the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Extending the utility of space-borne snow water equivalent observations over vegetated areas with data assimilation
Melissa L. Wrzesien
Sujay V. Kumar
Eunsang Cho
Kristi R. Arsenault
Paul R. Houser
Carrie M. Vuyovich
Abstract. Snow is a vital component of the Earth system. Yet, no snow-focused satellite remote sensing platform currently exists. In this study, we investigate how synthetic observations of snow water equivalent (SWE) representative of a synthetic aperture radar remote sensing platform could improve spatiotemporal estimates of snowpack. We use an Observation System Simulation Experiment, specifically investigating how much snow simulated using popular models and forcing data could be improved by assimilating synthetic observations of SWE. We focus this study across a 24°-by-37° domain in the Western United States and Canada, simulating snow at 250 m resolution and hourly timesteps in water-year 2019. We perform two data assimilation experiments, including: 1) a simulation excluding synthetic observations in forests where canopies obstruct remote sensing retrievals, and 2) a simulation inferring snow distribution in forested grid cells using synthetic observations from nearby canopy-free grid cells. Results found that assimilating synthetic SWE observations improved average SWE biases at peak snowpack timing in shrub, grass, crop, bare-ground, and wetland land cover types from 14 %, to within 1 %. However, forested grid cells contained a disproportionate amount of SWE volume. In forests, SWE mean absolute errors at peak snowpack were 111 mm, and average SWE biases were on the order of 150 %. Here, the data assimilation approach that estimated forest SWE using observations from the nearest canopy-free grid cells substantially improved these SWE biases (18 %) and the SWE mean absolute error (27 mm). Simulations employing data assimilation also improved estimates of the temporal evolution of both SWE and runoff, even in spring snowmelt periods when melting snow and high snow liquid water content prevented synthetic SWE retrievals. In fact, in the Upper Colorado River basin, melt-season SWE biases were improved from 63 % to within 1 %, and the Nash Sutcliffe Efficiency of runoff improved from –2.59 to 0.22. These results demonstrate the value of data assimilation and a snow-focused globally relevant remote sensing platform for improving the characterization of SWE and associated water availability.
- Preprint
(9299 KB) - Metadata XML
-
Supplement
(5071 KB) - BibTeX
- EndNote
Justin M. Pflug et al.
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2023-1603', Anonymous Referee #1, 28 Aug 2023
The manuscript presents results from an observing system simulation experiment investigating the extent to which synthetic SWE observations from a future SAR mission can improve SWE simulations in the western US. Prediction of SWE in the western snowpack remains a challenge, and from this angel, the contributions of the manuscript is novel and have potential to inform future measurements. The manuscript is very well written and I do not have any major comments or concerns. However, I do like the authors to consider following comments before the final publication.
1) The descriptions of "Open loop", "Nature Run", "Assimilation w/o Forest" and assimilation with forest are scattered and hard to follow. Readers not familiar with the concept of open loop and nature run will struggle and it will be helpful to provide a summary. I am unclear on what biases and uncertainties were considered in the open loop simulations.
2) How did you come up with the 750 m search radius for the forested and non-forested pairing?
3) Considering the fact that SWE magnitude and processes are highly variable among different years, how transferable are results from 2019 to a dry year when the role of snowpack is even greater for managing water? This point warrants a discussion.
4) You are using snow water equivalent and SWE interchangeably. use the abbreviation after you have defined it.
5) L58 missing reference
6) Fig. 1: Red contours are not really basins. They are boundaries of water resources regions 14 (upper Colorado), 16 (Great Basin), 17 (PNW), and 18 (CA).
7) L143: use of Western United States vs Western U.S.
8) Is it fair to use the same peak snow day, i.e., March 13 2009, for the entire study region? How variable this peak date was among the four regions?
Thank you and good luck!
Citation: https://doi.org/10.5194/egusphere-2023-1603-RC1 -
RC2: 'Comment on egusphere-2023-1603', Anonymous Referee #2, 01 Sep 2023
This study investigated a method to improve snow water equivalent (SWE) observations from space with a focus over the western US. A modeling approach was used to create synthetic SWE observations and then assimilate those into different modeling setups. One modeling setup provided a method to better simulate SWE in forested areas which make up a big portion of snow covered areas in the west. Results are promising and show large improvements using the runs with data assimilation and further improvements using the forest strategy.
Overall, the paper is well written, and results are clearly presented. I do have some suggestions to help improve the manuscript before it can be published.
- The way it is written it appears the “nature run” is used as the ground truth. At the large spatial extent of the Western US this seems appropriate. However, it is unknown how the “nature run” compares to ground based SWE observations. It would be helpful to provide some comparisons to observed SWE from SNOTEL stations in different regions to show that the whole modeling framework is actually representative of actual “ground truth”.
- Can you comment on why on a single year (2019) was used? I realize the modeling is likely a big lift computationally but snowpack characteristics can change an awful lot from year to year. You picked a big snowpack year but what about a shallow snowpack year?
- The introduction mentions a “future snow mission”. Can you clarify if this is an already planned mission or more of a hypothetical mission?
- In the Short Summary it says “250 m estimates of snow”. Can you be more specific and say snow water equivalent. It also states “snow water volume to within 4%”. Can you add what that is in reference to? Ground truth?
- Line 58: It looks like “Ref” needs to be filled in.
- Line 131: What does “SW” mean? Please define.
- Line 245: Is the word “snowy” out of place here?
Citation: https://doi.org/10.5194/egusphere-2023-1603-RC2 -
RC3: 'Comment on egusphere-2023-1603', Anonymous Referee #3, 08 Sep 2023
General comments
This manuscript describes an observing systems simulation experiment (OSSE) designed to determine the impact of satellite radar derived snow water equivalent (SWE) retrievals at moderate spatial resolution (250m). The major innovation is the focus on forested regions, and the determination of how SWE retrievals in open areas (which are viable with a volume-scattering based radar approach) can be used to improve SWE estimates in adjacent forested areas (for which volume scattering based radar retrievals are not viable). The large model domain also sets this study apart from most other recent snow-focused OSSE’s in the literature.
I have a number of questions below, focused on the experimental setup. In short, some additional details on: the radar mission configuration, sensitivity of the results to the 20% uncertainty value prescribed to the synthetic SWE data, and some assurances about the independence of the results from the nature run itself are needed before the manuscript is suitable for publication.
1. Section 2.2:
It would be helpful to have some further details on the exact mission configuration, or range of configurations that motivated this study would be helpful. There is only a parenthetical statement on line 83 that the focus is a volume-scattering based approach using X- and Ku-band measurements. Were a number of mission and orbital configurations explored? It’s not clear what swath width, or range of swath widths were applied and it is only stated that a 10 to 14 day revisit time was used.
2. Line 166: “Additionally, based on an error level of 20%, spatially and temporally uncorrelated random errors drawn from a Gaussian distribution were added to the synthetic SWE retrievals.”
Does this mean the 20% random error was applied directly to the nature run SWE values? While 20% isn’t overly conservative, there are systematic error considerations based on the frequency of radar measurement, and radar geometry considerations that result from the configuration of the mission. For instance, Ku-band retrievals will be biased low in deep snow areas while X-band will be more insensitive to shallow snow. Measurements near swath edges may have an incidence angle that results in systematic errors. Some additional details on how this 20% number was determined, whether you explored the sensitivity of the results to this value, and the potential impact of not considering more mission-specific systematic errors would be helpful.
3. Line 211 - 219: the approach to filling in forested areas with information from adjacent non-forested grid cells is clearly described and nicely illustrated in Figure 2. The text and image generally gives the impression that gridded SAR backscatter will either be from a clearing or a forest. But vegetation cover relevant to radar backscatter is not a binary influence. What thresholds were used to differentiate forest from clearing? How would mixed grid cells be treated? What about the influence of non-forest vegetation like shrubs within the snowpack?
4. Line 230: “when snowmelt is minimized and synthetic observations are masked by grid cells with liquid water content to the smallest degree”
This text suggests that wet snow grid cells were not assimilated, similar to how forest cells were masked. If this is the case, a description of how wet snow was treated needs to be added to Section 2.3 or 2.4.
5. Line 264: “This was driven by the expansive snow extents of the open loop simulation…”.
A feature of the open loop simulation is the smoother spatial pattern and clear lack of elevational influence on SWE. The high SWE areas are very smooth, unlike the nature run. In trying to understand this, I went back to Section 2, but could not find a clear description of the open loop simulation. Based on line 134, it is stated that Noah-MP (without SnowModel) was used for the open loop simulation but some further insight into the underlying differences in Figures 3 and 4 would be helpful.
6. Line 290: “In forested grid cells, SWE simulated by the open loop simulation were biased high by approximately 87 mm (+150%) on average (Fig. 6), with a mean absolute error of 111 mm (Table 2). These errors were propagated into the simulation with data assimilation without the forest strategy. Fortunately, the ratio between modeled SWE and synthetic SWE observations in forested grid cells and the nearest canopy-free grid cells had high levels of similarity. Therefore, estimating snow in forest regions using the nearest canopy-free pixels (Fig. 2) improved snow simulations significantly.”
I’m struggling a bit here to ensure that there is no impact based simply on the structure of the experiment. Forest SWE in the open loop simulation is too high. Assimilating the synthetic SWE retrievals lowers the forest SWE values so they agree better with the nature run, indicating a positive influence that the radar SWE product would deliver. But the synthetic radar data were generated from the nature run, perturbed with 20% random uncertainty. Because the nature run itself is used as the generator of the synthetic data and the reference, is this a case of the synthetic data adjusting the open loop back to itself? In this sense, the finding that the synthetic radar-retrieved SWE data is nudging the open loop back to the nature run does not prove the positive impact of the radar product but rather is just a mathematical adjustment achieved by using the nature run itself to ultimately adjust the open loop simulation. Presumably, prescribing a higher error value and considering radar-specific uncertainties on top of the 20% random error would result in less of an improvement. Conversely, applying a lower error value would result in more of an improvement. I think this issue of error characterization (see also comment #2 above), and independence of the results from the quality of the nature run itself needs to be addressed. Presumably the reduction in the forest SWE positive bias achieved via assimilation is what drives the reduction in positive bias in the streamflow estimate as well (line 346)?
7. Line 314: “mean SWE evolution tracked the nature run simulation significantly better than the open loop simulation in the spring snowmelt period.”
This is an interesting result. Essentially it shows that if you get peak SWE right, you can track SWE during the melt period even in the absence of many radar measurements (you had only 5% usable data during the melt period as stated on line 311). But I wonder how dependent this result is on the dynamics of the melt period. Presumably if there are additional snowfall events after peak SWE this may not be the case? Were the results in Figure 7 replicable for the other watersheds? And I wonder how replicable this result is from year to year (you would need to speculate here since this is outside the scope of this study)?
8. The two major limitations of volume scattering based radar approaches are forest and wet snow. The focus is on forest in this study, but synthetic SWE under wet snow conditions was also masked. Can a similar strategy as was developed to fill in forest areas be used to address the wet snow challenge? This is out of scope to do this in this study, but perhaps something for the discussion?
Editorial comments
Line 19: ‘popular’ seems like an odd word choice to me…how about ‘widely-used’?
Line 58: add the references for the limitations of passive microwave SWE datasets.
Line 60: I would specify that the Lievens et al study uses C-band SAR.
Line 63: “To overcome these limitations, modeling and data assimilation systems are needed that can extend the coverage and utility of available measurements to areas, times, and variables that are not directly observed.” Well said!
Lines 135-140: can some additional references be added for Noah-MP and TOPMODEL? I think Niu and Yang (2004) focuses only on the snow processes…
Figure 2: I find the dashed circle to be distracting and unnecessary. The gray shading shows the swath; the dashed line grid illustrates the SAR data. I don’t know what the circle means…
Citation: https://doi.org/10.5194/egusphere-2023-1603-RC3
Justin M. Pflug et al.
Data sets
Pflug et al. (2023) ‒ Model configuration and outputs Justin M. Pflug https://www.hydroshare.org/resource/e0ad80f818bf4062a335e9e0d7362834/
Justin M. Pflug et al.
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 163 | 83 | 13 | 259 | 24 | 5 | 5 |
- HTML: 163
- PDF: 83
- XML: 13
- Total: 259
- Supplement: 24
- BibTeX: 5
- EndNote: 5
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1