ArticlePDF Available

Supporting Long-Term Archaeological Research in Southern Romania Chalcolithic Sites Using Multi-Platform UAV Mapping

Authors:
  • Lower Danube Museum, Romania, Calarasi

Abstract and Figures

Spatial data play a crucial role in archaeological research, and orthophotos, digital elevation models, and 3D models are frequently used for the mapping, documentation, and monitoring of archaeological sites. Thanks to the availability of compact and low-cost uncrewed airborne vehicles, the use of UAV-based photogrammetry matured in this field over the past two decades. More recently, compact airborne systems are also available that allow the recording of thermal data, multispectral data, and airborne laser scanning. In this article, various platforms and sensors are applied at the Chalcolithic archaeological sites in the Mostiștea Basin and Danube Valley (Southern Romania). By analysing the performance of the systems and the resulting data, insight is given into the selection of the appropriate system for the right application. This analysis requires thorough knowledge of data acquisition and data processing, as well. As both laser scanning and photogrammetry typically result in very large amounts of data, a special focus is also required on the storage and publication of the data. Hence, the objective of this article is to provide a full overview of various aspects of 3D data acquisition for UAV-based mapping. Based on the conclusions drawn in this article, it is stated that photogrammetry and laser scanning can result in data with similar geometrical properties when acquisition parameters are appropriately set. On the one hand, the used ALS-based system outperforms the photogrammetric platforms in terms of operational time and the area covered. On the other hand, conventional photogrammetry provides flexibility that might be required for very low-altitude flights, or emergency mapping. Furthermore, as the used ALS sensor only provides a geometrical representation of the topography, photogrammetric sensors are still required to obtain true colour or false colour composites of the surface. Lastly, the variety of data, such as pre- and post-rendered raster data, 3D models, and point clouds, requires the implementation of multiple methods for the online publication of data. Various client-side and server-side solutions are presented to make the data available for other researchers.
Content may be subject to copyright.
Citation: Stal, C.; Covataru, C.;
Müller, J.; Parnic, V.; Ignat, T.;
Hofmann, R.; Lazar, C. Supporting
Long-Term Archaeological Research
in Southern Romania Chalcolithic
Sites Using Multi-Platform UAV
Mapping. Drones 2022,6, 277.
https://doi.org/10.3390/
drones6100277
Academic Editors: Diego
González-Aguilera, Parrinello
Sandro, Salvatore Barba,
Jesus Fernandez-Hernandez, Miguel
Angel Maté-González and Luis
Javier Sanchez-Aparicio
Received: 30 August 2022
Accepted: 20 September 2022
Published: 26 September 2022
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2022 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
drones
Article
Supporting Long-Term Archaeological Research in Southern
Romania Chalcolithic Sites Using Multi-Platform
UAV Mapping
Cornelis Stal 1,2 , Cristina Covataru 3, Johannes Müller 4, Valentin Parnic 5, Theodor Ignat 6,
Robert Hofmann 4and Catalin Lazar 3, *
1Department of Built Environment, HOGENT University of Applied Sciences and Arts, 9000 Ghent, Belgium
2Department of Geography, Ghent University, 9000 Ghent, Belgium
3
Research Institute of the University of Bucharest (ICUB), University of Bucharest, 050663 Bucharest, Romania
4Institute of Prehistoric and Protohistoric Archaeology, Kiel University, 24118 Kiel, Germany
5Lower Danube Museum Călăras
,i, 910001 Calarasi, Romania
6Bucharest Municipality Museum, 030167 Bucharest, Romania
*Correspondence: catalin.lazar@icub.unibuc.ro
Abstract:
Spatial data play a crucial role in archaeological research, and orthophotos, digital elevation
models, and 3D models are frequently used for the mapping, documentation, and monitoring of
archaeological sites. Thanks to the availability of compact and low-cost uncrewed airborne vehicles,
the use of UAV-based photogrammetry matured in this field over the past two decades. More recently,
compact airborne systems are also available that allow the recording of thermal data, multispectral
data, and airborne laser scanning. In this article, various platforms and sensors are applied at the
Chalcolithic archaeological sites in the Mosti
s
,
tea Basin and Danube Valley (Southern Romania). By
analysing the performance of the systems and the resulting data, insight is given into the selection
of the appropriate system for the right application. This analysis requires thorough knowledge of
data acquisition and data processing, as well. As both laser scanning and photogrammetry typically
result in very large amounts of data, a special focus is also required on the storage and publication
of the data. Hence, the objective of this article is to provide a full overview of various aspects of
3D data acquisition for UAV-based mapping. Based on the conclusions drawn in this article, it is
stated that photogrammetry and laser scanning can result in data with similar geometrical properties
when acquisition parameters are appropriately set. On the one hand, the used ALS-based system
outperforms the photogrammetric platforms in terms of operational time and the area covered. On
the other hand, conventional photogrammetry provides flexibility that might be required for very
low-altitude flights, or emergency mapping. Furthermore, as the used ALS sensor only provides a
geometrical representation of the topography, photogrammetric sensors are still required to obtain
true colour or false colour composites of the surface. Lastly, the variety of data, such as pre- and
post-rendered raster data, 3D models, and point clouds, requires the implementation of multiple
methods for the online publication of data. Various client-side and server-side solutions are presented
to make the data available for other researchers.
Keywords: southeastern Europe; Chalcolithic; archaeological sites; UAV; spatial data
1. Introduction
Image-based 3D modelling and laser scanning techniques are mature in cultural
heritage and archaeology [
1
,
2
]. Structure from motion and multi-view stereo algorithms
(SfM-MVS) are commonly used for image-based 3D modelling, allowing the processing of
large series of uncalibrated and unstructured photographs [
3
,
4
]. This processing generates
different results, such as orthophotos, digital elevation models (DEMs), or textured 3D
models. In archaeology, the technique is used for various applications, such as feature
Drones 2022,6, 277. https://doi.org/10.3390/drones6100277 https://www.mdpi.com/journal/drones
Drones 2022,6, 277 2 of 24
modelling, finds processing, and creating time series of excavations [
5
]. Furthermore, this
technique is used in many scientific disciplines thanks to the high geometric accuracy
and resolution of the resulting 3D models. In addition, SfM-MVS is appreciated for the
straightforward procedure regarding the acquisition and processing of data, the ability to
generate models with photorealistic texture, and the possibility of performing non-contact
measurements. Such a highly accurate scientific recording technique is required for all past
features since archaeological excavation is a destructive process [6], and revealed features
are exposed to many destructive environmental elements.
Additionally, traditional 2D documentation is complemented with a more realistic
and attractive 3D representation of the archaeological heritage [
7
]. However, the intro-
duction of uncrewed aerial systems (UAS) has allowed archaeologists to improve the
effectiveness of their modelling work with airborne data [
8
]. Traditionally, the necessary
images for these projects are obtained terrestrially by hand or tripod. The flexibility of
image-based modelling platforms, tools, and techniques enables the creation of high-quality
3D reconstructions of a various relics and artifacts. The availability of UASs suitable for
photogrammetric applications has increased significantly over the past decades, and the
technique is well established for topographic mapping and archaeological documenta-
tion [9,10].
Besides image-based 3D modelling, laser scanning is a valuable technique for archae-
ological applications [
11
]. Airborne laser scanning (ALS), in particular, is well known
for the topographic modelling of landscapes and offers a wide range of point densities
and accuracies. For georeferencing the final point cloud, each measurement combines
synchronised observations of a position and orientation measurement system (POS). After
post-processing, a relatively accurate point cloud is obtained, which can be used for further
analysis. Until recently, applications of ALS were limited to systems carried by piloted
airplanes or helicopters. The system’s portability was mainly hampered by the price and
size, as well as the reasonable energy consumption. However, UAV-based ALS has rapidly
evolved over the past five years, and the availability and applications of the systems are
increasing [12,13].
Archaeological research in the target area (Mosti
s
,
tea Basin and Danube Valley) strongly
focuses on applying modern methods for acquiring, processing, and interpreting various
parameters. Image-based 3D modelling techniques have been essential in many regional
projects. Combining these techniques with UAVs (uncrewed aerial vehicles), valuable data
sources can be tapped. Moreover, the construction of these data can be essential in the
spatial analysis of archaeological sites and cultural landscapes. The University of Bucharest
oversees the project, and various platforms and sensors are available to acquire UAV-based
data. Experiments and data acquisition are more specifically organised using conventional
true colour cameras (RGB), multispectral cameras (MS), thermal cameras, and ALS.
Considering this, the question arises about what type of system should be used for
a specific task in terms of possibilities and limitations. Hence, the characteristics of the
acquired and processed data should be investigated, and in-depth workflows should be
developed. Lastly, the storage and publication of the data should be considered. This aspect
is essential for sharing data with other researchers, and data sustainability and outreach.
This paper goes into greater detail on these crucial subjects, starting with an overview
of the study area and project outline, and a section dedicated to data acquisition, discussing
the used hardware and the flight plan preparation. The fourth section offers a data pro-
cessing workflow, emphasising enhancing the accuracy of picture geotags and the software
used to handle ALS and image data. The results of this research are discussed in the fifth
section, along with the quality of the data as well as its storage and publication. Final
remarks are given at the end of the paper to better understand the impact of using this kind
of technology.
Drones 2022,6, 277 3 of 24
2. Study Area and Project Outline
The targeted area is located in the southeast of Romania, Călăra
s
,
i County, Muntenia
Region, one of the most intensively investigated regions in terms of Chalcolithic sites in the
Northern Balkans. Geographically, we are dealing with two micro-regions, and the sites
investigated are spread across the Mosti
s
,
tea Basin and Danube Valley, from the cities of
Oltenit
,a to Călăras
,i (Figure 1).
Drones 2022, 6, x FOR PEER REVIEW 3 of 25
Final remarks are given at the end of the paper to better understand the impact of using
this kind of technology.
2. Study Area and Project Outline
The targeted area is located in the southeast of Romania, Călărași County, Muntenia
Region, one of the most intensively investigated regions in terms of Chalcolithic sites in
the Northern Balkans. Geographically, we are dealing with two micro-regions, and the
sites investigated are spread across the Mostiștea Basin and Danube Valley, from the cities
of Oltenița to Călări (Figure 1).
Figure 1. Situation of the study area and test sites in Romania (source: OpenTopoMap, coordinates
in EPSG:4326).
The target area was not chosen randomly, as it has been the most investigated region
in Romania since the beginning of the 20th century. Romanian archaeologists carried out
intensive excavations in the Chalcolithic and Middle/Late Neolithic sites of Călări
County, which formed the basis for the definition of two emblematic prehistoric cultures
in the Northern Balkans, the Boian and the Gumelnița [14,15]. Moreover, these investiga-
tions contributed to determining the relative chronology of archaeological cultures in Ro-
mania. On the other hand, the relief from Mostiștea Valley has changed drastically due to
the land and agricultural improvement activities from the Communist Period, which
makes the selected area an ideal place to apply the proposed technologies and quantify
the landscape changes.
Generally, due to natural processes combined with massive anthropic interventions
(planning and execution of land improvement work), a series of lakes were formed along
the Mostiștea Valley [16], while the Danube Meadow area has undergone major anthro-
pogenic disturbance with the making of flood protection systems and the recovery of land
for farming [17,18]. Along the way, a series of archaeological sites have been drastically
affected. In this respect, the current study is a part of a wider research project concerning
the integration of laser scanning and image-based 3D modelling methodologies for mon-
itoring and documenting cultural heritage sites in Southern Romania, and also to support
the current archaeological investigations.
Figure 1.
Situation of the study area and test sites in Romania (source: OpenTopoMap, coordinates in
EPSG:4326).
The target area was not chosen randomly, as it has been the most investigated region
in Romania since the beginning of the 20th century. Romanian archaeologists carried
out intensive excavations in the Chalcolithic and Middle/Late Neolithic sites of Călăra
s
,
i
County, which formed the basis for the definition of two emblematic prehistoric cultures in
the Northern Balkans, the Boian and the Gumelni
t
,
a [
14
,
15
]. Moreover, these investigations
contributed to determining the relative chronology of archaeological cultures in Romania.
On the other hand, the relief from Mosti
s
,
tea Valley has changed drastically due to the
land and agricultural improvement activities from the Communist Period, which makes
the selected area an ideal place to apply the proposed technologies and quantify the
landscape changes.
Generally, due to natural processes combined with massive anthropic interventions
(planning and execution of land improvement work), a series of lakes were formed along the
Mosti
s
,
tea Valley [
16
], while the Danube Meadow area has undergone major anthropogenic
disturbance with the making of flood protection systems and the recovery of land for
farming [
17
,
18
]. Along the way, a series of archaeological sites have been drastically
affected. In this respect, the current study is a part of a wider research project concerning the
integration of laser scanning and image-based 3D modelling methodologies for monitoring
and documenting cultural heritage sites in Southern Romania, and also to support the
current archaeological investigations.
Our attention was first focused on tell settlements, but the investigation was extended
to other kinds of sites belonging to the Chalcolithic period (c. 5000-4000 cal. BCE). Alongside
the Mosti
s
,
tea Basin, we investigated ten sites (three cemeteries, one flat settlement, and six
tell settlements). Moreover, from the Danube Meadow region, we investigated six other
Drones 2022,6, 277 4 of 24
tell settlements and one small settlement rediscovered during our 2019 survey (Figure 1).
This batch of sites provides a reasonable basis for analysis, which has enabled us to push
forward the knowledge of landscape changes (human impact and natural processes) in the
target area.
From an archaeological perspective, the 5th millennium BC represents the last stage
of the Neolithic period, and in Romania and Bulgaria, it is known as the Copper Age,
Eneolithic, or Chalcolithic period [
19
21
], or the “golden 5th millennium”. This timespan
was the most flourishing development period for the Chalcolithic civilisation from South-
eastern Europe, when significant changes and technological advances (e.g., metallurgy)
occurred in the human community. This time saw the rise of tell settlements as the main
form of habitation, completed by some flat settlements, cemeteries, water sources, and
settlements surrounding land used for agriculture, animal breeding, fishing, gathering, and
the procurement of other resources [19,20,22,23].
The areas of the Danube River Basin and Mosti
s
,
tea Valley are fascinating, as some
sites have been under archaeological investigations since the beginning of the Romanian
Archaeological School. In this respect, in 1923, under the guidance of V. Pârvan, several
archaeologists carried out field surveys to complete the few known data about prehistoric
communities in the Romanian Plain [
24
26
]. Thus, Sultana-Malu Ro
s
,
u was the first tell
settlement discovered and investigated in 1923 by Andrie
s
,
escu [
24
]. Moreover, R. Vulpe
and V. Dumitrescu’s investigation alongside the Mosti
s
,
tea River and between the village
of Doroban
t
,
u and the city of Călăra
s
,
i generated the first map of archaeological sites in
this region, when several sites in our study were identified (Cune
s
,
ti, Rasa, Vără
s
,
ti) [
27
].
The homonym settlement of the Gumelni
t
,
a culture was first investigated in 1924 by Vl.
Dumitrescu, and based on these investigations, the Gumelni
t
,
a culture was defined [
28
,
29
].
Moreover, some sites were identified and excavated by a German archaeologist team
(Chiselet site) led by Leo Frobenius at the end of the First World War [
25
]. Then, during
the interwar period, the tell settlement from Măgureni was investigated [
30
]. Various land
improvement works from the 1970s and 1980s performed in Mosti
s
,
tea Valley led to the
damage of other Chalcolithic sites (Măriu
t
,
a, Sultana-Valea Orbului,
S
,
einoiu, Vlădiceasca
I-II, Ulmeni). Other times, some natural disasters favoured the research of other sites in the
area (Spant
,ov) [31,32].
In 2006, a European project in collaboration with the AARG (Aerial Archaeology
Research Group) turned its attention to the Mosti
s
,
tea Valley [
33
]. The project’s intention
was to identify new areas of archaeological interest and determine the coordinates of the
sites mapped since the 1923 field investigation. Thus, multiple cartographic works, along
with satellite images, fieldwalking, and new flights for obtaining new aerial photos, were
performed in the targeted area. As a result, at the project’s end, it was possible to introduce
more than 200 new sites identified in the investigated area [34,35].
The excavations in the Sultana tell in 2001 and the year-by-year continuation of archae-
ological research in the area until now, without interruptions, led to a mapping program of
the sites in the Mosti
s
,
tea Basin, which led to the accumulation of new archaeological and
geospatial data [14].
In 2021, a new international project entitled “The dynamics of the prehistoric commu-
nities located in the Mosti
s
,
tea Valley and Danube Plain (between Olteni
t
,
a and Călăra
s
,
i)”
was started, organised by the ArchaeoSciences Division of the Research Institute of the Uni-
versity of Bucharest (ICUB) and Kiel University (Germany), in partnership with HOGENT,
University of Applied Sciences and Arts (Belgium), Museum of Bucharest, Museum of the
Lower Danube Călăra
s
,
i, Museum of Gumelni
t
,
a Civilization Olteni
t
,
a, and “Vasile Pârvan”
Institute of Archaeology (Romania).
This new project, in addition to the continuation of the geospatial data collection
activities, involves an intense program of non-intrusive prospecting (e.g., electrical, seismic,
and geomagnetic prospecting) together with coring of the sites in the Danube Basin and
the Mosti
s
,
tea Valley, which will allow a new step in the research program of Chalcolithic
sites in the target area.
Drones 2022,6, 277 5 of 24
All these circumstances—some beneficial, others catastrophic—created the context of
a complex and intense investigation in the targeted area, but also the accumulation of a set
of data that allowed us to move to the next level of analyses presented in this paper.
3. Data Acquisition
In this and other sections, data quality parameters are used in accordance with Uren
and Price [
36
]. Accuracy is defined as the systematic error of the data or the deviation of
the data from the true values. As such, this value corresponds to the mean error and is
also called the correctness of the data. The precision is used as a measure to quantify the
random error, corresponding to the standard deviation of the data. Finally, the resolution,
ground sampling distance (GSD), and point densities are used as parameters to describe
the minimum detectable object size.
3.1. Airborne Image-Based Sensors
Since the summer campaign of 2018, image-based 3D modelling has been applied in
the Mosti
s
,
tea Basin and the Danube Valley to measure in the visible range of the electro-
magnetic spectrum, resulting in true colour (RGB) composites and models. Regarding the
airborne aspect of these campaigns, a DJI Phantom 4 and DJI Mavic 2 Pro are intensively
used for the data acquisition (Table 1) [
37
]. These systems are commonly used for archaeo-
logical prospection and are appreciated by many users for their relatively low cost, flexible
deployment, operational time of 20 to 25 min, and reasonable camera specifications [
38
]. In
addition, both systems are compatible with autopilot software developed by the platform
producer, which facilitates the systematic execution of flights, given a series of predefined
parameters and constraints, such as coverage, flying speed, altitude, and GSD [39].
Table 1. Specifications of the used image acquisition hardware [37].
DJI Phantom 4 DJI Phantom 4
(MS) DJI Mavic 2 Pro DJI Mavic 2
Enterprise
Resolution (px) 4000 ×3000 1600 ×1300 5472 ×3078 4056 ×3040
Sensor size (mm)
6.2 ×4.9 4.9 ×4.0 13.8 ×7.7 6.17 ×4.55
Pixel size (µm) 1.56 3.04 2.53 1.51
Focal length
(mm) 3.6 5.7 10.0 4.5
In addition, new UAVs equipped with thermal and multispectral sensors were de-
ployed during the 2022 summer campaign. For archaeological research, both multispectral
and thermal imagery have the potential to identify areas of possible archaeological re-
search interest based on local differences in the spectral signature of the recorded echo,
e.g., through applied crop mark detection [
40
,
41
] and difference in thermal radiation [
42
].
The thermal sensor is an Uncooled VOx Microbolometer that allows measurements in the
8–14 µm
spectral band and is integrated into a dual-camera system of the DJI Mavic 2
Enterprise system (Table 1).
The specifications of the RGB sensor are not as good as the DJI Mavic 2 Pro, but
the system still allows the creation of orthophotos and DEMs with reasonable accuracy
(Figure 2left, middle). The white dots in this image represent the ground control point
(GCP) locations. The use of these points is discussed later in this paper. In combination
with the new thermal sensor, simultaneous true colour (RGB) and thermal images can
be obtained from the study area. Although the resolution of this sensor is limited to
1600 ×1200 pixels,
fixing both cameras into one camera body still allows the use of the
resulting thermal images in combination with conventional true colour 3D modelling
(Figure 2right). Data processing typically results in an orthorectified projection of pseudo-
thermal values presented in a false colour composite.
Drones 2022,6, 277 6 of 24
Drones 2022, 6, x FOR PEER REVIEW 6 of 25
fixing both cameras into one camera body still allows the use of the resulting thermal
images in combination with conventional true colour 3D modelling (Figure 2right). Data
processing typically results in an orthorectified projection of pseudo-thermal values pre-
sented in a false colour composite.
Figure 2. Sultana site: DEM (left), orthophoto (middle), and thermal composite (right) based on
Mavic 2 Enterprise data of the archaeological site of Sultana (illustrative, the white dots represent
the locations of the GCPs).
Another image-based system used in this project is the DJI Phantom 4 Multispectral
UAV, which was implemented for the first time during the summer campaign of 2022
(Table 1). Next to a regular visible light (RGB) camera, the mounted camera contains an
array of five additional and carefully aligned cameras. The geometrical camera specifica-
tions are identical for all six sensors, which are able to capture surface responses in specific
spectral bands:
- Near-infrared (NIR): 840 nm (±26 nm);
- Red edge (RE): 730 nm16 nm);
- Red (R): 650 nm (16 nm);
- Green (G): 560 nm (± 16 nm); and
- Blue (B): 450 (±16 nm).
The availability of these bands makes the system useful for vegetation mapping.
Therefore, the UAV is frequently used for environmental and agricultural monitoring ap-
plications (Candiago et al., 2015). The basis of this research is the calculation of vegetation
indices, such as the commonly used normalised difference vegetation index (NDVI). The
accurate calculation of these indices requires the correction of solar radiation fluctuations
during the acquisition, and an integrated spectral sunlight sensor provides their correc-
tion data.
3.2. Airborne Laser Scanning
Since early 2022, the research group has been able to implement its own UAV/ALS
combination. The main component of the newly obtained system is the Quantum Trinity
F90+ carrier (Figure 3, left). Furthermore, the system contains a reference station for post-
processing kinematic (PPK) global navigation satellite system (GNSS) positioning and a
control station for flight planning and monitoring. The UAV is a vertical take-off and land-
ing (VTOL) platform with an operational time of more than 90 min, a maximum ground
speed of 17 m/s, and a maximum altitude of 3500 m MSL. The platform can carry various
payloads, which are offered as different modules. Among others, an airborne laser scan-
ning module is available for this system. This module is a Qube 240 ALS sensor (Figure 3,
right). According to the specifications, the system can capture point clouds with a speed
of 240 k points per second, having a precision between 1.8 and 2.5 cm and an accuracy of
Figure 2.
Sultana site: DEM (
left
), orthophoto (
middle
), and thermal composite (
right
) based on
Mavic 2 Enterprise data of the archaeological site of Sultana (illustrative, the white dots represent the
locations of the GCPs).
Another image-based system used in this project is the DJI Phantom 4 Multispectral
UAV, which was implemented for the first time during the summer campaign of 2022
(Table 1).
Next to a regular visible light (RGB) camera, the mounted camera contains an
array of five additional and carefully aligned cameras. The geometrical camera specifica-
tions are identical for all six sensors, which are able to capture surface responses in specific
spectral bands:
- Near-infrared (NIR): 840 nm (±26 nm);
- Red edge (RE): 730 nm (±16 nm);
- Red (R): 650 nm (16 nm);
- Green (G): 560 nm (±16 nm); and
- Blue (B): 450 (±16 nm).
The availability of these bands makes the system useful for vegetation mapping. There-
fore, the UAV is frequently used for environmental and agricultural monitoring applications
(Candiago et al., 2015). The basis of this research is the calculation of vegetation indices,
such as the commonly used normalised difference vegetation index (NDVI). The accurate
calculation of these indices requires the correction of solar radiation fluctuations during the
acquisition, and an integrated spectral sunlight sensor provides their correction data.
3.2. Airborne Laser Scanning
Since early 2022, the research group has been able to implement its own UAV/ALS
combination. The main component of the newly obtained system is the Quantum Trinity
F90+ carrier (Figure 3left). Furthermore, the system contains a reference station for post-
processing kinematic (PPK) global navigation satellite system (GNSS) positioning and
a control station for flight planning and monitoring. The UAV is a vertical take-off and
landing (VTOL) platform with an operational time of more than 90 min, a maximum ground
speed of 17 m/s, and a maximum altitude of 3500 m MSL. The platform can carry various
payloads, which are offered as different modules. Among others, an airborne laser scanning
module is available for this system. This module is a Qube 240 ALS sensor (Figure 3right).
According to the specifications, the system can capture point clouds with a speed of 240 k
points per second, having a precision between 1.8 and 2.5 cm and an accuracy of less than
3.0 cm at a height of 140 m above ground level (AGL). The point density and accuracy
naturally decrease as a function of the ground speed and flying height [43].
Drones 2022,6, 277 7 of 24
Drones 2022, 6, x FOR PEER REVIEW 7 of 25
less than 3.0 cm at a height of 140 m above ground level (AGL). The point density and
accuracy naturally decrease as a function of the ground speed and flying height [43].
Figure 3. Quantum Trinity F90+ carrier (left) and the Qube 240 ALS sensor (right).
Given these specifications, the system opens opportunities for large-scale projects
that were initially challenging or technically impossible using image-based 3D modelling
[44,45]. The initial objectives for the use of this system are twofold. At first, the topo-
graphic documentation of larger areas around specific archaeological sites is envisioned.
Although reasonable areas have already been mapped using image-based 3D modelling,
the extension of the covered area allows researchers to understand the micro-topographic
aberrations in the landscape in a better way. Furthermore, with the interpretation of local
deviations of the NDVI, these topographic aberrations are a proxy for the configuration
of the hydrographic network in the Danube Valley. This might give insight into the loca-
tion of the new site.
A second objective is the documentation of the archaeological site that is covered by
vegetation, such as the sites of Spanțov (Figure 4, left) and Vărăști (Figure 4, right). Decid-
uous trees and low bushes cover both sites. Due to this vegetation, it was impossible to
make a DEM of this site during a previous site visit in 2018. However, given the canopy-
penetrating capabilities of ALS and the ability of the system to record three echoes per
pulse, it is expected that the new platform will allow the generation of a digital represen-
tation of the area [46,47]. These new data will facilitate a better understanding of the site’s
configuration.
Figure 3. Quantum Trinity F90+ carrier (left) and the Qube 240 ALS sensor (right).
Given these specifications, the system opens opportunities for large-scale projects that
were initially challenging or technically impossible using image-based 3D modelling [
44
,
45
].
The initial objectives for the use of this system are twofold. At first, the topographic
documentation of larger areas around specific archaeological sites is envisioned. Although
reasonable areas have already been mapped using image-based 3D modelling, the extension
of the covered area allows researchers to understand the micro-topographic aberrations in
the landscape in a better way. Furthermore, with the interpretation of local deviations of the
NDVI, these topographic aberrations are a proxy for the configuration of the hydrographic
network in the Danube Valley. This might give insight into the location of the new site.
A second objective is the documentation of the archaeological site that is covered
by vegetation, such as the sites of Span
t
,
ov (Figure 4left) and Vără
s
,
ti (Figure 4right).
Deciduous trees and low bushes cover both sites. Due to this vegetation, it was impossible
to make a DEM of this site during a previous site visit in 2018. However, given the canopy-
penetrating capabilities of ALS and the ability of the system to record three echoes per pulse,
it is expected that the new platform will allow the generation of a digital representation
of the area [
46
,
47
]. These new data will facilitate a better understanding of the site’s
configuration.
Drones 2022, 6, x FOR PEER REVIEW 8 of 25
Figure 4. The sites of Spanțov (left) and Vărăști (right) are covered by vegetation (source: Google
Satellite, coordinates in EPSG:3844).
3.3. Flight Plan Preparation and System Comparison
For both the photogrammetric flights, as well as for the ALS flights, specialised flight
planning software is used. Depending on the system, software developed by DJI (DJI Pi-
lot; Figure 5, top) or Quantum Systems (QBase; Figure 5, bottom) is used. Among other
things, the user is requested to define a study area, image overlap or strip overlap, and
ground sampling distance or point density for photogrammetric and ALS flights, respec-
tively. In combination with the knowledge of the used sensors and more specific data
acquisition parameters, this software gives useful insight into the coverage of the different
sensors and platforms.
Figure 4.
The sites of Span
t
,
ov (
left
) and Vără
s
,
ti (
right
) are covered by vegetation (source: Google
Satellite, coordinates in EPSG:3844).
Drones 2022,6, 277 8 of 24
3.3. Flight Plan Preparation and System Comparison
For both the photogrammetric flights, as well as for the ALS flights, specialised flight
planning software is used. Depending on the system, software developed by DJI (DJI Pilot;
Figure 5top) or Quantum Systems (QBase; Figure 5bottom) is used. Among other things,
the user is requested to define a study area, image overlap or strip overlap, and ground
sampling distance or point density for photogrammetric and ALS flights, respectively. In
combination with the knowledge of the used sensors and more specific data acquisition
parameters, this software gives useful insight into the coverage of the different sensors
and platforms.
Drones 2022, 6, x FOR PEER REVIEW 9 of 25
Figure 5. Gumelnița site. Screenshots of the used flight preparation and autopilot software: DJI Pilot
(top) and QBase (bottom).
In all cases, clearance of the study area by the local aviation authorities should be
taken into account. Furthermore, the flying height is limited to 120 m AGL for European
flights, including the flights in Romania. Otherwise, special licenses for the pilot and prior
permission are required. Given the advised operational height of 100 m AGL for the Qube
240 ALS system, this regulated maximum altitude does not play a sizeable restrictive role
in data acquisition. However, for image-based data acquisition, this constraint can be a
limiting factor for the coverage of a larger area where a relatively low GSD is acceptable.
Based on the specifications, a maximum flight time of 25 min is considered for the
DJI Mavic 2 Pro and the DJI Phantom Multispectral. For the ALS platform, the maximum
operation time is limited to 60 min. Given these values, combined with other values from
the specifications, a theoretical comparison can be made between the different systems
regarding the covered area and GSD or point density. For this comparison, two rectangu-
lar study areas of 1 × 1 km and 250 × 250 m are considered, situated at the site of Gumelnița.
Figure 5.
Gumelni
t
,
a site. Screenshots of the used flight preparation and autopilot software: DJI
Pilot (top) and QBase (bottom).
In all cases, clearance of the study area by the local aviation authorities should be
taken into account. Furthermore, the flying height is limited to 120 m AGL for European
flights, including the flights in Romania. Otherwise, special licenses for the pilot and prior
Drones 2022,6, 277 9 of 24
permission are required. Given the advised operational height of 100 m AGL for the Qube
240 ALS system, this regulated maximum altitude does not play a sizeable restrictive role
in data acquisition. However, for image-based data acquisition, this constraint can be a
limiting factor for the coverage of a larger area where a relatively low GSD is acceptable.
Based on the specifications, a maximum flight time of 25 min is considered for the
DJI Mavic 2 Pro and the DJI Phantom Multispectral. For the ALS platform, the maximum
operation time is limited to 60 min. Given these values, combined with other values from
the specifications, a theoretical comparison can be made between the different systems
regarding the covered area and GSD or point density. For this comparison, two rectangular
study areas of 1
×
1 km and 250
×
250 m are considered, situated at the site of Gumelni
t
,
a.
The extent of this area is imported into the various autopilot software as a KML file, where
a flight plan is calculated for 80 m, 100 m, and 120 m, using the default settings for the
systems (e.g., 80% and 70% frontal and side overlap, respectively, for the image sensors,
and 30% strip overlap for the ALS sensor). For image acquisition, the speed is set to 2.5 m/s
for all cases. The results are presented in Table 2for both the 1
×
1 km (top) and the 250
×
250 m (bottom) test sites.
Table 2.
Comparison of flight duration, number of batteries, and point density (PD)/GSD for two
test sites and various platforms.
1×1 km Trinity 90+ and Qube 240 DJI Mavic 2 Pro DJI Phantom 4 MS
Height AGL(m) 80 100 120 80 100 120 80 100 120
Duration (hh:mm:ss) 00:27:43 00:23:26 00:20:32 03:43:50 03:02:37 02:36:05 06:02:33 04:55:31 04:08:32
Number of batteries 1 1 1 9 8 7 15 12 10
PD/GSD 118 pts/m294 pts/m278 pts/m22.0 cm 2.5 cm 3.0 cm 4.3 cm 5.4 cm 6.5 cm
250 ×250 m Trinity 90+ and Qube 240 DJI Mavic 2 Pro DJI Phantom 4 MS
Height AGL. (m) 80 100 120 80 100 120 80 100 120
Duration (hh:mm:ss) 00:05:32 00:05:14 00:06:00 00:17:21 00:15:55 00:14:20 00:29:28 00:24:11 00:22:56
Number of batteries 1 1 1 1 1 1 2 1 1
PD/GSD 118 pts/m294 pts/m278 pts/m22.0 cm 2.5 cm 3.0 cm 4.3 cm 5.4 cm 6.5 cm
Based on these data, it seems evident that the ALS system is beneficial for the topo-
graphic mapping of large areas at relatively low flying heights. According to the speci-
fications of the Trinity 90+ and Qube 240, the minimal flying altitude of the platform is
limited to 40 m. ALS flights at this height will result in a point density of 240 pts/m
2
,
which is, therefore, the maximal point density obtained by the system. This value is lower
than the GSD of 1.0 cm and 2.2 cm when using the DJI Mavic 2 Pro and the DJI Phantom
4 MS, respectively. In contrast, the coverage of large areas using image-based systems
is time-consuming and requires many additional batteries. For study areas where the
maximum operational flying height is unconstrained or at least less constrained, more
similar results will be obtained for this comparison. For example, at a flying altitude of
395 m AGL, a GSD of 10.0 cm can be obtained using the DJI Mavic 2 Pro, which is similar
to the point density obtained at 90 m with the Trinity 90+ platform, and the Qube 240
ALS sensor might be considered inefficient. In that case, a study area of 1
×
1 km can be
accomplished using three batteries.
4. Data Processing
The different data processing workflows are presented in Figure 6and are discussed
in more detail in the following sections.
Drones 2022,6, 277 10 of 24
Drones 2022, 6, x FOR PEER REVIEW 11 of 25
Figure 6. Overview of the data processing workflow for the different data sources.
4.1. Georeferencing UAV Images
The DJI Mavic 2 Pro does not contain an advanced system for registering high-quality
GNSS data. As a result, a series of GCPs must be measured on-site using topographic
equipment. Black-and-white targets have been plotted on a series of plastic fabrics scat-
tered over the area to be documented before the flights take place. Depending on the avail-
ability of the hardware, these GCPs are measured using either a GNSS (Leica GS12) or a
total station (Leica TS06). These points are used as reference points to obtain absolute 3D
geo-referencing of the model. These control points are essential in order to be able to test
the accuracy of the model [48]. This approach reduces the absolute error from typically a
few meters to a few centimetres.
In contrast with the RGB images, the DJI Phantom 4 MS can use the raw RINEX data
and a separate file with a time stamp for each image obtained during the image acquisi-
tion. These data allow the correction of the standard geotags of the images with an accu-
racy of a few centimetres using reference data. These reference data can be extracted from
ALS data
MS imagery
RGB imagery Apache (JS)
RGB images
GCPs
3D model
DEM
Orthophoto
Point cloud
PLY to NXS
TIF
Agisoft Metashape
Apache (PHP)
Tile service
Geoserver
QGIS
Align photos
Assign GCPs to photos
Build dense point cloud
Build and texture mesh
Build DEM and orthophoto
WCS
MBTiles
3DHOP
MS images
Raw GNSS
Orthophoto
Agisoft Metashape
EUREF / REF
RTKLib (RTKPost)
Positioning QGIS
Assign geotags to photos
Align photos
Build dense point cloud
Build and texture mesh
Build DEM and orthophoto
True color composite
False color composite
MBTiles
Apache (PHP)
Tile service
Post-processing kinematic
ALS data
Raw GNSS
Point clouds
YellowScan
Apache (PHP)
EUREF / REF
Applanix POSPac UAV
Positioning and orientation
Raw INS
CloudCompare
QGIS
Apache (JS)
Point clouds
Combine ALS en POS data
Strip adjustment
CRS reprojection
Post-processing kinematic
SBET
Connected Component
Manual filering
PDAL
Classify
Rasterize
Clip
DEM TIF
Geoserver
WCS
Entwine
SkyViewFactor
MBTiles Tile service
PDAL
Cloud tiling
Figure 6. Overview of the data processing workflow for the different data sources.
4.1. Georeferencing UAV Images
The DJI Mavic 2 Pro does not contain an advanced system for registering high-quality
GNSS data. As a result, a series of GCPs must be measured on-site using topographic
equipment. Black-and-white targets have been plotted on a series of plastic fabrics scattered
over the area to be documented before the flights take place. Depending on the availability
of the hardware, these GCPs are measured using either a GNSS (Leica GS12) or a total
station (Leica TS06). These points are used as reference points to obtain absolute 3D geo-
referencing of the model. These control points are essential in order to be able to test the
accuracy of the model [
48
]. This approach reduces the absolute error from typically a few
meters to a few centimetres.
In contrast with the RGB images, the DJI Phantom 4 MS can use the raw RINEX data
and a separate file with a time stamp for each image obtained during the image acquisition.
These data allow the correction of the standard geotags of the images with an accuracy of a
Drones 2022,6, 277 11 of 24
few centimetres using reference data. These reference data can be extracted from a GNSS
reference station (e.g., a national implementation of a kinematic GNSS network such as
FLEPOS, or an international implementation such as EUREF). Another option is the use of
GNSS data acquired with an on-site reference station. In both cases, the reference station
coordinates are required in a known coordinate reference system (CRS). The correction
of the GNSS data acquired from the platform using PPK will minimise errors caused by
ionospheric and tropospheric aberrations of the speed of light. In addition, if the time
between the data acquisition and the data processing is extended by a few days, clock
errors and corrections of the satellite ephemerides can be considered [49].
For this study, the position of each photo is calculated using a post-processed kinematic
(PPK) based on reference data provided by EUREF. For the sites in the Danube Plains, the
reference station at Bucharest is situated at 50 to 70 km, which is still acceptable under the
assumption of similar meteorological conditions [
50
]. The coordinates of this station are
obtained, and GNSS observation and navigation data are downloaded from [
51
]. These
data are combined with the rover data obtained with the UAV using RTKPOST. This open-
source program within the RTKLIB suite allows the calculation of accurate GNSS-based
points [
52
]. The result of this procedure is a list of coordinates in WGS’84 with ellipsoidal
heights, accompanied by a time stamp and values for the accuracy of each point. The last
step is to align the newly generated series of coordinates with the file with time stamps for
each image. Since both the camera and GNSS sensor data are not completely temporally
coherent, a Python script was created to find the coordinate closest to the capture of a
particular image.
4.2. AGISOFT
After acquiring and downloading all UAV-based images, data processing took place
using Agisoft Metashape [
53
]. This commercial SfM-MVS-based 3D modelling software
is frequently used in comparative research [
4
,
7
,
54
]. The software semi-automatically
processes the photographs to produce a 3D reconstruction of the images. More details on
the processing steps and parameters to take into consideration are presented by Verhoeven
(2011). By default, the software used the geotags available in the images in WGS’84.
However, it should be mentioned that, in accordance with the previous section, images
from the Mavic 2 Pro are georeferenced using topographically measured GCPs. First, the
coordinates of these points are manually assigned to the corresponding pixels. Then, the
corrected geotags are directly assigned to the correct image within the software for the
multispectral imagery. Next to textured 3D models that can be used in dedicated software
(e.g., Meshlab) or web viewers (e.g., SketchFab), the data processing results in a series
of conventional orthophotos and DEMs. This orthophoto provides a multi-band image
made from multispectral data that enables the production of numerous composites. All
data are exported in the Romanian Pulkovo 1942(58)-Stereo70 coordinate reference system
(EPSG:3844) [
55
]. In existing cases, the elevations are converted from ellipsoidal heights
to Romanian orthometric heights using offsets obtained from the Romanian geoid model
(ROvT4).
4.3. ALS
The ALS data processing contains three steps. At first, the GNSS readings from the
sensor on board the platform should be corrected based on reference data using post-
processed kinematic (PPK). Then, the GNSS data for the ALS are processed following the
same methodology as the MS data, as described above.
The second step of the data post-processing encompasses the calculation of the Smooth
Best Estimated Trajectory (SBET) using the corrected GNSS data and INS readings. The
implemented algorithm allows the estimation of the position and orientation of the ALS
sensor at any moment during the data acquisition. Since the sensor measures points in an
intrinsically local coordinate system, the SBET is required as a starting condition for further
post-processing steps.
Drones 2022,6, 277 12 of 24
The third step comprises the actual extraction of a point cloud using the previously
calculated SBET and the raw ALS data. Initially, each local origin of a specific laser scanning
swath is assigned to the corresponding SBET position and orientation. Then, distances
between neighbouring strips are minimised using an iterative strip adjustment algorithm.
After applying this technique, the mean distance between points in consecutive strips is
typically limited to less than a decimetre. Individual points in the resulting optimised point
cloud can optionally be classified into ground points and non-ground points (vegetation,
buildings, etc.). Furthermore, digital elevation models (digital surface and digital terrain
models) can be extracted with a given resolution. It is also possible to export the processed
point clouds in the LAS file format.
For our study, Applanix POSPac UAV software was used for the GNSS and INS data
processing. For ALS data processing, YellowScan was used. As with the image-based data,
the ALS data are projected in the Pulkovo 1942(58)-Stereo70 coordinate reference system
(EPSG:3844) [
55
]. In addition, ellipsoidal elevations are converted to orthometric heights
using the ROvt4 geoidal model. Thus, it is assumed that all data are spatially coherent.
After exporting the point clouds, data cleaning was performed using CloudCompare [
56
]
using automated connected component labelling and manual data filtering.
A DEM was generated for each flight with a resolution of 25 cm using PDAL [
57
].
Instead of using a conventional interpolator, it was decided to consider the minimum
elevation value as a parameter to save in each pixel. Based on these models, an additional
raster with the sky view factor (SVF) [
58
] was generated using SAGA GIS [
59
]. Since SVF
is a frequently used topographic feature enhancement technique, it is handy to visualise
archaeological features.
5. Results and Quality Analysis
5.1. Sultana
Within the framework of the archaeological fieldwork in the Mosti
s
,
tea Basin and
Danube Valley, many sites have been recorded using airborne image-based 3D modelling
in previous years, such as Sultana Malu Ro
s
,
u, Gumelni
t
,
a, Ulmeni, Span
t
,
ov,
S
,
einoiu,
Măriu
t
,
a, and so forth. It is worth mentioning that two relatively large areas were covered
during the previous campaign in the summer of 2021. In Sultana, an update of the 2018
overview model was generated for the entire plateau containing the tell settlement, the flat
settlement, and the acropolis (Figure 7). The flight resulted in a series of photos with a GSD
of 2.5 cm, which is also the resolution of the final orthophoto. These data not only allow
the documentation of the archaeological fieldwork that took place during the campaign
but will also be included in the time series initiated in 2018 to investigate the magnitude of
the erosion processes of the cliff.
Figure 7. Sultana site: Overview of the 3D model of the archaeological habitation features.
Drones 2022,6, 277 13 of 24
5.2. Chiselet
Another significant contribution to the project concerns mapping the area around the
Chiselet tell settlements in the Danube Valley. An ALS flight was organised to cover an
area of approximately 1600
×
2550 m with a point density of 94 points/m
2
. The result
is converted to a rasterised DEM with a resolution of 25 cm. A hillshade of this DEM is
presented in Figure 8left. The multispectral sensor also covers the same area, which allowed
the calculation of true and false colour composites, as discussed earlier. An example of such
a false colour composite is presented in Figure 8right. The hillshade map and the false
colour composite cover the plane east of the settlement and reveal some old river channels
presented in the area before systematic cultivation in the last century. Thus, these data are
used to put the archaeological site in a larger spatial and historical context, which will be
helpful for future research (e.g., extending the results presented by Covataru et al. [15]).
Drones 2022, 6, x FOR PEER REVIEW 14 of 25
5.2. Chiselet
Another significant contribution to the project concerns mapping the area around the
Chiselet tell settlements in the Danube Valley. An ALS flight was organised to cover an
area of approximately 1600 × 2550 m with a point density of 94 points/m2. The result is
converted to a rasterised DEM with a resolution of 25 cm. A hillshade of this DEM is pre-
sented in Figure 8, left. The multispectral sensor also covers the same area, which allowed
the calculation of true and false colour composites, as discussed earlier. An example of
such a false colour composite is presented in Figure 8, right. The hillshade map and the
false colour composite cover the plane east of the settlement and reveal some old river
channels presented in the area before systematic cultivation in the last century. Thus, these
data are used to put the archaeological site in a larger spatial and historical context, which
will be helpful for future research (e.g., extending the results presented by Covataru et al.
[15]).
Figure 8. Chiselet site: Hillshade DEM (left) and false colour composite (right) of the larger extent
of the tell settlement (coordinates in EPSG:3844).
The tell settlement has also been covered by a series of images with a GSD of less
than 2 cm (Figure 9). For completeness, an illustrative cross-section was added to this fig-
ure, demonstrating a simple example of the usage of these DEMs. As with other flights
for this project, an extensive series of GCP have been measured in the area, facilitating the
spatial alignment of the different flights and the georeferencing of the data in the Pulkovo
1942(58)-Stereo70 coordinate reference system (EPSG:3844).
Figure 8.
Chiselet site: Hillshade DEM (
left
) and false colour composite (
right
) of the larger extent of
the tell settlement (coordinates in EPSG:3844).
The tell settlement has also been covered by a series of images with a GSD of less than
2 cm (Figure 9). For completeness, an illustrative cross-section was added to this figure,
demonstrating a simple example of the usage of these DEMs. As with other flights for
this project, an extensive series of GCP have been measured in the area, facilitating the
spatial alignment of the different flights and the georeferencing of the data in the Pulkovo
1942(58)-Stereo70 coordinate reference system (EPSG:3844).
Drones 2022,6, 277 14 of 24
Drones 2022, 6, x FOR PEER REVIEW 15 of 25
Figure 9. Chiselet site: Detailed DEM of the tell settlement with an illustrative profile. The white
dots represent the locations of the GCPs. The letters ‘A and B’ on the map correspond with the start
and end of the profile respectively (coordinates in EPSG:3844).
5.3. Overview of the Obtained Data
An overview of the various flights that will be considered for publication in the next
section is presented in Table 3.
Table 3. Overview of some of the data acquired within the framework of this research, with refer-
ence to the next section.
Location
Year
Technique
GSD/Resolution
Area (Ha)
Registration Error
(3D, cm)
Sultana
2018
RGB
2.3 cm
9
7.4
2019
RGB
2.7 cm
3
2.7
2021
RGB
2.5 cm
10
3.1
2022
MS
27.5 cm
219
9.5
2022
ALS
104 pts/m2
577
5.0
Chiselet
2019
RGB
2.7 cm
3
2.7
2021
RGB
2.5 cm
10
3.1
2022
MS
27.5 cm
219
9.5
2022
ALS
93 pts/m2
577
5
Valea Orbului
2022
MS
27.5 cm
219
9.5
2022
ALS
93 pts/m2
520
6.2
riuța
2022
MS
27.5 cm
219
9.5
2022
ALS
93 pts/m2
577
5
Gumelnița
2018
RGB
2.3 cm
9
7.4
2019
RGB
2.7 cm
3
2.7
2021
RGB
2.5 cm
10
3.1
2022
MS
27.5 cm
219
9.5
2022
ALS
93 pts/m2
577
5
Figure 9.
Chiselet site: Detailed DEM of the tell settlement with an illustrative profile. The white dots
represent the locations of the GCPs. The letters ‘A’ and ‘B’ on the map correspond with the start and
end of the profile respectively (coordinates in EPSG:3844).
5.3. Overview of the Obtained Data
An overview of the various flights that will be considered for publication in the next
section is presented in Table 3.
Table 3.
Overview of some of the data acquired within the framework of this research, with reference
to the next section.
Location Year Technique GSD/Resolution Area (Ha) Registration
Error (3D, cm)
Sultana
2018 RGB 2.3 cm 9 7.4
2019 RGB 2.7 cm 3 2.7
2021 RGB 2.5 cm 10 3.1
2022 MS 27.5 cm 219 9.5
2022 ALS 104 pts/m2577 5.0
Chiselet
2019 RGB 2.7 cm 3 2.7
2021 RGB 2.5 cm 10 3.1
2022 MS 27.5 cm 219 9.5
2022 ALS 93 pts/m2577 5
Valea
Orbului
2022 MS 27.5 cm 219 9.5
2022 ALS 93 pts/m2520 6.2
Măriut
,a2022 MS 27.5 cm 219 9.5
2022 ALS 93 pts/m2577 5
Gumelnit
,a
2018 RGB 2.3 cm 9 7.4
2019 RGB 2.7 cm 3 2.7
2021 RGB 2.5 cm 10 3.1
2022 MS 27.5 cm 219 9.5
2022 ALS 93 pts/m2577 5
Drones 2022,6, 277 15 of 24
5.4. Quality Analysis
5.4.1. ANOVA Using Checkpoints
The quality of the different elevation models is assessed using a series of 96 check-
points, which are measured using RTK GNSS at the site of Chiselet. One of these points
corresponds to the geodetic reference point located on the top of the tell settlement. To
minimise the effect of different implementations of the geoid model, resulting in a possible
offset of the altitudes, all altitudes are corrected as a function of this geodetic point. The
accuracy of the checkpoints is 0.03 m, which is higher than the theoretical accuracy of the
image-based DEMs, as well as the ALS-based DEM.
The corresponding elevations from the different DEMs are extracted for each check-
point using the point sampling tool in QGIS. The resulting values are evaluated using
a one-way analysis of variance (ANOVA) [
60
] using JASP [
61
,
62
]. In order to perform
the ANOVA, homogeneity of variances between the dataset should be assumed. This
assumption is tested by calculating Levene’s statistic [
63
]. The comparison of the variances
using JASP results gives (0.05 < p< 0.92), which states the assumption of homogeneity of
variance. The use of ANOVA is, therefore, fully justified.
According to the null hypothesis (H
0
), all mean elevation values (
µ
) of the different
datasets are equal. On the other hand, the alternative hypothesis (H
A
) states that, at least
for one dataset, the mean elevation is different from another dataset:
H0:µRTK-GNSS =µRGB =µMS =µALS
and
HA: at least one µvalue is different from the rest.
The results of the ANOVA are given in Table 4. Based on the results in the table, the null
hypothesis can be accepted with a 95% significance level for each study area. The calculated
F-values are smaller than the critical F-values, with a 95% confidence interval. This means
that no significant difference can be detected between the checkpoints generated by RTK-
GNSS, RGB-based photogrammetry, MS-based photogrammetry, and ALS
(0.05 < p< 0.95).
In other words, the variance of the different datasets is not significantly different.
Table 4.
Results of the ANOVA, where the four different datasets are compared with each other (SS is
the sum of squares; df is the number of degrees of freedom; MS is the mean square; F is the F-value;
pis the probability value; Fcrititcal is the critical F-value).
Source of
Variation SS df MS F pFcritical
Between groups 1.116 3 0.372 0.047 0.987 2.628
Within groups 3017.451 380 7.941
Total 3018.667 383
5.4.2. Qualitative Comparative Analysis of DEMs and Time Series Analysis
Assuming that the data are spatially coherent and the variance of the different datasets
is not significantly different, DEMs obtained in various years can be used to construct time
series. This concept is illustrated by the data obtained at the site of Sultana (Figure 10).
Drones 2022,6, 277 16 of 24
Drones 2022, 6, x FOR PEER REVIEW 17 of 25
where σ21 and σ22 correspond to the registration error of the DEMs obtained in 2018 (7.4
cm) and 2022 (5.0 cm), respectively, and where t corresponds to the Student t-values for
the required confidence interval (t = 1.96 for 95% confidence interval). As a result, all pixels
where the difference is within the interval [−17.5 cm; 17.5 cm] are caused by noise in the
model with a 95% confidence interval. These pixels are masked with a yellow colour in
Figures 10 and 11. The overall correspondence between the two models can be visually
stated, with most pixels being situated within this interval.
Figure 10. Sultana site: Time series analysis based on a DEM obtained in 2018 (left) and 2022 (right)
results in a difference model (middle) (coordinates in EPSG:3844).
Nonetheless, some artifacts are visible in the model, as well. In general, some scat-
tered patches of positive and negative differences can be observed, caused by the absence
or presence of new vegetation in the area. For some other exciting features, profiles are
presented in Figure 11. Profile A shows a section at the cliff below the site, which is also
clearly visible in the 3D model presented in Figure 7. This cliff is heavily affected by ero-
sion processes, endangering the site. Indeed, at this specific location, material loss be-
tween 2018 and 2022 can be observed. Profile B and profile C correspond to two excava-
tions organised in the summer of 2021 and 2022, respectively. The trenches (positive dif-
ference) and dump areas (negative difference) are visible in these two profiles.
Figure 10.
Sultana site: Time series analysis based on a DEM obtained in 2018 (
left
) and 2022 (
right
)
results in a difference model (middle) (coordinates in EPSG:3844).
For this analysis, the photogrammetry-based DEM from 2018 is compared with the
ALS-based DEM from 2022. Given the registration errors presented in Table 3, a mask
is applied to the resulting difference model where the threshold values are based on the
minimum level of detection (LoD). These values are calculated using [8]:
LODmin =t(σ21+σ22)
where
σ21
and
σ22
correspond to the registration error of the DEMs obtained in 2018 (7.4 cm)
and 2022 (5.0 cm), respectively, and where t corresponds to the Student t-values for the
required confidence interval (t = 1.96 for 95% confidence interval). As a result, all pixels
where the difference is within the interval [
17.5 cm; 17.5 cm] are caused by noise in the
model with a 95% confidence interval. These pixels are masked with a yellow colour in
Figures 10 and 11. The overall correspondence between the two models can be visually
stated, with most pixels being situated within this interval.
Drones 2022,6, 277 17 of 24
Drones 2022, 6, x FOR PEER REVIEW 18 of 25
Figure 11. Sultana site: Difference model (left) and various profiles illustrating topographic changes
between 2018 and 2022 (coordinates in EPSG:3844, profile units in meters).
5.4.3. Evaluating Point Clouds of Vegetated Sites
Among others, one of the objectives of the use of the UAV with ALS was the docu-
mentation of two archaeological sites at răști and Spanțov. Point clouds were generated
for both sites using the procedure described above. In addition, the ability to generate a
digital terrain model (DTM) of the topographic under the canopy was assessed for the site
at Vărăști. As mentioned above, this opportunity is made possible because the used ALS
sensor records multiple echoes per pulse. Various point cloud classifiers were applied to
the acquired point cloud using PDAL [57]. The used classifier uses a statistical outlier re-
moval and will distinguish between ground points and non-ground points using devia-
tions and variability within a specific kernel [64].
As visualised in Figure 12, the used ALS system indeed allows the penetration of the
canopy, revealing some topographic artifacts under the vegetation. However, the density
of the ground points is relatively low, resulting in spikes and large interpolated areas.
These are not only caused by the relatively low number of recorded echoes but also due
to the dense undergrowth vegetation in the area. As a result, it is advised to use scan
vegetated areas during winter, when the canopy is free of leaves. For the mapping of ag-
ricultural land, an additional advantage of this suggestion is the presence of crop-free par-
cels around the study area.
Figure 11.
Sultana site: Difference model (
left
) and various profiles illustrating topographic changes
between 2018 and 2022 (coordinates in EPSG:3844, profile units in meters).
Nonetheless, some artifacts are visible in the model, as well. In general, some scattered
patches of positive and negative differences can be observed, caused by the absence or
presence of new vegetation in the area. For some other exciting features, profiles are
presented in Figure 11. Profile A shows a section at the cliff below the site, which is also
clearly visible in the 3D model presented in Figure 7. This cliff is heavily affected by erosion
processes, endangering the site. Indeed, at this specific location, material loss between 2018
and 2022 can be observed. Profile B and profile C correspond to two excavations organised
in the summer of 2021 and 2022, respectively. The trenches (positive difference) and dump
areas (negative difference) are visible in these two profiles.
5.4.3. Evaluating Point Clouds of Vegetated Sites
Among others, one of the objectives of the use of the UAV with ALS was the docu-
mentation of two archaeological sites at Vără
s
,
ti and Span
t
,
ov. Point clouds were generated
for both sites using the procedure described above. In addition, the ability to generate a
digital terrain model (DTM) of the topographic under the canopy was assessed for the site
at Vără
s
,
ti. As mentioned above, this opportunity is made possible because the used ALS
sensor records multiple echoes per pulse. Various point cloud classifiers were applied to the
acquired point cloud using PDAL [
57
]. The used classifier uses a statistical outlier removal
and will distinguish between ground points and non-ground points using deviations and
variability within a specific kernel [64].
As visualised in Figure 12, the used ALS system indeed allows the penetration of the
canopy, revealing some topographic artifacts under the vegetation. However, the density of
the ground points is relatively low, resulting in spikes and large interpolated areas. These
are not only caused by the relatively low number of recorded echoes but also due to the
dense undergrowth vegetation in the area. As a result, it is advised to use scan vegetated
areas during winter, when the canopy is free of leaves. For the mapping of agricultural
land, an additional advantage of this suggestion is the presence of crop-free parcels around
the study area.
Drones 2022,6, 277 18 of 24
Drones 2022, 6, x FOR PEER REVIEW 19 of 25
Figure 12. Vărăști site: Digital surface model (DSM, top) and digital terrain model (DTM, bottom).
(coordinates in EPSG:3844)
5.5. Data Publication
Data sharing is a crucial element of open and sustainable research. Various tools and
platforms were selected to make the data obtained for this research available. The back-
bone of the implementation of this data sharing is the use of the Zenodo platform [65].
Zenodo is a multidisciplinary open access (OA) research data repository, assigning a dig-
ital object identifier (DOI) to any type of uploaded data. Data are available publicly and
are findable and reusable thanks to this unique identifier. The service has been operational
since 2013 and is recognised by the European Commission for referencing [66]. All gener-
ated DEMs are published on Zenodo (Dataset S1), as well as the MS-based orthophotos
(both true colour and false colour composites, Dataset S2) and the 3D models (Dataset S3).
Furthermore, various deliverables are available as XYZ-tile services, web coverage
services (WCS), or simply through front-end viewers. A website has been developed to
facilitate this data sharing, which can be accessed via https://geo.hogent.be/sultana (ac-
cessed on 10 September 2022). On this website, the following data are available for public
access (Dataset S4):
Image-based 3D models using 3DHOP: 3D models are made available using 3D
Heritage Online Presenter (3DHOP) as an interactive model viewer [67]. After exporting
the textured 3D model from Agisoft Metashape, the data should be converted to an NXS
multi-resolution model. This model is saved in a single file containing the geometry, the
texture maps, and the relations between these two elements. The resulting NSX file is
called by the 3DHOP JavaScript library, which organises the proper visualisation of the
model.
Raster data using MBTiles and XYZ-services: True colour composites, false compo-
sites, and SVF maps (sky view factor, calculated using unclassified ALS data) are pre-
sented using Leaflet [68]. All data are available through an XYZ-file service, which can
also be used in other desktop GIS software. These services are based on the conversion of
cartographic composites of the data in the Pseudo-Mercator CRS (EPSG:3857) to a system
of rows and columns. Each cell in this system is linked to a specific zoom level. Data are
Figure 12.
Vără
s
,
ti site: Digital surface model (DSM,
top
) and digital terrain model (DTM,
bottom
).
(coordinates in EPSG:3844).
5.5. Data Publication
Data sharing is a crucial element of open and sustainable research. Various tools and
platforms were selected to make the data obtained for this research available. The backbone
of the implementation of this data sharing is the use of the Zenodo platform [65]. Zenodo
is a multidisciplinary open access (OA) research data repository, assigning a digital object
identifier (DOI) to any type of uploaded data. Data are available publicly and are findable
and reusable thanks to this unique identifier. The service has been operational since 2013
and is recognised by the European Commission for referencing [
66
]. All generated DEMs
are published on Zenodo (Dataset S1), as well as the MS-based orthophotos (both true
colour and false colour composites, Dataset S2) and the 3D models (Dataset S3).
Furthermore, various deliverables are available as XYZ-tile services, web coverage
services (WCS), or simply through front-end viewers. A website has been developed
to facilitate this data sharing, which can be accessed via https://geo.hogent.be/sultana
(accessed on 10 September 2022). On this website, the fo