the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 14 Aug 2024
Benchmarking passive microwave satellite derived freeze/thaw datasets
Abstract. Satellite derived soil surface state has been identified to be of added value for a wide range of applications. Frozen versus unfrozen conditions are operationally mostly derived using passive microwave (PMW) measurements from various sensors and different frequencies. Products differ thematically as well as in spatial and temporal characteristics. All of them offer only comparably coarse spatial resolution in the order of several km to tens of km which limits their applicability. Quality assessment is usually limited to comparisons with in situ point records, but a regional benchmarking dataset is thus far missing. Synthetic Aperture Radar (SAR) offers high spatial detail and thus is potentially suitable for assessment of the operational products. Specifically, dual polarized C-band data acquired by Sentinel-1, operating in Interferometric Wide (IW) swath mode with a ground resolution of 5 x 20 m in range and azimuth, provide dense time series in some regions and are therefore suitable as basis for benchmarking. We developed a robust freeze/thaw (FT) detection approach, applying a constant threshold on the combined C-band VV and VH polarization ratios, that is suitable for tundra regions. The achieved performance (91.8 %) is similar to previous methods which apply an empirical local threshold on single polarized VV backscatter data.
All global products, tested with the resulting benchmarking dataset, are of value for freeze/thaw retrieval, although differences were found depending on seasons, in particular during spring and autumn transition. Fusion can improve the representation of thaw and freeze-up, but a multi-purpose applicability cannot be obtained since the transition periods are not fully captured by any of the operational coarse resolution products.
- Preprint
(23242 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 25 Sep 2024)
-
RC1: 'Comment on egusphere-2024-2518', Anonymous Referee #1, 23 Aug 2024
reply
This study evaluated the utility of relatively high resolution frozen area retrievals over tundra, derived from Sentinel-1 C-band SAR, for benchmarking coarser resolution freeze/thaw (FT) products from global passive microwave (PMW) satellite records. In situ air and soil temperature measurements from different locations across northern Alaska were used in calibrating the SAR FT retrievals for the benchmark dataset, while near surface soil temperature data from another site in Northern Finland were used for additional validation. Key findings show the SAR data to provide strong FT performance suitable for benchmarking the value and limitations of coarser global PMW FT records in the tundra region. The authors apply the framework to document differences in relative performance from the major PMW FT records in different seasons, while also demonstrating improved performance through a fusion of the different PMW products. The authors then provide an informative discussion of implications of differences between the datasets for a range of potential applications. Overall, the paper is well written and informative, providing a significant step forward in developing an effective benchmarking framework to gage and further advance the FT land parameter as an Essential Climate Variable. Despite the many strengths of the paper, moderate revisions are needed to address the following concerns and recommendations before the paper can be considered suitable for final publication.
Major comments:
The authors state that a target requirement for a FT product domain aiming at permafrost should be global (Ln 67-69). However, permafrost and FT events generally only occur in the high latitudes and high altitude regions. As such, more specifics and justification on the global domain requirement is needed here.
Ln 125: It should be noted that the MEaSUREs FT record is based on classification of Ka-band (V-Pol.) brightness temperature records from various PMW satellite missions (SMMR, SSM/I, AMSR). Clarification of the frequency of the FT retrieval is important given the strong link between frequency and the characteristic penetration and sensitivity of the retrieval to different landscape features (e.g. vegetation, snow, soil). Further, the MEaSUREs record includes calibrated brightness temperatures from the two AMSR missions (AMSR-E [2003-2011] and AMSR2 [2012-present]). This should be noted and the more general “AMSR” term used in place of “AMSR-E” throughout (e.g. Table 1 and elsewhere).
Table 1: Include additional information in the table summary stating the underlying frequency represented from the different FT products used given expected frequency dependent differences in sensitivity and performance. Also, indicate whether global or polar grid datasets are used since some products (e.g. MEaSUREs, SMAP) provide both.
Section 2: More justification of the particular FT data records and formats selected for this study is needed given the diversity of product formats available, which may be better suited to polar regions. For example, the MEaSUREs archive includes both global 25-km and polar 6-km grid FT products. The SMAP L3FT products (SPL3FTP[E]) include both global and polar projections in both 36-km and 9-km grid formats. The SMAP 9-km grid product is derived using a spatially optimized (Backus-Gilbert) interpolation of SMAP antenna temperatures and associated brightness temperatures intended to improve the effective spatial resolution. Because of the excessive stretching of the global cylindrical equal-area (EASE-grid) map projections toward the poles (e.g. illustrated by the elongated MEaSUREs grid in Fig 2.) the polar azimuthal EASE-grid products are likely better suited for applications over high latitude tundra, especially involving comparisons with in situ ground stations.
Figure 2 caption: More information is needed in the figure caption noting the different product grids represented in the image; e.g. polar vs global grid projections; 36-km vs 9-km Res.
Section 3.1: The authors state that the spatial frozen fraction is derived from the SAR data for each cell of the PMW FT products (e.g. Ln 192). This approach is a major strength of this study for validating coarser resolution land retrievals that may only contain a single representative value for each grid cell and then rely on validation against sparse in situ measurements that may not be fully representative of the complex landscape within the satellite footprint. The use of SAR FT data for benchmarking overcomes the limitations of sparse in situ measurements for being regionally representative and more congruent with the sampling footprint from other global FT records. This challenge is partially depicted by the varying FT grid sizes relative to the point measurements in Figure 2, but a more clear statement of this strength is needed.
Section 3.1: The PMW FT products (e.g. MEaSUREs, SMAP) include diurnal (AM and PM) FT retrievals from the satellite ascending and descending orbital passes. The authors should clarify which FT sampling period(s) were used in this study and provide justification for the selections (e.g. temporal optimization with in situ or SAR sampling, best accuracy/consistency, etc.).
Ln 250: The authors state that data from in situ measurements were used for assessing FT conditions indicated from the PMW FT product fusion. However, more information is needed describing how the spatial match-ups for the comparisons were made between in situ measurements representing local site conditions and the spatially gridded data, particularly from the SAR; e.g. were comparisons made between the in situ station measurements and the overlying SAR pixel or against a larger SAR subregion within the overlapping PMW grid cells?
Ln 406-409: It should also be noted here that MEaSUREs and SMAP FT performances strongly degrade with increasing fractional water (FW) cover common to tundra; in general the SMAP L-band data appear more strongly impacted than MEaSUREs Ka-band retrievals as documented from Kim et al. (2019). The latest SMAP product release also includes an improved water mask expected to enhance performance particularly in boreal-Arctic areas with abundant water bodies (Du et al. 2023). The PMW FT products include QC flags for FW, which could be used to help inform the stand-alone or Fusion approaches.
Section 4: The accuracy and utility of the PMW FT and fused products were assessed against the SAR frozen fraction and in situ observations. However, it is unclear whether additional ancillary data quality information provided with the PMW FT products, including pixel level quality (QC) information on expected performance and retrieval success, was used to help interpret and explain product behavior. For example, the MEaSUREs FT product contains annual maps of estimated quality along with per pixel QC data on the presence of known error factors including extensive FW, terrain complexity, and active precipitation days. The SMAP FT record includes similar per pixel QC flags. The tundra sites used for this study may be affected by these adverse conditions, which may help explain some of the degraded performance. The authors should include this additional information in the results discussion.
Figures 8, 9: It was unclear from the methods description whether the AM or PM overpass derived FT records were used for these comparisons. As noted in the results, some of the records have trouble fully delineating the transitional periods, but I wonder whether the transition would be better resolved using both the AM and PM FT series, as we might expect greater frequency of diurnal thawing and refreezing during these periods.
Section 5.3: More discussion is needed here regarding the potential utility of FT retrievals from different microwave frequencies for the desired application. In this study, C-band SAR data are used with in situ ground temperature observations to benchmark a fusion of FT products derived from Ka-, and L-band PMW observations. However, this approach neglects the frequency dependence of the FT products, which may represent different and potentially complementary aspects of FT heterogeneity from different landscape features (e.g. vegetation vs soil). The relative utility of FT products for the different applications described (Vegetation, soil moisture, etc..) is also expected to have strong frequency dependence. Given this dependence, is the use of the frozen area reference from C-band SAR data a constraint in benchmarking FT performance from different frequency retrievals? A more clear and concise statement of the study strengths and limitations in this regard is needed.
Section 5.3 Cont.: In this study the frozen fraction reference is defined in a spatial context and the FT condition is defined in accordance with binary FT states defined from temperature measurements. The simple binary FT classification approach likely contributes to frequency dependent differences in the PMW, SAR and site level FT comparisons. This is a constraint common to the FT products used, which should be mentioned. Further, more sophisticated FT algorithms may be able to detect changes in the amount unfrozen liquid water remaining in soil under sub-zero temperatures, rather than relying on simple binary FT thresholds (e.g. Wang et al. 2024). This requires advanced dielectric models to represent non-linear behavior in highly organic soils common in the arctic, but is more physically meaningful and may enhance application utility, which may be worth considering in future FT benchmarking efforts.
Minor comments:
Ln 20: “as proxy” should be “as a proxy”.
Ln 86: “this not” should be “this is not”.
Ln 191: “fulfil” should be “fulfill”.
Ln 144: Include Derksen et al. 2017 or Kim et al. 2019 (below) as more suitable citations on the stated SMAP accuracy relevant to this study since the Kraatz paper doesn’t involve tundra or other high northern latitude sites.
Ln 273: “higher then” should be “higher than”.
Ln 325: Sentence needs revising for better clarity.
Table 4 and Figure 11: “AMSR-E” should be restated more broadly as “AMSR” to encompass both AMSR-E and AMSR2 portions of the FT record.
Ln 368: “what differs” should be “which differs”.
Ln 436-437: “SSMR” should be “SMMR”; “what agrees” should be “which agrees”.
Ln 444: “fulfil” should be “fulfill”
Ln 471: “relevenace” should be “relevance”; also clarify what is meant by “fluxes” here; e.g. previous studies have used AM and PM FT retrievals to characterize transient nighttime frost events affecting vegetation growth and day-night FT transitions commonly occurring in the shoulder seasons.
Ln 476: “constrains” should be “constraints”.
References mentioned above:
Derksen, C., X. Xu, R.S. Dunbar, et al., 2017. Retrieving landscape freeze/thaw state from Soil Moisture Active Passive (SMAP) radar and radiometer measurements. Remote Sensing of Environment 194, 48-62.
Du, J., J.S. Kimball, S.K. Chan, et al., 2023. Assessment of surface fractional water impacts on SMAP soil moisture retrieval. IEEE JSTARS, 16, 4871-4881, DOI: 10.1109/JSTARS.2023.3278686.
Kim, Y., J.S. Kimball, X. Xu, et al., 2019. Global assessment of the SMAP freeze/thaw data record and regional applications for detecting spring onset and frost events. Remote Sensing 11, 11, 1317, https://doi.org/10.3390/rs11111317.
Wang, C., N. Yang, T. Zhao, et al. 2024. All-season liquid soil moisture retrieval from SMAP. IEEE JSTARS 17, doi:10.1109/JSTARS.2024.3382315.
Citation: https://doi.org/10.5194/egusphere-2024-2518-RC1 -
RC2: 'Comment on egusphere-2024-2518', Anonymous Referee #2, 10 Sep 2024
reply
Review on Benchmarking passive microwave satellite derived freeze/thaw datasets, by Bartsch et al., (egusphere-2024-2518).
This paper used a fusion approach to detect soil FT status over Alaska North slope and Northern Finland using three FT data records. The results demonstrated the combined ratio approach provides excellent overall agreement with in-situ measurements and a similar performance to a CNN approach. The results also show the potential capability of benchmarking PMW FT data products with Sentinel-1 FT retrievals. The paper covers a topic that is suitable to readers of The Cryosphere and should be of particular interest to those interested in FT classification algorithm development over the Cryosphere and permafrost monitoring under climate change. The suggested major revisions are as follows:
- Major concern is lack of detail in describing FT data fusion method and benchmark FT classification algorithm. Some references were included in this manuscript; however, it is not sufficient description for readers to get clear and repeat the FT data retrieval processing. For example, spatial and temporal resolutions for fused and benchmark FT data sets are not clearly described in the method section.
- In the accuracy agreement, authors show the temporal comparison with in-situ soil temperatures. However, differences in spatial FT patterns among FT products (with different frequency, resolution, and so on) would allow us to better understand sub-scale impacts on FT classification accuracy. Comparison of S1 frozen fraction maps and other FT state images (e.g., fused datasets) would be good to analyze their spatial patterns. Providing a couple of sample images would be enough.
- In Table 6, why does accuracy differs among the operational coarse resolution products (MEASURES, SMAP, SMOS)? Did you use high-quality FT flags? Several studies reported some possible reasons why there are differences particularly during the transitional periods. Authors should include the potential reasons briefly in the discussion section (e.g., different sensor specification) even though some of them are stated in the introduction section. If any unknown uncertainties are found in this study, you would include some challenges and probably suggest future study in the conclusion section to improve fused FT data products (stated shortly only in Line 479, page 25).
Additional edits are noted below:
Line 128, page 5: Did you use MEaSUREs FT data Version 5 in this study? Version 3 was introduced in line 124 , page 4 as well. Authors should clarify it.
Line 126 & 132, page 5; Which frequency and what algorithm was used for MEASURES FT datasets? Authors should include it.
Ine 131, page 5: Table 1 does not include a 6km resolution FT data sets. Authors should modify it.
Line 132, Page 5: The AMSR-E/AMSR2 data is available between 2002 and 2021 (not 2020).
Line 136, page 5: what FT algorithm was used for SMAP and SMOS products? Authors should describe briefly.
Line 145, page 5: How did authors determine the partially frozen state from SMOS? Does partially frozen mean thawing in Table 1? Is it thaw-refreeze condition? Is it different from SMAP transitional state?
Line 170-171, page 6: Authors should include the temperature data source (e.g., web link, references).
Line 182, page 6: Authors should include more details on ERA5 reanalysis (e.g., data source, spatial and temporal resolutions, which temperature?, soil or air temperature? and so on).
Line 192, page 8: It is not clear to me how to derive the frozen fraction from Sentinel-1. Authors should specify it.
Line 203-204, page 9: The equation, Rc should be included here to better understand.
Line 211, page 9: This Rc equation should be introduced first in Line 203-204.
Line 216, page 9: which reanalysis data? And which temperature was used? Is air or soil temperature? Daily or hourly?
Line 248, page 10: what are the other flags? Authors should specify it.
Line 254-255, page 10: is it daily mean air temperature?
Line 257, page 10: what is the fusion method? How to fuse Metop ASCAT and SMOS?
Line429, Page24: It (Finding … 2018) is not clearly stated. Authors should clarify it.
Line488,page 26: situ should be in-situ.
Figure 1: Legend describing permafrost zones (types) should be included inside the map, even though the permafrost zones are described in the caption.
Figure 2: It seems to me that MEaSUREs grids (in yellow) may not be 25km resolution. The length of horizontal lines looks like ~9km.
Figure 6 and 7: what do horizontal blue lines mean?
Table 6: ERA5 air or soil temperature?
Citation: https://doi.org/10.5194/egusphere-2024-2518-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 125 | 42 | 12 | 179 | 4 | 5 |
- HTML: 125
- PDF: 42
- XML: 12
- Total: 179
- BibTeX: 4
- EndNote: 5
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1