Integrating remotely sensed surface water dynamics into hydrologic signature modelling
Abstract. Extreme flow conditions in rivers have far-reaching environmental and economic consequences. The retention of surface water in lakes, wetlands, and floodplains can potentially modify the timing, duration, and magnitude of flow. However, efforts to explore the impact of surface water storage on discharge regimes have been limited in geographic extent. In this analysis, we calculated six hydrologic signatures, reflecting flashiness and high and low flow conditions, at 72 gaged watersheds across the conterminous United States. In addition to traditionally considered variables representing climate, land cover, topography, and soil, we incorporated a novel remote sensing (Sentinel-1 & 2) approach to study the contribution of surface water storage dynamics when modelling spatial variability in hydrologic signatures using random forest models. While climate variables explained much of the variability in the hydrologic signatures, models for five of the six signatures showed some degree of improvement in model performance when landscape characteristics were added with adjusted R2 improving 1.75 to 11.69 % and Akaike information criterions improving 0.24 % to 6.69 %. Automated variable selection can be indicative of the relative importance of certain variables over others. Using a forward selection process, five of the six signature models selected remotely sensed inundation variables with all five variables showing a significant (p<0.01) contribution to the respective model. More semi-permanent and permanent inundation within the floodplain (i.e., lakes along rivers), for example, was associated with lower wet season and annual flashiness. Further, greater seasonal floodplain inundation extent was associated with increases in peak flows, so that floodplain water storage was relevant to both flashiness and high flow signatures. Additionally, spatial variability in the amount of semi-permanent and permanent non-floodplain water significantly contributed to explaining spatial variability in the baseflow index. These findings suggest that surface water storage dynamics may help explain variability in streamflow signatures. Watershed management will benefit from an improved understanding of how surface water storage influences stream behaviour.
- Preprint
(1925 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 14 Feb 2025)
-
CC1: 'Comment on hess-2024-298', Nima Zafarmomen, 09 Nov 2024
reply
1) While the analysis touches on correlations between hydrological signatures over different periods (8 years versus 24 years), a more detailed uncertainty analysis on how sensitive the models are to temporal data selection could clarify the robustness of the findings.
2) The chosen hydrologic signatures seem well-grounded in the literature, but do they sufficiently capture all important dynamics relevant to flood and drought responses in varied watersheds? Exploring additional signatures or more dynamic metrics could provide a broader picture.
3) The paper uses random forest models, which are known for their predictive power but often lack interpretability. How do the authors address potential overfitting or the challenge of interpreting complex interactions among variables in these models? A comparison to simpler models might illustrate if and when the additional complexity of random forests yields meaningful benefits.
4) While the integration of Sentinel-1 and Sentinel-2 data for analyzing surface water dynamics represents an innovative advancement, the discussion would benefit from a more comprehensive comparison with earlier methodologies, particularly those leveraging Landsat and other datasets. Sentinel-1 and Sentinel-2 sensors bring distinct advantages, such as higher revisit frequencies and improved spatial resolution, which enable more accurate and consistent monitoring of water dynamics. However, a broader contextualization is needed to illustrate how these advantages surpass or complement previous approaches.
Notably, some studies that focus on similar data assimilation methods in related fields were not addressed, such as the "assimilation of Sentinel-based leaf area index for surface-groundwater interaction modeling in irrigation districts", "Analysis of surface water resources using Sentinel-2 imagery".
Citation: https://doi.org/10.5194/hess-2024-298-CC1 -
RC1: 'Comment on hess-2024-298', Vagner Ferreira, 30 Jan 2025
reply
General Comments:
The proposed paper by Vanderhoof et al. makes a valuable contribution to the field of hydrology and is highly appropriate for the Hydrology and Earth System Sciences (HESSD). It presents an interesting and novel approach to incorporating surface water dynamics into hydrologic signature analysis across multiple watersheds. Overall, the study is well-structured and makes a valuable contribution to understanding how surface water storage influences streamflow characteristics. However, some points need to be addressed before its publication in HESSD, and the comments provided below are intended to help the authors improve the paper even further.
Specific Comments:
1. Â Â Line 24: The improvement figures cited (e.g., 1.75% to 11.69% for adjusted R-sqaured are relatively modest. How meaningful are these improvements from a hydrologic modeling perspective? The authors should clarify whether these incremental gains justify the computational and methodological complexity of integrating remote sensing data.
2. Â Â Lines 26-28: The claim that remotely sensed inundation variables show a significant contribution (p<0.01) is important and a quantification of the practical hydrological implications of these findings could be addressed in the abstract. Consider adding a sentence stating how these findings could affect real-world watershed management decisions.
3.   Graphical Abstract: It seems too busy and is not meaningful at all. For example, it doesn’t illustrate the contribution of remote sensing data to model improvements Please, consider revising it.Â
4. Â Â Line 110: Are these watersheds representative of all streamflow regimes across the CONUS? Discuss this taking into account critical spatial and temporal representativeness issues.
5. Â Â Consider adding a conceptual figure showing the hypothesized relationships between surface water storage and different hydrologic signatures.
6. Â Â Lines 130-131: That is appropriate, however tidal and dammed systems are key contributors to surface water storage variability. So, how does their omission constrain the generalizability of findings?
7. Â Â There is no discussion of Sentinel-1/2 synergies. For example, how were discrepancies between SAR (Sentinel-1) and optical (Sentinel-2) water classifications resolved? Also, consider adding more details on how temporal or spatial mismatches were addressed.
8. Â Â No discussion of Sentinel-2 cloud cover gaps. How many watersheds had >20% missing optical data?
9. Â Â Line 250: The 14-day surface water composite lacks justification since this temporal resolution impacts the detection of short-term inundation events (e.g., flash floods).
10. Â Â Sub-section 2.5: The random forest modeling approach is appropriate, but the justification for choosing 72 watersheds (streamflow gages) seems arbitrary. Was this purely due to computational limitations as mentioned? Consider addressing if this sample size is statistically sufficient.
11.   Random forest variable importance analysis doesn’t account for SAR speckle noise or optical band correlations.
12. Â Â Flashiness Signatures: The decrease in flashiness associated with semi-permanent and permanent floodplain inundation is an expected finding. However, the discussion fails to address whether this relationship would hold under different climatic conditions (e.g., arid versus humid regions). A regional stratification of results would add robustness to these claims.
13. Â Â Peak Flow Signature: the model improvement metrics for MAX30/area and (Q10-Q95)/area are still relatively low (e.g., adjusted R-squared increase of only 5.44% for MAX30/area). What additional information might explain this lack of improvement despite integrating inundation data?
14. Â Â Lines 480-482: The discussion of model performance improvements notes they were "moderate at best" - consider providing context for what would constitute a more relevant improvement in such context.
15. Â Â Lines 501-505: The conclusion that dynamic inundation metrics are more informative than static datasets (e.g., NWI or floodplain maps) is convincing. However, the text should also briefly address the computational trade-offs of using remote sensing datasets.
16. Â Â Lines 522-524: The statement about process-based hydrologic models being limited to single watersheds needs stronger citation support.
17. Â Â Lines 524-526: The ability to explore spatial variability across multiple watersheds is valid, but the discussion does not sufficiently address the implications of spatial autocorrelation or site clustering. For example, did the Moran's I test adequately resolve these issues?
18. Â Â Sub-section 4.2: The uncertainty discussion is comprehensive but it lacks quantitative estimates of uncertainty where possible. Furthermore, this section presents various sources of uncertainty in a somewhat scattered manner. Consider reorganizing into clear categories: data uncertainty, methodological uncertainty, and model uncertainty and a including a table summarizing the major sources of uncertainty and their relative importance would be useful.
19. Â Â The regional analysis could be strengthened by examining if the relationships between surface water storage and signatures vary systematically by region
20. Â Â Sub-section 4.3: The management implications section could be expanded with more specific recommendations.
21. Â Â Conclusion: It focuses on the utility of novel inundation variables but does not discuss wide implications for advancing satellite-based hydrologic monitoring. For instance, how might this approach scale to global watersheds or align with state-of-art sensors (e.g., SWOT)?Citation: https://doi.org/10.5194/hess-2024-298-RC1 -
RC2: 'Comment on hess-2024-298', Anonymous Referee #2, 30 Jan 2025
reply
This paper calculates a suite of hydrologic signatures for 72 US catchments and assesses the relative ability of climate alone compared to other catchment characteristics (land cover, geology, topographic, and surface water storage input variables) to explain the variability in hydrologic signatures. Uniquely, it includes catchment characteristics describing surface water extent from remotely sensed products. It finds that surface water extent can help to explain the variability for certain hydrologic signatures.
Overall the paper makes a valuable contribution to understanding the controls on flow characteristics across large-samples of catchments and what types of surface water characteristics are useful to explain variability in hydrologic signatures. I list several comments below that I hope the authors find useful in revising the paper.
Comments
L82-83. I agree that analyses of ‘hydrologic signatures have rarely included or considered surface water storage capacity’. However it would be good to summarise the few studies that have done this to give the reader a sense of what has already been done.
L100-107. I don’t think this paragraph is that relevant for a data analysis study. I would recommend removing it.
L114. It is not clear to me what is novel about the remotely sensed surface water extent? Their inclusion in the large-sample analysis to understand hydrologic signatures is novel, but I don't think the metrics themselves are novel? Can you clarify?
L131-133. It is interesting that a study interested in surface water storage excludes dams – it would be useful to understand the justification for this.
Section 2.4.2. It would be good to add some justification on why these additional variables were chosen i.e. why these particular catchment characteristics?
L275-282. Again, it would be good to add some justification on why these particular metrics of surface water extent were chosen.
L316. It wasn’t clear to me whether any predictor variables were removed as part of the random forest and therefore not included in the results? It would be helpful to include the correlations between predictor variables in the supplementary information.
L348-349. Why were both these signatures included if they are so highly correlated? What additional information are they providing?
L542. 72 catchments is a relatively small sample for a random forest model. It would be good to comment on the robustness of the results given the relatively small sample size (for a large sample study!)
Citation: https://doi.org/10.5194/hess-2024-298-RC2
Data sets
Data release for integrating remotely sensed surface water dynamics in hydrologic signature modelling M. K. Vanderhoof, P. Nieuwlandt, H. E. Golden, C. R. Lane, J. R. Christensen, W. Keenan, and W. Dolan https://www.sciencebase.gov/catalog/item/652027f4d34e44db0e2e43b4
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
255 | 52 | 13 | 320 | 9 | 8 |
- HTML: 255
- PDF: 52
- XML: 13
- Total: 320
- BibTeX: 9
- EndNote: 8
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1