Benefit Assessment of Commercial Synthetic Aperture Radar Observations for NASA’s Surface Deformation and Change Mission Study

.


INTRODUCTION
Guided by the SATM (Horst et al. 2021, Khazendar et al. 2021), the SDC mission study is evaluating several architectures having different numbers of spacecraft, orbital configurations, sensor characteristics, and geometries.These configuration differences result in distinct core capabilities, which are evaluated in a consistent value framework.Some of these core capabilities can benefit from commercial SAR constellations filling in data gaps, especially in terms of high temporal and spatial resolutions.In recent years, commercial sector capabilities in SAR imaging have been expanding rapidly, however, the interferometric capability is currently still limited.The feasibility of commercial capabilities to support the GOs defined in SDC's SATM are being analysed to identify the strengths and weaknesses of commercial SAR data when applied to SDC observables.In addition to the parameters defined in the SATM, our analysis also looks at the interferometric capability of commercial constellations, along with their swath width and their global sampling coverage rate.The GOs are described in detail in the SATM, and a summary of the relevant parameters used in the scoring is provided in the appendix.To achieve a result that is resilient in the face of market fluctuations, the commercial assessment relies on studying industry capabilities as a whole instead of individually focusing on each constellation.

METHODOLOGY
The methodology is designed to assess possible strengths and weaknesses of commercial SAR data in meeting the GOs defined in SDC's SATM.The scoring equation uses eight key factors to parameterize the benefit of commercial data for each GO.

Parameters
All parameters defined in this assessment are defined as a percentage, represented as a number between 0 and 1 for low and high scores, respectively.Benefit scores are calculated using the following equations: (1) where is the interim score, N is the number of   parameters that are arithmetically averaged, and M is the number of multiplicative parameters (defined in Table 1).S and C represent the partial scores, is the final score, and   acc is expected measurement accuracy (relative to the desired capability in the SATM, see example in Table 4).
In (1), arithmetic averaging is used in the partial scores S and multiplication is used in the partial scores C due to the perceived difference in the impact of these parameters.For example, if data can be acquired every two weeks, but the desire is to collect data every week, the revisit score would be 50%.We equally weighted revisit, polarisation, resolution, and latency, therefore arithmetic averaging is used for a combined S score.On the other hand, coverage, large swath imaging, and InSAR capabilities are more limiting compared to parameters contributing to the S score.For example, if global coverage is needed every two weeks, and there is only enough commercial capacity to image 50% of the globe every two weeks, this should not result in a C score that is higher than the lowest factor, even if InSAR and large-swath capacity are fully met.Therefore the C scores are multiplicative as shown in (2).
As part of the commercial SAR feasibility assessments, the team identified the main observational parameters and associated representative capabilities that commercial SAR can currently provide, as shown in Table 2.These capabilities are derived from publicly available information from the commercial constellations operated by Capella Space, ICEYE, iQPS, NEC, Surrey Sat.Tech.Ltd, Synspective, and Umbra.One important parameter that is not considered in the above set of parameters is the satellite frequency band.This is due to the fact that the bulk of the available commercial data is in X-band.Rather than excluding GOs that desire a specific band, we estimate expected performance at X-band for all GOs with the assumption that any GO can be satisfied under specific observation scenarios or locations, even if they cannot be satisfied globally at X-band.However, depending on the target of interest, X-band sensors can fall short for Ecosystems and Solid Earth observables due to shallow penetration depth and rapid decorrelation (Zebker and Villasenor, 1992;Wei and Sandwell, 2010;Hajj et al., 2019).In some cases, imaging might still be possible, albeit with significantly less coverage and accuracy compared to a longer wavelength sensor; in other cases, however, the sensors may completely fail to retrieve any meaningful measurements.In particular, long-wavelength radar can be critical for observing change mapping in wetlands, forests, and coastal zones (Ottinger and Kuenzer, 2020), and for imaging subtle movements of faults in heavily vegetated areas (Nikolaeva et al., 2014;Tong et al., 2018).Here, we have not explicitly accounted for the cases in which X-band sensors would completely fail to return meaningful data, but these limitations are important to consider in the overall evaluation of commercial SAR capabilities.For the GOs that are not assessed by the performance tool, a literature review was conducted to approximate the measurement capability at X-band, and for heuristic scores the expert panel provided accuracy estimates considering X-band.Ultimately, a percentage of desired capability is calculated based on this value and the SATM is used in the assessment (see Table 4).

Parameter Value Justification
Coverage may require special attention in this calculation as this capability is defined by km 2 /day for the average commercial market, while the SATM did not define such a metric, assuming a global need.Unlike the value framework assessment for SDC architectures, a coverage area is defined in the commercial assessment for each geophysical observable, as shown in Table 3. Table 3. Coverage bins used in this study for each GO.
Table 3 is used to check the capability of the commercial sector to see if such an area can be covered during the required revisit time defined for the GO.
Even though the same SATM is used for both the value framework analysis and the commercial SAR assessment, the assessments are different to compensate for the differences between commercial capabilities and SDC architectures.For example, all SDC architectures provide coverage of all land and ice areas every 12 days or faster, and it is not feasible to acquire global coverage using commercial constellations due to coverage limitations and cost.Similarly, all SDC architectures are capable of interferometry, while only three out of seven commercial constellations provide that capability.Such differences necessitate the use of additional parameters for the commercial SAR assessment.

Scoring
The assessment scores each GO against the needs defined in the SATM.For example, the ninth ecosystem (E9) GO calls for "measuring inland coastal wetland areas" and mentions the need for dual polarisation imagery at 20 m resolution with a revisit and latency of 14 days (Table 4).Similarly, the need for coverage is classified as regional (9 million km 2 ) every 14 days, and neither large swath imagery nor interferometry are necessary (Table 5).Our panel of experts assessed the X-band performance of wetland mapping as 35%.   4 and Table 5.By design of (2) the final score is limited with the accuracy parameter.

RESULTS
This methodology was applied to all of the GOs in the SATM to produce a self-consistent set of commercial benefit scores.These results allow for identification of GOs that would benefit more than others in a relative sense.

Comparative Assessment
In order to better understand the benefit of commercial data, an architecture based on a simplified NISAR design (Kellog et al.The accuracy scores for L1A came from a Value Framework in development for assessing the SDC architectures with respect to Science Benefit and other components of Value.In the Value Framework, accuracy metrics for each GO can depend on either or both vertical and horizontal accuracy needs.In cases where more than a single accuracy score was provided, these scores were averaged to obtain GO specific L1A accuracy score.These scores then compare the capability of an architecture to the desired capability in the SATM for each GO, and quantify the degree to which that architecture achieves the SATM's needs. Figures 1 and 2 show the relationship between the commercial constellations vs. L1A scores, separating the GOs based on the source of the accuracy metric, heuristic, and performance tool, respectively.There are two reasons for this: 1) the performance tool accuracy calculation is holistic and accounts for parameters such as revisit rates as well; and 2) due to the very different nature of accuracy estimates and the strong control of accuracy on the final outcome, the behaviour of GOs for both cases needs to be investigated.Figures 1 and 2 demonstrate a logarithmic nature, with a sharp rise close to the origin and relative flattening at the far end of the x-axis.This behaviour is perhaps indicative of the fact that both SAR systems are capable of addressing the SATM albeit less so for commercial constellations.Furthermore, the different range of scores are evident, as commercial assessment scores achieve a high of only 0.43 while the L1A architecture does achieve a full score (1.0) for some GOs.On average, L1A scores an average of 0.49, while commercial constellations achieve a score of 0.15 in realising the SDC SATM.Despite this, there are some GOs that achieve higher assessment scores for the commercial constellation compared to L1A, such as cryospheric GOs C7 and C9, as well as solid Earth GOs such as S12, S13.

DISCUSSION
The effort documented here covers only an augmentation of a core SDC capability.The SDC team has also reached out to the industry through a Request For Information (RFI) and is planning to release a Request for Proposal (RFP) in the first quarter of 2023.The RFI/RFP activities include public-private partnership solutions, but these are outside the scope of this whitepaper.

Assessment Scores
The assessment scores obtained here are based on the commercial capabilities at the end of 2022.It is a fair assumption that commercial capabilities will increase significantly by the time of the SDC mission.However, it is difficult to project commercial capabilities into the 2030s, as the market is currently going through a rapid growth.It is unclear how the growth rate will change in the next ten or so years.Therefore, a conservative approach is taken by assessing the current capabilities and repeating the assessment over time before key decision points for SDC.
As it can be seen from Tables 4 and 6, the horizontal resolution capability of both commercial constellations and the L1A architecture is defined as 5 m.This is due to the fact that the SATM does not include any resolution needs beyond 5 m, and any architecture providing a resolution of 5 m or better automatically obtains the highest partial score of 1.
The Earth Science Decadal Survey (National Academies, 2019) emphasised the need for interferometric repeat pass at weekly to daily rates, resolution between 5-15 m, sensitivity to height changes between 1 and 10 mm, with a time series measurement accuracy of between 1 mm/week and 1 mm/year (depending on the duration of the time series) all the while providing continuous global monitoring of all land and coastal areas.
Architectures that obtain daily global monitoring of all land and coastal areas easily exceed the cost cap due to increased in orbit duty cycles (or number of satellites) necessary.Therefore, if an SDC architecture is selected such that it provides a generally accepted solution for global monitoring, albeit at longer end of the decadal survey revisit requirements, augmentation of SDC with commercial data can reduce the revisit for areas of interest whenever necessary (e.g.geohazards).
The GOs that score higher than SDC architectures indicate potential benefits to the science community beyond routine monitoring, all of which have a higher X-band accuracy score and are regional in nature (Table 7).In other words, if areas of interest can be narrowed down using non-commercial data, these GOs may benefit from routine commercial data collections for the specific applications.

Ice-Sheet Speckle Tracking and InSAR Analyses
Ice Interferometric SAR data are particularly useful for ice sheet science with respect to surface velocity and grounding line measurements.Ice velocity (C1, C2, C3), and strain fields (C4, C5) are typically measured using speckle tracking (Rignot et al., 2011), where surface parallel 2D motion can be measured using a pair of acquisitions.A more accurate method is to exploit the InSAR phase; this approach requires interferograms acquired in different viewing geometries to resolve 2D motion (Mouginot et al., 2019).Figure 3 shows an example of a feature tracking result of high resolution X-band data on an Antarctic glacier.In this case data were acquired in a near-repeat track orbit, but without controlling the interferometric baseline.The high resolution data provide sufficient surface features on the glacier itself for offset tracking to work.The ultimate goal for a comprehensive data assessment is to obtain InSAR data, however, to assess the full potential of a mission, as demonstrated by the following example.
The ice-ocean interface of a glacier is a critical boundary.The grounding line delineates where ice detaches from the bed and becomes afloat and frictionless at its base.Using double difference interferometry, the flexing of the ice shelf due to differential tide levels at four acquisition times results in a dense band of fringes in the interferogram due to the vertical displacement.The upstream boundary of this fringe band is interpreted as the InSAR grounding line (related to C7 in Table 7).The approach requires the availability of 2 interferograms with the same geometry.Fast glaciers can pose a problem for missions with a longer repeat orbit due to phase decorrelation.Commercial high resolution InSAR missions with short repeat orbits do not generally satisfy the spatial coverage requirement, however, they can be used to augment agency missions like SDC with a more global coverage approach.Early results highlighting the value of a commercial InSAR data for grounding line studies are presented in (Figure 3, Ciraci et al., 2022).

Analysis of Potential Persistent Scatterers
Preliminary analysis on the interferometric potential of commercial X-band data for deformation monitoring over semi-arid land and exposed rock was also conducted through the analysis of a set of available scenes from Capella Space over the West Angelas Mine in western Australia, shown in Figure 4.
The study area is a desert devoid of vegetation, which presents a good opportunity for image analysis without confounding issues of decorrelation and terrain type.Interferometric techniques can also supplement traditional change detection methods for monitoring environmental health in and around active mines; interferometry is more sensitive to small changes on the ground and is capable of monitoring hydrological activity such as groundwater extraction through measurement of deformation as a proxy.The potential performance of the scenes for use in persistent scatterer InSAR (PS-InSAR) was evaluated through the estimation of the density of potential persistent scatterer (PS) candidates.Because the scenes were described to be interferometry-feasible but were not acquired to be interferable, this preliminary analysis evaluated PS candidates not using phase-based methods but instead by using two common amplitude-based methods that are used in existing PS identification algorithms as initial filtering steps before detailed phase analysis.Results were compared with Sentinel-1B imagery that was acquired over the same area and over the same time period.Dataset parameters are shown in Table 8.
Capella imagery was corrected based on the WGS84 ellipsoid and geocoded to a 20 x 20 m grid using the GAMMA software (Wegnüller et al., 2016).Sentinel-1 imagery was processed using the 20 x 20 m Copernicus DEM for topographic correction and geocoding using a backprojection processor (Zebker, 2022).The Sentinel-1 imagery was then masked to the approximate size and area of the Capella acquisitions.Preliminary analysis of PS candidates was conducted by thresholding of calculated amplitude dispersion and signal-to-clutter ratio (SCR) of each stack, defined respectively as: (  8. Imaging parameters for Capella Space and Sentinel-1B datasets used in the mining case study. Figure 5 shows the comparison between the Capella and Sentinel stacks with an amplitude dispersion of 0.4 and 0.25, which are the thresholds used for initial pixel elimination during the first step of StaMPS and the original PSI algorithms, respectively (Hooper et al. 2004, Ferretti et al. 2001).Figure 6 shows the comparison between the two stacks with an SCR of 2, which is the threshold used for initial candidate selection in the STUN algorithm (Kampes, 2005), and an SCR of 10, a much more stringent threshold.The results are mixed, showing a higher estimated PS candidate density in Capella data when comparing SCR, but a higher estimated PS candidate density in Sentinel-1 data when comparing amplitude dispersion.The discrepancy between the two tested measures highlights the limitations of purely amplitude-based measures as effective phase proxies, given that PS candidate density is expected to scale with resolution regardless of wavelength (Huang and Zebker, 2018).The lower PS candidate density for Capella data using amplitude dispersion, in particular, is also likely due to the fact that the commercial imagery was not acquired for interferometry, resulting in an uncontrolled baseline that lowers phase stability.SCR, by contrast, is a more direct measure of relative scatterer brightness, which is higher in the higher-resolution Capella data compared to Sentinel-1 data.Capella data should show an improvement in PS candidate density detected using amplitude dispersion if acquired in orbit-controlled passes designed for interferometry.Ultimately, while this preliminary analysis indicates promise for the use of commercially acquired high-resolution SAR data in interferometric applications, future analysis with high-resolution DEMs as well as with an interferometric dataset, when available, will be required to make a further determination on the capabilities and limitations of commercially available data to augment observation of SDC GOs.

CONCLUSION
In conclusion, this quantitative feasibility study of the commercial sector indicates that, as is, the X-band commercial systems can have some role in meeting the observational capabilities defined in SDC's SATM.Some GOs in the SATM can benefit from commercial data either to supplement the observations of an SDC architecture or to fill a gap that is left by such an architecture.Assessing all of the SDC architectures through this framework would identify the relative benefits of commercial data for each architecture.It is important to note that, as the commercial constellation capability increases, many more GOs will benefit from this data-rich environment.

APPENDIX
In the table below, the needs are grouped for display purposes.

Figure 1 .
Figure 1.Commercial assessment scores against L1A architecture for GOs with heuristic accuracy.

Figure 2 .
Figure 2. Commercial assessment scores against L1A architecture for GOs with performance tool based accuracy.

Figure 3 .
Figure 3. Feature tracking result of imagery from Capella Space over Murphy Glacier, Antarctic Peninsula.

Figure 4 .
Figure 4. Coverage of imagery from Capella Space over the West Angelas Mine; area in western Australia shown in inset.
pixel across the image stack, and is the   amplitude dispersion; in (6), is the power of the dominant   scatterer, taken as the power of a single pixel, is the power   of the clutter, estimated by taking the average of the immediately surrounding pixels, and is the SCR.

Figure 5 .
Figure 5.Comparison between Capella and Sentinel image stacks with an amplitude dispersion threshold of < 0.4 (top) and < 0.25 (bottom).The number in parentheses indicates the percentage of total pixels identified as PS candidates.

Figure 6 .
Figure 6.Comparison between Capella and Sentinel image stacks with an SCR threshold of > 2 (top) and > 10 (bottom).The number in parentheses indicates the percentage of total pixels identified as PS candidates.

Table 1 .
Parameters used in the scores.
*The equation is only used for literature survey results when heuristic scores are not available.

Table 2 .
An overview of currently available representative commercial capabilities used in the commercial SAR feasibility estimation

Table 4 .
Example estimation of percent capabilities addressing the desired needs for parameters defined in the SATM (GO E9).

Table 5 .
Example estimation of percent capabilities addressing the desired needs for Commercial SAR assessment specific parameters (GO E9).

Table 6 .
Conservative capability estimate of L1A architecture for this analysis.
2020, Rosen et al. 2021), which we call as L1A, was assessed.The relevant parameters of this design are summarised in Table6.

Table 7 .
Comparative assessment of scores for the L1A architecture and commercial constellations (C.C.) for the GOs that do not follow the logarithmic trend.
models.Several national and international programs (NASA MEaSUREs, ESA Climate Change Initiative) fund efforts to generate high quality geoinformation products for Antarctica and Greenland based on satellite remote sensing data.The SDC SATM builds on the ECV requirements.
Yes if needed for the GO or No) "&" is entered for GOs that have multiple methods that and not all of them require InSAR.Same thing is true for Large area imaging.