Land deformation analysis using SBAS-InSAR time series with land cover class-specific coherence filtering for the Mississippi River Delta

Abstract

This study presents a comprehensive assessment of land deformation patterns across a 22,043 km² area of southeastern Louisiana and coastal Mississippi using Small Baseline Subset Interferometric Synthetic Aperture Radar (SBAS-InSAR) time series analysis from 2020 to 2024. The study processed 461 ascending-path interferometric Sentinel-1 SAR image pairs using the MintPy framework, implementing a minimum temporal coherence threshold and a land cover class-specific coherence thresholds filtering to optimize signal detection across diverse coastal environments. The analysis integrated National Land Cover Database (NLCD) 2021 data to characterize subsidence vulnerability patterns across different land cover types. The research results reveal extreme spatial heterogeneity in land surface deformation, with line-of-sight velocities ranging from − 43.71 mm/year to + 30.23 mm/year. Coherence-based filtering by land cover class (Developed Land, Pasture/Hay/Barren, Grassland/Shrub/Crops and Forest/Wetland/Marsh) produced LULC specific LOS velocity maps. Land cover class-specific coherence threshold filtering produced separate LOS velocity maps for Urban, Barren/Pasture, and Marsh/Agriculture land cover categories, enabling spatially refined deformation characterization across the delta. Results indicate that marsh and wetland areas experience the highest subsidence rates, particularly in organic-rich marshlands and low-elevation areas, while uplift signals are observed in southwestern coastal areas, related to sediment deposition processes. Validation against NOAA CORS GNSS stations (BVHS and GRIS) shows InSAR SBAS LOS velocities of -12.46 mm/yr and − 11.82 mm/yr respectively, compared to GNSS vertical-only rates of -2.5 mm/yr to -5.5 mm/yr. The integration of land cover data in this study proved crucial for understanding deformation mechanisms, as areas with rich organic soils and below-sea level elevations showed accelerate subsidence rates.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-55535-4.

Introduction

The Mississippi River Delta, also known as the Louisiana Delta, is a vast, dynamic region located in southeastern Louisiana where the Mississippi River meets the Gulf of Mexico. This Deltaic system, particularly the Bird’s Foot Delta, is characterized by its finger-like projections formed by sediment deposition as the river branches into multiple distributaries. It is one of the most active and ecologically rich deltas in North America, playing a vital role in coastal protection, wildlife habitat, and economic activity. This crucial Delta is highly vulnerable to land subsidence and threats from sea-level rise.

Subsidence in the Mississippi Delta is mainly caused by the deformation of deeper Pleistocene and older layers, not Holocene sediment compaction, except near the surface. GPS and strainmeter data support this and show a mean subsidence rate of 2.5 mm/yr¹. This region is moving southward at a rate of 2.2 ± 0.6 mm/yr, with subsidence rates reaching a mean of 5.2 ± 0.9 mm/yr relative to Earth’s center of mass². The Louisiana coastline is rapidly retreating due to sea-level rise and land subsidence, with hurricanes intensifying the process. For example, the Caminada-Moreau headland experienced an average shoreline retreat exceeding 21 m per year during frequent hurricane activity³. The special patterns of sediment deposition and erosion in the Delta are influenced by suspended sediment concentration and flow conditions. Advanced remote sensing techniques have been developed to map these sediment deposition patterns, providing insights into the Delta’s morphodynamics⁴. Remote sensing and InSAR (Interferometric Synthetic Aperture Radar) time series analysis have become pivotal in monitoring land deformation and identifying geohazard hotspots. InSAR data processing provides detailed insights into the spatiotemporal patterns of land subsidence and uplift, which are crucial for urban planning and disaster mitigation. Multi-sensor SAR integration ensures precise, continuous surface tracking, making InSAR a powerful tool for early warning and hotspot analysis. Global researchers from different countries have used the remote sensing technique of the InSAR time series for land deformation monitoring. InSAR Sentinel-1 time series data spanning 10 years combining with LiDAR & UAV used by the researcher for monitoring ground motion in central Mississippi region⁵. InSAR time series monitoring in Xi’an, China, reveals that land subsidence, driven by groundwater withdrawal and urban development, has caused subsidence in some areas while uplift in others, indicating potential geohazards^6–8. InSAR analysis revealed ongoing subsidence in reclaimed areas of Urayasu, Japan, caused by both natural and anthropogenic factors. The maximum subsidence rate was found to be 30 mm/year, demonstrating the application of InSAR in the coastal regions⁹. For land deformation analysis, by integrating InSAR time series with independent component analysis (ICA), researchers effectively distinguished between seasonal and long-term ground deformation, connecting these patterns to fluctuations in groundwater levels¹⁰¹¹., linked the natural processes like evaporation and landfill compaction with countrywide InSAR monitoring, identified areas of subsidence and uplift with deformation rates up to 35 mm/year. The integration of InSAR with other geospatial data and advanced modeling techniques can enhance the understanding of deformation mechanisms. Its application across diverse geographic regions demonstrates its versatility and effectiveness in providing critical data for urban planning and disaster risk reduction¹².

Among the different InSAR time series analyses, the Small Baseline Subset (SBAS) time series method is a powerful technique used in the field of remote sensing to monitor ground deformation over time. This method leverages synthetic aperture radar (SAR) data to generate deformation time-series results, which are crucial for understanding and managing the impacts of natural and human-induced surface changes. SBAS is particularly effective in regions characterized by significant ground displacement, including coastal zones, mining sites, volcanic regions, and tectonically active areas. The SBAS technique Leverages SAR data from multiple satellite platforms to generate a detailed time series of ground deformation. This integration allows for the monitoring of both large-scale and localized deformations with high accuracy^13,14. The SBAS time series can be combined with pixel offset tracking to monitor three-dimensional deformations. Providing insights into the directional movement as well as subsidence/uplift¹³. This method incorporates atmospheric delay instances to enhance the detection of sudden ground surface changes¹⁵. The InSAR SBAS is capable of analyzing deformation over large areas, for example, a 60,000 km ² test site in Nevada, demonstrating its effectiveness in temporal deformation over an extensive region¹⁶. The SBAS method provides high accuracy in deformation measurements, achieving sub-centimeter precision, which is beneficial for detailed deformation analysis, such as infrastructure monitoring¹⁷. The SBAS approach can fill temporal gaps in data by integrating different SAR acquisition modes, such as combining strip map and ScanSAR data for continuous monitoring even with irregular data availability¹⁷. There are challenges associated with data processing and interpretation. This method requires sophisticated algorithms and computational resources, especially when dealing with large datasets and complex terrains.

Many InSAR-based subsidence studies in deltaic environments have not fully integrated land cover, land use information, which limits the understanding of how different surface types of influence or experience subsidence. Deltas often have very subtle topographic gradients and are subject to flood risk due to land elevation, sediment deposition, or subsidence. Making accurate deformation measurements across all land cover types is a critical requirement for coastal hazard assessment. A fundamental challenge in InSAR analysis in these settings is the rapid temporal decorrelation of radar signals over vegetated surfaces, which degrades interferometric coherence and introduces spatially variable measurement uncertainty. Conventional approaches apply uniform coherence threshold across entire study domains fail to account for the inherently different scattering properties of diverse land cover types resulting in the exclusion of potentially informative low coherence pixels in vegetated areas or the retention of unreliable measurements in decorrelation-prone environments.

In the Mississippi Delta region, preliminary Sentinel-1 SBAS results found that much of the subsiding marsh areas had too low coherence for C-band interferograms to develop reliable deformation mapping¹⁸. Researchers have suggested several ideas, such as using longer-wavelength SAR(L-band), which maintains phase coherence better over vegetation, or increasing the temporal sampling intervals to reduce decorrelation.

This study addresses the coherence limitation within the existing Sentinel-1 C-band framework by implementing a land cover class-specific coherence threshold filtering strategy integrated with SBAS-InSAR time series analysis. The Annual National Land Cover Database (NLCD)2021 was incorporated to reclassify the study area into four broad categories for the Mississippi river delta areas- each exhibiting distinct interferometric coherence characteristics. By applying differentiated coherence threshold to each land cover category, the approach retains only pixels with sufficient phase stability for reliable deformation measurement within their respective land cover context, producing LULC-specific-line-of-sight (LOS) velocity maps which characterize deformation patterns across Urban, Baren/pasture, and Marsh/agriculture environments independently. The derived InSAR SBAS LOS velocities are validated against independently operated NOAA Continuously Operating Reference Stations (CORS) GNSS measurements at two stations (BVHS and GRIS) within the study domain. The coherence driven, land cover-conditioned filtering methodology directly addresses one of the principal limitations of the deltaic InSAR subsidence analyses and enhances the spatial deformation estimation across the Lower Mississippi River Delta.

Study area

The lower Mississippi River Delta, also known as the Birdsfoot Delta, is a vital deltaic region located in southeastern Louisiana, USA, where the Mississippi River meets the Gulf of America. The geographic extent of the center point of the study area is marked by longitudes 90.4866°W-88.8795°W and latitudes 28.9027°N-30.1401°N with a total of 22,043 km² (Fig. 1). This region’s average elevation is less than 1 m above sea level, with many areas even below sea level, making it vulnerable to subsidence and flooding due to climate change and sea-level rise. The lower Mississippi River Delta constitutes a priority area for InSAR-based time series land deformation monitoring due to the complex natural compaction, sediment loading and anthropogenic modification occurring across its diverse land cover types.

Fig. 1.

Lower Mississippi River (Birdsfoot) Delta study area. This map was created by the first author (R.B.) using QGIS (version 3.34; https://qgis.org). Basemap: © OpenStreetMap contributors (https://www.openstreetmap.org/copyright), accessed via the QGIS Quick Map Services plugin, licensed under the Open Data Commons Open Database License.

Within the deltaic landscape, three distinct deformation zones, three distinct deformation zones were delineated for analysis (Fig. 1). Zone A (Birdsfoot Delta) is the active delta front of the Mississippi River delta, extending from Venice through Pilot town to Port Eads. Zone B (Port Sulphur-Buras corridor) represents the intermediate deltaic plain with engineered levee systems and surrounding marshlands, leading to heterogeneous land cover conditions. Zone C (Grand Isle-Galliano Coastal zone) covers the western portion of the study area. This zone contains a mix of developed coastal infrastructure, barren and pasture lands, and extensive marsh environments, providing diverse surface scattering conditions of InSAR analysis. This SBAS-InSAR study investigates land cover-dependent deformation characteristics within these zones using coherence-based LULC filtering.

Datasets

For this study, ascending Synthetic Aperture Radar (SAR) datasets were acquired from ASF Vortex (Alaska Satellite Facility, https://search.asf.alaska.edu/) using on-demand HyP3 processing, covering the period from January 2020 to December 2024 (Table 1). The Small Baseline Subset (SBAS) InSAR time series method was applied to monitor and analyze land deformation dynamics. Although combining the ascending and descending path datasets gives a better 3d composition of the surface¹⁹. The analysis relied exclusively on ascending-track Sentinel-1 data, comprising 461 interferometric image pairs. Under the assumption of vertical deformation, the ascending line-of-sight (LOS) velocity provides a reliable proxy for vertical deformation rates without requiring ascending–descending decomposition.

Table 1.

Data description used in this study.

Parameters	Description
Sar satellite	Sentinel-1 A
Beam mode	Interferometric Wide (IW)
Orbit direction	Ascending
Track path number	165
Frame number	90
Polarization	VV + VH
Wavelength	5.546 cm (C-band)
Heading	−12.56 º
Repeat cycle	12 days
Time span	2020/01/01- 2024/12/31
Number of interferometric pairs	461
Processing Platform	ASF HyP3 (GAMMA)
InSAR time series software	MintPy

Data Type for GIS	Name	Resolution	Source
Land cover map	NLCD 2021 Land cover	10m spatial	https://www.mrlc.gov/data

These interferometric pairs were submitted to the HyP3 online platform for InSAR data processing. The number of looks was set to 20 × 4, allowing results with a pixel spacing of 80 m. After downloading, the processed products were manually checked the interferograms one-by-one to ensure their completeness and estimated the data error of each interferometric pair to ensure their availability.

The interferometric network comprised image pairs carefully selected based on their perpendicular baselines, constrained within ± 250 m, optimizing coherence and minimizing decorrelation effects over the study period (Fig. 3). The network spans (Fig. 2) from 2020 to 2024 with perpendicular baselines ranging from − 250 m to + 250 m, showing a dense network of interferometric pairs.

Fig. 3.

Perpendicular baseline history over the 5 years (2020–2024), ranging from − 250 m to + 250 m.

Fig. 2.

Interferogram network of paired images (2020 to 2024).

It should be noted that the Sentinel-1 IW mode interferograms generated by HyP3 are provided in range-azimuth geometry with non-square pixels (approximately 13.9 m range × 22.5 m azimuth spacing in SLC geometry). During MintPy processing, the geocoding module resamples these data to a regular geographic grid through bilinear interpolation, inherently addressing the non-square pixel geometry. Maximum coherence (Fig. 4a) consistently reaches 0.6–0.7 throughout the time series, with notable peaks in 2021 and 2022. Coherence shows seasonal variations and some degradation in late 2024, due to environmental factors such as vegetation changes or atmospheric conditions in the coastal Louisiana Regions. The matrix (Fig. 4b) shows coherence relationships between the SAR images.

Fig. 4.

(a) Coherence histogram (2020-2024) & (b) Coherence matrix.

Land use & land cover data

The National Land Cover Database (NLCD) 2021 at 10 m spatial resolution was obtained from the Multi-Resolution Land Characteristics Consortium (https://www.mrlc.gov/data) to characterise the land cover distribution across the study area. Figure 5 displays the NLCD 2021 land cover map, showing the spatial distribution of 14 distinct land cover classes across the Mississippi River Delta. The dominant land cover types include Emergent Herbaceous Wetlands and Woody Wetlands, followed by Developed land classes concentrated along the levee corridors and in the communities of Port Sulphur, Grand Isle, and Galliano.

Fig. 5.

NLCD 2021 Land Cover Map. Data source: https://www.mrlc.gov/data. This map was created by the first author (R.B.) using QGIS (version 3.34; https://qgis.org). Basemap: © OpenStreetMap contributors (https://www.openstreetmap.org/copyright), licensed under the Open Data Commons Open Database License.

For coherence threshold-based filtering analysis, the 14 NLCD classes were reclassified into four broad land cover categories based on their expected C-band interferometric coherence behavior. (1) Developed Land (Developed Open Space, Low Intensity, Medium Intensity, and High Intensity); (2) Pasture/Hay/Barren (Pasture/Hay, Barren Land); (3) Grassland/Shrub/Crops (Grassland/Herbaceous, Shrub/Scrub, Cultivated Crops); and (4) Forest/Wetland/Marsh (Deciduous Forest, Evergreen Forest, Mixed Forest, Woody Wetlands, Emergent Herbaceous Wetlands). This reclassification was guided by the distinct radar scattering properties of each category. For example, Developed Land exhibits stable, high-amplitude returns from built structures; Pasture/Hay/Barren provides moderately stable surface scattering; Grassland/Shrub/Crops introduces moderate temporal decorrelation due to vegetation growth cycles; and Forest/Wetland/Marsh produces the most variable scattering conditions due to canopy interaction, seasonal biomass fluctuations. These four categories formed the basis for the land cover class-specific coherence threshold filtering applied to the SBAS-InSAR LOS velocity results.

Methodology

The workflow for this study, illustrated in the Fig. 6 flowchart, consists of four main steps: (1) Data acquisition and SBAS Interferogram Generation from ASF Vertex, (2) Time series analysis using MintPy (Miami InSAR Time-Series), and (3) Land cover class-specific threshold filtering for LULC-based LOS velocity mapping, and (4)Validation against NOAA CORS GNSS stations.

Fig. 6.

Methodology flow chart of the study.

Step 1: Data acquisition and SBAS interferogram generation

Sentinel-1 Interferometric Wide (IW) mode SAR data acquired using the Alaska Satellite Facility’s (ASF) Vertex platform. The on-demand Hyp3 system performed co-registration, filtering, phase unwrapping, and analysis using InSAR GAMMA processing method. A total of 461 interferometric pairs were generated for the ascending path from January 2020 to December 2024. The Hyp3 SBAS job outputs included unwrapped phase, coherence maps and geometry files (incidence angle, look vectors, and DEM-aligned raster). The factor was set to 20 × 4 (range × azimuth), yielding a pixel spacing of approximately 80 m. After downloading the processed products, each interferogram was manually inspected for data quality prior to ingestion into MintPy.

Step 2: MintPy time series analysis

The core processing workflow implemented advanced time series analysis using the Mintpy framework. Mintpy’s smallbaselineapp.py was used to invert the line-of-sight velocity deformation timeseries from the HyP3 stack. The processing configuration was as follows:

Interferometric Network Configuration: The interferometric pairs were generated through ASF HyP3 with temporal baselines up to approximately 60 days (5 Sentinel-1 repeat cycles at the 12-day revisit interval). During MintPy network modification, a minimum coherence threshold of 0.12 was applied for network selection, with a minimum redundancy of 3 and a maximum of 15 connections per acquisition to ensure robust phase closure. This short-temporal-baseline strategy was intentionally adopted to maximize coherence retention in the vegetation-dominated coastal environment, where longer temporal separations lead to rapid decorrelation¹⁷.

Reference Point Selection: The reference point was automatically selected based on a minimum coherence criterion, ensuring that the reference pixel resides on a temporally stable scatterer. The selected reference point is located at 29.336561°N, −89.406648°W.

Phase Corrections: Tropospheric delay correction was applied using the PyAPS module with ERA5 global atmospheric reanalysis data, which provides spatially and temporally resolved estimates of tropospheric stratified and turbulent delay components. A linear phase ramp removal was applied to mitigate residual long-wavelength orbital errors while preserving the regional subsidence signal. A linear model was selected over quadratic deramping to avoid inadvertently removing legitimate large-scale deformation gradients characteristic of deltaic subsidence. Topographic residual correction was performed using a second-order polynomial fit to account for DEM errors propagating into the displacement estimates.

Network Inversion and Velocity Estimation: The SBAS network inversion employed variance-based weighting, which assigns higher weights to interferograms with lower phase variance, thereby reducing the influence of noisy pairs on the final velocity solution. A minimum temporal coherence threshold of 0.20 (minTempCoh ≥ 0.2) was applied to the inverted time-series to exclude pixels where phase information is unrecoverable. The initial interferometric dataset contained pixels with temporal coherence spanning the full range, including values well below 0.2 corresponding to open water, densely inundated marshlands, and areas of persistent decorrelation. This threshold represents the lower bound below which no meaningful displacement signal can be extracted and was applied to retain the maximum possible spatial coverage while ensuring measurement reliability. The residual noise assessment employed a linear ramp model consistent with the deramping strategy, and a residual threshold of 2.0 was applied to identify and flag anomalous pixels. The resulting output products include geocoded cumulative displacement time-series and mean LOS velocity maps for the study period 2020–2024, with velocities expressed in mm/year.

Step 3: Land cover class-specific coherence threshold filtering

Following the MintPy SBAS time series analysis, a land cover class-specific coherence threshold filtering strategy was implemented to produce LULC-based LOS velocity maps. This step constitutes the principal methodological contribution of this study, addressing the challenge of spatially heterogeneous coherence conditions across diverse deltaic land cover types. The land cover class-specific coherence analysis (Fig. 7) provides the empirical foundation for the filtering thresholds. Based on these statistical coherence distributions, the following class-specific average spatial coherence thresholds were applied to the MintPy velocity output using raster-based conditional filtering:

Fig. 7.

Land cover class-specific coherence analysis: (a) Average spatial coherence per interferogram by land cover class, (b) Distribution of average spatial coherence by land cover class, (c) Monthly average spatial coherence by land cover class, error bars in (c) represent ± standard deviation, (d) Mean coherence time-series (2020–2024).

The filtering (Table 2)was implemented through raster conditional equations applied in GIS, where for each land cover category, the LOS velocity value was retained only if the corresponding pixel’s average spatial coherence equaled or exceeded the class-specific threshold; otherwise, the pixel was assigned a no-data value (−9999). The NLCD 2021 reclassified land cover mask was used to spatially assign each pixel to its corresponding land cover category.

Table 2.

Coherence threshold values applied for different land cover types.

Land cover reclassification	Coherence threshold (avgSpatialCoh)	Justification
Developed land	≥ 0.45	Stable scatterers from built structures support a higher threshold, remaining only the most reliable urban pixels. The boxplot IQR upper quartile (0.40) and per-interferogram scatter confirm that Developed Land routinely exceeds 0.45 in most pairs.
Pasture/Hay/Barren	≥ 0.35	Moderately stable surfaces with intermediate scattering properties. The threshold is set at the approximate IQR upper quartile for this class, balancing spatial coverage against measurement reliability.
Grassland/Shrub/Crops/Agriculture	≥ 0.30	Vegetation-induced decorrelation lowers coherence; a threshold of 0.30 retains pixels above the median coherence for this class while excluding the most decorrelated measurements dominated by dense canopy interaction.
Forest/Wetland/Marsh	≥ 0.25	The lowest threshold reflects the inherently challenging coherence conditions in vegetated wetland environments. Setting the threshold at 0.25 above the minimum temporal coherence of 0.2 (applied during MintPy inversion) retains marsh and wetland pixels that maintain sufficient phase stability for deformation measurement while acknowledging the limited C-band penetration through dense canopies.

Step 4: GNSS validation

The InSAR SBAS LOS velocities were validated against independently operated NOAA Continuously Operating Reference Stations (CORS) GNSS measurements at two stations within the study domain: BVHS (Boothville) and GRIS (Grand Isle). The GNSS NAD83 multiyear three-dimensional velocity components (vertical Vu, north Vn, east Ve) were obtained from the published NOAA CORS solution. The GNSS three-dimensional velocity vector was projected into the Sentinel-1 ascending line-of-sight (LOS) direction using the following equation:

where V(_LOS) is the projected GNSS velocity in the satellite LOS direction, Vu is the vertical velocity component, V_N is the north velocity component, V_E is the east velocity component, θ is the local incidence angle, and α is the satellite heading angle (α = −12.56° for the ascending track used in this study).

For the vertical-only projection, the simplified equation was applied:

This assumes that horizontal deformation components are negligible, which provides a first-order comparison between InSAR LOS and GNSS vertical rates.

Result

The line of sight (LOS) velocity map (Fig. 8). SBAS-InSAR LOS velocity map (2020–2024) with temporal coherence ≥ 0.2) shows deformation rates ranging from − 43.71 mm/year to 30.23 mm/year with a mean subsidence rate of −14.36 mm/year. The limited spatial coverage reflects the challenging coherence conditions inherent to the vegetated deltaic environment, where open water and persistently decorrelated, surfaces were excluded by the minimum temporal coherence threshold (minTempCoh ≥ 0.2). The LOS velocity map (Fig. 8) reveals spatially heterogeneous deformation patterns across the Mississippi River Delta. Subsidence-indicating pixels (LOS velocity < −3 mm/yr) account for the valid measurements. The reference point, located at 29.337°N, −89.407°W near the NOAA CORS station BVHS, exhibits a coherence of 0.7 and a velocity of 0.0 mm/yr, confirming its suitability as a stable reference. The deformation field exhibits a clear spatial gradient, with subsidence rates increasing from inland developed areas toward the coastal marshlands and the active delta front, consistent with the regional pattern reported by²⁰ using PS-InSAR over the same region.

Fig. 8.

SBAS-InSAR LOS velocity map (2020–2024) with temporal coherence ≥ 0.20. The LOS velocity data were generated using MintPy (version 1.6; https://github.com/insarlab/MintPy). This map was created by the first author (R.B.) using QGIS (version 3.34; https://qgis.org).

Zone A (Birdsfoot Delta)

Zone A encompasses the active delta front extending from Venice through Pilottown to Port Eads. This zone exhibits the most intense and spatially variable deformation across the study area. The distributary channels of the Birdsfoot Delta show a complex mixture of subsidence and localized uplift signals. Along the main distributary levees, moderate subsidence rates of −10 to −25 mm/yr are observed. Localized uplift signals (+ 5 to + 15 mm/yr) are detected at the tips of the distributary channels, potentially attributable to active sediment deposition at the delta front. The panel for Zone A (Fig. 8) reveals that the deformation pattern closely follows the geomorphological framework of the Birdsfoot.

Zone B (Port Sulphur–Buras corridor)

Zone B occupies the intermediate deltaic plain along the lower Mississippi River channel. Developed areas along the river Port Sulphur, Empire, and Buras show moderate subsidence rates of −5 to −15 mm/yr, with high coherence pixels concentrated along the levee alignments and built structures.

Zone C (Grand Isle–Galliano Coastal Zone)

Zone C represents the western portion of the study area and exhibits the strongest subsidence signals observed in the entire study domain. The western coastal marshlands, extending from Galliano and Golden Meadow toward Port Fourchon and Grand Isle, show concentrated subsidence with LOS velocities frequently exceeding − 20 mm/yr. The intensity of subsidence in Zone C reflects a combination of organic marsh compaction, hydrocarbon extraction-related consolidation, and coastal erosion processes operating in this heavily industrialized coastal environment. The barrier island of Grand Isle and the coastal infrastructure at Port Fourchon show more moderate subsidence rates (−5 to −15 mm/yr), consistent with their built-on-compacted-fill construction.

LULC-based LOS velocity results

The land cover class-specific coherence threshold filtering produced four LULC-specific LOS velocity maps (Figs. 9, 10, 11 and 12), each retaining only pixels exceeding the class-specific coherence threshold. The Urban LOS velocity map (avgSpatialCoh ≥ 0.45) retains the highest-confidence measurements, concentrated along the levee corridor from Port Sulphur to Venice and in the communities of Galliano, Golden Meadow, and Grand Isle. Urban pixels predominantly exhibit moderate subsidence rates of −5 to −15 mm/yr, with localized areas near the Grand Isle coastal infrastructure showing rates approaching − 20 mm/yr. The sparse but high-coherence character of this map confirms that developed land provides the most reliable scattering targets for C-band InSAR in the deltaic environment. The Barren/Pasture LOS velocity map (avgSpatialCoh ≥ 0.35) extends the spatial coverage capturing measurements across pasture lands and barren surfaces. A notable concentration of Barren/Pasture pixels is visible in the Fig. 9 where agricultural and pasture lands show subsidence rates of −8 to −18 mm/yr. The Marsh/Agriculture LOS velocity map (avgSpatialCoh ≥ 0.25) provides the broadest spatial coverage, extending deformation measurements into the wetland and marsh environments. This map reveals the most spatially continuous subsidence patterns, particularly in Zone C where extensive marshlands show widespread subsidence exceeding − 15 mm/yr. The Integrated LULC-Based LOS velocity map combines all four filtered outputs, representation of deformation patterns across the study area (Figs. 9, 10, 11 and 12). The integrated map demonstrates that the LULC-based filtering approach increases spatial deformation information compared to any single land cover category, while maintaining measurement reliability through class-appropriate coherence thresholds.

Fig. 9.

Urban Land Cover Class LOS Velocity map (avgSpatialCoh ≥ 0.45). This map was created by the first author (R.B.) using QGIS (version 3.34; https://qgis.org). Basemap: © OpenStreetMap contributors (https://www.openstreetmap.org/copyright).

Fig. 10.

Barren & Pasture Land Cover Class LOS Velocity map (avgSpatialCoh ≥ 0.35). This map was created by the first author (R.B.) using QGIS (version 3.34; https://qgis.org). Basemap: © OpenStreetMap contributors (https://www.openstreetmap.org/copyright).

Fig. 11.

Marsh & Agriculture Land Cover Class LOS Velocity Map (avgSpatialCoh ≥ 0.25). This map was created by the first author (R.B.) using QGIS (version 3.34; https://qgis.org). Basemap: © OpenStreetMap contributors (https://www.openstreetmap.org/copyright).

Fig. 12.

Coherence Threshold Filtered LULC Class Based LOS Velocity Map. The LOS velocity data were generated using MintPy (version 1.6; https://github.com/insarlab/MintPy). This map was created by the first author (R.B.) using QGIS (version 3.34; https://qgis.org). Basemap: © OpenStreetMap contributors (https://www.openstreetmap.org/copyright).

Validation

The InSAR SBAS LOS velocities were validated against independently operated NOAA Continuously Operating Reference Stations (CORS) GNSS measurements at two stations within the study domain (Fig. 13). BVHS (Boothville, 29.34°N, −89.40°W), located near the SBAS reference point in Zone B, and GRIS (Grand Isle, 29.26°N, −89.96°W), situated on the Grand Isle in Zone C. The spatially averaged InSAR LOS velocities were extracted at each CORS station location using a 200 m radius buffer zone surrounding the station coordinates and computing the mean of all valid pixels within this neighborhood. This yields − 12.46 mm/yr at BVHS (median: −12.46, σ = 5.35 mm/yr) and − 11.82 mm/yr at GRIS (median: −11.82, σ = 5.69 mm/yr). These InSAR velocities are expressed relative to the SBAS reference point (29.337°N, −89.407°W), at which the velocity is zero by definition. The same 200 m buffer extraction applied at the reference point location yields a mean velocity of −1.14 mm/yr, indicating that the immediately surrounding pixels exhibit mild subsidence, which provides a measure of the local velocity gradient around the reference point.

Fig. 13.

NOAA continuously operating reference stations (CORS) and SBAS Reference Point Layout. Base map source: NOAA CORS Map (https://geodesy.noaa.gov/CORS_Map/), a publicly available U.S. federal government product. Station markers, SBAS reference point annotation, and north arrow were added by the first author (R.B.) using Microsoft Word.

The GNSS vertical subsidence rates (−4.2 mm/yr at BVHS, −7.6 mm/yr at GRIS) confirm that both stations are experiencing active subsidence, consistent with the InSAR observations. However, the InSAR SBAS LOS velocities are approximately 2–3 times larger in magnitude than the GNSS-projected LOS velocities (Fig. 14). Several factors contribute to this systematic discrepancy.

Fig. 14.

InSAR SBAS LOS velocities comparison with GNSS vertical-only and GNSS 3D LOS-projected velocities at NOAA CORS stations.

The GNSS-derived velocities and their LOS projections are summarized in Table 3.

Table 3.

Comparison of InSAR SBAS LOS velocities with GNSS-derived LOS projections.

Station name	InSAR LOS velocity (mm/yr)	GNSS NAD 83 Vu(mm/yr)	GNSS NAD 83 Vn(mm/yr)	GNSS NAD 83 Ve(mm/yr)	Incidence angle θ (°)
GRIS	−11.82	−7.6	0.1	0.7	37.64
BVHS	−12.46	−4.2	0.3	1.1	41.05

* Satellite heading angle (α) = − 12.56°. * Vu, Vn, and Ve are the GNSS vertical, north, and east velocity components, respectively.

The InSAR SBAS technique using distributed scatterers (DS) measures the total deformation from all layers at the ground surface, including the shallowest sediment compaction. In contrast, GNSS receivers at CORS stations are typically mounted on deep foundation structures and measure relatively deep subsidence that excludes the shallow compaction component. Shallow subsidence in the uppermost 1–2 m of organic marsh sediments can contribute substantially to total subsidence in coastal Louisiana^21,22 and this component would be captured by InSAR but not by deep-foundation GNSS. The spatial averaging of InSAR velocities around the CORS station locations encompasses pixels from multiple land cover types, including surrounding marsh and wetland areas that may exhibit higher subsidence rates than the station structure itself. The standard deviations of InSAR velocities within the extraction neighborhoods (Std σ = 5.35 mm/yr at BVHS, Std σ = 5.69 mm/yr at GRIS) reflect this spatial heterogeneity. Third, the different observation periods (InSAR: 2020–2024; GNSS: multi-year NAD83 solution) may introduce temporal discrepancies if subsidence rates are non-stationary.

Despite the absolute magnitude difference, the GNSS measurements independently confirm active subsidence at both validation sites. The relative patterns are comparable with InSAR LOS rates at both stations despite the absolute magnitude difference. The subsidence pattern is consistent with the LOS projection geometry, as the larger incidence angle at BVHS (41.05°) reduces the vertical sensitivity compared to GRIS (37.64°). These validation results are consistent with independent findings by²⁰, who reported maximum vertical subsidence rates of approximately 12 mm/yr at the Bird’s Foot area using PS-InSAR with Sentinel-1 data, and with the Louisiana CPRA 2023 Coastal Master Plan subsidence rate range of 4–12 mm/yr across the delta plain²³.

Discussion

The use of C-band Sentinel-1 data for deltaic subsidence monitoring is well-established in the published literature, despite the acknowledged coherence limitations over vegetated wetlands²⁰. Previous studies demonstrated the viability of Sentinel-1 C-band InSAR for systematic subsidence mapping across 40 major river deltas globally, including Mississippi^24,25. The SBAS-InSAR time series analysis in this study reveals pronounced spatial heterogeneity in land deformation across the lower Mississippi River Delta, with LOS velocities spanning − 43.71 to + 30.23 mm/yr and indicating active subsidence. The three delineated analysis zones exhibit deformation patterns that are spatially coherent with the underlying geomorphological and sediment deposition pattern of the delta. This study demonstrates that land cover class-specific coherence threshold filtering enhances the reliability of SBAS-InSAR deformation measurements in the heterogeneous deltaic environment of the lower Mississippi River Delta. The LOS velocity field provides robust evidence that the observed signals represent genuine displacement processes rather than coherence-related artefacts. The core contribution of this work lies in addressing a persistent methodological limitation in deltaic InSAR studies. Conventional approaches apply uniform coherence thresholds that systematically exclude the marsh and wetland surfaces experiencing the most rapid subsidence or retain unreliable measurements from decorrelated environments. The adaptive LULC-based filtering resolves this trade-off by calibrating the coherence acceptance criterion to the empirically characterized scattering properties of each land cover category (Developed Land ≥ 0.45, Pasture/Hay/Barren ≥ 0.35, Grassland/Shrub/Crops ≥ 0.30, Forest/Wetland/Marsh ≥ 0.25). The discrepancy between InSAR SBAS LOS velocities (−12.46 mm/yr at BVHS, −11.82 mm/yr at GRIS) and GNSS-projected LOS rates (−3.83 and − 6.42 mm/yr respectively) warrants careful interpretation. This 5–9 mm/yr difference is attributable to the fundamentally different measurement depths of the two techniques. InSAR distributed scatterers on natural or shallow-foundation surfaces capture total subsidence from all layers, including organic and unconsolidated sediment compaction within the uppermost strata. This finding carries significant implications for coastal restoration planning and flood risk assessment. The transferability of this methodology extends beyond the Mississippi Delta. Any deltaic or coastal environment characterized by heterogeneous land cover and spatially variable coherence conditions. The smooth velocity gradient from levee corridor to adjacent swamp confirms that the class-specific thresholds do not introduce artificial discontinuities, validating the physical consistency of the filtering approach.

Limitations

The present study acknowledges several limitations that may influence the interpretation of the SBAS-InSAR results. The Sentinel-1 C-band wavelength (5.546 cm) scatters off vegetation canopy surfaces rather than penetrating to the underlying soil, meaning that deformation estimates in marsh and wetland environments. The displacement at the canopy–soil interface with penetration limited to a few centimeters. This restricts direct comparison with subsurface deformation measurements from borehole extensometers. The overall valid pixel reflects the low coherence conditions across the deltaic landscape, where open water, persistent inundation, and dense vegetation preclude phase recovery over most of the study domain. The LULC-based coherence thresholds applied in this study were empirically derived from the spatial coherence distributions of the four reclassified land cover categories within the study area. They were not independently validated against co-located ground-truth deformation measurements across all land cover types. The NLCD 2021 land cover classification represents a single temporal data and does not account for land cover changes occurring during the 2020–2024 study period. The ERA5 tropospheric correction may not fully capture localized tropospheric anomalies in the humid subtropical coastal environment, potentially contributing residual atmospheric phase to the velocity estimates. Finally, the validation relied on only two NOAA CORS GNSS stations within the study domain, constraining the spatial representativeness of the InSAR–GNSS comparison.

Conclusions and future work

This study presents a SBAS-InSAR time series analysis of land deformation across the lower Mississippi River Delta using ascending-path Sentinel-1 interferometric pairs spanning 2020–2024, processed through the MintPy framework. The analysis introduces a land cover class-specific coherence threshold filtering methodology that extends deformation measurement capability from isolated infrastructure targets into the vegetation-dominated marshlands and wetlands constituting most of the deltaic landscape. The LULC-based adaptive coherence filtering, applying empirically calibrated thresholds to four reclassified land cover categories (Developed Land ≥ 0.45, Pasture/Hay/Barren ≥ 0.35, Grassland/Shrub/Crops ≥ 0.30, Forest/Wetland/Marsh ≥ 0.25), produces spatially continuous deformation maps that preserve velocity field continuity across land cover boundaries. The LOS velocity field exhibits deformation patterns that are spatially consistent with the geological and geomorphological framework of the delta. Zone A (Birdsfoot Delta) displays active sediment deposition signals. Zone B (Port Sulphur–Buras) reveals a sharp deformation boundary at the levee to swamp transition. Zone C (Grand Isle–Galliano) records the strongest sustained subsidence, exceeding − 20 mm/yr across extensive organic marshlands in the Barataria Basin margin. Validation against NOAA CORS GNSS stations at BVHS and GRIS confirms active subsidence at both sites. The methodology presented here addresses a recognized gap in deltaic InSAR applications by providing a reproducible, physically grounded framework for optimizing deformation measurement coverage across heterogeneous coastal land cover.

In future research, the integration of the LULC-based LOS velocity maps with flood susceptibility models and coastal inundation projections would enable spatially explicit identification of areas where active subsidence intersects with elevated flood exposure, providing directly actionable information for coastal protection prioritization. The forthcoming NISAR mission, providing L-band SAR data with superior vegetation penetration and a 12-day repeat cycle, offers the opportunity to validate C-band-derived velocities against L-band measurements in the same study area and to extend reliable deformation coverage into the currently decorrelated dense marsh environments. Additionally, the application of advanced multi-temporal techniques such as distributed scatterer interferometry within the LULC-conditioned framework may further increase measurement density in borderline-coherence environments.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

R.B. designed the study, collected and curated the data, performed the SBAS-InSAR analysis, and prepared the original draft of the manuscript, including figures and primary interpretation. R.S. contributed to the discussion and result validation. R.S. and S.K. reviewed the manuscript for technical accuracy and clarity. All authors participated in reviewing and approving the final version of the manuscript.

Funding

This research was supported by the Gulf Research Program of the National Academies of Sciences, Engineering, and Medicine under award number [SCON – 10000883]. The content is solely the responsibility of the authors and does not necessarily represent the official views of the Gulf Research Program or the National Academies of Sciences, Engineering, and Medicine.

Data availability

The Sentinel-1 SBAS datasets used in this study were obtained from the Alaska Satellite Facility (ASF) VERTEX portal (https://search.asf.alaska.edu/). The land elevation and land cover datasets for the Mississippi Delta region were obtained from the U.S. Geological Survey (USGS) National Map (https://apps.nationalmap.gov/downloader/) and National Land Cover Database (NLCD) repositories (https://www.mrlc.gov/data). All data were processed and analyzed using the open-source Mintpy (Miami InSAR Time series in Python) framework, and no commercial software was used in this research. The processed SBAS-InSAR datasets used in this study are publicly available at https://doi.org/10.5281/zenodo.20493999.

Declarations

Competing interests

The authors declare no competing interests.