the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 16 Aug 2024
Dynamic precipitation phase partitioning improves modeled simulations of snow across the Northwest US
Abstract. While the importance of dynamic precipitation phase partitioning to get accurate estimates of rain versus snow amounts has been established, hydrology models rely on simplistic static temperature-based partitioning. We evaluate model skill changes for a suite of snow metrics between static and dynamic partitioning. We used the VIC-CropSyst coupled crop hydrology model across the Pacific Northwest US as a case study. We found that transition to the dynamic method resulted in a better match between modeled and observed (a) peak snow water equivalent (SWE) magnitude and timing (~50 % mean error reduction), (b) daily SWE in winter months (reduction of relative bias from -30 % to -4 %), and (c) snow-start dates (mean reduction in bias from 7 days to 0 days) for a majority of the observational snow telemetry stations considered (depending on the metric, 75 % to 88 % of stations showed improvements). However, there was a degradation in model-observation agreement for snow-off dates, likely because errors in modeled snowmelt dynamics—which cannot be resolved by changing the precipitation partitioning—become important at the end of the cold season. Additionally, the transition from static to dynamic partitioning resulted in an 8 % mean increase in the snowmelt contribution to runoff. These results emphasize that the hydrological modeling community should transition to incorporating dynamic precipitation partitioning so we can better understand model behavior, improve model accuracies, and better support management decision support for water resources.
- Preprint
(1694 KB) - Metadata XML
-
Supplement
(199 KB) - BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on egusphere-2024-2284', Anonymous Referee #1, 25 Oct 2024
Summary
This manuscript examines the implications on simulated snowpack conditions associated with changing a precipitation partitioning model (partitioning precipitation into rain and snow fractions), from an approach considering only contemporary air temperature to an approach using both contemporary air temperature and humidity. The authors carry out a suite of numerical experiments in the Pacific Northwest region of the United States using the two alternative partitioning schemes and compare the approaches to each other and to observational data at USDA SNOTEL observation sites. Simulations and observations are compared on the basis of peak snow water equivalent (SWE), the timing of peak SWE, snow phenology (that is, the onset, conclusion, and duration of seasonal snow cover), and the fraction of runoff from snow at SNOTEL sites within the Columbia River Basin. Overall, the switch to a bivariate precipitation partitioning approach yields improvements in simulated snow conditions with respect to observations. Where there is degradation, the authors effectively argue that the degradations are likely the influence of factors beyond those that would be influenced by the change in partitioning approach. The paper is of interest to the readership of HESS and makes important, albeit primarily methodological, contributions to snowpack and hydrologic modeling in snow-dominated regions. I believe that it can be published with only minor revisions.
Major Comments:
- I can surmise why VIC-CropSyst is being used in this study, instead of VIC without the CropSyst model coupled. However, some readers might be left wondering why VIC-CropSyst is being used when, for example, value additive aspects of the crop model (e.g., dynamic crop yields, etc.) are not being examined and when the comparisons with SNOTEL observations are primarily in locations where there is little or no cultivation. As such, I would strongly encourage the authors to provide some additional context for using VIC-CropSyst for readers. For example, if the work summarized in the manuscript is part of larger and/or ongoing efforts to develop more sophisticated regional projections of how climate change might affect agriculture in the region, it would be helpful for readers to know.
- From a water balance perspective, the analysis of snowmelt contribution to streamflow is interesting. However, given that there is not really meaningful observational constraint at SNOTEL sites, we are left only with model-to-model comparisons. While the authors are clear that this is the case, it is also possible that they could select a few watersheds within the region and compare simulated SWE/Q to the USDA’s Basin Snow Water Equivalent estimates of SWE, normalized by runoff volume for the same watersheds. This represents some additional analysis, but it might provide a much more insightful comparison of the degree to which the change in precipitation partitioning influences whole-basin estimates of water supply.
Minor Comments:
- Line 84: I would quibble slightly that the Columbia River Basin is entirely snow dominated, given that some significantly large areas within the basin are, in fact, rain dominated. I’d suggest slightly rewording this to “The CRB encompasses multiple states and portions of Canada in North America, and snow comprises a substantial fraction of annual precipitation in much of the watershed, particularly high mountain areas.”
- Some of the figures (Figure 2, Figure 3) have some issues with the frames of the figure areas and seem misaligned with subfigures. I’d encourage the authors to address these minor formatting issues.
- For figure 9, the bin labels on the x-axis are slightly confusing. Is bin 4 really 10-20% bias? Or is it 0-20% bias? If the former, were there no results in the 0-10% bias range? Or is this what is meant by “Only bins with at least 8 stations are displayed.” If so, in the text, please indicate that “Fewer than 8 SNOTEL stations fall within the range of a 0-10% change in bias, therefore we exclude this range in bias change from these box plots.”
- Fir figure 9(d) the label of the plot could be more clear. I believe this is the fraction of precipitation to SWE, so perhaps P/SWE (-) would be clearer.
Citation: https://doi.org/10.5194/egusphere-2024-2284-RC1 -
AC1: 'Reply on RC1', Bhupinderjeet Singh, 26 Nov 2024
Thank you for the feedback and suggestions for improvements. We agree with all the suggestions, including the suggestion for additional analysis related to basin-level peak SWE/Q comparisons. All suggestions can be incorporated in the revised manuscript. Specific details for each comment are provided in the attached document.
--Bhupinderjeet Singh and co-authors
-
RC2: 'Comment on egusphere-2024-2284', Anonymous Referee #2, 31 Oct 2024
Singh et al. present a comprehensive analysis of changes to simulated snow cover evolution resulting from an updated dynamic rain-snow partitioning scheme in the VIC-CropSyst model in the Columbia River Basin (CRB) and other areas of the Pacific Northwest. They showed that a bivariate logistical regression method that predicts precipitation phase as a function of air temperature and relative humidity produces better simulated snow water equivalent (SWE) than the default VIC method that uses air temperature alone. The authors noted that improved snow outcomes led to a higher proportion of simulated streamflow coming from snowmelt versus the baseline case using the default VIC method.
I commend the authors on the hard work they put into this paper. The methods are clearly described, and results are straightforward and easy to follow. However, there are a few major shortcomings that I believe preclude this article from publication in HESS. I detail these below.
The first, and most important, issue is that the authors use the default VIC rain-snow partitioning scheme as the benchmark to which they compare the dynamic method. This default is a dual-threshold method with a lower, all-snow threshold of -0.5°C and an upper, all-rain threshold of 0.5°C, with mixed precipitation falling in between. Previous modeling work at multiple sites in the western US has shown this method to produce highly negative biases and low r2 values in both SWE and snow depth (Jennings and Molotch, 2019). More recent observational work from the Sierra Nevada in the western US further highlighted the poor performance of the default VIC method compared to visual reports of rain, snow, and mixed precipitation. It was the second-worst partitioning method, only correctly predicting rain, snow, and mixed precipitation 47.1% of the time (Jennings et al., 2023). In other words, the VIC method got over half of its precipitation phase predictions wrong.
Thus, the premise of the research—that a modified rain-snow partitioning scheme incorporating humidity and air temperature would improve on the default VIC method—is not a particularly evocative one. In fact, the researchers would be hard pressed to find a worse rain-snow partitioning method than the VIC default. This has the unsatisfying effect of producing results that are practically pre-ordained. In my opinion, this makes the work more appropriate as a case study for a different journal.
The second major shortcoming is the novelty of the work. Wang et al. (2024) recently demonstrated similar findings when implementing the VIC model with a new rain-snow partitioning scheme (wet bulb temperature, TW) and comparing the outcomes to the default dual-threshold method. They showed “improved performance of the TW scheme in simulating snowfall fraction (SF) and snow water equivalent (SWE) in relation to in situ observations and a gridded SWE product.” They also took the research a step further, analyzing the effect of method selection on simulated changes to snowpack and streamflow under future climate conditions.
A minor issue I had with the work was the use of VIC-CropSyst versus VIC. It was never made clear the motivation for using the coupled crop model when focusing on the mountain regions of the CRB. The authors did note “While the study primarily focuses on snow dynamics simulated by the VIC component of the model, the full VIC-CropSyst model was used because the streamflow calibration was performed on the coupled model version.” This sounds like maybe the baseline simulations using VIC-CropSyst were already on hand, so the authors used the same coupled setup for the dynamic method for easier comparisons. Additionally, the authors said the model was calibrated for “five soil parameters” despite their focus on snow outcomes. This is perplexing to me.
I would again like to acknowledge the effort the authors put into this manuscript. It is not easy to wrangle this amount of data and write a straightforward, clearly described manuscript. While I don’t believe the paper in its current form is suitable for publication in HESS, I do think its appropriateness could be enhanced with a few modifications:
- There is a lot of work out there that makes large conclusions about hydroclimatology in the western US using VIC and its now demonstrably poor default rain-snow partitioning method. This work has been published in top-tier journals like Nature and Water Resources Research, to name just two. The authors may wish to reframe their paper around this issue. Have our previous VIC simulations misled us about the hydroclimatology of the Pacific Northwest? What does this mean for water resources modeling and hydrologic forecasting? Do we understand the potential impacts of climate change on snow cover evolution and streamflow?
- Reconsider the use of VIC-CropSyst. If the purpose is to compare snow simulations in mountain regions, it is hard to justify the coupled model versus VIC alone. Similarly, calibrate the model’s snow parameters and see what happens.
- Consider other rain-snow partitioning methods. Beating the default VIC method, as demonstrated above, is no challenge given its poor track record. What about spatially variable air temperature thresholds, wet bulb temperature thresholds, etc.?
I thank the authors for their time and wish them luck with their manuscript!
References
Jennings, K.S. and Molotch, N.P. (2019). The sensitivity of modeled snow accumulation and melt to precipitation phase methods across a climatic gradient. Hydrology and Earth System Sciences. https://doi.org/10.5194/hess-23-3765-2019
Jennings, K.S., Arienzo, M.A., Collins, M., Hatchett, B., Nolin, A.W., and Aggett, G.R. (2023). Crowdsourced Data Highlight Precipitation Phase Partitioning Variability in Rain-Snow Transition Zone. Earth and Space Science. https://doi.org/10.1029/2022EA002714
Wang, Z., Vivoni, E. R., Whitney, K. M., Xiao, M., & Mascaro, G. (2024). On the sensitivity of future hydrology in the Colorado River to the selection of the precipitation partitioning method. Water Resources Research, 60(6), e2023WR035801. https://doi.org/10.1029/2023WR035801
Citation: https://doi.org/10.5194/egusphere-2024-2284-RC2 -
AC2: 'Reply on RC2', Bhupinderjeet Singh, 26 Nov 2024
Thank you for the feedback, suggestions for improvements, and for bringing the recent Wang et al. (2024) paper to our attention. Specific responses are provided in the attached document and we have provided clarification related to the novelty of this work in comparison to other related work.
--Bhupinderjeet Singh and co-authors
Status: closed
-
RC1: 'Comment on egusphere-2024-2284', Anonymous Referee #1, 25 Oct 2024
Summary
This manuscript examines the implications on simulated snowpack conditions associated with changing a precipitation partitioning model (partitioning precipitation into rain and snow fractions), from an approach considering only contemporary air temperature to an approach using both contemporary air temperature and humidity. The authors carry out a suite of numerical experiments in the Pacific Northwest region of the United States using the two alternative partitioning schemes and compare the approaches to each other and to observational data at USDA SNOTEL observation sites. Simulations and observations are compared on the basis of peak snow water equivalent (SWE), the timing of peak SWE, snow phenology (that is, the onset, conclusion, and duration of seasonal snow cover), and the fraction of runoff from snow at SNOTEL sites within the Columbia River Basin. Overall, the switch to a bivariate precipitation partitioning approach yields improvements in simulated snow conditions with respect to observations. Where there is degradation, the authors effectively argue that the degradations are likely the influence of factors beyond those that would be influenced by the change in partitioning approach. The paper is of interest to the readership of HESS and makes important, albeit primarily methodological, contributions to snowpack and hydrologic modeling in snow-dominated regions. I believe that it can be published with only minor revisions.
Major Comments:
- I can surmise why VIC-CropSyst is being used in this study, instead of VIC without the CropSyst model coupled. However, some readers might be left wondering why VIC-CropSyst is being used when, for example, value additive aspects of the crop model (e.g., dynamic crop yields, etc.) are not being examined and when the comparisons with SNOTEL observations are primarily in locations where there is little or no cultivation. As such, I would strongly encourage the authors to provide some additional context for using VIC-CropSyst for readers. For example, if the work summarized in the manuscript is part of larger and/or ongoing efforts to develop more sophisticated regional projections of how climate change might affect agriculture in the region, it would be helpful for readers to know.
- From a water balance perspective, the analysis of snowmelt contribution to streamflow is interesting. However, given that there is not really meaningful observational constraint at SNOTEL sites, we are left only with model-to-model comparisons. While the authors are clear that this is the case, it is also possible that they could select a few watersheds within the region and compare simulated SWE/Q to the USDA’s Basin Snow Water Equivalent estimates of SWE, normalized by runoff volume for the same watersheds. This represents some additional analysis, but it might provide a much more insightful comparison of the degree to which the change in precipitation partitioning influences whole-basin estimates of water supply.
Minor Comments:
- Line 84: I would quibble slightly that the Columbia River Basin is entirely snow dominated, given that some significantly large areas within the basin are, in fact, rain dominated. I’d suggest slightly rewording this to “The CRB encompasses multiple states and portions of Canada in North America, and snow comprises a substantial fraction of annual precipitation in much of the watershed, particularly high mountain areas.”
- Some of the figures (Figure 2, Figure 3) have some issues with the frames of the figure areas and seem misaligned with subfigures. I’d encourage the authors to address these minor formatting issues.
- For figure 9, the bin labels on the x-axis are slightly confusing. Is bin 4 really 10-20% bias? Or is it 0-20% bias? If the former, were there no results in the 0-10% bias range? Or is this what is meant by “Only bins with at least 8 stations are displayed.” If so, in the text, please indicate that “Fewer than 8 SNOTEL stations fall within the range of a 0-10% change in bias, therefore we exclude this range in bias change from these box plots.”
- Fir figure 9(d) the label of the plot could be more clear. I believe this is the fraction of precipitation to SWE, so perhaps P/SWE (-) would be clearer.
Citation: https://doi.org/10.5194/egusphere-2024-2284-RC1 -
AC1: 'Reply on RC1', Bhupinderjeet Singh, 26 Nov 2024
Thank you for the feedback and suggestions for improvements. We agree with all the suggestions, including the suggestion for additional analysis related to basin-level peak SWE/Q comparisons. All suggestions can be incorporated in the revised manuscript. Specific details for each comment are provided in the attached document.
--Bhupinderjeet Singh and co-authors
-
RC2: 'Comment on egusphere-2024-2284', Anonymous Referee #2, 31 Oct 2024
Singh et al. present a comprehensive analysis of changes to simulated snow cover evolution resulting from an updated dynamic rain-snow partitioning scheme in the VIC-CropSyst model in the Columbia River Basin (CRB) and other areas of the Pacific Northwest. They showed that a bivariate logistical regression method that predicts precipitation phase as a function of air temperature and relative humidity produces better simulated snow water equivalent (SWE) than the default VIC method that uses air temperature alone. The authors noted that improved snow outcomes led to a higher proportion of simulated streamflow coming from snowmelt versus the baseline case using the default VIC method.
I commend the authors on the hard work they put into this paper. The methods are clearly described, and results are straightforward and easy to follow. However, there are a few major shortcomings that I believe preclude this article from publication in HESS. I detail these below.
The first, and most important, issue is that the authors use the default VIC rain-snow partitioning scheme as the benchmark to which they compare the dynamic method. This default is a dual-threshold method with a lower, all-snow threshold of -0.5°C and an upper, all-rain threshold of 0.5°C, with mixed precipitation falling in between. Previous modeling work at multiple sites in the western US has shown this method to produce highly negative biases and low r2 values in both SWE and snow depth (Jennings and Molotch, 2019). More recent observational work from the Sierra Nevada in the western US further highlighted the poor performance of the default VIC method compared to visual reports of rain, snow, and mixed precipitation. It was the second-worst partitioning method, only correctly predicting rain, snow, and mixed precipitation 47.1% of the time (Jennings et al., 2023). In other words, the VIC method got over half of its precipitation phase predictions wrong.
Thus, the premise of the research—that a modified rain-snow partitioning scheme incorporating humidity and air temperature would improve on the default VIC method—is not a particularly evocative one. In fact, the researchers would be hard pressed to find a worse rain-snow partitioning method than the VIC default. This has the unsatisfying effect of producing results that are practically pre-ordained. In my opinion, this makes the work more appropriate as a case study for a different journal.
The second major shortcoming is the novelty of the work. Wang et al. (2024) recently demonstrated similar findings when implementing the VIC model with a new rain-snow partitioning scheme (wet bulb temperature, TW) and comparing the outcomes to the default dual-threshold method. They showed “improved performance of the TW scheme in simulating snowfall fraction (SF) and snow water equivalent (SWE) in relation to in situ observations and a gridded SWE product.” They also took the research a step further, analyzing the effect of method selection on simulated changes to snowpack and streamflow under future climate conditions.
A minor issue I had with the work was the use of VIC-CropSyst versus VIC. It was never made clear the motivation for using the coupled crop model when focusing on the mountain regions of the CRB. The authors did note “While the study primarily focuses on snow dynamics simulated by the VIC component of the model, the full VIC-CropSyst model was used because the streamflow calibration was performed on the coupled model version.” This sounds like maybe the baseline simulations using VIC-CropSyst were already on hand, so the authors used the same coupled setup for the dynamic method for easier comparisons. Additionally, the authors said the model was calibrated for “five soil parameters” despite their focus on snow outcomes. This is perplexing to me.
I would again like to acknowledge the effort the authors put into this manuscript. It is not easy to wrangle this amount of data and write a straightforward, clearly described manuscript. While I don’t believe the paper in its current form is suitable for publication in HESS, I do think its appropriateness could be enhanced with a few modifications:
- There is a lot of work out there that makes large conclusions about hydroclimatology in the western US using VIC and its now demonstrably poor default rain-snow partitioning method. This work has been published in top-tier journals like Nature and Water Resources Research, to name just two. The authors may wish to reframe their paper around this issue. Have our previous VIC simulations misled us about the hydroclimatology of the Pacific Northwest? What does this mean for water resources modeling and hydrologic forecasting? Do we understand the potential impacts of climate change on snow cover evolution and streamflow?
- Reconsider the use of VIC-CropSyst. If the purpose is to compare snow simulations in mountain regions, it is hard to justify the coupled model versus VIC alone. Similarly, calibrate the model’s snow parameters and see what happens.
- Consider other rain-snow partitioning methods. Beating the default VIC method, as demonstrated above, is no challenge given its poor track record. What about spatially variable air temperature thresholds, wet bulb temperature thresholds, etc.?
I thank the authors for their time and wish them luck with their manuscript!
References
Jennings, K.S. and Molotch, N.P. (2019). The sensitivity of modeled snow accumulation and melt to precipitation phase methods across a climatic gradient. Hydrology and Earth System Sciences. https://doi.org/10.5194/hess-23-3765-2019
Jennings, K.S., Arienzo, M.A., Collins, M., Hatchett, B., Nolin, A.W., and Aggett, G.R. (2023). Crowdsourced Data Highlight Precipitation Phase Partitioning Variability in Rain-Snow Transition Zone. Earth and Space Science. https://doi.org/10.1029/2022EA002714
Wang, Z., Vivoni, E. R., Whitney, K. M., Xiao, M., & Mascaro, G. (2024). On the sensitivity of future hydrology in the Colorado River to the selection of the precipitation partitioning method. Water Resources Research, 60(6), e2023WR035801. https://doi.org/10.1029/2023WR035801
Citation: https://doi.org/10.5194/egusphere-2024-2284-RC2 -
AC2: 'Reply on RC2', Bhupinderjeet Singh, 26 Nov 2024
Thank you for the feedback, suggestions for improvements, and for bringing the recent Wang et al. (2024) paper to our attention. Specific responses are provided in the attached document and we have provided clarification related to the novelty of this work in comparison to other related work.
--Bhupinderjeet Singh and co-authors
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 322 | 124 | 16 | 462 | 29 | 6 | 10 |
- HTML: 322
- PDF: 124
- XML: 16
- Total: 462
- Supplement: 29
- BibTeX: 6
- EndNote: 10
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1