Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

Combining Multispectral & Thermal UAV Imagery in

archaeological sites. The case study of the Iron Age Hillfort of


Villasviejas del Tamuja (Cáceres, Spain)

Summary and main impression of the article

The application of high-resolution UAV mounted multispectral and Thermal


imagery to archaeological sites is indeed a rapidly growing area of research.
However, examples of both multispectral and thermal imagery collected over the
same site are rare and this paper, therefore, makes a significant contribution to
this. It also covers a variety of the complexities surrounding the successful use of
both these sensors and it is ambitious in its scope.

The article focuses on a single case study at the protohistoric Hillfort at Villasviejas
in central Spain, consisting of two walled enclosures dating to the fourth to the
first century BC. The site has been the focus of research prior to this investigation
and most recently the authors (with others) have carried out extensive non-
invasive techniques, which have been published elsewhere (Mayoral et al., 2019).
This data provides a very clear layout of the streets and buildings in the settlement
and is used as an important comparison with the multispectral and thermal
imagery. This is most clearly visualised in the magnetic gradiometer data, which
is helpfully included as a KML file. However, the analysis in this paper is restricted
to two small sectors of the site. Sector 1 located in the southern end of the ‘A’
enclosure, which embraces the defensive wall and part of its interior and Sector 2
covers the NW corner of the ‘B’ enclosure, thought to have contained a defensive
tower and one of the main gates of the settlement.

For the multispectral data, a wide range of spectral indices were calculated,
including those focussed on vegetation and soil characteristics. Many of these have
been previously shown to be of use in archaeological applications, but as is stated
by the authors, these have been generally applied to satellite imagery. Although
some indices are regularly calculated for UAV-borne imagery (e.g. NDVI), there
are only a small number of investigations that incorporate a wider range of these
(e.g. Moriarty et al., 2018). As a result, this paper provides a further valuable case
study of how such measures can be applied to these high spatial resolution
datasets.

Similarly, the capability of thermal imagery to reveal archaeological features has


been known for some time, but effective use of this on satellite and airborne
systems is rare. This is primarily because its success relies on data collection
during optimum times, which is difficult to predict and often dependant on the
properties of the features, diurnal heat flux and temperature changes as well as
other environmental factors. However, with the development of thermal sensors
that can be mounted on UAVs, there is an increased flexibility in choosing the time
of survey, as well as the possibility to carry out multiple surveys across a defined
time period. As the authors have noted, there is still a great deal of experimental
work to be done to determine ideal conditions and the results of this research
certainly add to this. detailed
It is stated that ‘The main purpose of the paper is to assess to what extent these
[multispectral and thermal] sensors can provide, objective empirical foundations
for a clear categorization and discrimination of the presence or absence of
archaeological features (PG 5 – Lines 115-7). In order to achieve this, the paper
attempts to establish whether there is a statistically significant difference in the
responses of various spectral indices. It is suggested that this could then enable
semi-automatic feature detection procedures.

A very brief discussion of the results of the original bands of the multispectral
sensor is provided, but the focus of the paper is dedicated to the efficacy of these
spectral indices, which were tested through three separability tests. These
analyses are commonly used in remote sensing applications to quantify the
separability of spectral classes define different types of land classifications or
themes, and its use is therefore appropriate.

The results show that both the multispectral and thermal sensors were effective
in the detection of buried archaeological structures and correlated well with the
result of the previous geophysical surveys and excavation. Through the statistical
analysis it was shown that some of the indices produce good contrasts, but these
do not allow categorical responses that could enable semi-automatic feature
detection procedures. Thus, although this does not necessarily pave the way
towards such an automated approach, it does provide a useful case study of how
such methods could be applied. However, the paper stops short of providing a
broader discussion of why automating this process would be a desirable outcome.

Recommendation

Minor Revisions

- Following the correction of typos, minor clarifications and more detailed


explanation of some of the themes covered, this paper should be accepted.

Comments and recommendations for the authors

Overall, I found the paper to be well written, with an appropriate level of detail.
Although some of the key principles of the techniques are only very briefly
discussed, it highlights key texts where more thorough discussion can be found
(e.g. Agapiou et al. 2013; Verhoeven). I think the authors are correct to avoid
going into too much depth here, but the paper would benefit from further
discussion of the spectral reflectance characteristics of vegetation/baresoil. This
would provide a firmer platform from which to consider the relative advantages of
the various spectral indices that have been deployed.

The figures, graphical abstract and additional items are generally of a high
standard and are informative. Figure 3 provides aerial shots of sector 1 and 2 and
is very helpful for understanding the layout and conditions of the site. The
separability graphs are a little blurry at their present size, but I suspect these will
be fine once reduced to fit within the format of the Journal. For increased clarity,
I think it would be beneficial to provide more labels and indications on the captions
of the figures 4-6, and these are referred to in the ‘minor typos and corrections
section’ below.
The methodology used in this research is clearly outlined. This enables the reader
to appreciate decisions that were taken to ensure accurate image capture. Many
of these are specific to the sensors and software utilised but can be appropriately
understood by users of other systems.

In terms of the indices for the multispectral data, the use of both vegetation and
soil indices is very interesting. The use of separability analysis also enables a
thorough assessment of their efficacy. However, I think an explanation of why
certain indices have performed better than others would be beneficial. As each
index is designed to detail certain ground cover characteristics you may be able
to go further in your analysis and perhaps comment further on how these relate
to features located in the subsurface. For example, why does the RENDVI perform
so poorly on this site? Why do the SR and MSR perform better? Is it because the
bands used in the algebraic calculations are those where the difference between
chlorophyll absorption / reflectance is most notable?

Regarding the thermal imagery, it is curious that you note that the noon flight
provided the clearest and largest number of traces. It is normally held that a pre-
dawn image provides most clarity, but I agree that there is still much work to be
done on this matter.

The authors may wish to consider the relevant section from ‘Scollar et al. 1990.
Archaeological Prospecting and Remote Sensing. Topics in Remote Sensing 2,
Cambridge University Press (PG 601 – 609 specifically)’. This refers to the role of
‘inhomogeneity in the surface layer’ and shows results obtained at three different
times of day (07 h 47, 13 h 32 and 17 h 08 – Figure 10.8). It suggests that
features closer to the surface (e.g. modern field boundaries) predominate in the
afternoon and can eclipse other, deeper features that were identifiable in the
images from the morning. It is suggested that it is better to collect imagery in the
morning so that deeper, weak anomalies are not masked. However, given that in
your case, the structural remains are located relatively close to the surface, and
that it is unlikely that any deeper features will be identifiable, this might provide
some reasoning for the outcome you have experienced at Villasviejas del Tamuja.

Throughout the paper a variation in ‘terrain conditions’ is referred to, but as the
study details two ostensibly similar and relatively small areas it is difficult to see
how this variety is represented. In the discussion it is stated that changing terrain
conditions and other factors determine the varying degrees of success in the
detection of thermal and multispectral anomalies, but it is not entirely clear how.
I assume this is probably related to the variation in the overlying stratigraphic
thickness that is referred to, but I think a clarification of this would be appropriate.

The given the amount of archaeological remains that are clearly visible in the
geophysical survey of the area, I am unclear as to why the specific small areas of
the site were selected. Perhaps this was to cut down on processing time, but I
wonder whether this could have been applied to the whole site, or at least the
area within the hillfort (once a mask surrounding densely vegetated areas and
walls etc. was created). However, I am not suggesting that this is undertaken at
this stage, as there is plenty of content for a paper of this length.
A central consideration of this paper is how best to produce categorical responses
that could enable semi-automatic feature detection procedures. However, there is
no discussion of why this may be desirable. Some papers suggest that the main
advantage of digital classification is that it is not subject to the biases of human
interpretation of archaeological features. In addition, it is also often stated that
such an approach is useful for large datasets or ‘big-data’. However, this study
covers relatively small areas of a site, and it is difficult to compare the success of
a manual interpretation. As a result, it is not made explicit why an automated or
semi-automated might be beneficial. Thus, the authors may consider adding a
brief discussion relating to why automation might be sought after and what
potential benefit this may have for the interpretation of the archaeological record.

Minor typos and corrections

Title

As the subject of the paper does not necessarily involve combining multispectral
and thermal UAV imagery, it might be better to avoid the term. Perhaps an
alternative could be;

Assessing the potential of Multispectral and Thermal UAV imagery


from archaeological sites. A case study from the Iron Age Hillfort of
Villasviejas del Tamuja (Cáceres, Spain).

This is of course merely a minor suggestion and is up to the authors to decide.

Abstract

You state that ‘This work explores the potential of low altitude remote sensing
(multispectral and thermal infrared imaging) for the study of large and complex
archaeological zones…’ But I would argue that the areas you have investigated are
actually rather small, especially in remote sensing terms. This does not negate
the value of the work, but I feel that the term ‘large’ is not representative of the
study and could be removed.

Similarly, you also state that the ‘response of the sensors in different types of
terrain’ are analysed, but as the study is limited to a single site, with very similar
terrain types within the two sectors, it is difficult to justify this. I understand that
later you refer to the variable depth of the overlying soils/sediments as a possible
factor affecting the results, which is very interesting, but I don’t think that a
variety of terrain types have indeed been covered.

PG 6

Line 166 – 167

It is stated that two years of fieldwork and ‘multiple campaigns of data collection
have taken place’. However, the analysed multispectral imagery is from a single
flight of the area, captured on May 15, 2018.

It is not clear whether multiple flights of the same area were undertaken? If you
just have the multispectral imagery from May, I think it would be useful to clarify
this. However, if imagery has been captured from other times of year, it would be
interesting to note whether there are any differences in the rates of detection of
archaeological features or what determined the selection of this specific dataset.

PG 7

Lines 198-199

In the case of the Parrot Sequoia, the sunlight sensor enables a radiometric
calibration of all the images taken in different light conditions. It is often achieved
post-flight in software such as Pix4d - which I can see that you have used. It would
be useful to further clarify this if possible.

Lines 200-201

‘The vegetation conditions were good enough after a rainy winter and spring’.
Please provide a bit more context and explain what ‘good enough’ means? And
why a rainy winter and spring would be beneficial or detrimental to the results
here?

Line 213

You reference Bennett et al., 2012 – You may wish to also see more recently
Kalayci, T.; Lasaponara, R.; Wainwright, J.; Masini, N. Multispectral Contrast of
Archaeological Features: A Quantitative Evaluation. Remote Sens. 2019, 11, 913.

PG 8

Lines 220 - 222

With reference to soil indices… ‘These can be very useful in Mediterranean-type


landscapes, where rainfall is very irregular and the cycles of greenery in
herbaceous plants can be very short’. It would be good to refer to an example of
this if possible? Perhaps 'Agapiou et al 2013. Optimum temporal and spectral
window for monitoring crop marks over archaeological remains in the
Mediterranean region, Journal of Archaeological Science, Volume 40, Issue 3,
https://doi.org/10.1016/j.jas.2012.10.036.' would be appropriate.

Lines 222 – 223

Are there any examples of soil indices providing good results in the identification
of archaeological structures? If not perhaps rephrase or clarify this.

PG 10

Within the table, the description for RENDVI there it says ‘Modfication to NDli’ This
should be NDVI.

PG 11

Line 238

‘MDS’ – I think it would be worth explaining the acronym.

PG 12

Line 264
Please replace ‘C’ with °C when referring to temperature.

PG 13

Line 317

I am not sure that you meant to refer to Figure 7 here. It should probably be
Figure 4? If so, consider adding an additional indicator in the caption for clarity

e.g. Figure 4: Graphic of spectral separability for vegetation indices


in a) Sector 1 and b) Sector 2.

Line 323

Possibly add a reference to Figure 5 at the end of the paragraph. Also consider
adding a further indicator to figure 5 as per my suggestion for Figure 4 and as you
have already done for Figure 6.

Lines 333 – 334

You refer to visual identification as a complimentary method to digital


classification. I would consider it to be more of a traditional approach. It is also
unclear whether a visual identification would out perform these analyses?

Figure 4 and 5

Please add ‘A’ and ‘B’ to Sector 1 and 2 as discussed above.

PG 14

Line 343

‘… is interrupted in its central part by the presence of a narrower linear traces’ –


It would be useful to describe the orientation of these features to help clarify which
features you are referring to. Also consider adding further labels to the images.

Lines 371-372

‘One particularly clear element is a wide, straight street that divides the enclosure
into two halves’. I think it would be very beneficial to add a label to this image
and again describe its orientation to make it easier to understand which feature is
being referred to.

Figure 6

The labels that have been provided in the first image are very difficult to see.
Perhaps illustrate it in an alternative colour, change it to white or add a mask
surrounding the text.

FIgure 7

Consider adding further labels discussed below

PG 15

Line 384 – 385


‘… the thermography captured at nightfall adds valuable information to more
clearly define the limits of the outer edge of the fortification below the modern
dry-stone walls’ - Another label highlighting this on the figure would be very
useful.

PG 17

Lines 483 - 484

‘… the signal captured by multispectral and thermal sensors does not allow us to
offer categorical responses.’ I would refer back to your introduction here and add
‘that could enable semi-automatic detection procedures’ to the end of this
sentence. As the issue of automatic feature detection is central to the theme of
this paper, further discussion of why this may be beneficial should be made more
explicit.

You might also like