Review of "Assimilation of surface pressure observations from personal

weather stations in AROME-France"

General points

The manuscript is generally well written. Like a number of other similar 

studies the sample size is rather limited, and on its own too short to 

prove that PWS data is beneficial. 

There is a comparison of impact in the 3DVar and 3DEnVar systems (the description/comparison is at lines 154-162). The PWS data has very little 

impact in the former, but some benefit (on one extreme case) in the latter. 

As I understand it 3DVar is the operational system and 3DEnVar is a test 

system that may replace it at some point. In section 6 I would like to see 

some discussion of likely future work/changes. When might 3DEnVar become 

operational? When might use of Netatmo data become operational (or what 

further testing is required)? Also what are the typical horizontal scales 

in the analysis of surface pressure of the two systems. 

Using very high density data can cause convergence problems in a variational 

analysis system. Wsa there any sign of such problems (eg in number of 

iterations required and/or condition number)? What was the additional cost of 

assimilating the PWS data? 

Personally I would have preferred to see the statistics in pressure units 

(rather than geopotential units) throughout, but I can see the logic of using the 

assimilated variable. 

I have made some suggestions to improve the English (but it is better than my French).

Detailed comments

line 1 [PWSs] 'have proved their ability' This seems too definite to me, if 

fully proved why do we need another paper on the subject? Reword. 

Section 1

I suggest adding at least two of the four references below (two of which use Netatmo 

data): 

Nipen, T.N.; Seierstad, I.A.; Lussana, C.; Kristiansen, J.; Hov, Ø. Adopting Citizen Observations in Operational Weather Prediction. Bull. Am. Meteorol. Soc. 2020, 101, E43-E57.

Sgoff, C., Acevedo, W., Paschalidi, Z., Ulbrich, S., Bauernschubert, E., Kratzsch, T., et al. (2022) Assimilation of crowd-sourced surface observations over Germany in a regional weather prediction system. Q J R Meteorol Soc, 148(745), 1752-1767. Available from: https://doi.org/10.1002/qj.4276

Hahn, C.; Garcia-Marti, I.; Sugier, J.; Emsley, F.; Beaulant, A.-L.; Oram, L.; Strandberg, E.; Lindgren, E.; Sunter, M.; Ziska, F. Observations from Personal Weather Stations--EUMETNET Interests and Experience. Climate 2022, 10, 192. https://doi.org/10.3390/cli10120192 

Bell, S.; Cornford, D.; Bastin, L. The state of automated amateur weather observations. Weather 2013, 68, 36-41

24 'combined to SWS' - 'combined with SWS'

26-27 'MC20 showed that PWS surface pressure observations allow multiplying by an

order of magnitude of 100 ... current state.' - 

'MC20 showed that use of PWS allows approximately 100 times more surface pressure observations to be assimilated.' (Better English.) As noted in the general comments 

in practice it is possible to have 'too many' observations and this should be 

addressed somewhere in the manuscript. 

28 'These observations, when simply spatialized' - unclear, 'spatialized' not a word? 

Possibly 'when regularly spaced'? 

36 'World Meteorological Organization (WMO) Regional Basic Observing Network (WMO, 2021)'

Instead of RBON I would refer to GBON (Global Basic Observing Network) 

https://community.wmo.int/en/activity-areas/wigos/gbon 

- a relatively recent concept. 

82 'Sect. 3' perhaps personal preference, I would write Section in full.

87 'represents 717 cycled hours of simulation for each experiment (without two time steps' - 

'represents 717 hours for each experiment (excluding two time steps' suggestion

 

96 'on ships (sending SHIP reports)' - 'on ships and buoys'?

Figure 1 caption: 

'emitting' - 'providing' 

'metropolitan France' - in English this means Paris or 'urban France', not the intention. Just 'France'. 

137 'J is incremental' - is it calculated at full forecast model resolution? 

Sometimes the increments are at lower resolution. Please carify in the main text. 

160 'reduces the propagation of the innovation at a shorter spatial distance' - 

'reduces the propagation of the innovation at larger distances' or 

'limits the propagation of the innovation to shorter distances'. 

200 'Andersson et al., 1991' I would remove this reference, old OSEs are not 

really useful now.

201 'Making OSEs allows us to' - 'OSEs allow us to'

212 'interpolated at round hours, which concomitantly removes' 

'interpolated to round hours, which also removes' 

I don't understand the bit about removing sparse observations. 

217 'concomitantly' - delete

224 'the Figure 3' - delete 'the'

227ff 'To implement a geopotential bias correction for PWS data based on AROME ... 

changes the altitude of the station in the assimilation scheme.'

One of the main causes of surface pressure biases is errors in the station altitude. 

One cause of error in station altitude is use of uncorrected GPS heights: 

"GNSS systems use a reference ellipsoid, and so the 'undulation of the geoid' 

must be taken into account to get heights relative to sea level. (Appendix

2 describes a nearly 40 m error at a radiosonde site where this was not taken into account.)"

from 

Pauley P, Ingleby B (2022) Assimilation of in-situ observations. In: Park SK, Xu L (eds) Data Assimilation for Atmospheric, Oceanic and Hydrologic Applications (Vol. IV). Springer.

Pages 293-371 in https://link.springer.com/book/10.1007/978-3-030-77722-7 

248 '1 is randomly drawn' - 'one is randomly drawn'

248 'generally never' - 'not usually'

289 'perturbed' - 'unsettled'?

312 'meridian' - 'meridional'

Figure 6 caption 'improvement' - presumably with respect to the control? 

Figure 7. The horizontal (average) lines partially obscure the score values. 

They should either be removed entirely or restricted to the side of the figure. 

Figure 7 caption. Add that higher values are better. 

359 'developed' - 'shown'

264 'dumping' - 'giving'

387 'which' - 'this'

Figure 8 caption 'black contours' - 'black edges' or 'black outlines'

300 'Once known ...' awkward. Perhaps 'Given the pressure increments the influence 

on other variables is explored.'

410 'Spalial size ...' - 'The surface pressure differences in Fig. 10e have 

smaller scale than the analysis increments in Fig. 10a.'

'5.4 Impact on forecasted rainfall' - remove 'ed' from forecasted

473/4 'a VarBC method could be ... correct the bias of PWS observations.' 

There are some caveats to this. Ideally there should be a good proportion 

of (uncorrected) anchor observations, but SWS are swamped by PWS data. 

How much this matters in practice for surface pressure isn't clear. 

Eyre, J.R. (2016), Observation bias correction schemes in data assimilation

systems: a theoretical study of some of their properties. Q.J.R. Meteorol. Soc.,

142: 2284-2291. https://doi.org/10.1002/qj.2819

NB. I was pleased to see the exploration of the bias correction in Appendix A. 

This manuscript describes the impacts of assimilating surface pressure observations from personal weather stations over France for one month cycling experiments using both 3DVar and 3DEnvVar. It includes diagnosis of a specific heavy rain case study.

I think this is a very well-written, clear description of a very important set of experiments. The number of non-professionally sited surface instruments is exploding. One would hope that the number of obs could counterbalance their less than optimal siting or poor measurement error characteristics. One of the easiest obs to do this with is surface pressure since it is much less severely impacted by poor siting.

This study describes a carefully designed set of experiments. Details of bias correction, and thinning are important and are presented clearly. A number of summary verification statistics for both analyses, and forecasts of varying length are included. The diagnostics of the heavy rain event are good choices and provide insight into how the personal surface pressure obs may have improved things. I have no major concerns about the manuscript, but I will append a short list of minor points. Thanks to the authors for presenting such a nice study on a subject I have been interested in for decades.

Minor stuff:

1. Line 94: replace ‘gather’ with ‘include’
2. Sentence starting at line 99: It was unclear to me how these numbers differed from those in the previous sentence. Please clarify.
3. Sentence starting at line 126: I wasn’t sure what the SURFEX model was here and what the relation of the Meso-NH was to other models discussed.
4. Line 128: ‘thereafter’ to ‘hereafter’
5. Line 135: I don’t think ‘determines’ is quite the right word. Maybe ‘estimates’ or ‘approximates’?
6. Line 143: ‘multiple observations during the assimilation window’. I was confused here. I thought you were averaging the two nearest obs around each hour so that there could not be multiple obs during an assimilation window of 1-hour?
7. Line 148: This sudden mention of geopotential obs was confusing. I suggest having the discussion of the use of geopotential versus pressure before you introduce the associated error constant. I will admit that I was never quite clear on the geopotential vs surface pressure uses either. I suspect that the paper describes it correctly, but maybe not quite as clearly in one place as it could.
8. Sentence starting at line 160: The discussion of the localization needs more clarity, and I suggest including an equation to make it clear. Is the localization something like Gaspari-Cohn with a halfwidth that varies linearly with height? Or is the localization itself linear?
9. Sentence starting on line 163: This seems out of place in the middle of the geopotential discussion and might be better placed somewhere else.
10. Sentence starting at line 167: I struggle to understand exactly what this sentence implies. If possible, try to make this distinction a bit more clear.
11. Line 174: ‘reduce this correlation’ to ‘reduce the negative impact of this correlation’. In fact, your later results suggest that there is almost no correlation and that the real benefit of thinning is to reduce the dominance of the PWS surface pressure observations.
12. Line 175: You introduce the term ‘innovation’ but then confuse the definition with the mention of OmA and OmB in the next sentence. I don’t think you ever use the term innovation again, so maybe you don’t need to define it?
13. Line 201 and subsequent places: I have a long history of criticizing the use of analysis error statistics because of the potential problem of overfitting. I don’t think that is a huge problem here, but I caution the authors to be very careful when looking at analysis errors, especially when there is a large number of a particular kind of observation.
14. Line 204: The description of the ‘monitoring’ experiment was not as clear as it could be. If I understand correctly, it does not cycle because it uses the background from one of the cycling cases. You say here that the raw pressures are assimilated, but is that really the purpose of the monitor? Isn’t it only used to look at OmB? My confusion might suggest a way to clarify the description.
15. Line 249-50: You are correct that these are independent instruments. I guess the only reason to expect correlated error is instrument systematic bias (all instruments always to high) or correlation of the altitude error of the model grid at the subgrid scale?
16. Line 287: Phrase starting with “which” was unclear to me.
17. Line 376: First phrase starting with “As” was unclear to me.
18. Figure 8: There is a lot of information here and I had trouble reading it. It might be better to separate into two panels, one with the obs and the other with the radar and the wind info. However, even the current panel mostly supports your points.
19. Line 439: Any ideas on why there is a degradation for the small threshold?