Katarzyna Osródka, Jan Szturc, Anna Jurczyk, Agnieszka Kurcz

Adaptation of RainGaugeQC algorithms for quality control of rain gauge data from professional and non-professional measurement networks

https://doi.org/10.5194/amt-2024-204

Summary

The article describes a number of performant quality control functions in order to assess the quality of precipitation measured by ground based stations. The overall system is detailed and provides a good understanding of different observation problems. The final outcome of the RainGaugeQC system is a quality index for each station.

General remarks

lines 67 and 73: There are numerous stations which are operated by former meteorologists or even active meteorologists on their private ground which provide data of a high quality. Often, operators of such stations are organised in amateur meteorological clubs. These clubs often publish data on the internet and togther with the conditions of the measurement locations, so that they are documented. Therefore, private stations do not guarantee a good data quality, but a considerable number of them is regularly monitored and their data of high quality. This needs to be verified for each private station, though.

line 82: internationally, dual-sensor rain gauges are not the rule, even in national weather services. The WMO classifies station location (e.g. GIMO Guide of WMO (2021)) and has performed several gauge intercomparison exercises (Lanza / Vuerich, 2009).

Specific remarks

line 33: "... highly distorted" - depending on the operator, usually radar measurements are quantitatively less accurate, but not highly distorted (see e.g. WMO, 2024)

line 33: "Rain gauge measurements are still considered ..." - maybe you could be more specific, like: "In hydrology, rain gauge measurements are considered ..."

line 42: it is correct that the authors have published relevant papers in this context. However, the reader would also appreciate more general publications, such as WMO BPG on radar data quality and Lanza / Vuerich (2009) on the WMO rain gauge intercomparison. 

lines 110 and 115: radar data should only be incorporated after their quality control - else this is not state of the art (WMO, 2024).

line 117: If you do not set a minimum threshold, your correlation risks to give you random results - this should be pointed out more clearly here.

line 131: the adjustment to rain gauges modifies the radar time series temporally and thus bears the risk that a comparison to a rain gauge time series becomes more difficult. A time series with corrected radar data and without gauge adjustment might give better results, in particular if the gauge data to be analysed have been integrated into the adjustment. 

lines 132 and 133: please give an indication of the length of the required time interval, e.g. one year.

table 1: the column "type of rain gauges" would merit additional information: what is the minimum volume of the tipping bucket gauges? Are they unheated for the two first rows? Which measurement type is the third row? "heated" does not tell much ...

line 181: please comment on the accuracy of the daily gauges if there is rainfall at the time of the day change. How accurate can the daily precipitation amount be under such conditions? Manually operated gauges often have a time interval for the operator to read and check the rainfall amount which may be in the order of 15 minutes. Such information should be communicated, additionally to the formulation that their data "are believed to be more accurate".

line 189: Please comment on the range of 250 km. Please note that for hydrological quantitative use, distances beyond 120 km range are subject to higher uncertainties in the radar measurement due to the measurement height and measurement geometry. Which range is practically used for your applications?

Figure 3: Why 215 km ranges? Can you please elaborate on this?

line 210: please give more details on the satellite data used in the system! Which is the data source, how is it quantitatively transformed to precipitation? Which is the original resolution in time and space and how is it mapped to fit to the ground measured data?

Table 2: What do you understand with "gross errors"? Please explain more in detail or refer to the correspondent section in this paper

Table 2: TCC - over which time interval does the comparison take place?

Table 2: SCC - please provide more details on the definition of outliers in this context!

line 255: what would be a typical "specific time interval"? Please provide a range in minutes!

line 256: do you request a minimum amount of rainfall in radar and gauge? 

line 256: When do you consider a correlation coefficient to be "good"? Please give more details in the assessment of the correlation quality!

line 263: how do you carry out the unbiasing procedure? Please provide more details!

line 272: when do you consider a time series to be long? What is the minimum duration for this?

line 275: the amount of 0.025 mm is per which time period?

line 336: the formula implies that the QI value is reduced even for perfect data. Is this intended?

Formulas (9) and (11): it is unusual to work with a bias in this way - more often, a multiplicative approach is used (see chapter 3.3.5 of WMO, 2024). Your approach penalises a deviation of 5 mm equally for a rain gauge sum of 10 mm and of 50 mm where in the first case, this represents 50%, in the second one 10%.

Figure 4: (a) please indicate the number of dry time spells for each of the months - for some statistics this plays an important role

          (b) how did you take into account the systematic bias of extreme values due to the interpolation of the gauges? Does the distance of each gauge play a role? How would the result have looked when comparing to radar data?

line 422: how can you state that the reliability of both data sources is comparable, if your reference is biased? This is, also in the light of the important and illustrative discussion in this paragraph, a statement without foundation.

line 458: Your finding that gauge data in Junauary are the least reliable is at least surprising since point rainfall data in summer are less representative in space. It would therefore be beneficial to read a discussion on these findings, in particular considering the predominant rainfall types and their spatial variability. Is this influenced by low temperatures in winter?

line 521: is this a finding for this day only or are the non-professional gauges always biased towards higher values?

line 540: do you consider this high value to be an outlier or a true value? If a true value, please discuss the discrepancies that you can see in figure 7.

line 556: do you have an explanation for the underestimation? Was there snowfall? 

Technical details

line 11: replace "10-min" by "10 min".

line 49: delete "often"

lines 76, 133: replace "np" by "e.g."

line 79: delete "very"

line 122: I suggest to replace "underestimated" by "underestimating the true rainfall"

line 141: replace "were" by "are"

lines 147 - 148: please provide the number of stations here or omit them from the two following bullet points in order to give a uniform information

line 292: please add "r" after "correlation coefficient" for clarity purposes!

References:

WMO (2024) Guide to Operational Weather Radar Best Practices (WMO-No. 1257). Volume VI: Weather Radar Data Processing. Provisional version at https://community.wmo.int/en/activity-areas/weather-radar-observations/best-practices-guidance

Lanza, L., Vuerich, E. (2009) The WMO Field Intercomparison of Rain Intensity Gauges. Atmospheric Research, Volume 94, Issue 4, December 2009, Pages 534-543.

WMO (2021) Guide to Instruments and Methods of Observation (WMO No. 8). https://community.wmo.int/en/activity-areas/imop/wmo-no_8.

The authors modify an existing rain gauge quality control (QC) algorithm, to make it better able to handle data from non-professional gauges. The modified algorithm is used to quantify the accuracy of both professional and non-professional data. The added value of including the non-professional data in multi-source precipitation products is demonstrated, both for a longer period and for individual events. 

There is a growing interest in using non-professional precipitaton observations to complement professional observations from national or regional meteorological and hydrological services. A major challenge concerns the higher degree of uncertainties and errors in the non-professional data. Therefore, efforts to develop and share practically applicable QC algorithms are timely and must be encouraged. I find the approach presented overall sensible (although some further clarification is needed; see below). The presentation is clear and the calculations and analyses appear well performed, as far as I can judge, but some revision is needed in light of the following comments.

General:

• Much of the results and conclusions are based on meesurements from four single months, considered typical for their corresponding season. This is a quite heavy assumption, firstly because one month is very short in a rainfall/precipitation context – it may contain just a few single events – and secondly because a single month may differ substantially from seasonal climatology. I would suggest to include as much “seasonal data” as you have available for this analysis. Alternatively, but much less preferably, show that the selected months are sufficiently good “seasonal representatives” for supporting the conclusions made.
• On lines 347-349 is written that (sufficiently dense) PWS data is not available at IMGW, and therefor tests have not been made on PWS data. I wonder whether the EUMETNET Sandbox with Netatmo data (Netatmo, 2021), which cover Poland, could be used for this purpose. If so, this would be a very interesting addition to the paper.
• Parts of the description of the updated algorithm are a bit hard to follow (see examples below and also comments from RC1), more explanation and justification needed.
• The English is overall understandable and not a big problem for me, but there are examples of curious expressions that could be improved by a “native check”, I think (some examples below). 

Specific:

•    68: I think the P in PWS usually stands for Personal (although I do have seen also Private).

•    76 and others: Change np. to e.g.

•    79: Example of sub-optimal English (in my opinion): “relatively very large”.

•    124: et al. is missing.

•    Table 1: What type is the DLP gauge? Weighing?

•    171: This expression is rather for the Introduction.

•    Figs. 2 and 3: I suggest combine into one.

•    204-207: Some more details here are needed to understand the following applications.

•     238: Is this done at each time step? Do you mean the sensor with highest quality?

•    246: “proved unsuitable”, in what way?

•    255: Which time step (or interval)?

•    268: English: “degree of outlying”

•    276: How was 5 and 10 days selected? They are not that different, can you justify that they are short and long enough?

•    Section 3.3: Generally, many numbers/thresholds here that are given without explanation or motivation. I assume they have been carefully set, but some clarification would be good.

•    285: Why different limits for R and G?

•    295: Has c in this eq. been defined?

•    323: This eq. is one example of something that needs more clarification.

•    336: So, PWS is always biased? Is this a reasonable assumption?

•    359: English: “additionally non-outlier”

•    373: How is this classification made? More explanation needed here.

•    411: Insert “, respectively,” between “gauges” and “is”.

•    Fig. 5: The font is a bit small.

•    458: Any influence of snow?

•    570-571: A bit strong statement, in my opinion, that these two cases show that non-professional are useful “in most cases”. Probably/hopefully they are, but the statement would require more cases to be evaluated.

•    597-599: Is this a conclusion from the present study? I do not really see this.

•    602-604: This is clearly not a conclusion from this work. I suggest extend this to a final paragraph about future efforts, remaining issues, etc. 

Reference:

Netatmo (2021): EUMETNET Sandbox: Netatmo observing network data v1. NERC EDS Centre for Environmental Data Analysis, 2025-03-15. https://catalogue.ceda.ac.uk/uuid/e8793d74a651426692faa100e3b2acd3/