the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Leveraging a Disdrometer Network to Develop a Probabilistic Precipitation Phase Model in Eastern Canada
Abstract. This study presents a probabilistic model that partitions the precipitation phase based on hourly measurements from a network of radar-based disdrometers in eastern Canada. The network consists of 27 meteorological stations located in a boreal climate for the years 2020–2023. Precipitation phase observations showed a 2-m air temperature interval between 0–4 °C where probabilities of occurrence of solid, liquid, or mixed precipitation significantly overlapped. Single-phase precipitation was also found to occur more frequently than mixed-phase precipitation. Probabilistic phase-guided partitioning (PGP) models of increasing complexity using random forest algorithms were developed. The PGP models classified the precipitation phase and partitioned the precipitation accordingly into solid and liquid amounts. PGP_basic is based on 2-m air temperature and site elevation, while PGP_hydromet integrates relative humidity. PGP_full includes all the above data plus atmospheric reanalysis data. The PGP models were compared to benchmark precipitation phase partitioning methods. These included a single temperature threshold model set at 1.5 °C, a linear transition model with dual temperature thresholds of –0.38 and 5 °C, and a psychrometric balance model. Among the benchmark models, the single temperature threshold had the best classification performance (F1 score of 0.74) due to a low count of mixed-phase events. The other benchmark models tended to over-predict mixed-phase precipitation in order to decrease partitioning error. All PGP models showed significant phase classification improvement by reproducing the observed overlapping precipitation phases based on 2-m air temperature. PGP_hydromet and PGP_full displayed the best classification performance (F1 score of 0.84). In terms of partitioning error, PGP_full had the lowest RMSE (0.27 mm) and the least variability in performance. The RMSE of the single temperature threshold model was the highest (0.40 mm) and showed the greatest performance variability. An input variable importance analysis revealed that the additional data used in the more complex PGP models mainly improved mixed-phase precipitation prediction. The improvement of mixed-phase prediction remains a challenge. Relative humidity was deemed the least important input variable used, due to consistent near water vapor saturation conditions. Additionally, the reanalysis atmospheric data proved to be an important factor to increase the robustness of the partitioning process. This study establishes a basis for integrating automated phase observations into a hydrometeorological observation network and developing probabilistic precipitation phase models.
- Preprint
(2797 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on hess-2024-78', Anonymous Referee #1, 28 May 2024
This study of predicting the phase of precipitation is rather regional, but it is worthwhile for the large amount of novel data used and different conclusions on the importance of humidity to earlier studies in other regions. My comments are minor or technical.
Specific comments
Considering that there are several types of disdrometer, it might be worth getting the word “radar” into the title.
21
Specify which reanalysis data are in PGP_full (thickness, temperature lapse rate and surface pressure).
74
The difference in computational resources between threshold and curvilinear functions is trivial in application.
189
The WMO definition of freezing rain is supercooled liquid drops that freeze on impact with the ground. Is this what the WS100 records as “freezing rain”? How does its diameter-fall velocity relationship differ from warm rain?
Figure 2
Picking up some hints in the text, would scatter plots of these variables be interesting?
240
Thickness does not indicate the travel time unless the hydrometeor fall velocity is also known.
264
Aggregation of freezing rain with snow rather than rain is common in previous studies and is justified by the hydrological influence, but it seems to be a misclassification of the phase of the hydrometeors.
Figure 3
After reading the text many times, I think that the “mix of snow and rain/drizzle” in Figure 3a combines what the disdrometers class as “mix of snow and rain/drizzle” (which gets aggregated with the liquid phase in 3b) and hours with 15-minute periods classed as both snow and rain (which remain classed as “mixed” in 3b). But I am not confident in that interpretation (and I have no idea why the peak in snowfall close to 0C appears to go down slightly when classed as solid precipitation). Please don’t make the reader work so hard on something simple.
Figure 9
The colours do not convey any information, so I would not use them.
572
What is “Improving PGP models’ ability to accurately predict the mixed phase is manifold” meant to mean?
Figure 10
The diagonal is redundant. Removing it would allow making the rather small labels a bit bigger. The colour scale should have a label.
825
Why exclude snowfall when there is not already snow on the ground?
Appendix C
Does equation C4 not give the temperature for unventilated hydrometeors? The ice bulb temperature may be a more appropriate predictor (or there may be little difference due to the high relative humidities in this study).
Technical corrections
74
“yield a more accurate”
86-89
“it” becomes “they” over the course of this sentence.
107
“Therefore” seems incorrect at the start of this sentence.
168
“elevations range”
207
“with sensors”
208
Delete “ground”.
238 (and subsequently)
Prevent automatic capitalization of the first word after a display equation when there is not a new sentence.
314
“harshly penalizes a poor score in either” has already been said.
316
“the model partitioning performances are”
359
“the predicted phase is either solid or liquid,”
453
“PGP_basic overpredicts the mixed phase”
512-513
Spurious line break
658
“importance of 1000-850 hPa layer thickness”
679
“such as laser disdrometers”
686
“at a site sheltered from the wind”
Citation: https://doi.org/10.5194/hess-2024-78-RC1 - AC1: 'Reply on RC1', Alexis Bédard-Therrien, 01 Aug 2024
-
RC2: 'Comment on hess-2024-78', Anonymous Referee #2, 30 May 2024
General comment
This is a very nice paper and it is written well. It includes a lot of detailed analyses and discussions that make the paper very informative. I think it fits the journal well. I recommend major revision as I have several minor comments that need to be addressed before I can accept the paper for publication. Otherwise, the paper is in good shape. Also, as you revise the paper please make sure to clarify how the automatic measurements enables mixed phase classification/ partitioning?
Also I suggest adding “radar-based” in the title, to clarify this is not for laser-based dendrometer.
Specific comments:
- Lines 21, list the atmospheric variables used
- Abstract: partitioning is not an obvious term. I think by partitioning you mean the amount of each phase versus the type (classification). It would be helpful to define partitioning upfront.
- Line 65, you may want to add one or both of these also to further support your point: https://doi.org/10.1016/j.jhydrol.2022.127884; https://doi.org/10.1029/2019GL084221
- Line 145, I thought the use of “However” instead of “Additional” might fit the sentence better.
- Line 153, I thought it would be useful to discuss the difference between phase classification and partitioning here as this may not be obvious for readers.
- 185-186, so my understanding is that disdrometer does not “observe” phase. As you said in line 190 precipitation phase is identified according to the hydrometeor diameter-fall velocity relationships for water droplets and solid particles. So maybe you should replace “observation” with “estimation” or something similar.
- Lines 280-285, it is not clear to me how the aggregation of the mix of snow and rain/drizzle
with rain is performed? How you decided to convert fractions to solid or liquid phase?
- Add reference for the performance metrics used, on the other hand POD, FAR, HSS, CSI, etc. are also very popular. I also like BIAS as it shows over or under detection. How can you say you over or underpredicted?
- Line 316, don’t you have an extra “model” in the sentence?
- L323-324: I am not sure if I understand why RMSE is the same for liquid and solid phase. Explain.
- Section 3.3. it is not clear. Did you use the same number of solid and liquid phase for training? How about testing? For example you say 60% solid and 26% liquid were there. Did you reduce number of solid in training to match the number of liquid phase samples?
- Section 3.5. line 390, in the PB method, is the hydromet temperature similar to Wet bulb temperature. If not what’s the difference between PB and wet bulb temperature?
- Line 395, not sure what you mean the simple threshold method also gives binary and there is no mixed phase. The probability approach can be converted to binary rain/snow. I am not sure if I understand your point.
- Line 421, from Fig 6 I don’t see how median RH is around 97%. Should be lower.
- Lines 445-455: I don’t see any evidence in supporting your performance evaluation. Do you have a graphic or statistics in support of what you stated here about the performance evaluation? If so, refer to them in your text.
- Can you show the observation reference in Fig. 7?
- Line 522, add “S” . Table 4 show”s”
- REMOVE lines 512-521. This is a repeat ! later in Lines 530-538:
- Line 571, I don’t understand this sentence “ Improving PGP models’ ability to accurately predict the mixed phase is manifold”
- Lines 627-628: Do you have references to back up your statement here?
- Line 635, could be helpful if you refer to the figure or table that supports your overprediction claim
- Line 785, can you remind based on which figure the “phase overlap between 1.5 and 3.5°C” was concluded?
Citation: https://doi.org/10.5194/hess-2024-78-RC2 - AC2: 'Reply on RC2', Alexis Bédard-Therrien, 01 Aug 2024
-
RC3: 'Comment on hess-2024-78', James Feiccabrino, 11 Jun 2024
My recommendation (old standard) would be accept with minor corrections.
I overall enjoyed reading through the article and commend the writers for a well written thorough overview of the state of precipitation phase determination in hydrological or hydro-met models. They did a very good job of stating the problem (why it's important, and reviewing past work on the issue). They have an interesting and well thought out method and do a good job explaining their results and how it fits into the current state of work on the issue. They also attempt to identify why some results don't agree with all previous work and how it could be used for future studies. I would consider this a solid write-up.
Major corrections:Lines 530-539 are an exact copy of lines 513-521. One of these paragraphs should be deleted, and may affect the final location of Table 4.
Line 254 - (ECCC, 2024) is the reference used, but does not appear in that form in the reference list. I believe this is the reference on line 932, but there is no easy way to link the reference in the article to the reference named "climate glossary".
Line 225 - Hersback 2023 is missing et al., in the article reference.Minor Corrections and things to consider (not necessary changes):
Line 72 - grammar - consider switching "while occurring" with "that occur"
Line 205 - grammar - you are missing a word in "The weather stations measure hourly air temperature and relative humidity ___ sensors mounted at 2 m above ground level" some possibilities are rewording or filling ___ with (using/from...)
Line 201 - it could be cleaner for the reader if you edited to make "from weather stations near the disdrometer stations" the last statement in the paragraph lines 201-209, moved it to the beginning of the next paragraph lines 210-219. As is, it leaves the reader wondering How close is near? - However, not a major issue since the answer is found in the next paragraph
Line 220 - Formatting - Table 1 seems to be the last two words of the previous paragraph rather than needing a new line.
Lines 419 - 423 as it relates to Figure 6d - Something to check. It seems a bit odd, but possible that the mean precipitation rate 0.9mm is higher than the medians of 0.8mm, 0.7mm, and 0.6mm for mixed phase, liquid phase, and solid phase respectfully. It can be a correct statement, but figure 6d makes this less likely given the very low numbers of heavy precipitation events depicted in the chart... It's not highly important to the paper, but does look a bit off.
Line 452 - grammar - "the most the mixed phase" - perhaps PGP_basic has the greatest overprediction in mixed phase (plenty of options, but right now the grammar is incorrect).
Lines 566 - 568 - wording is a bit tricky, no issues with the beginning " The layer thickness is affected by environmental temperatures, as air temperature is inversely proportional to its temperature" however the end needs to indicate temperatures increasing or decreasing to finish the thought "which increases the distance between two pressure levels." A suggestion would be "... temperature. Therefore as temperatures increase, the distance between two pressure levels also increases".
Line 679-680 - grammar (missing word) - "options such __ laser disdrometers", looks like it should be "such as".
Line 766 - 767 - I would suggest consulting co-authors to make sure this is the final consensus on why "The longer time-step may lead to a lower critical threshold because the energy needed to melt the precipitation can be supplied over a longer period" I can't attach this thought with anything in the paper and would not personally agree with this statement.
Line 792 - 793 - should double check this, might be 850-1000mb height difference according to figures 10 and 11 - "According to the input variable importance analysis, *atmospheric pressure* was the second most important hydrometeorological variable for phase classification" - It is the second greatest reanalysis variable (bright blue in figure 11) but the 4th most important variable if considering all data in Figure 11. This statement could be correct depending on the intended meaning of "hydrometeorological variable".Lines 811 - 814 (Appendix A) - Longitude and Latitude, some values are given to 5 decimal points and others to 6, usually these values all have similar accuracy. I would suggest either rounding to 5 decimals, or if it is dropping a sixth decimal if = to 0, reformatting to show the 0 to show all coordinates having accuracy to 6 decimal points.
Other notes:I really liked how you explained the difference in outcomes between your study and other studies citing RH as an important factor for precipitation phase determination lines 605 - 609, and 641-644. I wish more papers included notes like this.Citation: https://doi.org/10.5194/hess-2024-78-RC3 - AC3: 'Reply on RC3', Alexis Bédard-Therrien, 01 Aug 2024
Data sets
Data for "Leveraging a Disdrometer Network to Develop a Probabilistic Precipitation Phase Model in Eastern Canada" Alexis Bédard-Therrien et al. https://zenodo.org/records/10790810
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
365 | 158 | 123 | 646 | 16 | 20 |
- HTML: 365
- PDF: 158
- XML: 123
- Total: 646
- BibTeX: 16
- EndNote: 20
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1