Background

The Earth's surface is never truly still. From gradual tectonic drift to sudden ground subsidence, these subtle movements can have serious consequences for infrastructure, ecosystems, and human safety. Monitoring such changes is essential, not only to understand the Earth's behavior but also to reduce risk and prepare for potential hazards. For decades, tools like GNSS sensors and total station surveys have offered precise measurements. However, their reach is limited. These systems require physical installations and only capture data at specific points, leaving large areas unobserved. Today, Synthetic Aperture Radar (SAR) interferometry presents a powerful alternative. It enables accurate and wide-area ground monitoring from space, all without the need for on-site equipment.

By tracking how radar waves bounce back from the Earth’s surface over time, advanced multi-temporal InSAR techniques can now measure ground movement with millimeter-level precision. These techniques help overcome many limitations of earlier methods, such as atmospheric noise, spatial decorrelation, and even some aspects of temporal decorrelation, by using long time series stacks of satellite images. The result is continuous, wide-area monitoring without the need for any ground based equipment. So how does this actually work? It comes down to two powerful approaches at the heart of modern InSAR: Persistent Scatterer (PS) and Distributed Scatterer (DS) techniques.

Think of structures like skyscrapers, boulders, or bridges as fixed markers in a moving world. Persistent Scatterer (PS) techniques identify these stable, radar-reflective targets, objects that maintain a consistent signal over time. Because they remain unchanged for years, they serve as highly reliable reference points. This allows scientists to track even the slightest ground movement, such as the slow tilt of a dam or gradual sinking in an urban neighborhood, with remarkable precision.

Not all landscapes have stable features to rely on. Rural areas, farmlands, and forests are often covered with natural surfaces like vegetation, soil, or gravel materials that reflect radar signals inconsistently. Distributed Scatterer (DS) techniques address this by analyzing clusters of moderately coherent scatterers instead of focusing on single points. Through advanced statistical processing of dense interferogram networks, DS methods reduce noise from environmental factors such as crop changes or seasonal snow, revealing subtle ground deformation patterns. This allows InSAR to effectively monitor regions that were once considered too noisy or variable for reliable analysis.

Together, Persistent Scatterer and Distributed Scatterer methods offer a comprehensive view of ground deformation. PS techniques provide precise measurements in areas with stable reflectors, while DS methods capture broader trends across more dynamic, natural landscapes. This combined approach helps scientists monitor phenomena ranging from volcanic inflation to groundwater related subsidence, offering valuable insights for disaster management, urban development, and environmental protection.

At Solafune, we decode the Earth’s movements and turn them into decisions that matter. By using both Persistent Scatterer and Distributed Scatterer techniques, we monitor ground uplift, subsidence, and deformation across large and diverse landscapes. This allows us to detect risks early, giving industries, governments, and communities the time they need to respond with confidence. From preventing infrastructure damage in cities to protecting natural habitats from the effects of mining, our technology connects the quiet movements of the Earth with the power of informed decision making. In a world where the ground is always shifting, timely knowledge is not just an advantage, it is prevention.

How to Analyse Time Series Deformation using SAR Data?

Time series analysis of SAR data utilizes sophisticated techniques to observe and measure ground deformation at an incredibly precise millimeter scale. Traditional methods, such as Small Baseline Subset (SBAS) and Persistent Scatterer Interferometry (PSInSAR), enhance the effectiveness of interferogram networks by employing constrained baselines. These constraints, which limit temporal baselines to less than 120 days and spatial baselines to less than 300 meters, are crucial in mitigating decorrelation and atmospheric noise. Through these optimizations, these conventional methods can achieve impressive accuracy levels in the range of 1 to 2 millimeters per year, which is better compared to traditional DInSAR (Differential InSAR).

Some salient features of available techniques are listed below

• DInSAR

  • Uses just two SAR images.

  • Measures short-term deformation (≥5 cm) with 6–14 cm accuracy.

  • Suitable for rapid events like earthquakes or mining collapses.

  • Suffers from temporal/spatial decorrelation and atmospheric artifacts.

• SBAS

  • Used to measure ground deformation over large areas, particularly when deformation is distributed across the surface (e.g., landslides, subsidence, or tectonic activity)

  • It reduces spatial decorrelation by 30-50% compared to single-pair DInSAR,and enables temporal filtering to separate atmospheric artifacts from true deformation signals, achieving 1-2 mm/yr velocity accuracy versus DInSAR's 5-10 mm/yr.

  • Additionally, SBAS allows SVD-based inversion to solve phase unwrapping ambiguities across multiple acquisitions.

• PSInSAR

  • Used to monitor localized and precise ground deformation by identifying stable points (persistent scatterers) on the surface, such as buildings, rocks, or infrastructure.

  • Leverages phase-stable targets (like buildings or rocks).

    • Requires analysis of ≥25 SAR acquisitions.

    • Achieves 1–2 mm/yr velocity accuracy.

    • Corrects atmospheric distortions through temporal filtering and spatial correlation of phase residuals.

    • Reliance on linear deformation models limits detection of rapid displacements (>12 cm/year).

    • Coverage is sparse in rural areas due to scatterer scarcity.

• Hybrid Approaches that combine PSInSAR and SBAS: Combine PS and SBAS techniques to maximize total spatial coverage by utilising both permanent and distributed pixels or targets.

In this blog, we will use SBAS technique to analyse the time series surface movement of Sakurajima, an active stratovolcano located on a peninsula in Japan. Utilizing approximately eight months of Sentinel-1 Synthetic Aperture Radar (SAR) data, our main objective is to extract precise displacement information for any location within the designated Area of Interest (AOI) shown in Fig. 1. 

SBAS

Fig. 1: Location of Sakurajima active stratovolcano in Kagoshima Prefecture, Japan.

Data specifications are listed below.

• Satellite: Sentinel-1 SAR 

• Wavelength: C-Band

• Dates: September 2023 – April 2024

• No. of images: 19 

• Flight Direction: Descending

• Polarization: VV+VH

• Beam Mode: Interferrometric Wiseswath (IW)

• Relative orbit number: 163

Data Selection

Selecting the optimal SAR dataset for surface movement analysis is a crucial step. Most of the inaccuracies in the analysis will be suppressed if we select image pairs optimal requirements of spatial baseline, temporal baseline, and relative orbit of data capture.  For the surface movement (deformation or uplift)  analysis we need the time series data in repeat pass configuration, captured from the same pass (Ascending or Descending), and the same relative orbit. The B (Perpendicular Baseline) is a crucial parameter in selecting the pair of images for surface movement analysis. 

The baseline is defined as a physical distance or separation (m or km) between the instance in space from which the images have been captured. There are two types of baselines- parallel and perpendicular. We are particularly interested in perpendicular baseline (B) defined as the distance between two acquisition spots perpendicular to the satellite viewing direction.

Fig. 2: A geometry of InSAR image capturing is shown with two images taken at two different instance points P1 and P2. The satellite is travelling along the Y axis (along-track or azimuth direction), and X axis is the direction perpendicular to satellite motion (cross-track or range direction). The two satellite passes capture the same area on the ground shown by Image 1 and Image 2. The effective baseline is shown by ‘B’ which is decomposed into B⊥ and B∥. Credit: ASF

• A small perpendicular baseline (100-200 m) is crucial to maximize the coherence of an InSAR pair. Coherence is a measure of the similarity between the two SAR images (ranges between 0 and 1), and it is dependent on the perpendicular baseline. Typically, we need high coherence( greater than 0.5) to be able to trust the phase measurements. 

• An excessively large B range can lead to data inconsistencies, incomplete noise removal, and reduced accuracy in interferometric processing.

• The temporal baseline refers to the time difference between two SAR acquisitions used to create an interferogram.

• A shorter temporal baseline (e.g., 60-100 days) can help reduce temporal decorrelation, which is the loss of coherence due to changes in the Earth's surface. 

• However, a very short temporal baseline might not capture enough deformation, while a longer baseline might lead to more atmospheric and orbital errors. 

• A common practice is to use a temporal baseline of around 90-100 days, which balances the need to capture deformation and minimize decorrelation.

Critical Baseline in Interferometric SAR

• The critical baseline is the maximum perpendicular baseline beyond which interferometry is no longer feasible due to loss of coherence.

• It depends on factors such as sensor altitude, wavelength, and viewing geometry.

• For Sentinel-1, the critical baseline is approximately 5 km.

Impact on Interferometry

• Topographic Mapping:

  • Larger baselines improve topographic accuracy but reduce coherence.

  • If the perpendicular baseline exceeds 75% of the critical baseline, interferogram quality deteriorates due to noise.

• Deformation Mapping:

  • Minimizing the perpendicular baseline helps maintain coherence.

  • In practice, baselines used for deformation studies are much smaller than the critical threshold.

  • Trade-offs may exist between selecting optimal temporal and perpendicular baselines

SBAS Processing Flow

Figure shows SBAS processing workflow, let’s dive into details!

Fig.3 : SBAS Pre-processing Steps

Fig. 4: SBAS Time Series Processing

1. Generating a Network of Unwrapped Interferograms

The SBAS time series process begins with a stack of unwrapped interferograms created using pre-processing steps shown in Fig. 3. The pre-processing starts with getting the list of images that satisfy the SBAS criteria and are visualised using “Connection Graph”. 

The connection graph is a visual or mathematical representation of how SAR acquisitions (dates) are interconnected via interferograms. The connection graphs consist of “Nodes” that represent the individual SAR acquisition, “Edges” representing interferograms formed between pairs of SAR images. Based on spatial and temporal baselines, one can select the most suitable pairs for SBAS processing. A key point to remember is that, the small baselines reduce the geometric decorrelation and time gap between the acquisition minimize the temporal decorrelation. 

The formation of dense networks creates redundancy to improve phase unwrapping and error correction. Redundant interferograms allow statistical averaging to suppress atmospheric phase delays (e.g., weather effects). A regular sampling of acquisition is recommended even if some acquisitions are excluded due to large baselines. The graph forms a solvable system of equations to estimate deformation rates and temporal evolution at each pixel. Even if some SAR images are excluded, the remaining connections ensure continuous time-series recosntruction. Once the formation of pairs is finalised, then the DInSAR processing is applied to individual pairs and a stack of unwrapped phase is generated, these interferograms represent the phase difference between pairs of SAR images taken at different times. The stack is then fetched as input to the subsequent processing pipeline (Fig. 4).

2. Preprocessing the Interferograms

Several steps refine the interferogram stack before estimating displacement:

• Network modification: The interferogram network may be adjusted to improve stability and connectivity, ensuring a robust analysis.

• Selecting a reference point: Since radar measures relative displacement, a stable reference point is chosen as a baseline.

• Unwrap error correction: This step corrects any phase unwrapping errors that may have occurred.

• Network inversion: Finally, the network undergoes mathematical inversion to derive the displacement signal.

A water mask is sometimes applied to filter out water bodies, which can introduce noise due to phase inconsistencies.

Fig. 5: The SBAS connection network is shown in (a), It has additional color coded information on average spatial coherence along (Z-axis) with temporal (X-axis) and spatial baseline (Y-axis). Averaged temporal coherence is shown in (b), it shows how stable is the coherence of the pixel over time. The red box in (b) shows high coherence and the yellow box shows low coherence. By modifying the network (skip the pairs with low coherence value from analysis)  shown in (a), phase unwrapping errors and noise can be minimised (this step is optional)

3. Constructing the Raw Phase Time-Series

Once the interferogram network is refined, the raw phase time-series and temporal coherence are extracted. This represents the uncorrected phase information over time. Similar plots shown in Fig. 5 are generated by removing unwanted image pairs. After the refinement, we expect that the temporal coherence is consistently very high and there are no or very few dark regions (the region within the yellow box in Figure 5b will now have high coherence similar to the red box).

The average phase velocity is computed with following steps:

• For each pixel in every interferogram, the unwrapped phase (Φ) is converted to a deformation rate by dividing it by the temporal baseline (Δt) between the two SAR acquisitions

• Rates from all interferograms are averaged to produce a single velocity map (shown in Figure 7a)

The average phase velocity map provides a quick first-pass estimate of ground deformation without complex time-series modeling, and highlights regions of significant uplift/subsidence (e.g., volcanoes, sinkholes).

Fig. 6b and 6c are quality control maps for phase unwrapping. It identifies pixels with persistent unwrapping errors (e.g., noisy region, rapid deformation). High values in Fig. 6a show pixels with frequent unwrapping errors (e.g., vegetated areas, steep terrain) and are unreliable. Reliable pixels have lower values (shades of blue in the map (6b) shows reliable pixels). Fig. 6c shows the histogram of map shown in 6b, and it gives a total count of pixels suffering from phase unwrapping errors.

Fig. 6: (a) Average phase velocity map, (b) Number of times a pixel encounters the persistent unwrapping errors, and  (c) Histogram Tint

4. Correcting for Atmospheric and Topographic Effects

The time series deformation needs to be corrected for atmospheric, topographic and DEM related errors. Fig. 7 shows the time series of deformation before any correction applied. It accounts for orbital and geometric errors between SAR acqusitions. 

Fig. 7: The foundational time series of ground deformation derived from InSAR phase measurements. 

Several corrections are necessary to remove unwanted signals from the raw phase data:

• Tropospheric delay correction: Atmospheric disturbances, particularly in the troposphere, can distort the phase signal. These are corrected using global atmospheric models and Digital Elevation Models (DEM).

• Phase deramping: Any long-wavelength phase trends are removed to focus on local displacement.

• Topographic residual correction: Errors due to inaccurate DEM data are adjusted to improve precision.

By incorporating the ERA5 atmospheric data from ECMWF, tropospheric delays (caused by water vapor/ pressure variations) in InSAR deformation phase are mitigated that results in improved signal-to-noise ratio and reduction of noise in the deformation time series. The output after tropospheric correction is shown in Fig. 7 . Note the difference between Fig. 6 and 7, we can see changes after correction.

Fig. 8: Time series deformation after the atmospheric correction using ERA5 data. Note that, 2024/01/09 is taken as a reference point, hence it will have zero value.

The residual orbital errors or long-wavelength noise manifest as spatial ramps in deformation maps, removing these isolated localized deformation signals. A nest-fit ramp (linear or quadratic) is subtracted from each interferogram of an atmospherically corrected deformation time series to generate the ramp removal output shown in Fig. 9.

Fig. 9:  Ramp removal correction appointed on atmospherically corrected time series data.

Inaccuracies in the DEM cause phase residuals proportional to the perpendicular baseline. Correcting these refines elevation-related phase contributions. By further minimizing the DEM related inaccuracies, further reduction in noise, especially in regions with steep topography or poor DEM resolution can be achieved as shown in Fig. 10.

Fig. 10: DEM error adjusted deformation time series 

5. Evaluating Noise and Estimating Displacement

After applying the necessary corrections, the signal is further refined:

• Noise evaluation: Any remaining noise in the dataset is assessed.

• Noisy acquisitions filtering: Particularly noisy acquisitions are identified and potentially removed.

  • Select the date with the minimum RMS value as the optimal reference date.

  • Detect the noisy acquisitions with RMS beyond the outlier detection threshold.

Fig. 11: Detection of noisy acquisitions using outlier detection threshold (shown as dotted horizontal line). All the RMS values are below the threshold and hence no outlier is detected.

6. Extracting Displacement and Velocity Information

The final steps involve calculating displacement over time and estimating ground velocity:

• Displacement time-series: This captures how the ground moves at different time intervals.

• Velocity estimation: Using the displacement time-series, the average velocity of ground deformation is determined.

At the end of the SBAS workflow, we obtain average velocity maps (shown in Fig. 12) and detailed time-series displacement data 

Fig. 12: Velocity and its standard deviation without any correction is shown in (a) and (b), respectively. The final corrected output is shown in (c) and (d). 

Visualisation of Results

The final result is stored in .kmz format compatible with Google Earth Engine. It is divided into three grids- 20 by 20, 5 by 5, and 2 by 2 as shown in Fig. 13.

Fig. 13: Final deformation time series is shown at different zoom levels 

These grid sizes help in visualizing deformation trends at different scales. In a KMZ file, different layers may correspond to different resolutions, allowing users to toggle between broad and detailed deformation patterns. These numbers represent the size of the grid cells (in kilometers) used to average the deformation data. A 20 by 20 grid provides lower resolution data for regional studies, while a 2 by 2 grid offers higher resolution for local analysis. Different grid sizes allow users to view deformation patterns at various scales.

Reasons for Data Gaps in 2 km × 2 km Resolution:

• Coherence Loss: In areas with dense vegetation, water bodies, or rapid surface changes, SAR coherence decreases. This results in missing data due to these low-coherence regions lacking reliable phase information.

• Atmospheric & Geometric Effects: Phase consistency is impacted by tropospheric delays, orbital errors, and terrain-related distortions. These effects become more pronounced at smaller grid scales, leading to data gaps.

• Sparse Valid Pixels: High-resolution grids need a dense distribution of valid pixels. Fewer reliable pixels are available in areas with low coherence, causing gaps that are less noticeable in coarser grids (e.g., 20 × 20 km) due to spatial averaging.

• Filtering & Masking: Unreliable deformation estimates are removed to maintain accuracy in the processing pipeline. As higher-resolution grids are more sensitive to noise, this leads to increased data gaps after filtering.

Every point in the .kmz file stores the time series deformation information as shown in Fig. 14.

Fig. 14: A plot of Line of Sight (LOS) displacement vs Time. Additional information on the location, and mean LOS velocity, coherence, and cumulative displacement is also displayed. 

These data points are more than just numbers. They show how the ground has moved over time in a quiet and gradual way. This becomes especially important in volcanic regions where even small changes at the surface can reveal early signs of activity below. A clear shift from the long term trend may suggest magma movement, rising pressure, or changes in the underground structure.

Such subtle motion is often invisible to the naked eye but becomes clear through InSAR analysis. By using Persistent Scatterer and Distributed Scatterer techniques, we gain both accuracy and wide area coverage. This helps us monitor different types of terrain and detect early warning signs. It allows scientists, decision makers, and local communities to prepare ahead of time and act before small movements turn into larger events.

We hope this article offered a clear introduction to ground motion detection using InSAR technology. If you’re curious to learn more, explore real-world applications, or collaborate on a project, feel free to reach out to the authors at abhijeet.saroj@solafune.com or akshay.patil@solafune.com. We’d be happy to connect.

Credits: Special thanks to Ritwika Dasgupta for her significant contribution to the codebase used in this blog, and to Pushkar Kopparla and Charles Genki Shimokura for their insightful feedback on the draft, which greatly helped enhance the article.