Hydrologic implications of projected changes in rain-on-snow  melt for Great Lakes Basin watersheds

Myers, Daniel T.; Ficklin, Darren L.; Robeson, Scott M.

doi:https://doi.org/10.5194/hess-27-1755-2023

Articles | Volume 27, issue 9

https://doi.org/10.5194/hess-27-1755-2023

© Author(s) 2023. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/hess-27-1755-2023

© Author(s) 2023. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 27, issue 9

Research article

|

04 May 2023

Research article |

| 04 May 2023

Hydrologic implications of projected changes in rain-on-snow melt for Great Lakes Basin watersheds

Daniel T. Myers, Darren L. Ficklin, and Scott M. Robeson

Download

Final revised paper (published on 04 May 2023)
Supplement to the final revised paper
Preprint (discussion started on 16 Sep 2022)
Supplement to the preprint

Interactive discussion

Status: closed

RC1:
'Comment on egusphere-2022-657', Anonymous Referee #1, 10 Oct 2022
General comments:

This study is a novel investigation that is of interest to the professional community and in-line with the aims and scope of the journal. The topic is appropriately introduced with justification provided for the specific objectives. While some additional details on the statistical testing could be added (see below), the methodological approach appears logical and reproduceable. The results are organized around specific themes with figures that enhance understanding and are aligned with the final conclusions. Prior to supporting acceptance and publication, there are a small number of outstanding concerns with the manuscript that are addressed below as specific comments.

Specific comments:

The proportions of historical ROS melt [to total melt] is larger here than a variety of previous findings for the region. For instance, Welty and Zeng (2021) find extreme ROS occurrence is approximately 24% for the Great Lakes basin, similar to the value the authors give on line 34 at over 25% of extreme ablation events being ROS. Looking at all ROS events, not just extreme, the maximum value to date I am aware of for this region is found in Suriano (2022). This notes between 30-50% of ablation is ROS in the eastern lakes, compared to less than 20% in the extreme northern/western regions. While the results here have a similar spatial pattern to Suriano (2022), with more ROS in the eastern lakes and less to the north and west, the magnitudes are rather different. Given one of the primary results of this study is the detection of large decreases in ROS events under the RCP4.5 scenario relative to historical period, it is warranted to provide further discussion on the robustness of the historical model values relative to observations. This appears absent from the manuscript currently and should be incorporated into the discussion section of the revision.

The authors acknowledge on line 126 the threshold used for statistical significance for their correlation tests. However, it is unclear if any significance testing was conducted for the rest of the study. Was any sort of t-test or difference of means testing conducted for the results comparing the historical period to the mid-century period? If not, this should be considered by the authors to aid in differentiating meaningful changes from ones still within the noise.
Citation: https://doi.org/10.5194/egusphere-2022-657-RC1
- AC1: 'Reply on RC1', Dan Myers, 09 Nov 2022
  
  Dear Reviewer 1,
  Thank you for pointing out that the estimates of historic ROS melt differ from previous studies including Welty and Zeng (2021) and Suriano (2022). We are looking into how differences in each study’s definition of what makes a ROS event could affect the historic estimates. For instance, the Suriano definition seems more restrictive than ours and could result in a lower estimate of historic ROS. We are going to investigate this further and incorporate a discussion of the robustness of our historical model values relative to observations and possible implications into our revision.
  We are also considering how best to incorporate significance testing into the revision, in addition to our analyses of the sizes of the effects from the model results.
  Again, we appreciate the thorough review, and thank you for your thoughtful comments. We will submit a full response with revisions after the public comment period closes.
  Best,
  Dan, Darren, and Scott
  
  References:
  Jeong, D. Il and Sushama, L.: Rain-on-snow events over North America based on two Canadian regional climate models, Clim. Dyn., 50, 303–316, https://doi.org/10.1007/s00382-017-3609-x, 2018.
  Suriano, Z. J.: North American rain-on-snow ablation climatology, Clim. Res., 87, 133–145, https://doi.org/10.3354/CR01687, 2022.
  
  Citation: https://doi.org/10.5194/egusphere-2022-657-AC1
- AC4: 'Reply on RC1', Dan Myers, 30 Nov 2022
  
  Reviewer 1,
  Thank you for your helpful comments. We have attached our final responses.
  Best,
  Dan, Darren, and Scott
  
  Citation: https://doi.org/10.5194/egusphere-2022-657-AC4
RC2:
'Comment on egusphere-2022-657', Anonymous Referee #2, 27 Oct 2022

A representative simulation of ROS melt events is important for improving hydrological modeling practice in snow dominated region. It is valuable to look into the future impact of ROS melt events under climate change. This is exactly what this work intends to address. However, the current manuscript is not yet ready for publication, due to two points :

1. This work utilized a calibrated SWAT ROS model to simulate the hydrological process using CMIP5 climate projections. All analyses are based on the assumption that this calibrated model is representative. However, as described “The SWAT ROS model for the Great Lakes Basin simulated historic streamflow at the daily time step with an NSE of 0.38 (with 29% of stations greater than 0.5) and a dr of 0.62 (Myers et al., 2021b). The model simulated historic snowpack SWE at the daily time step with an MAE of 26 mm”, the model cannot be well considered well-calibrated with a low NSE of 0.38 for discharge simulation. Moreover, 26 mm MAE for daily SWE is a considerable high bias in comparison to the SWE value of the study area (e.g. Figure 4). The median SWE value of many months is around 50 mm or lower. GCM climate projections are highly uncertain already. A hydrological model with high bias will make the combination much worse. As a consequence, it is not reasonable to trust the analyses of this work about future climate change impact, even the analysis strategy is comprehensive. Therefore, the authors should implement the climate change investigation based on a reasonably well-calibrated SWAT ROS model. Moreover, detailed information about the rationality of the calibrated SWAT model is necessary but missing. Such information should be properly added to this paper or its supplementary material for its readers. The authors simply cited the paper that developed and evaluated the SWAT ROS model (reference below). But it is not open-access.

Myers, D. T., Ficklin, D. L., and Robeson, S. M.: Incorporating rain-on-snow into the SWAT model results in more accurate simulations of hydrologic extremes, J. Hydrol., 603, 126972, https://doi.org/10.1016/J.JHYDROL.2021.126972, 2021b.

2. Future climate projects have large uncertainty. When evaluating climate change impacts, it is more reasonable to discuss the trend or relative changes rather than absolute quantities. The authors should shorten such contents and keep the necessary ones only. Besides, Figure 2 shows different behaviors of climate driving force during different future periods. It would be interesting to investigate the corresponding hydrological signatures of different future periods. Although, as described in section 2.4, the analyses of future period include mid-21st century and late-21st century. Throughout the paper, the result and analysis of late-21st century is almost none. Please complete such missing parts.

Citation: https://doi.org/10.5194/egusphere-2022-657-RC2
- AC2: 'Reply on RC2', Dan Myers, 09 Nov 2022
  
  Dear Reviewer 2,
  Thank you for pointing out that the model performance is lower than often reported in hydrologic studies. We are currently experimenting with different ways to adjust and recalibrate our model to achieve better performance, such as modifying potential parameter ranges within reason. Our goal is to have our best model ready for the revision.
  Our study simultaneously evaluates 99 streamflow and 50 snowpack stations at the daily time step spread over a large hydrologically diverse region. This has the benefit of allowing us to verify that our model is performing as well as possible for simulating many watersheds and hydrological processes. However, it has created an extra challenge to achieve the error metrics associated with a more geographically limited system, which could fit well to one watershed or process but perform less well with other watersheds or processes (e.g. snowpack). We currently include several hydrologically diverse stations where our model performs well (NSE > 0.5) and where our model performs not-so-well, so that the model can be as representative as possible of the entire basin.
  Also, we want to use the same model parameter set across the Great Lakes Basin so we can know that any spatial differences in ROS represent actual differences due to spatial variation of climate forcings, rather than artifacts of regionally calibrated parameter sets simulating processes differently. Thus, with this single global parameter set, some stations perform well while others perform not-so-well, but we believe that the benefits of this approach in providing confidence in our interpretations of spatial ROS variation outweigh the sacrifice of model performance for some stations.
  We apologize that the Myers et al. 2021 paper that the ROS model is based on is not open access. We are happy to share this paper (through the editor if you prefer) to provide the best background for our study. We will also incorporate substantially more information about the ROS model into our current paper so that other readers who encounter the same problem can access the information.
  With regards to the comparison of different time periods (mid-21st century and late-21st century), we decided to focus on the mid-21st century because that is the period that the models and RCP's generally agree about climate changes, and because we expect that it will be the most meaningful focus for water resource managers in the Great Lakes Basin. By late-21st century, the climate projections also have much larger uncertainty. Thus, we prefer to stick with the mid-21st century for most of our analyses. We will certainly incorporate more relative rather than absolute changes into the manuscript to best evaluate the climate change impacts.
  Thank you again for the thorough review and helpful comments that are improving the quality and reliability of our paper. We will submit a full response with revisions after the public comment period closes.
  Best,
  Dan, Darren, and Scott
  
  Citation: https://doi.org/10.5194/egusphere-2022-657-AC2
- AC5: 'Reply on RC2', Dan Myers, 30 Nov 2022
  
  Reviewer 2,
  Thank you for the helpful comments. We have attached our final responses.
  Best,
  Dan, Darren, and Scott
  
  Citation: https://doi.org/10.5194/egusphere-2022-657-AC5
CC1:
'Comment on egusphere-2022-657', Sandra Akkermans, 09 Nov 2022
This review was prepared as part of graduate program Earth & Environment (course Integrated Topics in Earth & Environment) at Wageningen University, and has been produced under supervision of dr Ryan Teuling. The review has been posted because of its potential usefulness to the authors and editor. Although it has the format of a regular review as was requested by the course, this review was not solicited by the journal, and should be seen as a regular comment. We leave it up to the author’s and editor which points will be addressed.
Title paper: “Hydrologic implications of projected changes in rain-on-snow melt for Great Lakes Basin watersheds”
Overall impression
Rain on snow (ROS) melt events can have a big influence on their surroundings and can be either big or small. This manuscript investigates the impact of the climate change on the ROS events in the Great Lakes Basin in northern United States and few states of Canada for the period of 1960 to 2099, focussing on the period of 2040-2069 by the use of the model Soil and Water Assessment Tool (SWAT) with the addition of an energy budget equation to project the ROS events. With the results they looked at relationships and correlations between the different obtained variables. In general, the ROS events tend to happen earlier in the year by mid-21st century compared to the historical 1960-1999 values. The rain to snow ratio changes from around 1.5 historically to 2.0 at the end of the 21st century. This all also has influence on the water yield of the basins which also shift to earlier in the year.
The paper is nicely written and is important in these times of changing climate. The text has a good structure as the results are divided in understandable blocks. The variables are also captured in some good figures, although some changes should be made. Based on these comments, I would recommend publication with minor revisions.
General comments,
Firstly, as definition of a rain-on-snow event in the analysis section is stated that an event occurs on days with >1 mm rainfall on >1 mm snowpack SWE. By reading the Jeong and Sushama, 2018 paper this definition is not complete. It should be days with >1 mm rainfall on >1 mm snowpack SWE and decreasing SWE. By using the wrong definition rain-on-snow events could be wrongfully depicted in the results, and thus possible differences in conclusions. This should be solved by correctly using the definition and changing this in the paper.
Secondly, in the research question is stated that the change of rain-on-snow melt and hydrology due to climate change will be assessed for the 21^st century. In the method an argument is made about that for informing water resources management and because of better agreement with the models the primary focus will be on the mid-21^st century (2040-2069). In the rest of the paper late-21^st century is only mentioned for the change in ratio of area-weighted winter+spring rain-to-snow. Which has only a 0.1 change to the ratio change of the mid-21^st century where the mid-21^st century has a better agreement to the models and thus will probably be more accurate. Also in the conclusion at the end is stated “could help prepare ….. for the climatic changes of the 21^st century and beyond” where nothing is known for this latter time period with results from this paper. Thus, either the research question has to be rewritten to only the mid-21^st century together with the conclusion or the late-21^st century should be included in the rest of the analysis.
Lastly, again a comment on the research question but now because change in hydrology due to climate change is asked in the question. When stating hydrology I expect more variables to be analysed then only water yield, the rest of the analysed variables either belong to the change in climatic values such as rain and temperature or change in rain-on-snow events. Also, in the methods groundwater is mentioned to be modelled (line 65) but later not analysed in the results. So, either the phrasing of the research question should be altered or more hydrological variables should be assessed in the paper. This is important as the goal of this paper is to inform water managements to prepare for the changes due to climate change. For them the groundwater or runoff variables are also very important, and as they change with changing ROS melt (as said in line 39-41) this should be addressed.
Comments,
In lines 312 -320, Myers mentions the difference in findings with Surianon mentioning that the studied times differ. But as the simulation used in this study is done for 1960-2099, the same time periods could be compared as the data will be present after simulations. Why not analyse the same time period (1960-2009) as Suriano to make this comparison possible, to diminish this suggestive difference.

Lines 338-340, the speculation of influence of rain-to-snow ratio on size and timing of spring snowmelt and summer baseflow is made. This could be analysed by simply calculating the correlation between the COV (center of volume) and the rain-to-snow ratio which are variables present in the results.

Line 99, “thus. Nineteen climate models …… were used”. There is a missing argument about why you use 19 models instead of more or less. Please add an argument.

Line 165, the word “drastic” is used. Drastic is not a quantitative value as it is more an subjective use when not supported with arguments. Either rewrite the sentence or add an argument.

Lines 103-105, this information is a good/ better argument for the statement in line 118-119. As it now seems as random added information but potentially better used in the latter argument.

Line 167 , it is as a reader unclear whether the value mentioned for March is also the maximum or the value at the time of the maximum of April. This should be clarified.

Line 172, it is unclear if “proportion of melt” means temperature based melt or ROS based melt. This should be clarified.

Comment on figures in general: in the captions some abbreviations are written out where others are only posted as abbreviation. For the consistency of the paper this should be the same in all figures.

Line 212, why is 2050s mentioned here instead of the same formulation used in the rest of the paper: “mid-21^st century”? This could confuse the reader.

Figure 6, Instead of “high flows” state “high water yields” in figure title of 6 e and f as stated in the caption for clarification.

Figure 7, could be better depicted when the figures with frequencies (b, d and f) are on the right and with melt (a, c and e) on the left.

Specific comments,
Line 168, “a 54% April decrease” should be rephrased, for example”54% decrease in April”.

Line 197, figure 5c should be changed in 5d.

Figure 7, d is never mentioned in text.

In figure 3 add in title of figure 3 b, d and f that it is for the period of the mid-century for clarity.

In figure 4 some alterations on the axis titles can clarify the graphs. In 4b “snowmelt” can be clarified by writing “total snowmelt” and for c “proportion ROS” can be “proportion melt by ROS”.

References
Jeong, D. Il and Sushama, L.: Rain-on-snow events over North America based on two Canadian regional climate models, Clim. Dyn., 50, 303–316, https://doi.org/10.1007/s00382-017-3609-x, 2018
Citation: https://doi.org/10.5194/egusphere-2022-657-CC1
- AC3: 'Reply on CC1', Dan Myers, 10 Nov 2022
  
  Dear Sandra and members of the Integrated Topics in Earth & Environment course,
  This is very valuable feedback and we appreciate the input! We will incorporate these comments and aim to write you a response as well.
  First, thank you for pointing out our error in stating how we defined ROS events. You are correct that the definition must include snowmelt occurring, which we included in our script but hadn't mentioned in the text. We will be sure this is addressed in our revision.
  Second, thank you for identifying the need for clarity in the time periods our study is applicable to. We will follow your recommendations to make these more clear.
  Third, we focused on water yield because it is normalized streamflow, which our model was calibrated and evaluated for. Similarly, we report snowpack that our model was calibrated and evaluated for. Other variables such as groundwater and soil water clearly are important, but we provided limited results for them because of space or data limitations. This is definitely an area of future research for us.
  Thank you also for providing the other general comments and specific fixes, which we will incorporate into our revision. We are also adding you to the acknowledgements in the revision.
  Best,
  Dan, Darren, and Scott
  
  Citation: https://doi.org/10.5194/egusphere-2022-657-AC3
- AC6: 'Reply on CC1', Dan Myers, 30 Nov 2022
  
  Sandra and the Integrated Topics in Earth & Environment course,
  Thank you for the helpful comments on our manuscript. We have attached our final responses.
  Best,
  Dan, Darren, and Scott
  
  Citation: https://doi.org/10.5194/egusphere-2022-657-AC6

Peer review completion

AR: Author's response | RR: Referee report | ED: Editor decision | EF: Editorial file upload

ED: Reconsider after major revisions (further review by editor and referees) (06 Dec 2022) by Daniel Viviroli

AR by Dan Myers on behalf of the Authors (09 Jan 2023) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (10 Jan 2023) by Daniel Viviroli

RR by Anonymous Referee #1 (16 Jan 2023)

RR by Anonymous Referee #2 (14 Feb 2023)

Suggestions for revision or reasons for rejection

The manuscript has been improved according to the comments in round 1. Especially, the climate projection sections read more reasonable now. But (there is a but), the model rationationty part, as indicated in my first comment of last round, is still not convincing. This is the basis for whatever outcomes of the future projection investigations in this work. Because ROS is such an interesting research topic and this work could be beneficial for the community, I dig into the methodology paper (Myers et al., 2021b) that this manuscript is based on. To make it clear, let me zoom into the specific points:
1. Low average NSE for discharge simulation. “The SWAT ROS model for the Great Lakes Basin simulated historic streamflow at the daily time step with an average NSE of 0.38 (with 29% of stations greater than 0.5, 48% greater than 0.4, and a maximum NSE of 0.71)”. It is not acceptable, in particular, the author wants to address the effects on extreme high water yield (as defined by the authors: “Finally, the SWAT model outputs for water yield represent the area-averaged water export through the outlet in mm.”). with such a definition, streamflow is more or less equivalent to or contributes a lot to the water yield of this work. Mathematically, NSE is highly affected by peak flow which is highly relevant to extreme high water yield. Low NSE largely indicates a bad fitting of the peaks. Thus, it is not reasonable to discuss the implications on extreme water yield, if the model cannot represent them properly.
Thoughts: After checking the methodology paper (Myers et al., 2021b), it seems that the model simulated somehow descent streamflow in winter/spring period but performed bad in summer period. If ROS is more important for winter/spring period (I am not a snow scientist. This statement should be double checked), perhaps calculating the metrics and proving the reasonable representation of streamflow in these seasons is a breakthrough point.
Minor suggestion: higher temporal resolution does not necessarily mean using a lower criterion for evaluating the model performance. And the spatial side should not be ignored. In this work, the spatial resolution is coarse. I suggest to remove the statement “for instance that an NSE between 0.3 and 0.5 could fit criteria for satisfactory model performance in some contexts”.
2. High MAE for daily SWE. The authors added some descriptions of snow melt performance as supporting references of reasonable snow simulations. However, acceptable MAE snowmelt simulations do not mean reasonable performance of SWE outputs. It may indicate the melt module works, and something has to be improved in the snow accumulation module. The authors should clarify this point: is 26 mm MAE for daily SWE is an acceptable bias in the study region or not? Particularly, as shown in Figure 5, the median SWE value of many winter months is around 50 mm or lower. As a reader, I will interpreter 26 mm MAE as a large error.
3. The authors added some information about the modified ROS-SWAT in this manuscript. “simulates ROS melt based on a function of air temperature, precipitation, wind, saturated vapor pressure, and atmospheric pressure”. In your current descriptions, it means wind, saturated vapor pressure, and atmospheric pressure datasets are required. It could indicate your method cannot work in many snow process dominated remote areas due to data scarcity. But I found the answers in your supplementary source code snom.f, those variables are simulated internally based on elevation (atmospheric pressure), air temperature (saturated vapor pressure) and fixed value of 0.15 (for wind variable UADJ). To avoid misleading, please clarify briefly what exactly is needed as inputs for your module in the manuscript.
Many thanks to the editors for sharing me the methodology literature: Myers, D. T., Ficklin, D. L., and Robeson, S. M.: Incorporating rain-on-snow into the SWAT model results in more accurate simulations of hydrologic extremes, J. Hydrol., 603, 126972, https://doi.org/10.1016/J.JHYDROL.2021.126972, 2021b.

Hide

ED: Reconsider after major revisions (further review by editor and referees) (27 Feb 2023) by Daniel Viviroli

AR by Dan Myers on behalf of the Authors (09 Mar 2023) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (15 Mar 2023) by Daniel Viviroli

RR by Anonymous Referee #2 (16 Apr 2023)

ED: Publish as is (17 Apr 2023) by Daniel Viviroli

AR by Dan Myers on behalf of the Authors (17 Apr 2023) Manuscript

Short summary

We projected climate change impacts to rain-on-snow (ROS) melt events in the Great Lakes Basin. Decreases in snowpack limit future ROS melt. Areas with mean winter/spring air temperatures near freezing are most sensitive to ROS changes. The projected proportion of total monthly snowmelt from ROS decreases. The timing for ROS melt is projected to be 2 weeks earlier by the mid-21st century and affects spring streamflow. This could affect freshwater resources management.