since there has been substantial work on the quality of PWS rainfall data as well, especially since rainfall data is quite strongly spatially distributed.
• Most PWS stations are urban – how is that dealt with in the DA schemes, given urban roughness parameters and so on? Since these areas are not WMO compliant as would be the case for standard DA of observations.
• RH as prognostic variable calculated whereas it is directly measured by PWS (with inherent autocorrelation to the Tair measurements). How does that influence DA procedure? Also, RH depends on temperature, meaning biases are autocorrelated between the two, I didn’t see that point discussed (or perhaps I’ve missed it).
• Where does the sigma-o level in the OmB calculation come from? What is its influence on the bias correction successes? In L. 110: I don’t quite understand how the rejection threshold is calculated in the given examples – what does it mean that the threshold is “up to 6.5 °C”? Is the threshold something different than the OmB?
• 154: the number of observations is given, but does that mean that the amount of stations is equal to that? Or do stations give multiple obs. per hour?
• P 6/7: do you also have an indication on the amount of assimilated Netatmo stations, as you do for the other networks? Would put things into perspective and point out the importance of these data, as these numbers can be huge from my own experience. I see figure 3 contains frequency diagrams but a (rough estimation of) number of stations would be helpful.
• The assumption that observational errors are not correlated is an interesting one, as there can be apparent errors caused by urban morphology, i.e. microclimatical variations caused by the urban environment that are only picked up by a set of sensors at the specific location, whereas all neighbours would not find it, therefore classifying it as an error. How robust is the QC system that it does not dismiss microclimatical variations as sources of error? And how important is this for the DA procedure?
• Section 4.1: this could benefit from a spatial image as well to complement figure 8, since I’m curious what the effect of PWS density on forecast quality is.
• Figure 11: It’s not quite clear to me what is meant with ‘temperature increment’ (which is also not correctly labelled on the colour bar, as a side note).