Application of UAV, GNSS and InSAR Techniques in the Raw

Material Supply Chain

Tanja Grabrijan

, Krištof Oštir

, Klemen Kozmus Trajkovski

, Dejan Grigillo

Veronika Grabrovec Horvat

, Polona Pavlovčič Prešeren

, Veton Hamza

, Antonio Pepe

Fabiana Calò

, Francesco Falabella

, Enoc Sanz-Ablanedo

, Mihaela Gheorghe

, Teodora Selea

and Ana Cláudia Teodoro

University of Ljubljana, Faculty of Civil and Geodetic Engineering, Jamova cesta 2, Ljubljana, Slovenia

Italian National Research Council, Institute for Electromagnetic Sensing of the Environment (IREA),

328, Diocleziano, Napoli, Italy

Grupo de Investigación en Geomática e Ingeniería Cartográfica (GEOINCA), Universidad de León, León, Spain

GMV Innovating Solutions SRL, Bucharest, Romania

University of Porto, Department of Geosciences, Environment and Spatial Planning, and Institute of Earth Sciences (ICT),

Rua do Campo Alegre 687, Porto, Portugal

{tanja.grabrijan, kristof.ostir, klemen.kozmus-trajkovski, dejan.grigillo, veronika.grabrovec-horvat,

Keywords: Mining, UAV, Tri Stereo, SfM Photogrammetry, Low-Cost GNSS, InSAR, Super-Resolution.

Abstract: The marble quarry, located in southern Austria produces high-quality marble, both in open-pit and

underground extraction sites. Extraction, transportation and accumulation of material require close monitoring

to maintain the stability of the whole area and to observe changes in waste dumps. In this study, we have

shown how different methodologies can be used to support the monitoring of the entire raw material supply

chain. Several unmanned aerial vehicle (UAV) surveys were performed to compute high-resolution digital

elevation models (DEM) to serve as a reference for comparing stereo and tri-stereo DEMs calculated from

satellite imagery. A low-cost global navigation satellite system (GNSS)-based monitoring system was set up

to estimate horizontal and vertical displacements. In addition, the state-of-the-art technique of Interferometric

synthetic aperture radar (InSAR) provided displacement time series for a broader area.

1 INTRODUCTION

The Secure and Sustainable Supply of Raw Materials

for EU Industry (S34I) project aims to increase

European autonomy over raw materials resources

through research and development of new data-driven

methods for analysing Earth observation (EO) data. It

will improve the systematic exploration of minerals

and ensure continuous monitoring of all mining

activities, i.e., extraction, closure, and post-closure.

This paper presents the preliminary results using

various techniques and methods, including satellite

data, airborne systems, unmanned aerial vehicles

(UAVs), ground-based measurements, and traditional

in-situ methods. Improved volume maps of mining

https://orcid.org/0009-0009-7374-9251

waste deposits using a multispectral sensor mounted

on a UAV have been used at the extraction site.

Ground instability maps are obtained through new

InSAR methods for the Sentinel-1 radar data to

compute long-term, i.e. 2014-2023 ground

displacement time series. Mineral stockpile volume is

calculated through satellite photogrammetry from

optical data. Displacement time series are measured

and analysed using low-cost global navigation

satellite system (GNSS) receivers. The results

presented are preliminary and will require further

validation to ensure their accuracy and reliability for

long-term monitoring applications.

Grabrijan, T., Oštir, K., Trajkovski, K. K., Grigillo, D., Horvat, V. G., Prešeren, P. P., Hamza, V., Pepe, A., Calò, F., Falabella, F., Sanz-Ablanedo, E., Gheorghe, M., Selea, T. and Teodoro, A. C.

Application of UAV, GNSS and InSAR Techniques in the Raw Material Supply Chain.

DOI: 10.5220/0013493100003935

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 11th International Conference on Geographical Information Systems Theory, Applications and Management (GISTAM 2025), pages 331-338

ISBN: 978-989-758-741-2; ISSN: 2184-500X

331

2 STUDY AREA

The extraction site is located in southern Austria and

produces high-quality marble, both in open-pit and

underground extraction sites. This marble formed

during the Devonian period, around 350-400 million

years ago, and has a colour ranging from light grey to

white. This material is used to produce mineral fillers

and pigments. In this study, only the open pit mine is

considered. The slopes are represented by hard rock,

which requires high-precision measurements to

assess their stability. The area of interest is about 1.2

and includes a waste dump and an active quarry

(Figure 1).

Figure 1: The area of interest: waste dump and active

quarry.

3 InSAR DISPLACEMENT

MAPPING

Multi-frequency and multi-orbit SAR satellite data

were used for Interferometric SAR (InSAR)

displacement mapping purposes. C-Band Sentinel-1

A/B ascending and descending complex SAR images

gathered from October 2014 to March 2023 were

used. Complementary to medium-resolution

Sentinel-1 images, high-resolution X-Band COSMO-

SkyMed SAR images acquired along the descending

orbit were also used.

Sentinel-1 images were independently processed

by following the standard steps of SAR image

registration, topography removal using an external

one arc-second Shuttle Radar Topography Mission

(SRTM) Digital Elevation Model (DEM), orbit

compensation via precise Sentinel-1 supplementary

information, additional fine co-registration procedure

on the images, and finally, interferograms generation.

The interferograms were then unwrapped using the

multigrid algorithm (Falabella et al., 2022) to obtain

the high-resolution unwrapped phases. Line-of-sight

(LOS)-projected ground displacement time series

were obtained by inverting the unwrapped single-

look interferograms via Small BAseline Subset

(SBAS) algorithm (Berardino et al., 2002).

The same processing operations, up to the

unwrapping stage, were also performed on COSMO-

SkyMed datasets. Subsequently computed Sentinel-1

vertical (up-down) and horizontal (east-west) ground

displacement time-series were used in feedback to the

multigrid unwrapping operation for flattening the

COSMO-SkyMed interferograms. This auxiliary step

was incorporated into the processing flow to facilitate

phase-unwrapping by reducing low-frequency

displacement contributions, thereby minimizing

ambiguity numbers during the unwrapping process.

Once they were unwrapped, the COSMO-SkyMed

LOS displacements were obtained, as well as the

enhanced combined Sentinel-1 (resampled onto the

spatial resolution grid of COSMO-SkyMed) and

COSMO-SkyMed vertical (up-down) and horizontal

(east-west) displacement time-series.

The multisensor data combination aims to

enhance the temporal sampling of measurements

rather than perform cross-calibration. Independent

ground displacement time-series are computed and

geocoded before solving a constrained linear problem

to derive 2-D ground displacement time-series,

following Pepe et al. (2016).

Results are shown in Figure 2 and Figure 3. Figure

2 shows the 2-D InSAR products obtained using the

detailed multigrid InSAR methodology, while Figure

3 shows displacement time-series related to the SAR

measure point indicated in Figure 2.

4 DEEP LEARNING FOR SAR

IMAGE ENHANCEMENT

We investigated several deep learning models using

Sentinel-1, Sentinel-2 and COSMO-SkyMed (CSK)

data to improve the resolution of Synthetic Aperture

Radar (SAR) images. The dataset used in this study

comprises 135,013 patches extracted these images.

The CSK data preprocessing involved using a half-

band filter to generate the corresponding low-

resolution (LR) images for training purposes. This

filter applies downsampling while preserving phase

information,

which

essential

for

SAR

applications

S34I 2025 - Special Session on S34I - From the Sky to the Soil

332

Figure 2: Deformation velocity maps generated through advanced InSAR processing technique show vertical (left) and

horizontal (right) components of displacements.

Figure 3: Displacement time-series related for points indicated in the upper map. From the top: first two time series show the

subsidence (Up-Down) and Horizontal (E-W) deformation trend. At the bottom: time series related to stable measurement

points. Also, the velocity of deformation is reported in the graphs.

such as interferometry. Each SAR CSK patch consists

of real and imaginary components, and the filter was

applied independently to each channel. Sentinel-1

patches from the Sen1-2 dataset, and its optical

Sentinel-2 equivalents acted as supplementary inputs

in several models.

We evaluated the following deep learning

architectures for SAR super-resolution:

1. Residual-in-Residual Dense Block (RRDB)

(Wang et al., 2018): A convolutional neural

network that incorporates multiple residual

blocks to enhance feature extraction and avoid

information loss.

2. Super-Resolution U-Net (SRUN) (Ronneberger

et al., 2015): A modified U-Net architecture

optimized for SAR data, integrating

downsampling and upsampling layers for

effective feature recovery.

3. Optical-Guided Super-Resolution Network

(OGSRN) (Li et al., 2015): A dual-network

system utilizing both SAR and optical images to

enhance SAR resolution.

Application of UAV, GNSS and InSAR Techniques in the Raw Material Supply Chain

333

4. DC2SCN (Deep CNN with Skip Connection and

Network in Network): A model designed

specifically for SAR super-resolution, handling

the real and imaginary parts of SAR data

separately to preserve phase integrity.

Each model was trained using a combination of

Sentinel-1, COSMO-SkyMed, and Sen1-2 dataset

patches. The training process involved Z-score

normalization, where pixel intensities were

standardized to ensure optimal network convergence.

The models were assessed using Peak Signal-to-

Noise Ratio (PSNR) and Structural Similarity Index

Measure (SSIM). As a baseline, we compared the

deep learning models against bicubic interpolation, a

commonly used upscaling method in image

processing.

The experimental results, evaluated at a global

scale using the Sen1-2 dataset, indicate significant

improvements in SAR image resolution using deep

learning techniques. The RRDB model outperformed

the other architectures, achieving the highest PSNR

and SSIM scores. The OGSRN model, which

incorporates Sentinel-2 optical data, performed worse

than RRDB despite its additional guidance input. This

could be explained by variations in domain between

SAR and optical images, which might cause feature

extraction conflicts.

Overall, the results show that deep learning-based

super-resolution can enhance SAR imagery quality

significantly. Compared to bicubic interpolation, the

best-performing model demonstrated an 18% PSNR

increase (e.g., ~25 dB to ~29.5 dB) and a 2% SSIM

improvement (e.g., 0.85 to 0.87), with RRDB

outperforming OGSRN by 2-3 dB, highlighting SAR-

only models’ strengths. The visual result presented in

Figure 4 (e.g., original vs. filtered COSMO-SkyMed

patches) reinforced these findings, showing sharper

edges and textures. This approach enhances SAR

imagery for ground instability mapping in Gummern,

aligning with S34I goals for extraction and closure

phases of mining.

These findings underscore the potential of deep

learning for SAR image enhancement and highlight

future directions for improving model architecture

and dataset quality.

Figure 4: Original patch (first row) and filtered patch

(second row) of COSMO-SkyMed sample.

5 UAV SURVEY

UAV Photogrammetry was used to perform several

UAV surveys at the quarry with Phantom 4 Real Time

Kinematic (RTK) + RGB camera. Because of

variable terrain heights, terrain following method, i.e.

flying at the same height above ground, proved to be

the most efficient. The steepest part of the quarry was

additionally surveyed with so-called angled flight

mode. Ground Control Points (GCPs) were

established for each survey and measured with RTK

GNSS equipment to use later in image processing,

ensuring better product accuracy. The latter are dense

point clouds, 3D models, digital surface models

(DSM), digital terrain models (DTM), and

orthophotos. The flight height above ground was 100

m, resulting in a Ground Sampling Distance (GSD) of

3 cm.

Since multiple surveys were conducted,

performing change detection for the same product

across different dates is possible. One of the

possibilities is monitoring deposited material in waste

dumps by comparing two or more DEMs, which is

especially useful in the northern part of the quarry at

waste dumps. Figure 5 represents a calculated

difference between DEMs, created from two different

surveys.

High-resolution UAV data (especially DEM)

served as a reference for less detailed remote sensing

methods for terrain mapping, such as tri-stereo

mapping, which is discussed in the Chapter 6.

S34I 2025 - Special Session on S34I - From the Sky to the Soil

334

Figure 5: Differences between two DEMs produced from

UAV.

6 TRI-STEREO MAPPING

Pléiades Neo tri-stereo consists of images that were

collected during the same pass from different sensor-

viewing angles (along-track stereo): forward (F),

close to nadir (N) and backward (B). In the study,

only panchromatic bands are used. Table 1

summarises the main acquisition parameters and

characteristics of the images used.

Table 1: Acquisition parameters of the used Pléiades Neo

tri-stereo images: incident angles (across-track, along-track

and overall), GS), area, the computed [5] stereo intersection

angles between two Pléiades scenes and their base-to-

height ratio (B/H).

View F N B

Across-track (

) 6.34 4.56 2.76

Along-track (

) -7.67 -0.29 7.10

Overall (

) 9.90 4.57 7.61

GSD (m) 0.31 x 0.31 0.30 x 0.30 0.30 x 0.30

Area (km

) 130.89 130.84 130.31

Convergence (

) 7.55 (FN) 7.60 (NB) 15.14 (FB)

B/H Ratio 0.13 (FN) 0.13 (NB) 0.27 (FB)

The initial orientation of the images was set using the

Rational Function Model (RFM) containing 80

Rational Polynomial Coefficients (RPCs). RPCs

allow the transformation between object and image

space. Additionally, 25 tie points were automatically

measured to improve the relative orientation of the

images. GCPs, which were evenly distributed across

the area, avoiding hilly and overgrown areas, were

then measured on all tri-stereo images to improve

their georeferencing. Multiple rounds of bundle block

adjustment were performed using different GCPs as

checkpoints. The maximum difference in the

individual coordinate components of the checkpoints

was 31 cm. In the final block adjustment, we used all

GCPs surveyed in the field. Final Root Mean Square

Error (RMSE) was 0.35 on image level (given in

pixels) and 0,21 m on GCPs (0.14 m in Easting, 0.15

m in Northing and 0.03 m in height). The dense image

matching algorithms were applied to oriented images

to extract surface points, from which the DEM with a

spatial resolution of 0.5 m was interpolated. Figure 6

shows the terrain shaded relief of DEM produced

from Pléiades Neo tri-stereo panchromatic images.

Figure 6: Terrain shaded relief of digital elevation model

produced from tri-stereo imagery.

The accuracy of the produced DEM was estimated

by comparing it to those produced with UAV surveys

that were performed before and after the tri-stereo

acquisition. 190 control points were extracted from

UAV DEMs in the areas where no changes were

detected between two consecutive surveys.

Additionally, the satellite DEM was registered to

those un-changed areas using Iterative Closest Point

(ICP). The high accuracy of DEM, produced from

satellite images was evaluated using the methodology

proposed by EuroSDR (Höhle & Potuckova, 2011).

We have achieved a mean value of 4.9 cm, RMSE of

30.1 cm, and standard deviation of 29.7 cm. Figure 7

shows the absolute distances between Pléiades Neo

DEM and UAV DEM. The most extensive distances

occur at the steep terraces in the quarry. The RMSE

size is in the range of spatial resolution of the images

and comparable to the results obtained by other

authors (e.g. Loghin et al., 2020).

Pléiades Neo-derived DEM was additionally

compared to DEM, created from WorldView-2 data.

The latter captures images of the same area from

different angles (stereo images). The slight

displacement between the stereo pairs allows for

depth perception and DEMs creation. The volume

changes in the areas where waste dumping occurred

are summarized in Figure 8.

Application of UAV, GNSS and InSAR Techniques in the Raw Material Supply Chain

335

Figure 7: The absolute distances between Pléiades Neo

DEM and UAV DEM after the registration.

Figure 8: Volumes of waste dumps between Pléiades Neo-

derived DEM and from WorldView-2 DEM.

7 GNSS MONITORING

Low-Cost GNSS Monitoring Station (LGMS) was

developed using low-cost GNSS receivers and

antennas to continuously monitor displacement with

high precision and decreased costs (Hamza et al.,

2021, 2023). The main components of the LGMS are

the ZED-F9P GNSS module and survey-calibrated

antenna (AS-ANT2BCAL). The LGMS can work in

remote areas without electricity connection since it

uses solar panels to charge the battery from which the

systems get the electricity (Figure 9). It acquires and

stores GNSS observations continuously, which are

later post-processed to estimate the daily

displacement of the monitoring locations.

Four LGMS stations were deployed in Gummern,

with two stations, F1 and F2 (Figure 10), installed on

stable ground to be used as reference stations. These

reference stations were used to estimate

displacements at ST1 and ST2 (Figure 11 and Figure

12), which were positioned on the waste dump where

displacements were expected.

Figure 9: LGMS installed in Gummern.

Geodetic monitoring of displacements started on

03.04.2023 and is still ongoing, with the LGMS

operating continuously for almost two years. The

open-source software RTKLIB (demo33_g) was used

to post-process GNSS observations, with further

details on the processing strategy provided in Table 2.

Table 2: Processing parameters used in RTKLIB.

Positioning mode Static

Frequencies L1+L2/E5b

Filter type Forward

Cutt-off angle 10

Sampling rate 10s

Iono/Troposphere

correction

Broadcast/Sastamoinen

Satellite ephemeris Broadcast

Satellite systems GPS+GLO+Galileo

Ambiguity Continuous (Lambda)

GNSS Antenna AS-ANT2BCAL(igs14)

The results indicated that in ST1 horizontal and

vertical displacements of 20 and 40 mm occurred,

respectively, while larger displacements of 40 mm

horizontally and 70 mm vertically were noticed in

ST2.

S34I 2025 - Special Session on S34I - From the Sky to the Soil

336

Figure 10: Locations of GNSS measuring stations.

Figure 11: Horizontal and vertical displacement in ST1.

Figure 12: Horizontal and vertical displacement in ST2.

8 CONCLUSIONS

This work presents preliminary results of several EO

methodologies designed to address specific needs in

monitoring mining activities. We produced DEMs

using photogrammetric data and satellite imagery,

specifically from stereo and tri-stereo images. The

results demonstrate that the quality of the DEM

generated from satellite data is comparable to that of

UAV-derived DEMs. Additionally, we found that

both InSAR and low-cost GNSS data capture the

same ground motion trends. While GNSS provides

daily displacement measurements, which InSAR does

not, InSAR offers a much denser result with broader

spatial coverage. Furthermore, we showed that SAR

imagery can enhance spatial resolution when

processed with higher-resolution images.

By evaluating these methods, we provide decision

makers with important information on the trade-offs

between accuracy, spatial coverage and cost-

effectiveness. The results emphasize the importance

of a multi-sensor approach for comprehensive

monitoring. However, further work is needed to

validate these results. We are focusing on defining

pipelines that will utilize the most effective methods

and develop them into tools and services. This

includes estimating the cost and value that the

products will have for users so that operational

services can be offered commercially.

ACKNOWLEDGEMENTS

This study is funded by the European Union under

grant agreement no. 101091616

(https://doi.org/10.3030/101091616), project S34I –

SECURE AND SUSTAINABLE SUPPLY OF RAW

MATERIALS FOR EU INDUSTRY.

REFERENCES

Berardino, P., Fornaro, G., Lanari, R., & Sansosti, E.

(2002). A new algorithm for surface deformation

monitoring based on small baseline differential SAR

interferograms. IEEE Transactions on Geoscience and

Remote Sensing, 40(11), 2375–2383. https://doi.org/

10.1109/TGRS.2002.803792

Falabella, F., Serio, C., Masiello, G., Zhao, Q., & Pepe, A.

(2022). A Multigrid InSAR Technique for Joint

Analyses at Single-Look and Multi-Look Scales. IEEE

Geoscience and Remote Sensing Letters, 19.

https://doi.org/10.1109/LGRS.2021.3086271

Hamza, V., Stopar, B., Ambrožič, T., & Sterle, O. (2021).

Performance Evaluation of Low-Cost Multi-Frequency

GNSS Receivers and Antennas for Displacement

Detection. Applied Sciences, 11(4), 6666.

https://doi.org/10.3390/app11146666

Hamza, V., Stopar, B., Sterle, O., & Pavlovčič-Prešeren, P.

(2023). A Cost-Effective GNSS Solution for

Continuous Monitoring of Landslides. Remote Sensing,

15(9), 2287. https://doi.org/10.3390/rs15092287

Höhle, J. & Potuckova, M. (2011). Assessment of the

Quality of Digital Terrain Models. European Spatial

Data Research. https://www.eurosdr.net/sites/default/

files/uploaded_files/eurosdr_publication_ndeg_60.pdf

Li, Y., Zhou, L., Xu, F., & Chen, S. (2022). OGSRN:

Optical-guided super-resolution network for SAR

image. Chinese Journal of Aeronautics, 35(5), 204–219.

https://doi.org/10.1016/j.cja.2021.08.036

Loghin, A., Otepka-Schremmer, J., & Pfeifer, N. (2020).

Potential of Pléiades and WorldView-3 Tri-Stereo

DSMs to Represent Heights of Small Isolated Objects.

Application of UAV, GNSS and InSAR Techniques in the Raw Material Supply Chain

337

Sensors, 20(9), 2695. https://doi.org/10.3390/

s20092695

Pepe, A., Solaro, G., Calò, F., & Dema, C. (2016). A

minimum acceleration approach for the retrieval of

multiplatform InSAR deformation time series. IEEE

Journal of Selected Topics in Applied Earth

Observations and Remote Sensing. 9(8), 3883–3898.

https://doi.org/10.1109/JSTARS.2016.2577878

Ronneberger, O., Fischer, P., & Brox, T. (2015). U-Net:

Convolutional Networks for Biomedical Image

Segmentation. https://doi.org/10.48550/

arXiv.1505.04597

Wang, X., Yu, K., Wu, S., Gu, J., Liu, Y., Dong, C., Loy,

C. C., Qiao, Y., & Tang, X. (2018). ESRGAN:

Enhanced Super-Resolution Generative Adversarial

Networks. http://arxiv.org/abs/1809.00219

S34I 2025 - Special Session on S34I - From the Sky to the Soil

338