the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 14 Mar 2023
Uncertainty estimation for a new exponential filter-based long-term root-zone soil moisture dataset from C3S surface observations
Abstract. Soil moisture is a key variable in monitoring climate and an important component of the hydrological, carbon, and energy cycles. Satellite products ameliorate the sparsity of field measurements but are inherently limited to observing the near-surface layer, while water available in the unobserved root zone controls critical processes like plant water uptake and evapotranspiration. A variety of approaches exists for modelling root-zone soil moisture (RZSM), including approximating it from surface layer observations. While the number of available RZSM datasets is growing, they usually do not contain estimates of their uncertainty. In this paper we derive a long-term RZSM dataset (2002–2020) from the Copernicus Climate Change Service (C3S) surface soil moisture (SSM) COMBINED product via the exponential filter (EF) method. We identify the optimal value of the method’s model parameter T , which controls the level of smoothing and delaying applied to the surface observations, by maximizing the correlation of RZSM estimates with field measurements from the International Soil Moisture Network (ISMN). Optimized T-parameter values were calculated for four soil depth layers (0–10 cm, 10–40 cm, 40–100 cm, and 100–200 cm) and used to calculate a global RZSM dataset. The quality of this dataset is then globally evaluated against RZSM estimates of the ERA5-Land reanalysis. Results of the product comparison show satisfactory skill in all four layers with median Pearson correlation ranging from 0.54 in the topmost to 0.28 in the deepest soil layer. Temporally-dynamic product uncertainties for each of the RZSM product layers are estimated by applying standard uncertainty propagation to SSM input data and by estimating structural uncertainties of the EF method from ISMN ground reference measurements taken at the surface and in varying depths. Uncertainty estimates were found to exhibit both realistic absolute magnitudes as well as temporal variations. The product described here is, to our best knowledge, the first global, long-term, uncertainty-characterized, and purely observation-based product for RZSM estimates up to 2 m depth.
-
Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
-
Preprint
(1909 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(1909 KB) - Metadata XML
- BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-47', Anonymous Referee #1, 11 Apr 2023
The MS titled, 'Uncertanity estimation for a new exponential filter-based long-term root-zone soil moisture dataset from C3S surface observations' by Pasik et al. describes a methodology to deduce rootzone SM using only satellite derived surface soil moisture using a well-known exponential filter approach with its error characterization. In general, I found this study lacking in novetly factor. The exponential model has been around for a while now and is shown to have limited success in estimating SM at deeper layers (>40cm) as also shown in this study. Furthermore, there are some glaring gaps in background infromation, method descriptions etc. Therefore, despite tackling a critical issue in sub-surface hydrology, I would recommend rejection in its current form. However, I would be more than happy to see it re-submitted with significant revisions.
Major comments:
- As mentioned earlier, the exponential filters have been studied extensively (as also acknowledged in the MS) in the past. Also has well docmented issues such as poor performance in the deeper layers, struggles with widely different soil types between surface and lower layers (no mention or discussion around this in the MS), summer season decoupling etc. This study acknowledges and re-affirms most of these issues but presents no path forward in trying to solve them. Its does not seem to be taking the scinece forward either by reducing the model known limitations, or sheding new lights on model performance with its global implementation or discussion. Therefore, in my opition, in the current form, this study lack the novelty factor and may need significant revisions.
- I agree with the authors that their product could be the covering the longest period of record for observation-based RZSM, but shoud have at least acknowledged other global products such as SMAP and SMOS L4 products in the MS. In fact, SMOS L4 product is similar to this study, they are also using EF (https://sextant.ifremer.fr/record/316e77af-cb72-4312-96a3-3011cc5068d4/) whereas SMAP L4 uses data assimialtion approach to merge SMAP with the Catchment Land surface Model with detailed uncertanity analysis and published ATBDs.
- Furthertmore, I think the MS could be improved further with some more background information on other rootzone SM estimation techniques, currently the MS does not talk about other methods and why EF might be better than others, in that context an best average correlation of 0.56 doesnt not inspire too much confidence. The authors simply skipped over some of the detailes of satellite SM estiamtions like difference in bands (X&C for AMSR-E vs L for SMAP and SMAOS etc.), I think it would improve MS.
- If I understand correctly, most of the in-situ sites are in open fields (usually near agricultural land). Therefore, Topt obtained may only be able to represent (presumabely) a particular landcover type. Is there any analysis authors have performed to assess the model perfromance at other locations? While reading the MS, I could not figure out if the model was implemented at gridded scale or only at the ISMN sites. Perhaps, this could be made more clearer.
- Finally, I think the discussoin could be further improved (especially if study is simply focused in implementing at larger scale with its limitations intact) by talking about if there is any regional pattern in model performance (arid vs humid conditions); tropical vs sub-tropical region? Does soil types play any role in Topt and uncertanity estimations? What is the dominent landcover type and can these different rooting systems (barren soils vs cropland vs deep rooted trees) explain some of the issues beign faced?
Specific Comments:
- Figure 1, there seems to be huge overlaps between Topt for different layers (25th-75th percentile box), how would this impact the results? Have the authors consider perhaps running the model woth those as upper and lowes limits on Topt to see the impact on performance?
- Typically EF is implemented in a normaliezed SM scale (SWI either by scaling from 0-1 using min/max or using soil characteristic properties). In the MS, it's not mentioend which expecific method was used (if at all).
- Lines 45-48, could use more detailes on existing methods for rootzzone SM estiamtions and their challenges.
- Line 69, EF typically is used to estimate SM at specific layer depth not a composites like 0-10 or 0-40. It's either 0 or 10 or 40 cm (± few cms)
- Section 2.1, I would suggest to include a table showsing the timeline of various satellite SM products being part of C3S with band information that would help understanding the dataset better. Also, I would liked to see some examples of mentioned structural breaks either as timeseries.
- Figure 3, I dont think there is any need to compare all the layers at each depth. For instance, deeper layers should not be compared with 0-7 cm modeled SM. They are not the same thing to be comapared and does not add any value to the MS. Similarly, at 100-200 cm depth, surface SM correlations and their discussions (Lines 241-250) could be avoided.
- Line 259, the structural uncertanity mentioend here is it same as the structural breaks (Line 87)? if not, perhaps use some other terminology to differentiate is further.
- Line 268, please add exact location of the data.
- Figure 6, very busy plot. Hard to read.
- Figure 7, how is the sample size in the subset bigger than the whole dataset (Fig d vs h).
Citation: https://doi.org/10.5194/egusphere-2023-47-RC1 - AC1: 'Reply on RC1', Adam Pasik, 07 Jun 2023
-
RC2: 'Comment on egusphere-2023-47', Anonymous Referee #2, 01 May 2023
The manuscript “Uncertainty estimation for a new exponential filter-based long-term root-zone soil moisture dataset from C3S surface observations” by A. Pasik et al. describes the development of a root-zone soil moisture (RZSM) dataset based on the C3S near-surface soil moisture using the exponential filter method. The newly derived product contains estimates of root-zone soil moisture for different depths, but also their uncertainty estimates considering several sources of uncertainty and their propagation in time. The data product and the code to generate this dataset are available online. I enjoyed reading this well written and clear paper. Please read below my comments which hopefully will help to further improve the manuscript.
Major comments:
1. The root-zone soil moisture is defined here as the water present in the top meter of the soil column (p2, lines 38-39). However, I would argue that the root-zone soil moisture represents the water in the subsurface which is accessible to the roots of the vegetation for transpiration. The depth thereof is highly variable and may depend on climatic (de Boer-Euser et al., 2016; Wang-Erlandsson et al., 2016) and topographic indicators (Fan et al., 2017). Now the derived product provides estimates of soil moisture at different depth intervals, but not really an integrated root-zone soil moisture estimation, which depends on the rooting depth. Would such an addition in the dataset be feasible?
2. When comparing the newly derived product with ERA5L and local data, it would be interesting to see plots of timeseries at specific locations and a spatial map showing where both products have high or low correlations. This would give potential users of the dataset some guidance on where the product may and may not be used. Perhaps the authors can work out some correlations to relate skill with catchment characteristics including climate, land use, topography and soil types.
Minor comments:
3. line 64: I would specify here: “does not consider the model structural error of the EF method”. Although this becomes clear later in the paper, it was not directly clear to me at this stage.
4. line 85: “uncertainties [..] were then calculated from the law of propagation of uncertainties”. Could you explain how this was done in more detail?
5. section 3.2.3: could you explain the presented formulas in more detail? What are the units and what are all parameters? E.g. what does G represent? What does delta represent?
6. line 246: could you quantify with numbers in the text how substantial the difference is between the correlation between E5L and RZSM versus E5L and SSM?
7. Figure 5: in the legend I read “GCOS required uncertainty” but it is not entirely clear to me what this threshold refers to exactly, could you please elaborate?
8. line 280: could you elaborate more on why we expect uncertainties to be amplified during transitions between wet and dry conditions? Which processes play a role which are not well represented in the EF method? Now you briefly refer to Fig6, but it does not provide a clear explanation on why this is expected.
9. line 292 “and highlighting 20% of data with the highest uncertainty”. At this stage, this reads a bit confusing as the previous paragraph describes masking out data with the highest uncertainty and here (if I understood correctly) you are instead plotting data with high uncertainty. Perhaps good to clarify what you mean with “highlighting 20% of data with the highest uncertainty”.
10. line 299: the described difference in uncertainty from 0.008 m3/m3 to 0.004 m3/m3 is very hard to see in the Figure using the applied scale.
11. line 302: why are the uncertainties related to structural breaks not clearly seen in the MAD Topt approach. Can you reflect (here or later in the discussion) a little bit more on this result. It seems to me that the change in uncertainty related to a change in sensor is an important change that you would also want to see in the improved methodology for uncertainty estimation.
12. line 314-320: here, it is not clear to me why using Topt/10 as T parameter uncertainty yields more realistic estimates of temporal uncertainty variations than using MAD Topt in the case of using the time series which includes a structural break (and the opposite in case a shorter time series is used). Which aspects in figure 7 suggest these findings?
13. line 320-324: Again, could you elaborate why the uncertainty estimations of temporal uncertainty variations are no longer accurate for deeper layers?
14. line 351: here, you forgot to add the units of the mentioned uncertainties.
15. In addition, I also downloaded the netcdf files and checked the github page. The nc files contain all the necessary meta data. However, the github page does not include extensive documentation on how to use the different methods within the package. Would it be feasible to elaborate on this further?References:
de Boer-Euser, T., McMillan, H. K., Hrachowitz, M., Winsemius, H. C., & Savenije, H. H. G. (2016). Influence of soil and climate on root zone storage capacity. Water Resources Research, 52, 2009–2024. https://doi.org/10.1002/2015WR018115
Fan, Y., Miguez-Macho, G., Jobbágy, E.G., Jackson, R.B., & Otero-Casal, C. (2017). Hydrologic regulation of plant rooting depth. Proceedings of the National Academy of Sciences, 114(40), 10572-10577.
Wang-Erlandsson, L., Bastiaanssen, W. G. M., Gao, H., Jägermeyr, J., Senay, G. B., Van Dijk, A. I. J. M., et al. (2016). Global root zone storage capacity from satellite-based evaporation. Hydrology and Earth System Sciences, 20(4), 1459–1481. https://doi.org/10.5194/hess-20-1459-2016
Citation: https://doi.org/10.5194/egusphere-2023-47-RC2 - AC2: 'Reply on RC2', Adam Pasik, 07 Jun 2023
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-47', Anonymous Referee #1, 11 Apr 2023
The MS titled, 'Uncertanity estimation for a new exponential filter-based long-term root-zone soil moisture dataset from C3S surface observations' by Pasik et al. describes a methodology to deduce rootzone SM using only satellite derived surface soil moisture using a well-known exponential filter approach with its error characterization. In general, I found this study lacking in novetly factor. The exponential model has been around for a while now and is shown to have limited success in estimating SM at deeper layers (>40cm) as also shown in this study. Furthermore, there are some glaring gaps in background infromation, method descriptions etc. Therefore, despite tackling a critical issue in sub-surface hydrology, I would recommend rejection in its current form. However, I would be more than happy to see it re-submitted with significant revisions.
Major comments:
- As mentioned earlier, the exponential filters have been studied extensively (as also acknowledged in the MS) in the past. Also has well docmented issues such as poor performance in the deeper layers, struggles with widely different soil types between surface and lower layers (no mention or discussion around this in the MS), summer season decoupling etc. This study acknowledges and re-affirms most of these issues but presents no path forward in trying to solve them. Its does not seem to be taking the scinece forward either by reducing the model known limitations, or sheding new lights on model performance with its global implementation or discussion. Therefore, in my opition, in the current form, this study lack the novelty factor and may need significant revisions.
- I agree with the authors that their product could be the covering the longest period of record for observation-based RZSM, but shoud have at least acknowledged other global products such as SMAP and SMOS L4 products in the MS. In fact, SMOS L4 product is similar to this study, they are also using EF (https://sextant.ifremer.fr/record/316e77af-cb72-4312-96a3-3011cc5068d4/) whereas SMAP L4 uses data assimialtion approach to merge SMAP with the Catchment Land surface Model with detailed uncertanity analysis and published ATBDs.
- Furthertmore, I think the MS could be improved further with some more background information on other rootzone SM estimation techniques, currently the MS does not talk about other methods and why EF might be better than others, in that context an best average correlation of 0.56 doesnt not inspire too much confidence. The authors simply skipped over some of the detailes of satellite SM estiamtions like difference in bands (X&C for AMSR-E vs L for SMAP and SMAOS etc.), I think it would improve MS.
- If I understand correctly, most of the in-situ sites are in open fields (usually near agricultural land). Therefore, Topt obtained may only be able to represent (presumabely) a particular landcover type. Is there any analysis authors have performed to assess the model perfromance at other locations? While reading the MS, I could not figure out if the model was implemented at gridded scale or only at the ISMN sites. Perhaps, this could be made more clearer.
- Finally, I think the discussoin could be further improved (especially if study is simply focused in implementing at larger scale with its limitations intact) by talking about if there is any regional pattern in model performance (arid vs humid conditions); tropical vs sub-tropical region? Does soil types play any role in Topt and uncertanity estimations? What is the dominent landcover type and can these different rooting systems (barren soils vs cropland vs deep rooted trees) explain some of the issues beign faced?
Specific Comments:
- Figure 1, there seems to be huge overlaps between Topt for different layers (25th-75th percentile box), how would this impact the results? Have the authors consider perhaps running the model woth those as upper and lowes limits on Topt to see the impact on performance?
- Typically EF is implemented in a normaliezed SM scale (SWI either by scaling from 0-1 using min/max or using soil characteristic properties). In the MS, it's not mentioend which expecific method was used (if at all).
- Lines 45-48, could use more detailes on existing methods for rootzzone SM estiamtions and their challenges.
- Line 69, EF typically is used to estimate SM at specific layer depth not a composites like 0-10 or 0-40. It's either 0 or 10 or 40 cm (± few cms)
- Section 2.1, I would suggest to include a table showsing the timeline of various satellite SM products being part of C3S with band information that would help understanding the dataset better. Also, I would liked to see some examples of mentioned structural breaks either as timeseries.
- Figure 3, I dont think there is any need to compare all the layers at each depth. For instance, deeper layers should not be compared with 0-7 cm modeled SM. They are not the same thing to be comapared and does not add any value to the MS. Similarly, at 100-200 cm depth, surface SM correlations and their discussions (Lines 241-250) could be avoided.
- Line 259, the structural uncertanity mentioend here is it same as the structural breaks (Line 87)? if not, perhaps use some other terminology to differentiate is further.
- Line 268, please add exact location of the data.
- Figure 6, very busy plot. Hard to read.
- Figure 7, how is the sample size in the subset bigger than the whole dataset (Fig d vs h).
Citation: https://doi.org/10.5194/egusphere-2023-47-RC1 - AC1: 'Reply on RC1', Adam Pasik, 07 Jun 2023
-
RC2: 'Comment on egusphere-2023-47', Anonymous Referee #2, 01 May 2023
The manuscript “Uncertainty estimation for a new exponential filter-based long-term root-zone soil moisture dataset from C3S surface observations” by A. Pasik et al. describes the development of a root-zone soil moisture (RZSM) dataset based on the C3S near-surface soil moisture using the exponential filter method. The newly derived product contains estimates of root-zone soil moisture for different depths, but also their uncertainty estimates considering several sources of uncertainty and their propagation in time. The data product and the code to generate this dataset are available online. I enjoyed reading this well written and clear paper. Please read below my comments which hopefully will help to further improve the manuscript.
Major comments:
1. The root-zone soil moisture is defined here as the water present in the top meter of the soil column (p2, lines 38-39). However, I would argue that the root-zone soil moisture represents the water in the subsurface which is accessible to the roots of the vegetation for transpiration. The depth thereof is highly variable and may depend on climatic (de Boer-Euser et al., 2016; Wang-Erlandsson et al., 2016) and topographic indicators (Fan et al., 2017). Now the derived product provides estimates of soil moisture at different depth intervals, but not really an integrated root-zone soil moisture estimation, which depends on the rooting depth. Would such an addition in the dataset be feasible?
2. When comparing the newly derived product with ERA5L and local data, it would be interesting to see plots of timeseries at specific locations and a spatial map showing where both products have high or low correlations. This would give potential users of the dataset some guidance on where the product may and may not be used. Perhaps the authors can work out some correlations to relate skill with catchment characteristics including climate, land use, topography and soil types.
Minor comments:
3. line 64: I would specify here: “does not consider the model structural error of the EF method”. Although this becomes clear later in the paper, it was not directly clear to me at this stage.
4. line 85: “uncertainties [..] were then calculated from the law of propagation of uncertainties”. Could you explain how this was done in more detail?
5. section 3.2.3: could you explain the presented formulas in more detail? What are the units and what are all parameters? E.g. what does G represent? What does delta represent?
6. line 246: could you quantify with numbers in the text how substantial the difference is between the correlation between E5L and RZSM versus E5L and SSM?
7. Figure 5: in the legend I read “GCOS required uncertainty” but it is not entirely clear to me what this threshold refers to exactly, could you please elaborate?
8. line 280: could you elaborate more on why we expect uncertainties to be amplified during transitions between wet and dry conditions? Which processes play a role which are not well represented in the EF method? Now you briefly refer to Fig6, but it does not provide a clear explanation on why this is expected.
9. line 292 “and highlighting 20% of data with the highest uncertainty”. At this stage, this reads a bit confusing as the previous paragraph describes masking out data with the highest uncertainty and here (if I understood correctly) you are instead plotting data with high uncertainty. Perhaps good to clarify what you mean with “highlighting 20% of data with the highest uncertainty”.
10. line 299: the described difference in uncertainty from 0.008 m3/m3 to 0.004 m3/m3 is very hard to see in the Figure using the applied scale.
11. line 302: why are the uncertainties related to structural breaks not clearly seen in the MAD Topt approach. Can you reflect (here or later in the discussion) a little bit more on this result. It seems to me that the change in uncertainty related to a change in sensor is an important change that you would also want to see in the improved methodology for uncertainty estimation.
12. line 314-320: here, it is not clear to me why using Topt/10 as T parameter uncertainty yields more realistic estimates of temporal uncertainty variations than using MAD Topt in the case of using the time series which includes a structural break (and the opposite in case a shorter time series is used). Which aspects in figure 7 suggest these findings?
13. line 320-324: Again, could you elaborate why the uncertainty estimations of temporal uncertainty variations are no longer accurate for deeper layers?
14. line 351: here, you forgot to add the units of the mentioned uncertainties.
15. In addition, I also downloaded the netcdf files and checked the github page. The nc files contain all the necessary meta data. However, the github page does not include extensive documentation on how to use the different methods within the package. Would it be feasible to elaborate on this further?References:
de Boer-Euser, T., McMillan, H. K., Hrachowitz, M., Winsemius, H. C., & Savenije, H. H. G. (2016). Influence of soil and climate on root zone storage capacity. Water Resources Research, 52, 2009–2024. https://doi.org/10.1002/2015WR018115
Fan, Y., Miguez-Macho, G., Jobbágy, E.G., Jackson, R.B., & Otero-Casal, C. (2017). Hydrologic regulation of plant rooting depth. Proceedings of the National Academy of Sciences, 114(40), 10572-10577.
Wang-Erlandsson, L., Bastiaanssen, W. G. M., Gao, H., Jägermeyr, J., Senay, G. B., Van Dijk, A. I. J. M., et al. (2016). Global root zone storage capacity from satellite-based evaporation. Hydrology and Earth System Sciences, 20(4), 1459–1481. https://doi.org/10.5194/hess-20-1459-2016
Citation: https://doi.org/10.5194/egusphere-2023-47-RC2 - AC2: 'Reply on RC2', Adam Pasik, 07 Jun 2023
Peer review completion
Journal article(s) based on this preprint
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 336 | 113 | 14 | 463 | 6 | 8 |
- HTML: 336
- PDF: 113
- XML: 14
- Total: 463
- BibTeX: 6
- EndNote: 8
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Adam Pasik
Alexander Gruber
Wolfgang Preimesberger
Domenico De Santis
Wouter Dorigo
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(1909 KB) - Metadata XML