From Satellites to Safety: An Open-Source SBAS Workflow for Ground Deformation Monitoring

Highlights

What are the main findings?

• A robust and semi-automated Small Baseline Subset (SBAS) workflow using the open source hyp3_sbas Python library has been implemented for large-scale ground deformation monitoring in mining environments.
• The workflow efficiently detects millimetric-scale deformation patterns at the Björkdal gold mine using Sentinel-1 data.

What are the implications of the main findings?

• The use of the hyp3_sbas Python implementation reduces the learning curve of the complex SBAS methodology, enabling non-expert users to perform analyses.
• The proposed Interferometric Synthetic Aperture Radar (InSAR) workflow provides reproducible, scalable, and operationally ready satellite-based monitoring of mining infrastructure, facilitating the integration of InSAR techniques into risk prevention strategies and supporting sustainable mining management.

Abstract

Ground deformation monitoring is critical for safety and environmental management in modern mining. Active mining sites are highly exposed to terrain instabilities and subsidence, risking infrastructure integrity, disrupting operations, and posing hazards to communities. In this context, Differential Synthetic Aperture Radar Interferometry (DInSAR) techniques provide an effective and non-invasive tool capable of detecting millimetric surface displacements. This study implements the Small Baseline Subset (SBAS) technique through an open-source workflow based on the Python package hyp3_sbas, enabling semi-automated and reproducible interferometric processing by combining HyP3 with MintPy. The workflow is applied to the Björkdal gold mine (Sweden), a pilot site of the Horizon Europe XTRACT project focused on enhancing resilience in critical raw material supply chains. Integrating Sentinel-1 viewing geometries resolves the true vertical deformation field, yielding an overall mean velocity of −3.99 mm/year across the mining complex, with significant displacement rates concentrated below the 25th percentile (Q1) at −11.07 mm/year. Sector-specific analysis reveals localised subsidence accelerating over underground footprints and tailings storage facilities (mean velocities of −6.56 and −3.98 mm/year; Q1 thresholds near −13.00 mm/year), contrasting with the geomechanical stability observed at the open-pit area (mean: −0.45 mm/year). The proposed open-source framework shows strong potential for operational satellite-based monitoring, supporting predictive maintenance and early-warning strategies for risk management in mining environments while simplifying and standardising the interferometric processing workflow.

1. Introduction

Mining activities are widely recognised as major drivers of ground deformation and subsidence, affecting both natural and built environments [1,2]. The increasing complexity of the built environment, together with the expansion of settlements into former mining and undeveloped areas, has significantly increased exposure to ground instability. This situation highlights the need for reliable and scalable tools capable of detecting deformation processes and supporting sustainable spatial planning, infrastructure protection, and resource management [1,3]. Beyond safety considerations, effective deformation monitoring is also essential to ensure the continuity and resilience of supply chains for critical raw materials, which are fundamental to the energy transition and modern industrial systems. Within this context, Interferometric Synthetic Aperture Radar (InSAR) has become a key technology for large-scale ground deformation monitoring, offering millimetric sensitivity, wide spatial coverage, and independence from in situ instrumentation [3,4].

The versatility of multi-temporal InSAR (MT-InSAR) and radar time series analysis has been extensively demonstrated across a broad spectrum of anthropogenic and natural hazard assessments, moving beyond traditional geological applications. In urban and conflict-affected regions, Sentinel-1 radar data, often integrated with optical imagery or machine learning frameworks, has become fundamental for the rapid and large-scale assessment of building damage and infrastructure destruction, providing essential automated indicators in inaccessible zones [5,6]. Furthermore, MT-InSAR algorithms combined with numerical modelling have advanced structural health monitoring, enabling the evaluation of displacement trends to perform forensic analyses on infrastructure collapses and simulate risks induced by urbanization and underground construction [7,8].

In parallel, InSAR has emerged as a critical tool for environmental monitoring and the safety management of large-scale energy and water infrastructure. Radar interferometry is widely employed to ensure the security of Carbon Capture and Storage (CCS) initiatives by monitoring ground stability and surface deformation [9]. This technology also provides a continuous diagnostic framework for evaluating the long-term integrity of hydroelectric plants and reservoir dams [10]. Specifically within the mining sector, recent forensic assessments emphasize Sentinel-1 InSAR as a powerful hazard-screening tool capable of detecting millimeter-scale precursor displacements in tailings storage facilities, thereby guiding geotechnical investigations and mitigating risks of structural instability [11]. Building upon these advancements, its application in mining environments enables the detection of subsidence, slope instability, and surface deformation processes evolving over different spatial and temporal scales [3,4,12,13]. Numerous studies have demonstrated the value of InSAR-based monitoring for mitigating structural damage, preventing infrastructure failure, and reducing social impacts such as housing loss or population displacement in mining-affected areas [1,2,4,13].

Since its early development in the late twentieth century, InSAR has evolved from an experimental geodetic technique into a mature and operational Earth observation tool. Advances in radar sensor technology, orbital control, and processing algorithms have significantly improved measurement accuracy, temporal sampling, and spatial coverage [14,15]. The availability of satellite missions such as Sentinel-1 [16], TerraSAR-X [17], and COSMO-SkyMed [18] has further expanded the applicability of InSAR for continuous monitoring of mining areas and critical infrastructures [19,20]. In parallel, open-access initiatives and cloud-based processing platforms promoted under the Copernicus programme have contributed to the widespread adoption of InSAR products at regional and continental scales [21].

To overcome the limitations of conventional two-image interferometry, several multi-temporal Differential InSAR (DInSAR) methodologies have been developed. Persistent Scatterer Interferometry (PSI) exploits phase-stable targets to achieve high measurement accuracy and long-term deformation monitoring, particularly in urban or infrastructure-dense environments [19,22]. However, its dependence on persistent scatterers (PS) often results in sparse spatial coverage in non-urban, vegetated, or dynamically changing areas such as active mining sites [23]. To address this limitation, SBAS techniques combine multiple interferometric pairs with small spatial and temporal baselines, enabling the reconstruction of continuous deformation time series and the detection of non-linear displacement patterns [20]. Other advanced approaches, such as SqueeSAR, extend the PSI framework by incorporating both PS and distributed scatterers (DS), improving measurement density in low-coherence environments, including tailings storage facilities and open-pit mines [1,19,24]. More recently, polarimetric tomographic interferometry (PolTSI) has integrated polarimetric information to enhance coherence and phase stability in complex geological settings, albeit at the cost of increased computational complexity and processing demands [23]. Each of these approaches presents specific advantages and limitations depending on terrain characteristics, deformation magnitude, and temporal evolution.

Among the available DInSAR techniques, SBAS has proven to be particularly well-suited for mining environments [25,26,27]. Its capacity to maintain sufficient spatial coverage in partially coherent areas allows the monitoring of gradual subsidence and episodic deformation associated with excavation activities, waste deposits, and underground voids [1,20]. While strong deformation gradients may lead to local underestimation due to phase unwrapping issues [28], SBAS remains highly effective for operational deformation monitoring and hazard assessment in mining contexts.

Despite the demonstrated potential of SBAS and other DInSAR techniques, their operational integration into mining workflows remains limited. Large-scale services such as the European Ground Motion Service (EGMS) provide standardised and freely accessible deformation products [29,30,31,32], but offer limited flexibility in processing parameter control and workflow customisation. Moreover, many SBAS implementations rely on fragmented software tools, expert-driven processing chains, and high computational resources, limiting reproducibility and accessibility for non-specialist users [31,32]. Frequent surface changes associated with active extraction further reduce interferometric coherence precisely in areas of highest interest, reinforcing the need for expert interpretation and complementary monitoring strategies [33].

In response to these challenges, this study presents a reproducible workflow for SBAS-based ground deformation analysis using Sentinel-1 data. The proposed approach is implemented through an open-source Python 3.11 framework that integrates data selection, automated download, interferometric pair generation, SBAS processing, and the production of georeferenced deformation products suitable for GIS-based analysis. By consolidating the complete processing chain into a modular and transparent workflow, the framework aims to reduce technical barriers, enhance reproducibility, and facilitate the operational use of SBAS in mining environments. The novelty of this work does not lie in the SBAS technique itself, which is already well established, but in the implementation of a reproducible processing workflow based on open-source tools that streamlines the processing chain and supports the practical adoption of DInSAR monitoring in mining contexts. Accordingly, the applicability of the proposed methodology is demonstrated through a case study at the Björkdal gold mine, a pilot site within the European XTRACT project focused on sustainable resource recovery and advanced monitoring solutions for complex mining environments. While the case study serves as validation, the primary contribution of this work lies in the framework itself, which is transferable to other mining sites and geotechnical contexts.

The remainder of this article is structured as follows: Section 2 presents the materials and methods, including the description of the study area, the Sentinel–1 satellite datasets, the HyP3-derived products used for interferometric processing, and the SBAS-based deformation analysis workflow. This section also provides an overview of DInSAR interferometry, the theoretical background of the SBAS technique, and the automated processing workflow implemented through the hyp3_sbas Python library. Section 3 presents the deformation results obtained for the Björkdal mine, including both the overall deformation patterns across the mining area and a detailed analysis of the two most relevant zones. Section 4 discusses the implications of the results, as well as the methodological limitations and the potential transferability of the proposed workflow to other mining environments. Finally, Section 5 summarises the main conclusions of the study.

2. Materials and Methods

2.1. Study Area: Björkdal Mine

The study area corresponds to the active Björkdal gold mine (64°56′09″N, 20°36′15″E) [34], located approximately 28 km northwest of the municipality of Skellefteå and about 750 km north of Stockholm (Figure 1). The operation lies within the Boliden mining district and comprises four main subzones: an open pit (orange outline), an underground mine (yellow dotted line), two waste piles (blue outline), and a tailings storage facility with a containment dam (purple outline), together with associated processing and administrative infrastructure (black outline) [35]. The mining production and extraction began in 1988 with open-pit extraction, which continued until 2019, while underground mining started in 2008 and currently represents the main production method [34].

Geologically, the Björkdal mine is situated within the Fennoscandian Shield [36], specifically in the Skellefteå volcanic belt [37], one of the most mineralised regions in northern Europe. This geological setting is characterised by metasedimentary and metavolcanic rocks intruded by syn-orogenic granitoids [38]. Surface geology is predominantly covered by Quaternary deposits, with glacial till representing the dominant surficial material, accompanied by localized peatlands and post-glacial sediments, as mapped by the Geological Survey of Sweden (SGU) [39]. Hydrogeologically, groundwater conditions are strongly controlled by both regional crystalline aquifers and local fracture systems. In addition, multiple groundwater reservoirs may occur at different depths beneath the study area. According to SGU records, a groundwater reservoir associated with sedimentary deposits has been identified southeast of the mining complex; however, this reservoir is located outside the main area of mining influence and is not expected to directly affect the mine site (Figure 2).

Furthermore, although no significant groundwater reservoir is present within the immediate mining area, active mining operations require continuous dewatering processes [34]. In addition, the regional climatic conditions, characterised by subarctic-boreal permafrost-related freeze–thaw cycles, also significantly influence local groundwater dynamics. According to the Swedish Meteorological and Hydrological Institute (SMHI), the area experiences a subarctic/boreal climate, characterized by an average annual precipitation of approximately 550–650 mm [40]. This climate imposes severe winter freezing conditions and rapid spring snowmelt cycles, leading to dynamic ground frost heave and thaw processes that can significantly influence seasonal and superficial ground deformation patterns. Regarding population distribution, the study area is located in a remote, sparsely populated territory of the Västerbotten county, dominated by forestry and industrial mining activities. The nearest small residential settlements, such as Sandfors, are situated several kilometers from the main open pit and tailings storage facilities (TSFs) [41], indicating that the millimetric deformations monitored do not present an immediate geohazard risk to high-density urban infrastructure.

Each subzone of the Björkdal mine exhibits distinct geotechnical and operational characteristics relevant to ground deformation monitoring [42]. The open pit is the main excavation area, where past activities produced significant surface deformation. The tailings dam is a critical structure for subsidence and stability monitoring. Underground workings may induce complex subsurface deformation patterns, while the waste piles, composed of overburden and non-ore material, are prone to gradual settlement and slope movement. Monitoring these four subzones provides a comprehensive view of the mine’s deformation dynamics, integrating both anthropogenic and natural influences on ground stability. The coexistence of surface and underground operations makes Björkdal an ideal test site for evaluating the performance of SBAS techniques in active mining environments.

2.2. Satellite Data: Sentinel–1

The Copernicus Programme [21], coordinated by the European Commission (EC) in collaboration with the European Space Agency (ESA), provides open and free access to a wide range of geospatial datasets acquired through its Sentinel satellite missions [43]. Among these missions, Sentinel–1 supplies C-band SAR imagery (λ = 5.55 cm) in both ascending and descending orbits [44], offering consistent spatial and temporal coverage that is independent of weather and daylight. This capability is essential for continuous monitoring in mining environments.

For interferometric applications targeting ground deformation analysis, such as SBAS, Sentinel–1 Level–1 Single Look Complex (SLC) products are the most suitable, as they preserve phase information necessary for detecting millimetric surface displacements. The standard Interferometric Wide (IW) acquisition mode, optimised for land applications, provides coverage through three sub-swaths composed of synchronised bursts, ensuring consistent interferometric pairing across acquisition periods [44,45].

In this study, Sentinel-1 SLC products constitute the primary input for the hyp3_sbas processing workflow. These datasets provide the interferometric foundation from which deformation time series and displacement maps are generated, forming the basis for subsequent geospatial analysis and interpretation.

2.3. Data Processing: Hyp3 Products

The Hybrid Pluggable Processing Pipeline (HyP3), developed by the Alaska Satellite Facility (ASF), is a cloud-native processing platform designed to efficiently execute computationally intensive SAR workflows. It addresses key challenges in SAR data processing, including high computational demands, reliance on complex or proprietary software, and the need for expert knowledge to transform raw data into analysis-ready formats [46,47]. Using an on-demand processing service for Sentinel-1 data, HyP3 generates standardised, GIS-ready products with improved efficiency, reducing tasks that typically require one to two days to a few hours [47].

HyP3 products are generated using the GLO-30 Copernicus Digital Elevation Model (DEM) and are primarily processed with the GAMMA and ISCE2 software suites. The platform provides two main categories of products:

• Radiometric Terrain Correction (RTC) outputs, derived from Sentinel-1 IW mode acquisitions, which correct geometric and radiometric distortions inherent to SAR side-looking geometry. Specifically, radiometric correction normalizes the backscatter intensity by accounting for the local terrain slope, ensuring that variations in pixel brightness reflect surface properties rather than topographic orientation. RTC products are delivered as UTM-projected GeoTIFFs with configurable parameters such as pixel spacing (10–30 m) and radiometry type, providing either σ0 or the more robust γ0 backscatter, the latter being essential for eliminating terrain-induced intensity variations [46,48,49]
• InSAR products, essential for deformation analysis, including full-scene interferograms (GAMMA software [49]) with 40 or 80 m pixel spacing; burst-based interferograms (ISCE2 [50]) with 20, 40, or 80 m pixel spacing; ARIA Sentinel-1 geocoded unwrapped interferograms (ISCE2) with 90 m pixel spacing in netCDF format; and autoRIFT products, which compute velocity maps via feature tracking.

HyP3’s modular and scalable architecture enables the integration of advanced interferometric time-series methods such as SBAS. SBAS typically employs Singular Value Decomposition (SVD) to derive robust estimates of linear and non-linear surface deformation rates from sequential interferometric pairs [2], supporting high-precision ground motion monitoring in mining environments.

For the SBAS analysis of the Björkdal mine, Sentinel-1 data from both ascending and descending orbits were processed to capture deformation from complementary viewing geometries. The ascending dataset corresponds to orbit 58 and includes 165 HyP3 interferometric pairs, while the descending dataset corresponds to orbit 95 with 151 HyP3 pairs, providing comparable spatial coverage. Perpendicular and temporal baselines were set to 300 m and 12 days, respectively, covering the period from 1 April 2021 to 1 April 2025.

To ensure a robust and stable inversion of the time series, the SBAS network was constructed to maintain high phase coherence and multi-reference redundancy. Following the hardware anomaly and subsequent decommissioning of the Sentinel-1B satellite in late December 2021, the nominal revisit period over the study area shifted from 6 to 12 days, relying solely on Sentinel-1A acquisitions. Consequently, while 12 days represents the minimum temporal step between consecutive images, the interferometric pair generation was not strictly limited to this single-interval forward step. Instead, the network configuration allowed for extended temporal connections (e.g., 24 or 36 days), ensuring that each SAR acquisition was linked to multiple subsequent images. The perpendicular baseline threshold was constrained to a maximum of 300 m to minimise geometric decorrelation.

The resulting interferogram networks for both the ascending and descending datasets are presented in Figure 3. In these graphs, the yellow circles represent the individual SAR acquisitions (nodes) plotted against their specific temporal and perpendicular baselines. The solid lines denote the successfully processed interferometric pairs (edges), which are colour-coded based on their average spatial coherence, whereas dashed lines indicate pairs that were dropped during the network inversion due to low coherence. As visually confirmed by this multi-connected topology, the network prevents vulnerable single-chain dependencies, ensuring that continuous temporal overlap was maintained throughout the monitoring period without isolated clusters or disconnections.

It should be noted that these parameters can be adjusted by the user depending on the specific analysis requirements. The products used were SLC InSAR-Gamma. A sample of HyP3 pairs for the ascending orbit is shown in Figure 4.

2.4. SBAS-Based Deformation Analysis Workflow

This section describes the workflow implemented for the extraction of ground deformation time series using the SBAS technique applied to Sentinel-1 SAR data. The proposed approach integrates the fundamental principles of DInSAR with a structured processing chain designed to ensure temporal consistency, phase stability, and reliable deformation retrieval in complex mining environments. Particular emphasis is placed on the implementation of the workflow through the hyp3_sbas Python framework, which enables the automation of key processing steps, including interferometric pair selection, time-series inversion, and generation of georeferenced deformation products. The following subsections provide an overview of the DInSAR interferometric principles, the theoretical background of the SBAS method, and the specific implementation of the hyp3_sbas processing workflow, the highlight of this manuscript.

2.4.1. DInSAR Interferometry Overview

InSAR is a radar remote sensing technique that exploits the phase information of two or more SAR images acquired over the same area to measure topographic height or surface deformation [14,15,51]. The fundamental principle relies on the analysis of the phase difference (interferometric phase) between two complex SAR images:

$${\Delta }_{\phi }={\phi }_2-{\phi }_1$$

(1)

where φ₁ and φ₂ denote the phases of the radar signals acquired at times t₁ and t₂, respectively. The interferometric phase Δ_φ contains contributions from topography (Δ_φtopo), surface displacement (Δ_φdef), atmospheric effects (Δ_φatm), orbital inaccuracies (Δ_φorb), and noise (Δ_φnoise):

$${\Delta }_{\phi }={\Delta }_{\phi topo}+{\Delta }_{\phi def}+{\Delta }_{\phi atm}+{\Delta }_{\phi orb}+{\Delta }_{\phi noise}$$

(2)

If the topography is known, Δ_φtopo can be simulated and subtracted, isolating the deformation component Δ_φdef. This defines DInSAR and allows the quantification of ground displacement along the radar’s Line of Sight (LOS) as:

$${\delta }_r=\lambda ·{\Delta }_{\phi def}/4\pi$$

(3)

where λ is the radar wavelength.

Conventional DInSAR is limited by temporal and spatial decorrelation, atmospheric artefacts, and residual orbital or DEM errors. To mitigate these effects, advanced DInSAR time-series techniques have been developed [20,22,52,53]. These methods process long temporal sequences of SAR acquisitions to reconstruct deformation histories, enhance phase stability, and reduce noise.

Among these, PSI and SBAS are the most widely implemented. PSI focuses on stable, point-like targets (PS) that maintain phase stability over time, making it effective in urban or infrastructure-dense areas [20,23]. SBAS, in contrast, maximises the number of interferometric pairs by constraining temporal and spatial baselines, improving temporal sampling and performance in non-urban areas dominated by DS. SBAS applies SVD to reconstruct the cumulative deformation time series by linking independent SAR acquisition subsets. From this reconstructed series, both linear and non-linear deformation components can be decoupled, which is critical for mining contexts characterized by spatially extensive and temporally non-linear surface displacements.

Further methodological advances aim to increase measurement density and robustness in low-coherence terrains. SqueeSAR extends the PSI framework by jointly exploiting PS and DS pixels using Maximum Likelihood Estimation (MLE) over homogeneous areas, increasing coherent point density and reducing phase noise. However, its performance may deteriorate under high spatial gradients or rapid deformation. PolTSI integrates dual-polarisation channels (Vertical–Vertical (VV) and Vertical–Horizontal (VH)), applying amplitude dispersion-based stabilisation for PS pixels and polarimetric Minimum Mean Square Error (MMSE) filtering for DS pixels, improving coherence and measurement density at a higher computational cost.

Table 1. Comparative overview of the main characteristics, advantages, and limitations of four widely used advanced DInSAR techniques: PSI, SBAS, SqueeSAR, and PolTSI.

Technique	Main Scatter Type	Key Advantages	Key Disadvantages
PSI	PS	High precision in urban or artificial areas	Sparse point density in vegetated or low-coherence zones
SBAS	DS	Reduced spatial decorrelation; improved temporal sampling	Lower accuracy in low-deformation areas
SqueeSAR	PS and DS	Increased measurement density; reduced decoherence noise	Performance loss under high deformation gradients
PolTSI	PS and DS	Enhanced coherence and density in complex, low-coherence terrains; dual-polarisation optimization	Computationally intensive; requires dual-polarisation data

provides a schematic comparison of the advantages and disadvantages of PSI, SBAS, SqueeSAR, and PolTSI. These four methods represent the foundational DInSAR approaches evaluated within the XTRACT project as reference algorithms for comparative analysis.

Despite the specialisation of advanced DInSAR methods, PSI and SBAS remain the most widely applied techniques due to their robustness and mature processing software. While PSI is effective for monitoring stable, point-like targets in urban environments, SBAS is preferred in regions dominated by DS, such as mining sites. The large, non-linear, and spatially extensive deformation typical of mining subsidence, combined with the scarcity of dense, permanent structures, makes SBAS particularly suitable for capturing continuous ground motion while mitigating decorrelation effects. Accordingly, SBAS was adopted in this study to analyse surface deformation dynamics in the Björkdal gold mine, providing a real-world case study for the workflow proposed.

2.4.2. Theoretical Background of SBAS

The SBAS technique exploits a temporal series of SAR images to derive surface deformation histories through the joint inversion of multiple interferograms [20]. SBAS builds a network of interferograms from pairs of SAR acquisitions separated by small spatial and temporal baselines, minimising decorrelation and ensuring high phase coherence [20]. Each acquisition connects to multiple others, forming a multi-reference interferometric network that allows deformation estimation even when some acquisitions are missing or degraded. The SBAS network is defined by limiting the perpendicular (B_⊥) and temporal (B_t) baselines of each interferometric pair according to predefined thresholds (B_⊥ < B_⊥^max, B_t < B_t^max).

Unlike PSI [19,54], which focuses on stable, isolated reflectors, SBAS derives information from DS, surface areas composed of multiple small scattering elements whose combined backscatter remains coherent over time [20,52]. Consequently, retrieved deformation represents the average displacement within each SAR resolution cell.

For each differential interferogram $i$, generated from two SAR acquisitions at different times, the unwrapped phase difference $\Delta {\varphi }_i$ can be expressed as [20]:

$$\Delta {\varphi }_i={\displaystyle \frac{4\pi }{\lambda }}·A_{i}v+{\Delta \varphi }_i^{topo}+{\Delta \varphi }_i^{atm}+{\Delta \varphi }_i^{noise}$$

(4)

where λ is the radar wavelength, $A_i$ is the design matrix linking acquisition times to the deformation velocity vector $v$, ${\varphi }_i^{topo}$ represents the residual topographic phase error due to Digital Elevation Model (DEM) inaccuracies, ${\varphi }_i^{atm}$ represents the difference in the atmospheric phase component between two acquisitions, and ${\varphi }_i^{noise}$ the residual noise. The system is solved through least-squares inversion or SVD, providing temporally regularised displacement time series and average velocity maps along the radar LOS with millimetric precision [13].

By integrating multiple interferometric pairs, SBAS improves robustness against decorrelation and atmospheric artefacts, enabling millimetric deformation retrieval over extended periods [52]. It has been successfully applied to monitor large-scale phenomena such as volcanic activity, tectonic motion, landslides, and mining-induced subsidence [55,56]. Atmospheric delay correction was applied using the ERA5 reanalysis model from the European Centre for Medium-Range Weather Forecasts (ECMWF) which provides hourly global data on temperature, pressure, and humidity at a spatial resolution of approximately 31 km (0.25° grid). This allows for the mitigation of stratified tropospheric delays and large-scale atmospheric biases in InSAR measurements [57].

Commencing with a coregistered network of unwrapped differential interferograms, the SBAS time-series analysis follows these steps [58]: (1) selection of DS based on spatial or temporal coherence to ensure reliable unwrapped phases; (2) network inversion of the unwrapped phase using Ordinary Least Squares (OLS) or SVD to retrieve the raw cumulative displacement time series [20]; (3) estimation and correction of the residual topographic phase (${\Delta \varphi }_i^{topo}$) induced by DEM inaccuracies, exploiting its proportionality to the perpendicular baseline; (4) mitigation of atmospheric artifacts (${\Delta \varphi }_i^{atm}$) using tropospheric delay correction from external reanalysis models (e.g., ERA5) or filtering techniques; (5) definition of a stable reference point to anchor displacement estimates; (6) evaluation of phase noise and exclusion of low-quality acquisitions; and (7) estimation of mean LOS deformation velocity as the slope of the corrected displacement trend over time. This framework ensures reliable deformation estimation and provides the methodological foundation for the implementation of the hyp3_sbas workflow in the Björkdal gold mine case study.

2.4.3. SBAS Processing with hyp3_sbas

Building on the theoretical framework presented above, the SBAS technique was implemented for Sentinel-1 SAR data processing through the development of an open-source Python library (v3.9.16) named hyp3_sbas [59]. As stated before, this tool provides a modular workflow that encompasses the entire processing chain, from data acquisition to the generation of deformation time series and GIS-compatible outputs. Its design ensures flexibility, reproducibility, and scalability, facilitating operational use by both expert and non-expert users in radar interferometry.

The methodological workflow followed is encapsulated within a modular architecture designed for the hyp3_sbas library components, as can be seen in Figure 5.

The specific modules are:

• download: Manages the automated acquisition of Sentinel–1 SLC products confined to a designated Area of Interest (AOI), based on parameters specified by the user. This function is underpinned by the capabilities of the ASF Search API for Python [60,61].
• pairs: Executes the systematic generation and retrieval of the interferometric pairs requisite for SBAS analysis, delivered in the HyP3-compatible data format. Its implementation leverages the HyP3 SDK for Python [62], which further enables the optimal selection of pairs considering predefined constraints on baseline and temporal separation.
• processing: Facilitates the execution of the complete SBAS workflow. This implementation adopts a class-based, object-oriented paradigm, ensuring the inherent modularity, high reusability, and robust maintenance of all integrated processing routines.
• sbas: Applies the core SBAS methodology through the MintPy v1.5.3 library [63]. This encompasses the precise execution of interferogram stacking, temporal series inversion, and the subsequent estimation of LOS deformation rates. A critical feature is its capacity to perform the decomposition of displacements into East–West and vertical components, achieved by integrating data derived from both ascending and descending orbits.
• shapefile: Translates the derived SBAS time series results into a SHP format. This conversion operation rigorously associates each deformation measurement point with its corresponding, complete temporal displacement record. The module additionally supports the execution of geometric operations on SHP data and ensures its exportability to CSV format for seamless integration and visualisation within both conventional GIS platforms and WebGIS environments.
• utils: Constitutes a reservoir of essential ancillary functions whose purpose is to provide requisite support and ensure operational interoperability among the primary processing modules.

The hyp3_sbas library integrates custom functions built on established external tools, enabling the execution of the complete SBAS processing chain. Its main advantages are the semi-automation of complex interferometric tasks and the provision of an intuitive, user-friendly interface that facilitates its use even by non-expert practitioners. Established SBAS processing pipelines, including MintPy, LiCSBAS, ARIA tools, and SNAP-based workflows, offer robust and highly optimised frameworks for InSAR analysis. However, their effective configuration and operation can be complex, as they typically require substantial technical expertise, posing a significant learning curve for non-specialist users. The hyp3_sbas package is not intended to supersede or compete with these advanced tools. Instead, it functions as an integration and accessibility layer, orchestrating existing, well-validated components within a unified, programmatic Python interface.

The principal contribution of hyp3_sbas lies not in algorithmic innovation, but in workflow simplification and enhanced reproducibility. It significantly reduces the manual steps required to transition from raw data acquisition to displacement time-series generation, thereby lowering the entry barrier for researchers without extensive DInSAR processing experience. Furthermore, hyp3_sbas promotes transparent, script-based reproducibility of complex SBAS analyses. Consequently, it occupies a distinct niche relative to existing pipelines, serving as an operational bridge that democratises access to state-of-the-art DInSAR processing, particularly in research environments where SAR expertise is not the primary focus.

The data processing sequence followed for this application in a mining environment is segmented into four main steps, as shown in Figure 6:

Step 1. Product Identification and Selection (Figure 7): Based on user-defined search parameters, Sentinel–1 SLC products within the AOI are retrieved through the ASF Search API. Once the products are identified, an optimal set of interferometric pairs (sbas_pairs) is generated according to spatial and temporal baselines criteria. These pairs form the input for step 2.

Step 2. Interferogram Generation (Figure 8): Using the optimised list of interferometric pairs from Step 1 (sbas_pairs), interferograms are generated and downloaded according to the HyP3 standard. Download performance depends on ASF Search service response times. A preliminary query verifies successful generation of all interferograms before the final data acquisition.

Step 3. Interferogram Processing: The interferograms are processed using the SBAS methodology described in Section 2.4.2. This stage requires the preparation of the MintPy-compatible data structure and the execution of the sequential processing steps illustrated in Figure 7. The ERA5 model was used for tropospheric correction. The output consists of a georeferenced SHP file that stores all deformation points and their associated displacement time series.

Step 4. Post-processing and Analysis (Figure 9): The resulting SHP layer undergoes several geometric operations using the shapefile module within hyp3_sbas. The processed data are then exported as a CSV file containing displacement time series and mean velocity values.

The outputs obtained from this process are subsequently used to derive vertical and East–West displacement components by integrating information from both ascending and descending orbits. This step is fundamental because a single viewing geometry is inherently insufficient to fully describe the actual three-dimensional motion of the ground. The decomposition is achieved by projecting the LOS vectors associated with both orbital passes, allowing the retrieval of predominantly vertical movements (e.g., subsidence or uplift) and horizontal displacements oriented mainly in the East–West direction.

In mining environments, distinguishing between these motion components is particularly critical. Vertical components typically indicate ground settlement related to underground voids or material extraction, whereas East–West components often correspond to slope instabilities or lateral ground spreading. Therefore, the joint interpretation of the derived vertical and East–West deformation fields provides a more comprehensive and physically meaningful characterisation of the complex terrain instability processes commonly observed in mining scenarios [64].

To better understand the true kinematic behaviour of the monitored mining hazards, the Line-of-Sight displacement rates obtained from the independent ascending and descending tracks were geometrically decomposed into two-dimensional components. As detailed in the geometric configuration in Figure 10, the InSAR acquisition geometry combines two distinct views to resolve the true ground displacements. Given that Sentinel-1 operates on a near-polar, sun-synchronous orbit, its look geometry is inherently insensitive to north–south ground movements (typically capturing less than 10% of such signal). Consequently, the horizontal displacement vector can be mathematically approximated as a strict East–West component ($U_E$), while the vertical vector represents the Up–Down component ($U_V$).

Assuming a negligible North–South contribution, the relationship between the observed LOS velocities ($d_{asc}, d_{desc}$) and the true 2D displacement vectors is governed by the following system of linear equations [65,66]:

$$d_{asc}=U_V·cos{\theta }_{asc}-U_E·sin{\theta }_{asc}·cos{\alpha }_{asc}$$

(5)

$$d_{desc}=U_V·cos{\theta }_{desc}-U_E·sin{\theta }_{desc}·cos{\alpha }_{desc}$$

(6)

where ${\theta }_{asc}$ and ${\theta }_{desc}$ represent the satellite incidence angles (the angle between the LOS vector and the vertical), and ${\alpha }_{asc}$ and ${\alpha }_{desc}$ are the satellite heading (azimuth) angles for the ascending and descending orbits, respectively. By solving this system via matrix inversion, the independent vertical ($U_V$) and East–West ($U_E$) velocity fields can be isolated. A detailed geometric breakdown of this vector decomposition is illustrated in Figure 10, visually demonstrating this decomposition mechanism. It depicts the ascending satellite pass (left) and descending pass (right), showing how their local reference frames ($U, N, E$) relate to the global reference frame on the ground, and how the dashed LOS vectors project onto the vertical and horizontal components. By solving this system via matrix inversion, the independent vertical ($U_V$) and East–West ($U$_E) velocity fields can be accurately isolated.

Figure 10. Schematic diagram illustrates the ascending and descending InSAR geometric configuration for 2D deformation decomposition. The combination of independent LOS observations allows for the extraction of the true vertical (U_V) and East–West (U_E) displacement vectors, neglecting the North–South component. In this figure, U corresponds to Uv and E corresponds to Ue, where vertical motion can be interpreted as uplift or subsidence, and horizontal motion as displacement towards the East or West. Adapted from [66].

Figure 10. Schematic diagram illustrates the ascending and descending InSAR geometric configuration for 2D deformation decomposition. The combination of independent LOS observations allows for the extraction of the true vertical (U_V) and East–West (U_E) displacement vectors, neglecting the North–South component. In this figure, U corresponds to Uv and E corresponds to Ue, where vertical motion can be interpreted as uplift or subsidence, and horizontal motion as displacement towards the East or West. Adapted from [66].

This decomposition step is crucial to prevent misinterpretation of deformation signals since a point undergoing actual subsidence (i.e., downward vertical motion) may appear to move toward the satellite in ascending orbit data due to the viewing geometry. Only through the simultaneous use of both geometries can the true motion be accurately determined. Consequently, the integration of ascending and descending Sentinel-1 datasets and their subsequent decomposition into vertical and horizontal components is not merely a methodological enhancement but an essential requirement for producing reliable and physically consistent interpretations of ground deformation. This approach is widely applicable across different operational scenarios, including geological hazard monitoring, infrastructure assessment, and land stability analysis [67].

3. Results

Upon completing the SBAS processing chain and the decomposition of LOS measurements into East–West and vertical components, the deformation products are ready for spatial and temporal interpretation. The results obtained over the Björkdal mining area are presented in this section, structured according to the main operational and geomorphological domains of the site. It should be noted that, in the context of SBAS analysis, negative values indicate motion away from the satellite along the line-of-sight direction (e.g., subsidence or eastward horizontal displacement after decomposition), while positive values indicate motion towards the sensor along the LOS direction (e.g., uplift or westward horizontal displacement after decomposition).

3.1. Björkdal Mine

A total of 3345 coherent points were obtained from the ascending orbit, showing velocity values ranging from −76.34 to 63.58 mm/year. In the descending orbit, 3223 coherent points were identified, with velocities between −40.23 and 92.98 mm/year. The spatial distribution of the coherent points and the histograms of the calculated velocities for both viewing geometries are displayed in Figure 11 and Figure 12. Additionally,

Table 2. Summary statistics of SBAS LOS velocity estimates derived from Sentinel–1 ascending (ASC) and descending (DESC) datasets (mm/year).

Orbit	Mean	Std	Min	25%	50%	75%	Max
ASC	−2.63	19.04	−76.34	−14.27	−1.46	10.42	63.58
DESC	28.84	18.08	−40.23	17.495	29.55	40.69	92.98

reports the key statistical metrics of the velocity estimates derived from each orbit.

The velocity rates derived from both viewing geometries confirm the presence of significant ground deformation within the study area. Following standard InSAR convention, positive values correspond to motion toward the satellite (typically indicating uplift or horizontal movement towards the sensor), while negative values denote motion away from the satellite (suggesting subsidence or horizontal movement away from the sensor). This spatial distribution reveals the coexistence of complex 3D deformation processes. The statistical parameters derived from the SBAS analysis further highlight contrasting deformation behaviours between the ascending and descending geometries. In the ascending orbit, the slightly negative mean velocity suggests a predominant motion away from the satellite, whereas in the descending geometry, the positive mean velocity indicates motion towards the sensor. This stark contrast is a classic signature of mining-induced activity, where strong inward horizontal movements (towards the centre of the subsidence basin) combine with vertical settlement, projecting with opposite signs in the ascending and descending LOS vectors.

Given the SBAS complexity, interpreting ascending and descending datasets independently is insufficient to determine the dominant displacement directions, especially in the most prominent areas, which are the tailings dam and waste piles. Therefore, a decomposition of the LOS measurements into vertical and East–West displacement components is needed. In the context of this study, the vertical component is the most relevant, as it directly reflects subsidence and uplift processes, and is thus the one retained for interpretation.

The vertical component of deformation achieved is based on 3081 monitoring points and shows velocities ranging from −48.74 to +35.42 mm/year (

Table 3. Summary statistics of vertical deformation velocities derived from the combination of ascending and descending SBAS datasets (mm/year).

Orbit	Mean	Std	Min	25%	50%	75%	Max
VERT	−3.99	11.43	−48.74	−11.07	−3.30	3.84	35.42

). The vertical deformation map reveals a heterogeneous but well-defined deformation pattern across the Björkdal mine (Figure 13). The green tones represent stable areas where vertical motion is minimal, corresponding to relatively undisturbed ground. The vertical decomposition is much closer in sign and central tendency to the ascending-orbit results than to the descending ones: the vertical mean (≈−4.0 mm/yr) closely matches the slightly negative mean found in the ascending dataset (≈−2.6 mm/yr), whereas the descending mean is strongly positive. However, the vertical component shows a reduced amplitude compared with the single-orbit values: the extreme minima and maxima in the vertical map (Table 3) are noticeably smaller than those reported for the ascending orbit (Table 2). This compression of the range indicates that the vertical solution removes much of the geometry-dependent projection and East–West deformation contributions that inflate apparent velocities in single-look analyses.

As shown in Figure 13, most points fall within the green range, indicating vertical stability with deformation rates close to 0 mm/year, particularly in the open-pit zone. However, several localised zones deviate from this stable behaviour. The red to orange areas, corresponding to subsidence rates between approximately −20 mm/year and −50 mm/year, are mainly concentrated in the central-west sector (underground footprint) and northeastern portions of the mine (tailing dams).

Table 4. Statistical summary of vertical deformation velocities for the main operational sectors of the Björkdal mine derived from SBAS analysis (mm/year).

Area	Mean	Std	Min	25%	50%	75%	Max
Openpit	−0.45	6.54	−21.44	−4.88	−0.29	4.29	17.81
Tailings	−3.98	12.75	−45.52	−12.82	−3.35	5.13	35.42
Underground	−6.56	10.56	−39.63	−13.01	−5.32	1.21	20.12
Waste piles	−2.85	9.73	−48.74	−9.07	−2.70	3.43	34.48

provides a detailed spatial breakdown of the velocity metrics derived for the four constituent sectors of the mine site. However, because SBAS only captures motion along the satellite’s LOS and relies on surface coherence, it may not fully represent deep-seated or complex deformation processes. To improve spatial and temporal control of subsurface displacements, future monitoring should combine SBAS with complementary geodetic techniques.

From an operational standpoint, the tailings dam and waste pile areas also represent highly dynamic deformation zones. These features display a pronounced propensity for subsidence and are characterised by the greatest magnitude and variability (range of values) in the derived surface velocity measurements. This outcome underscores the necessity for intensive monitoring of these high-variability zones due to their critical role in material management and potential for slope instability.

3.2. Tailings Dam

The vertical velocity map of the tailings dam area reveals a clear spatial differentiation in deformation behaviour (Figure 14). The orange to red zones, concentrated mainly along the central and eastern margins of the dam, indicate active subsidence with velocities reaching up to −50 mm/year. These areas are likely associated with material compaction, consolidation of fine sediments, and variations in moisture or pore pressure within the deposited tailings. The extent and intensity of these subsiding zones suggest ongoing settling processes that may continue as deposition and drainage evolve over time.

In contrast, the blue-toned regions located primarily in the southern and western sectors exhibit positive vertical displacements, interpreted as surface uplift. These could result from differential drainage, localised heaves due to hydrological recharge, or small-scale stress redistribution in the underlying materials. The predominance of green tones across the dam crest and peripheral zones reflects relative stability, with velocities near zero. This spatial distribution aligns with the expected behaviour of a tailings storage facility where active settling occurs in depositional centres, while marginal and structurally reinforced areas remain comparatively stable or experience slight rebound.

The vertical deformation field definitely establishes the tailings dam as one of the most dynamically active sectors within the Björkdal mining complex. The dam exhibits a marked spatial contrast, displaying distinct zones of both pronounced settlement and localised rebound (or uplift). This highly heterogeneous mechanical response directly evidences the dam’s internal structural variability and the differential material behaviour under continuous static and dynamic loading, coupled with evolving drainage conditions. Given the critical implications for structural integrity, water retention efficacy, and long-term geotechnical stability, this area necessitates continuous, high-resolution monitoring to promptly detect any potential acceleration or spatial propagation of the observed deformation patterns.

To enhance the temporal characterisation of the area, the cumulative LOS displacement time-series for three representative monitoring points (P1, P2, and P3) were extracted (Figure 15). The selection of these points was based on their representativeness of the main deformation regimes observed in the study area: P1 is located within a subsidence trough (water body) and was selected to capture active ground lowering processes; P2 corresponds to an area affected by terracing and material emplacement, representative of engineered or anthropogenic uplift; and P3 is situated in a relatively stable zone, used as a reference for background deformation behaviour. Point P1 exhibits a severe and persistent linear subsidence trend throughout the monitoring period. Starting from an initial value of approximately +50 mm in early 2021, the series experiences a steady decline, reaching between −100 mm and −110 mm by mid-2025. This continuous downward displacement is a clear indicator of active subsidence, suggesting ongoing ground consolidation in the presence of water.

Conversely, points P2 and P3 display an upward dynamic behaviour, albeit with varying intensities. Point P2 experiences a strong and sustained positive trend, progressing from slightly negative initial values to a maximum displacement of nearly +75 mm by the end of the series. This marked uplift strongly points to a dynamic topography driven by the continuous accumulation of material, such as waste rock deposition on upper terraces or sub-benches, although minor contributions from elastic rock mass rebound cannot be entirely ruled out. Meanwhile, point P3 exhibits an intermediate and relatively stable behaviour characterized by pronounced high-frequency cyclical oscillations, likely seasonal ground volume changes driven by annual precipitation cycles, ultimately achieving a net positive displacement of approximately +30 mm.

3.3. Waste Piles

The vertical deformation velocity map is shown in Figure 16 and reveals predominantly stable behaviour, with most of the measured velocities clustered around the near-zero range, as indicated by a mean value of approximately −2.8 mm/year in the histogram. However, localized deformation is observed in specific sectors. The most pronounced subsidence zones (orange to red tones) are mainly concentrated along the eastern and southeastern flanks, likely associated with ongoing material compaction, heterogeneous loading, and differential drainage processes typical of unconsolidated waste deposits.

Conversely, minor uplift patterns (green to blue areas) appear scattered throughout the central and lower sections of the deposit. These uplift zones may result from seasonal variations in moisture content, internal pore pressure redistribution, or minor structural adjustments due to the settling of adjacent areas. The overall deformation pattern suggests that, while the waste piles exhibit relatively moderate vertical dynamics compared with the tailings dam, their intrinsic instability potential warrants periodic monitoring. Detecting early changes in vertical motion remains critical to ensure slope safety and the long-term geotechnical performance of these anthropogenic landforms.

Thus, the upper waste pile area shows marked subsidence, whereas the lower pile exhibits pronounced uplift, reflecting contrasting processes of material compaction and drainage. These patterns reveal the heterogeneous mechanical response of the deposits and their potential instability, underscoring the need for continuous monitoring and targeted geomechanical analyses to support stability control and operational planning. To illustrate these dynamics, the time series of three representative points are analysed, following the same case-based approach as for the tailings dam: one point located in a stable area, one exhibiting moderate deformation signals, and one affected by pronounced subsidence. The InSAR time-series analysis presented in Figure 17 illustrates these geomechanical processes between 2021 and 2025, revealing distinct stages of operational activity and consolidation. Point P1, located on the southernmost section, exhibits a severe and continuous subsidence trend, reaching a cumulative deformation of approximately −90 mm. In the context of waste dumps, this pronounced negative LOS displacement is highly characteristic of primary and secondary settlement, indicating significant mechanical compaction of recently deposited, unconsolidated waste material or potential localized slope yielding.

In contrast, the remaining monitored sectors demonstrate entirely different structural behaviours driven by their specific operational statuses. Point P2, situated on the central terraces, displays a significant upward trajectory, culminating in a positive displacement of over +100 mm. This marked elevation gain serves as a clear structural signature of active dumping operations, where the continuous deposition of new waste rock progressively raises the topographic surface. Meanwhile, point P3 on the western terraces remains tightly constrained near the zero-displacement baseline throughout the multi-year period, suggesting that this specific waste dump is currently inactive and has reached a high degree of material consolidation and geomechanical stability.

3.4. Methodological Limitations and Framework Reliability

A recognised limitation of the current study is the absence of independent, in situ geodetic measurements, such as continuous GNSS or high-precision levelling data, for an absolute quantitative validation of the derived deformation velocities. Such ground-truth datasets were not available from the mining operators and fell outside the logistical scope of the XTRACT research framework. This challenge is common in large-scale remote sensing studies where direct ground validation across extensive areas is often impractical or cost-prohibitive.

The operational reliability of this workflow is constrained by the physical limits of the Sentinel–1 C–band sensor. A primary limitation is the maximum detectable deformation gradient: InSAR phase unwrapping becomes unreliable if the differential displacement between adjacent pixels exceeds a quarter of the radar wavelength (~1.4 cm) between consecutive acquisitions. In active mines, steep deformation gradients or sudden collapses easily exceed this threshold, causing severe coherence degradation. To prevent the propagation of unwrapping errors, our MintPy SBAS workflow inherently masks out these chaotic areas. Furthermore, the 6– to 12–day temporal sampling interval restricts the accurate reconstruction of highly non-linear, instantaneous structural failures. Consequently, while highly robust for mapping the progressive, long-term kinematics of the broader subsidence basin, this framework operates fundamentally as a regional monitoring tool and remains structurally blind to localized, abrupt collapses.

Nevertheless, the reliability of the detected deformation patterns is robustly supported by several internal validation mechanisms. Foremost, a high degree of spatial and temporal consistency was observed between the independent ascending and descending Sentinel-1 datasets over the primary monitoring zones. This inter-track agreement, particularly in areas exhibiting significant deformation, provides strong evidence for the geophysical reality of the observed displacements, mitigating concerns regarding potential sensor-specific biases or atmospheric artifacts. This stringent criterion was crucial for effectively rejecting noise-dominated pixels. Furthermore, strict data quality criteria were rigorously implemented during the MintPy time series inversion, including a conservative temporal coherence threshold (>0.7). and minimising the influence of decorrelation effects, thereby ensuring that the reported deformation signals are statistically significant and representative of actual ground motion.

Within this context, the proposed workflow is not intended as a direct replacement for high-precision local geodetic monitoring. Instead, it serves as an operational, low-barrier hazard-screening tool. This framework provides mine managers with an efficient first-pass diagnostic mechanism to rapidly detect critical subsidence hotspots and optimise the strategic deployment of localised, often cost-intensive, geotechnical instrumentation. Its primary value lies in its ability to provide comprehensive spatial coverage and frequent updates, enabling proactive risk management across vast and dynamic mining environments.

4. Discussion

The application of the SBAS approach enabled the detection and characterisation of surface displacement patterns across the study area throughout the observation period. Since SAR measurements are inherently sensitive to motion along the LOS, the integration of both ascending (south-to-north) and descending (north-to-south) viewing geometries was essential to capture the multi-dimensional nature of the deformation. To ensure the fidelity of these measurements, atmospheric phase delays were mitigated using ERA5 reanalysis data. The integration of ERA5 was particularly critical because its high spatial and temporal resolution allowed for the simultaneous correction of topography-correlated delays and turbulent atmospheric fluctuations. By effectively isolating these artefacts from the tectonic or anthropogenic signals, the correction significantly enhanced interferometric coherence and reduced noise variance. Therefore, this refinement was a key factor in increasing the reliability and precision of the final SBAS displacement estimates [68], providing a more robust basis for interpreting the observed mining-related processes.

The adoption of this hyp3_sbas workflow, combining cloud-based HyP3 processing with MintPy SBAS time-series analysis, is uniquely suited to the operational constraints of active mining environments. By utilising standardised SBAS networks, the framework effectively circumvents the high computational bottlenecks and local storage dependencies traditionally associated with MT-InSAR. This methodology optimizes phase coherence over the heavily modified and fragmented surfaces typical of open-pit margins and tailings infrastructure, directly fulfilling the study’s objective: providing an accessible, reproducible hazard-screening tool for mine risk management.

Variations in the magnitude or spatial distribution of displacement between ascending and descending orbital configurations, as evidenced by the different values shown in Table 2, do not indicate inconsistencies but instead reflect the directional nature of the radar acquisition geometry and the multi-component character of mining-related deformation processes. These tracks observe the terrain from opposite sides, meaning that each geometry projects the vertical and east–west components of motion differently onto the LOS. Consequently, apparent differences in magnitude or spatial distribution are expected when deformation has a multi-component character, as is typically the case in mining environments.

The SBAS-derived velocity map from the ascending orbit reveals concentrated zones of intense deformation within the Björkdal mine (Figure 11). The most pronounced subsidence, shown in dark red, occurs in the central-right and northwestern sectors, corresponding to the tailings dam and waste piles (bottom part) extraction areas, and as expected, in the western part of the underground section. However, it is in this latter area where the highest subsidence values are observed, undoubtedly related to its constant mining activity, and can be further explained by the characteristic underground mining method applied at the site, based on sublevel drilling, material extraction, and subsequent backfilling, which promotes localised void generation and progressive ground readjustment during the extraction–filling cycle. In general, this downward motion likely results from ground compaction and void collapse associated with continuous material extraction and deposition activities. In the case of tailings deposits and waste rock piles, these deformation patterns may be further influenced by the climatic conditions characteristic of northern European environments, where seasonal freeze–thaw cycles and the potential development of discontinuous permafrost can induce additional ground instability. The freezing of pore water and subsequent thawing phases can modify the mechanical properties of the material, promoting consolidation, settlement, or localised deformation within these unconsolidated deposits. Conversely, uplift signals (dark blue) are mainly detected in the central and southeastern margins, also linked to the tailings dam, possibly due to slope adjustments, stress redistribution, or minor elastic rebound near active subsidence zones. The close coexistence of uplift and subsidence indicates a highly heterogeneous deformation regime, suggesting localised instability and ground fracturing driven by mining-induced stresses and topographic influences. On the contrary, the green areas, characterised by velocities between −5 and +10 mm/year, indicate zones of ground stability within the Björkdal mine. These stable regions are predominantly located in the eastern (tailings dam) and southwestern sectors (open pit outline and top of waste piles) of the mine, extending toward the processing facilities and peripheral access zones, where mining activity and excavation are less intense. The spatial continuity of these green zones suggests that these parts of the mine experience minimal deformation (particularly in the open pit area), likely corresponding to undisturbed or rehabilitated areas with limited subsurface void evolution or material displacement. Their presence provides a useful reference baseline for assessing the magnitude and spatial extent of the surrounding deformation zones detected by SBAS.

Regarding the descending orbit velocity map (Figure 11), the deformation pattern of the Björkdal mine presents notable spatial contrasts that complement the results from the ascending geometry. The orange tones (with a maximum of −40 mm) indicate areas of mild subsidence. These zones are mainly distributed along the north–west of the Tailings Dam area. The limited magnitude of these negative velocities suggests gradual ground settlement or slow compaction processes, rather than pronounced or abrupt subsidence. The coexistence of these low-intensity subsidence zones with stable areas (in green) and localised deformation elsewhere confirms the spatial heterogeneity of the deformation field. Most of the coherent points are concentrated within the blue range of the colour scale, corresponding to positive LOS velocities. These points are mainly located in the central area of the tailings dam and the upper part of the waste piles. These values indicate motion toward the satellite, interpreted as apparent uplift of the ground surface. Such behaviour may be related to slope readjustments, stress redistribution, or partial elastic rebound following ground compaction in nearby sectors affected by extraction activity.

The vertical deformation map (Figure 13), derived from the integration of ascending and descending SBAS datasets, provides a more physically representative model of the Björkdal mine’s dynamics by isolating vertical motion from East–West displacement components. The close alignment between the vertical decomposition and the ascending-orbit results, both in sign and central tendency, suggests that the ascending geometry was already capturing the dominant subsidence signal more effectively than the descending configuration. However, the observed “range compression” in the vertical solution is a critical finding; it demonstrates that single-look analyses likely overstate deformation magnitudes due to the conflation of vertical settlement and East–West slope movement. The spatial distribution of vertical displacement highlights a clear correlation between surface response and mining intensity. The pronounced subsidence clusters (−20 to −50 mm/year) identified in the central–west and northeastern sectors coincide with active underground footprints and tailings storage facilities. These patterns likely reflect a combination of ground compaction, the evolution of subsurface voids, and pore-water drainage processes common in mining environments. In contrast, the stability observed in the open-pit zone confirms its relative geomechanical maturity and the efficacy of current slope stabilization measures, which is consistent with the high-quality rock mass conditions reported for the Björkdal site, characterized by Rock Quality Designation (RQD) values on the order of 90% and Geological Strength Index (GSI) values in the range of 70–80, and is further supported by the use of relatively short bench heights (approximately 5 m), compared to the larger bench configurations typically employed in open-pit mining (12–15 m [69,70,71]), which contributes to enhanced slope stability and reduced deformation. Conversely, the localized areas of positive vertical velocity (uplift) in the central and northwestern sectors warrant careful interpretation. Rather than true “growth,” these signals are likely attributable to elastic rebound following significant overburden removal or complex geomechanical readjustments at the boundaries of extraction zones. The presence of these contrasting movements, subsidence in active zones and moderate uplift at the peripheries, confirms that the Björkdal mine is undergoing differential vertical deformation.

The statistical variance captured in Table 4 underscores the tailings dams (Figure 14) and waste piles (Figure 16) as the most geochemically dynamic features. Their high variability (standard deviations exceeding 10 mm/year) reflects the heterogeneous nature of unconsolidated material management. Ultimately, while the SBAS-derived vertical component resolves much of the ambiguity inherent in radar look-direction, the reliance on surface coherence means that deep-seated, non-linear deformation might be underrepresented. Therefore, these results should be viewed as a baseline for a multi-geodetic monitoring framework, combining InSAR with sub-surface sensors to fully characterise the mine’s 3D deformation field.

These identified deformation magnitudes and spatial patterns demonstrate strong scientific consistency when critically compared with previous literature. Specifically, the millimetric precursor displacements and minor slope creeping trends detected along the slopes of the Björkdal tailings storage facility align closely with the geotechnical indicators reported by Rana et al. (2024) [11], who verified that Sentinel–1 SBAS configurations are highly effective for screening early-stage structural instability in tailings dams. Furthermore, the localised, non-linear deformation histories observed near active mining sectors match the complex loading and subsidence dynamics documented by Ma et al. (2022) [8] and Rodríguez-Antuñano et al. (2023) [7] in anthropogenically stressed infrastructure, confirming that our results reflect genuine structural behaviours rather than uncompensated atmospheric noise.

Beyond the accuracy of the derived datasets, the implementation of this workflow highlights the significant methodological strengths of using the introduced open-source hyp3_sbas library. Unlike “black-box” proprietary software or standardised regional services, this approach provides a flexible parameterisation framework that can be tailored to the specific geotechnical challenges of a given mine site. At the same time, the framework is designed to abstract much of the underlying processing complexity, enabling users with predominantly mining or geotechnical backgrounds to execute the workflow without requiring advanced expertise in command-line environments or SAR-specific parameter tuning. Its cloud-native scalability further ensures that large Sentinel-1 stacks can be processed efficiently, facilitating a systematic and reproducible surveillance routine that is essential for operational safety. In this sense, the library bridges the gap between domain experts and advanced InSAR processing, as users are not required to manually define critical interferometric parameters, which are internally managed within the workflow. In that way, the implementation of the hyp3_sbas tool represents a significant advancement in making advanced DInSAR accessible for mining experts, facilitating the transition from episodic academic studies to continuous operational monitoring. However, the efficacy of the SBAS method remains intrinsically linked to signal quality, facing critical limitations such as coherence loss and temporal decorrelation, particularly in active mining sectors characterised by constant earthwork or rapid surface changes. Although the integration of ERA5 reanalysis data effectively mitigated topography-correlated delays, residual atmospheric artefacts or turbulent fluctuations may persist, necessitating expert interpretation to distinguish genuine subsidence from processing noise.

In the context of the European InSAR landscape and the XTRACT project framework, this study serves as a necessary bridge between global satellite products, like SAR Sentinel–1, and site-specific high-resolution metric needs. By integrating ascending and descending geometries to resolve the vertical deformation field, the inherent ambiguities of single-orbit LOS measurements are overcome, aligning this work with the current state-of-the-art in mining remote sensing. This analytical capability holds direct operational relevance, as it allows for the seamless integration of deformation maps into predictive maintenance and geotechnical risk assessment frameworks. The early detection of settlement in critical infrastructure provides mine operators with a transparent and verifiable monitoring tool. The transition from static deformation maps to dynamic, predictive models represents a paradigm shift in mining geotechnics. Future research within the XTRACT framework should focus on the deployment of Advanced Deep Learning architectures, such as Long Short-Term Memory (LSTM) networks or Graph Convolutional Networks (GCNs), which are uniquely suited for interpreting the non-linear temporal evolution of SBAS time series. By training these models on historical displacement patterns and triggering factors, such as rainfall intensity or extraction rates, it becomes possible to move beyond simple detection toward probabilistic forecasting of ground failure.

Furthermore, the synergy between satellite-derived vertical fields and in situ geotechnical instrumentation (e.g., automated total stations or MEMS-based inclinometers) is crucial for resolving the current ‘surface-only’ limitation of DInSAR. In complex mining environments like Björkdal, AI algorithms can act as a data-fusion engine, cross-referencing surface settlement velocities with deep-seated strain measurements. This multi-layered analytical framework could not only enhance the reliability of early-warning systems but also provide the high-fidelity data required for 3D geomechanical numerical modelling, ultimately bridging the gap between remote sensing observations and structural engineering requirements.

5. Conclusions

This study has successfully demonstrated the following points: (I) the SBAS technique is highly effective for characterising surface deformation at the Björkdal mine; (II) the efficient Python implementation of the methodological workflow using the hyp3_sbas library; and (III) the library significantly streamlines the SBAS analysis workflow. Specifically, the first point confirms that SBAS reliably captures both gradual and localised deformation patterns; the second point highlights the efficiency and semi-automation provided by the Python implementation, which also contributes to reducing the learning curve of the complex SBAS methodology for non-expert users; and the third point highlights that automating the workflow reduces processing time and facilitates reproducible analyses.

The main findings, considering that SBAS is a robust, well-established, and widely used technique in mining environments, reveal a complex and heterogeneous deformation field. In this context, integrating ascending and descending orbital configurations proved essential to isolate the vertical displacement component and correct the geometric distortions inherent in single-look LOS measurements. The resulting vertical deformation maps accurately identified subsidence clusters in active extraction zones and tailings facilities while also confirming the geomechanical stability of the open-pit area.

The implementation of the open source hyp3_sbas library underscores its significant operational potential for the mining industry. Its cloud-native scalability, flexible parameterisation, and open accessibility allow for a degree of transparency and reproducibility that proprietary solutions often lack. By automating the transition from raw Sentinel-1 data to GIS-ready displacement time series, this workflow enables mine operators to perform systematic, high-frequency surveillance of critical infrastructure without the need for prohibitive computational resources. Consequently, this work recommends the broad adoption of open-source DInSAR workflows within European risk management frameworks. Standardising these methodologies would not only enhance regulatory oversight but also provide a cost-effective early-warning tool to mitigate geotechnical hazards on a continental scale.

Regarding future research directions, several pathways will be pursued within the framework of the XTRACT project to improve and extend the capabilities of the proposed open-source workflow. First, future work will focus on integrating independent, in situ geodetic measurements such as continuous GNSS campaigns, high-precision levelling, or corner reflectors to perform a rigorous quantitative validation and absolute accuracy assessment of the satellite-derived deformation rates, bridging the current gap in ground-truth data. Second, the current hyp3_sbas processing framework will be extended through multi-sensor data fusion, integrating C-band Sentinel–1 datasets with high-resolution L-band (e.g., NISAR) or X-band (e.g., TerraSAR-X) radar imagery to improve phase coherence over challenging vegetated slopes and capture highly localized structural deformations in greater detail. Ultimately, these steps aim to evolve this satellite-based baseline into a comprehensive 3D multi-geodetic monitoring ecosystem by coupling SBAS displacement fields with geotechnical sensors like extensometers. Furthermore, embedding advanced Artificial Intelligence (AI) and machine learning algorithms into the time-series analysis will be crucial to automatically detect and flag non-linear acceleration anomalies. This integration will support the development of predictive maintenance models, digital twins, and automated early-warning systems, transforming the current hazard-screening methodology into a proactive, risk-based management strategy for sustainable mining infrastructure.