the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 28 Apr 2026
Monitoring Land Subsidence at San Francisco International Airport Using Satellite Radar Interferometry
Abstract. Coastal airports are increasingly vulnerable to infrastructure degradation from land subsidence exacerbated by sea-level rise (SLR) and extreme weather events. San Francisco International Airport (SFO), situated on a reclaimed land overlying thick compressible Young Bay Mud, provides a representative case study for understanding the implications of land subsidence on infrastructure resilience. This study employs advanced Interferometric Synthetic Aperture Radar (InSAR) techniques to measure and analyze spatially detailed subsidence at SFO from 2017 to 2024. We integrate the InSAR data with subsurface stratigraphy derived from geotechnical investigations and historical construction records to identify and quantify patterns and drivers of subsidence. The results indicate spatially heterogeneous subsidence rates, with rates exceeding −10.0 ± 0.1 mm/yr, concentrated primarily under the airfield's infrastructure, notably along Runway 10R/28L. Temporal analyses of deformation time series reveal significant variability and nonlinear trends, likely due to seasonal groundwater fluctuations, construction activities, and heterogeneous subsurface stratigraphy. Areas with older hydraulically placed fills demonstrate higher rates of long-term compaction, emphasizing the critical role of historical construction practices and sediment properties. This study's findings underscore the urgent need for comprehensive and continuous ground deformation monitoring at coastal airports. The implications for infrastructure resilience planning at SFO serve as a valuable model for other coastal airports facing similar geotechnical and climatic challenges.
- Preprint
(1744 KB) - Metadata XML
-
Supplement
(962 KB) - BibTeX
- EndNote
Status: open (until 09 Jun 2026)
-
RC1: 'Comment on egusphere-2026-1929', Anonymous Referee #1, 10 May 2026
reply
-
AC1: 'Reply on RC1', Oluwaseyi Dasho, 20 May 2026
reply
Response to Reviewer 1
We thank the reviewer for the thorough and constructive evaluation. We have carefully considered each comment and provided detailed responses below. Changes to the manuscript are described where applicable.
Comment 1.1 — Contradicting Results: VLM rate discrepancy with Dasho & Shirzaei (2025) and NASA OPERA
Reviewer concern: The current manuscript reports VLM rates of up to ~12 mm/yr for 2017–2024, which appears inconsistent with the ~5 mm/yr reported in Dasho & Shirzaei (2025) for 2017–2021, particularly given the stated deceleration in subsidence during 2022–2024. The reviewer also notes a ~3–4 mm/yr discrepancy with the NASA OPERA VLM dataset and raises concerns about result reliability.
Response: We appreciate the opportunity to clarify this apparent inconsistency. We respectfully note that Dasho & Shirzaei (2025) reported subsidence rates of up to 9.2 ± 0.2 mm/yr along the SFO runways, not ~5 mm/yr as characterized in the review. The maximum rate of ~12 mm/yr reported in the current study is observed at a location outside the runway footprint (visible in Figure 1a), in an area not covered by the prior analysis in Dasho & Shirzaei (2025). Along the runways themselves, maximum subsidence rates in the present study remain consistent with those previously reported by Dasho & Shirzaei (2025). This spatial distinction is now made explicit in the revised manuscript to prevent misinterpretation.
Regarding the discrepancy with NASA OPERA: the ~3–4 mm/yr difference is attributable primarily to differences in spatial averaging and methodological differences. The OPERA VLM product is generated at a coarser spatial resolution and involves spatial aggregation that smooths localized deformation signals. The peak rates reported in this study are derived from spatially dense InSAR measurements at the pixel scale (~15 × 15 m), which resolve localized subsidence hotspots that are effectively diluted in gridded, regionally averaged products such as OPERA. This is a well-recognized limitation of coarser-resolution VLM datasets when applied to sites with sharp spatial gradients in deformation, as is the case at SFO. Additionally, the strategy for generating LOS time series differs between the two methods: OPERA products rely on two consecutive pairs, whereas the analysis presented here uses short, medium, and long temporal baseline pairs to reliably retrieve the deformation field. It is well known that accounting for only short pairs leads to artifacts and reduced accuracy in LOS deformation time series. We have added clarifying language to the manuscript to explicitly contextualize this comparison.
Comment 1.2 — InSAR Methodology: Incorrect citation for ascending/descending LOS combination
Reviewer concern: The reference to Miller and Shirzaei (2019) for the LOS combination methodology is inappropriate, as that study presents separate VLM maps rather than describing a combination methodology.
Response: The reviewer is correct. This was a citation error; we incorrectly cited Miller and Shirzaei (2019) when the intended reference was Miller and Shirzaei (2015), which describes the methodology for combining ascending and descending LOS displacement velocities to derive vertical motion. This has been corrected in the revised manuscript.
Comment 1.3 — GNSS Referencing: Temporal non-overlap between HSIB GNSS (2007–2012) and InSAR (2017–2024)
Reviewer concern: It is unclear how InSAR-derived VLM for 2017–2024 was referenced to IGS14 using HSIB GNSS observations available only for 2007–2012, given the absence of temporal overlap.
Response: We acknowledge this as a legitimate methodological limitation. HSIB is the only GNSS station within our InSAR data footprint, and we applied a linear projection of the GNSS-derived velocity to the InSAR period, assuming temporally stable long-term VLM rates at the reference site. While this assumption introduces potential uncertainty, particularly if rates at the reference site changed between the two periods, the consolidation-driven subsidence at SFO is expected to evolve slowly and quasi-linearly over decadal timescales, making linear extrapolation a reasonable first-order approximation. We acknowledge, however, that this represents a limitation of the current study, as verification of the reference frame stability during the InSAR period is not possible with the available GNSS record. This limitation, along with its implications for absolute rate uncertainty, is now explicitly stated in the Methods section and discussed in the limitations paragraph.
Comment 1.4 — Statistical Analysis: Insufficient explanation of methodology and interpretation
Reviewer concern: The manuscript does not sufficiently explain how the Geographical Detector (GD) analysis was performed using the construction year, paleoenvironment, and stratigraphy datasets, nor how the resulting statistical relationships explain the observed VLM rates.
Response: We agree that the statistical methodology section warranted greater clarity. In the revised manuscript, we have expanded the description of the GD analysis to include: (1) explicit specification of the dependent variable (pixel-level VLM rates along the four runways) and the independent variables (construction year, paleoenvironmental condition, and touchdown zone designation); (2) the stratification scheme applied to each factor; (3) the sample size and spatial extent of the analysis; and (4) a more explicit interpretive link between the q-statistics and the observed spatial VLM patterns. Specifically, we clarify that construction year (q = 0.186) and paleoenvironmental condition (q = 0.143) together explain a meaningful share of the spatial variance in runway VLM, and that the residual unexplained variance (~65%) reflects the dominant role of subsurface stratigraphy, particularly lateral variability in Young Bay Mud thickness, which is discussed qualitatively through the cross-sectional analysis in Figure 2 but is not represented as a discrete categorical variable in the GD framework along the runways. We have now added a GD framework along with the AA’ that includes Young Bay Mud (YBM) thickness as an independent variable. The GD results demonstrate that YBM thickness is the dominant control on VLM rates along the profile (q = 0.478).
Comment 2 — Scientific Novelty
Reviewer concern: Given prior VLM studies at SFO by the same authors using ALOS and Sentinel-1 datasets, and the existing 1-D compaction modeling in Shirzaei et al. (2018), the scientific novelty of the current manuscript is limited. The reviewer recommends integrating multi-dataset VLM results and incorporating physical modeling.
Response:
We thank the reviewer for this constructive comment and have substantively expanded the manuscript to address both points raised. We respectfully clarify the distinct contribution of this work and, in response to the reviewer's recommendation, have now incorporated targeted physical modeling that directly strengthens our central finding.
The novelty of this study does not rest solely on the InSAR-derived VLM measurements. We acknowledge that vertical land motion at SFO has been characterized in our prior work using ALOS (2007–2010) and Sentinel-1 A/B (2017–2024) datasets. Rather, this manuscript is the first to integrate Sentinel-1 InSAR with high-resolution subsurface stratigraphy from CPTs, paleoenvironmental classification, and runway-by-runway construction chronology within a unified spatial attribution framework using the Geographical Detector method. To address the reviewer's first point regarding multi-dataset integration, we have added a brief comparison of VLM magnitudes and spatial patterns, identifying regions of intensifying and slowing subsidence.
The most significant scientific finding of this study, that maximum deformation rates do not spatially coincide with maximum compressible sediment (Young Bay Mud) thickness, is not explained by stratigraphy alone and was not identified in Shirzaei et al. (2018), which examined deformation at the Bay-wide scale without per-runway stratigraphic attribution. This spatial decoupling is the central novel observation of the manuscript.
In response to the reviewer's recommendation for physical modeling, we have added a targeted 1-D compaction modeling analysis at the 11 CPT locations along profile AA′ (new Section 3.6). We calibrated the Gibson & Lo (1961) rheological creep 1-D model against the InSAR displacement time series at each CPT location using the measured per-CPT layer thicknesses (fill, YBM, alluvium, and Old Bay Mud). This goes substantially beyond the spatially uniform compaction model of Shirzaei et al. (2018) in three ways. First, we resolve the spatial heterogeneity of YBM creep compliance (b_YBM) and retardation time (τ_YBM) at the CPT scale. Second, although YBM thickness emerges as the dominant single factor in the GD analysis, substantial intra-stratum variance remains. The 1-D consolidation modeling provides a direct physical explanation. We show that b_YBM/τ_YBM spans two orders of magnitude along the profile (2.6×10⁻⁹ to 1.2×10⁻⁷ Pa⁻¹ yr⁻¹), controlled by drivers other than applied stress or layer thickness alone. The CPT calibration demonstrates that pixels in tidal marsh and tidal flat paleoenvironments have b_YBM = 3.6–4.0×10⁻⁶ Pa⁻¹, 4–8× higher than YBM beneath shallow bay deposits, driven by elevated organic content and high initial void ratio. This paleoenvironment-controlled spatial gradient in creep compliance is a new physical result that mechanistically explains the statistical GD patterns and was not recoverable from the spatially uniform model of Shirzaei et al. (2018). Third, the profile-median τ_YBM = 89.9 yr independently validates the bulk geodetic estimate of ~80 yr from Shirzaei et al. (2018), confirming the physical consistency of our approach while revealing its spatial structure for the first time. This paleoenvironment-controlled spatial gradient in creep compliance is a new physical result that mechanistically explains the statistical GD patterns and was not recoverable from the spatially uniform model of Shirzaei et al. (2018).
Together, the expanded GD methodology and the supporting 1-D compaction modeling provide both statistical attribution and mechanistic interpretation of the spatial VLM pattern at SFO. We are grateful to the reviewer for prompting this strengthening of the analysis.
Citation: https://doi.org/10.5194/egusphere-2026-1929-AC1
-
AC1: 'Reply on RC1', Oluwaseyi Dasho, 20 May 2026
reply
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 92 | 36 | 19 | 147 | 19 | 15 | 16 |
- HTML: 92
- PDF: 36
- XML: 19
- Total: 147
- Supplement: 19
- BibTeX: 15
- EndNote: 16
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This manuscript presents InSAR-derived vertical land motion (VLM) measurements for the period 2017-2024 and investigates the relative contributions of historical land reclamation practices, paleoenvironmental setting, and operational aircraft loading to the observed VLM using statistical approaches. While the study addresses an important topic and takes a useful approach of examining subsurface profiles to explain VLM, the current manuscript significantly lacks the scientific analysis to support the reported VLM rates and does not provide sufficient scientific novelty in its present form. Therefore, I recommend major revisions, which should help the authors improve both the scientific rigor and novelty of the study. My detailed comments are provided below.