Post on 03-Oct-2016
* Correspondence to: R.Bournemouth University,E-mail: rbennett@bournem
Copyright © 2012 John
Archaeological ProspectionArchaeol. Prospect. (2012)Published online in Wiley Online Library(wileyonlinelibrary.com) DOI: 10.1002/arp.1429
Short Report
The Application of Vegetation Indices for theProspection of Archaeological Features inGrass-dominated Environments
REBECCABENNETT*,KATEWELHAM,ROSSA.HILLANDANDREWL. J. FORD
School of Applied Sciences, Bournemouth University, Talbot Campus, Poole BH12 5BB, UK
ABSTRACT The identification of archaeological remain
s via the capture of localized soil and vegetation change in aerial imageryis a widely used technique for the prospection of new features. The near infrared (NIR) region has been shown byenvironmental applications to exhibit the signs of vigour and stress better than reflectance in the visible region, andthis has led to interest in the application of digital spectral data for archaeological prospection. In this study we assessquantitatively the application of 12 common vegetation indices to archive Compact Airborne Spectrographic Imagerdigital spectral data acquisitions from January and May 2001 in a grassland environment. The indices are comparedwith the true colour composite (TCC), best performing spectral band (711.2� 4.9 nm NIR) and the transcription of theaerial photographic archive. The results of the study illustrate that the calculation of a number of vegetation indices canassist with the identification of archaeological features in spectral data. However, the performance of the indices variesby season and although the features detected are shown to be complementary to those detected by the TCC, fewindices out-perform the TCC in terms of feature numbers identified. It was also shown that the Normalised DifferenceVegetation Index (NDVI), the most commonly applied index in archaeological prospection to date, performed poorly incomparison to indices such as the Modified Red Edge Simple Ratio Index, Simple Ratio Index and Modified Red EdgeNormalized Difference Vegetation Index. It is therefore recommended that the application of appropriate vegetationindices can enhance archaeological feature detection when combined with the TCC but that the calculation of theNDVI alone is insufficient to detect additional features. Copyright © 2012 John Wiley & Sons, Ltd.Key words: Vegetation indices; grassland; multispectral; CASI; NDVI; archaeology
Introduction
Identifying remains of past human interaction with thelandscape using aerial imagery is a long held tenet ofarchaeological prospection across the world. Tradition-ally panchromatic or colour photography has been usedto identify proxy indications of anthropogenic activity.Some of the most commonly detected proxy featuresare changes in vegetation growth patterns caused byarchaeological features in the top and subsoils (Wilson,1975;Maxwell, 1983; Brophy andCowley, 2005; Hejcmanet al., 2011). These changes are generically referred to as
Bennett, School of Applied Sciences,Talbot Campus, Poole BH12 5BB, UK.outh.ac.uk
Wiley & Sons, Ltd.
crop marks, as the high susceptibility of many arablecrops to stress means these features are frequently,though not exclusively, observed in agricultural cropstands.Those working with aerial imagery in the historic
environment sector are becoming increasingly awareof the importance of the non-visible regions of thespectrum for the detection of archaeological features(see Verhoeven and Doneus, 2011; Verhoeven 2012),leading to a growing number of studies using digitalspectral data for archaeological prospection. Initiallyapplications focused on feature detection from satelliteimagery of predominantly bare earth conditions insouthern Europe and the Middle East (Wilkinson,1993; Mumford and Parcak, 2002; Gheyle et al., 2004;Beck et al., 2007; Lasaponara and Masini, 2007). More
Received 18 June 2012Accepted 30 July 2012
Table 1. Wavelengths of the vegetation bandset of all the digitalspectral data supplied for the Everleigh study area.
CASIband
Wavelengthrange(nm)
Mid-pointwavelength
(nm)
Interpretation
1 446.2� 6.6 446.2 Blue vegetationresponse
2 470.1� 6.6 470.1 Blue vegetationresponse
3 490.4� 6.7 490.4 Green vegetationresponse
4 550.1� 6.7 550.1 Green vegetationmaximum
5 671.1� 6.8 671.1 Red vegetationabsorptionmaximum
6 683.5� 4.0 683.5 Red edge7 700.7� 5.9 700.7 Red edge8 711.2� 4.9 711.2 Red edge9 721.7� 5.9 721.7 Red edge10 751.3� 6.8 751.3 Near infrared
plateau11 763.7� 4.0 763.7 Vegetation
reflection12 780.9� 5.9 780.9 Water absorption13 860.2� 6.8 860.2 Near infrared
plateau14 880.2� 11.6 880.2 Near infrared
plateau
R. Bennett et al.
recently airborne sensors have been applied success-fully to detect direct changes associated with archaeo-logical features in the Mediterranean region wherethe fraction of soil to vegetation is low and archaeo-logical remains are often stone built and upstandingfrom the ground surface, resulting in high contrastbetween background and archaeological spectralresponse (Rowlands and Sarris, 2007; Pascucci et al.,2010; Traviglia and Cottica, 2011). An increasing bodyof work has illustrated the usefulness of airbornespectral data in temperate zones such as the UK(Donoghue and Shennan, 1988; Winterbottom andDawson, 2005; Powlesland et al., 2006; Challis et al.,2009; Bennett et al., 2011; Aqdus et al., 2012). In contrastto the Mediterranean examples, archaeological featuresin the UK are more often detected through changes ingrowth patterns caused to the overlying vegetationrather than direct observations of soil. However,analysis of spectral imagery to date has largely ignoredthe vegetation indices used in environmental studies,focusing instead on the assessment of true and falsecolour imagery. In a handful of studies the NormalisedDifference Vegetation Index (NDVI) (Rouse et al., 1973)has been calculated and noted to be of use for thedetection of features (Winterbottom and Dawson,2005; Challis et al., 2009; Aqdus et al., 2012), however,the application of this index has not been justified withregard to the aims of the research. The potential ofvegetation indices in general to highlight parameterssuch as plant quality, vigour or stress that are relatedto localized differences caused by underlying anthropo-genic features has been illustrated by work undertakenby Traviglia (2005, 2008) in a Mediterranean context,but has yet to be explored for temperate zones.This paper presents the results of a quantitative
comparison of 12 indices selected due to their provenability to illustrate biophysical parameters of vegetation,for an area of previously recorded archaeologicalfeatures on the calcareous grassland of the SalisburyPlain, Wiltshire (latitude 51.27, longitude �1.74). Indoing so it provides the first assessment of a range ofvegetation indices for archaeological prospection in atemperate environment. The aim of this research wasto conduct a quantitative comparison of the detectionof archaeological features using vegetation indices andrelate these results to the feature detection rates from atrue colour composite (TCC), the single best performingband of the spectral data and the archive of archaeo-logical features as recorded from aerial photography inthe Wiltshire Historic Environment Record (HER).While the record held in the HER is a collection of theevidence from many aerial photographs over manyyears and therefore cannot be considered directly
Copyright © 2012 John Wiley & Sons, Ltd.
comparable to a single spectral collection, it providesa useful baseline to assess the effectiveness of anytechnique compared with the cumulative informationfrom aerial photography. The evaluation of relativesuccess of these indices when detecting archaeologicalfeatures in the study area will inform future work.
Method
Two archive airborne multispectral datasets collectedusing the ITRES Compact Airborne SpectrographicImager (CASI) in January and May 2001 were usedfor this study. The CASI data were collected in 14bands ranging from 440 nm to 891 nm as shown inTable 1 and had a ground resolution of 1.5m. Thedata were geometrically and atmospherically correctedby the Environment Agency of England and Wales(henceforth EA) prior to acquisition by this projectand no further preprocessing was applied.Over 150 vegetation indices have been published in
remote sensing literature, many based on the premisethat algebraic combination of spectral bands canhighlight useful attributes of vegetation health andgrowth better than the study of either individual bandsor true/false colour RGB images (Ray, 1994). As so littlework has been done to establish the use of vegetation
Archaeol. Prospect. (2012)DOI: 10.1002/arp
Table2.
Detailsof
thevege
tationindices
appliedin
thisstud
y.
Index
Abbreviation
Form
ula
Categ
ory
Description
Normalized
Difference
Vege
tation
NDVI
NVDI¼r N
IR�r
RED
r NIRþr
RED
Broad
band
greenn
ess
Normalized
difference
ofgreenleaf
scatterin
gin
near
infrared
and
chlorophyllab
sorptionin
RED(Rou
seet
al.,19
73)
SimpleRatio
SRI
SR¼
r NIR
r RED
Broad
band
greenn
ess
Ratio
ofgreenleaf
scatterin
gin
near
infrared
andchlorophyll
absorptionin
RED(Tucker,19
79)
Enh
ancedVege
tationI
EVI
EVI¼
2:5
r NIR�r
RED
r NIRþ6
r RED�7
:5r B
LUEþ1
��
Broad
band
greenn
ess
Anenhancem
enton
theNDVItobetteracco
untforsoilbackgroun
dandatmospheric
aerosoleffe
cts(Heute
etal.,19
97)
Atm
ospherically
Resistant
Vege
tation
ARVI
ARVI¼r N
IR�
2rRED�r
BLUE
ðÞ
r NIRþ
2rRED�r
BLUE
ðÞ
Broad
band
greenn
ess
Anenhancem
entof
theNDVItobetteracco
untforatmospheric
scatterin
g(Kaufm
anandTanre,
1996
)
Red
Edge
Normalized
Difference
Vege
tation
RENDVI
NVDI 705
¼r 7
50�r
705
r 750þr
705
Narrowband
greenn
ess
Amod
ificationof
theNDVIu
sing
reflectancemeasurements
alon
gthe
reded
ge(Gitelson
andMerzlyak,
1994
;SimsandGam
on,20
02)
Mod
ified
Red
Edge
SimpleRatio
MRESRI
mSR
705¼r 7
50�r
445
r 705�r
445
Narrowband
greenn
ess
Aratio
ofreflectancealon
gthereded
gewith
bluereflectionco
rrection
(Datt,19
99;SimsandGam
on,20
02)
Mod
ified
Red
Edge
Normalized
Difference
Vege
tation
MRENDVI
mNDVI 705
¼r 7
50�r
705
r 750þr
705�2
r 445
Narrowband
greenn
ess
Amod
ificationof
theRed
Edge
NDVIu
sing
blueto
compensate
for
scatteredlight
(Datt,19
99;SimsandGam
on,20
02)
Red
Edge
Position
REPI
REP¼
700þ40
Rreded
ge�R
700
R740�R
700
��
Narrowband
greenn
ess
Thelocationof
themaximum
derivativein
near-infraredtransition,
which
issensitive
tochlorophyllco
ncentration(Curranet
al.,19
95)
Structure
Insensitive
Pigment
SIPI
SIPI
¼r 8
00�r
445
r 800�r
680
Ligh
tuse
efficiency
Areflectancemeasurementdesignedto
maximizethesensitivity
oftheindex
totheratio
ofbulkcaroteno
ids(e.g.alpha-caroteneand
beta-carotene)to
chlorophyllwhiledecreasingsensitivity
tovariation
incano
pystructure(e.g.leaf
area
index)(Penuelaset
al.,19
95)
Plant
Senescence
Reflectance
PSRI
PSRI¼r 6
80�r
500
r 750
Dry
orsenescent
carbon
Designedto
maximizethesensitivity
oftheindex
totheratio
ofbulk
caroteno
ids(e.g.alpha-caroteneandbeta-carotene)tochlorophyll.
(Merzlyaket
al.,19
99)
Antho
cyanin
Reflectance1
ARI1
ARI1
¼1
r 550
�� �
1r 7
00
��
Leaf
pigments
Chang
esingreenab
sorptionrelativeto
redindicateleafanthoc
yanins.
(Gitelson
etal.,20
01)
Antho
cyanin
Reflectance2
ARI2
ARI2
¼r 8
001
r 550
�� �
1r 7
00
��
hi
Leaf
pigments
Avariant
oftheARI1,which
issensitive
tochange
sin
green
absorptionrelativeto
red,indicatingleaf
anthoc
yanins.(Gitelson
etal.,20
01)
Application of Vegetation Indices in Grass-dominated Environments
Copyright © 2012 John Wiley & Sons, Ltd. Archaeol. Prospect. (2012)DOI: 10.1002/arp
R. Bennett et al.
indices for prospection of biophysical parametersrelating to archaeological features, guidance was takenfrom the selection of indices made by Asner (2008). Theindices that are potentially appropriate to identifyingvegetation stress caused by archaeological features canbe grouped into five categories: broadband greenness,
Figure 1. Examples of imagery used in the study: (a) May, TCC; (b) May, SRI; (cThis figure is available in colour online at wileyonlinelibrary.com/journal/arp
Copyright © 2012 John Wiley & Sons, Ltd.
narrowband greenness, light use efficiency, dry orsenescent carbon and leaf pigments, as detailed inTable 2. All indices are given in full in Table 2, but arereferred to by their acronyms in the text.In previous work by the authors, each band of
the CASI data was assessed for its ability to detect
) May, NDVI; (d) January, TCC; (e) January, MRESRI; (d) January, NDVI
Archaeol. Prospect. (2012DOI: 10.1002/arp
.
)
Figure 3. Relative feature detection rates from the vegetation indicesapplied to the May spectral data.
Application of Vegetation Indices in Grass-dominated Environments
archaeological features, and the near infrared (NIR)range of 711.2� 4.9 nm (band 8) was shown to be themost useful for feature detection (Bennett, 2012). Thevegetation indices were calculated using ENVI 4.7 withthe selection of bands that is standard to the software,while the TCC was composed of bands 5 (~671nm), 4(~550nm) and 2 (~490nm). Archaeological featureswere then mapped as a vector layer indicating thelocation and extent from each image following NationalMapping Programme protocol for aerial photographtranscription (English Heritage, 2006) at a scale of1:4000 or less. Examples of imagery are shown inFigure 1. Length was automatically calculated for eachvector transcribed and, on comparison of the digitizedvector data with the Wiltshire HER record, a maximumdetected length from each feature was calculated. Thefeature lengths recorded by each visualization techniquewere then converted to a percentage of the maximumdetected by anymethod. Thismeasurewas termed aver-age percentage feature length (APFL) of all features in agiven visualization. Thus both binary visibility (presentor not present) and percentage recovery were measuredas an indication of the detection rate of archaeologicalfeatures in each visualization.
Results
The number of features mapped from each of thevegetation indices for both spectral datasets are shownin Figures 2 (January) and 3 (May). The figures areillustrated alongside the number of known archaeo-logical features in the study area (Wiltshire HER) andthe number mapped from the TCC image and mostsensitive single band (ca. 711 nm). To illustrate the
Figure 2. Relative feature detection rates from the vegetation indicesapplied to the January spectral data.
Copyright © 2012 John Wiley & Sons, Ltd.
depiction of archaeological features typical in thisenvironment, a field system of lynchets and banks thatruns through three different land-use types is shownin Figure 4.The clearest result of the study is that the single
NIR band of the spectral data outperformed boththe TCC and vegetation indices in the January andMay data; illustrating the heightened sensitivity ofthe NIR to changes in plant growth in this environ-ment. This is also highlighted by the fact that inboth datasets the TCC recovered fewer of the fea-tures known from the aerial photographic archivethan the best performing NIR band. It was alsoshown that a number of vegetation indices can beused to detect archaeological features in the grassdominated environment of the study area. Thegenerally lower feature recovery in the spectral datais unsurprising given its lower spatial resolutionwhen compared with the aerial photography andthe fact that the HER archive covers more than50 yr of data acquisitions at different times of yearand under different crop conditions.The analysis shows that the relative performance of
the vegetation indices in comparison with the TCCand single best band varies by season. For the Januarydata, both the MRESRI and the MRENDVI were seento provide better binary and APFL detection offeatures than the TCC (Figures 2 and 5). This wasnot the case for the May data where no indexoutperformed the TCC in terms of number of featuresdetected (Figure 3), but the MRESRI and SRI had aslightly higher APFL (Figure 6). This concurs with theresults of Verhoeven et al. (2009), who tested theperformance of the SRI created using NIR photog-raphy. Of particular note in both datasets is therelatively poor performance of the NDVI, ranking
Archaeol. Prospect. (2012)DOI: 10.1002/arp
Figure 4. An extensive field system, running through three types of land use (from north to south: scheduled monument (grazed), heavily ploughedfield, ungrazed grassland) as depicted in the vegetation indices.
R. Bennett et al.
Copyright © 2012 John Wiley & Sons, Ltd. Archaeol. Prospect. (2012)DOI: 10.1002/arp
Figure 5. Average percentage feature length detection rates fromthe vegetation indices applied to the January spectral data.
Application of Vegetation Indices in Grass-dominated Environments
seventh in January and ninth in May out of all theindices evaluated.A comparison was made to see if the same features
were being detected in the vegetation indices andthe TCC (Table 3). This exercise demonstrated thecomplementarity of the different visualization tech-niques showing that the top performing vegetationindices recorded a significant number of extra featuresto the TCC of the same data. For example in the Maydata when the number of features recovered from theTCC, MRESRI and SRI was identical (40 features), thevegetation indices allowed the detection of at least15 extra features that were not detectable in the TCC.The results from both the January and May dataindicate that the use of the best performing vegetationindex can lead to a 30–49% increase in featuredetection over the TCC alone and a 24–30% increaseover the best performing single band.To summarize the usefulness of each of the indices
more clearly with respect to the factors assessed,
Figure 6. Average percentage feature length detection rates fromthe vegetation indices applied to the May spectral data.
Copyright © 2012 John Wiley & Sons, Ltd.
a ranked scoring system has been used to combinetotal number of features detected and level ofcomplementarity to the TCC and best performingsingle band (Table 4).
Discussion
The results of a quantitative comparison of 12 indicesselected for their empirical basis and applied to archiveCASI spectral data showed that none of the indices gavedetection rates comparable to those attained from tran-scription of the aerial photographic archive. Variationin feature detection rates relating to season wasillustrated in this study, both in the TCC and vegetationindices, with more features being detectable in the Janu-ary data than the May. Assessing the causes of thisobserved difference is complex due to the lack ofcontemporary ground observations, but it is suggestedthat the shallow root system of hardy vegetation suchas grass is less likely to exhibit stress or variationassociated with underlying archaeological features inits peak growing season (May) under non-droughtconditions (such as were captured in the data used forthis study).The best performing indices varied across the
spectral datasets of different dates, with only theMRESRI narrowband greenness index consistentlyperforming well in this grassland environment. Thestudy also illustrated that binary feature counts alonedo not provide the best assessment of the usefulnessof a particular index for archaeological prospection,as this measure does not quantify differences in theaverage percentage length of a feature that can bedetected or the level of complementarity to the TCCin terms of additional features detected. An awarenessof these factors, along with the quantitative compari-son, will enable historic environment professionalsto improve the application of vegetation indices. Insummary, the best performing indices allowed thedetection of a number of features that were notdetectable in the TCC/best performing single bandand therefore should be considered as complementaryvisualizations for archaeological feature detection.Although it must be emphasized that the NDVI is a
formula that can be applied to a number of bandcombinations in multispectral data and therefore thereis no “standard” NDVI calculation, this study showedthat the NDVI automatically calculated for these databy ENVI, was in fact one of the worst performing indi-ces for archaeological feature detection. This is due toits use of a broad NIR band designed for satellite datawhen the spectral resolution of the airborne data lends
Archaeol. Prospect. (2012)DOI: 10.1002/arp
Table 3. Comparison of complementarity or additional features detected by the vegetation indices over the TCC and best performing NIRband.
Vegetationindex
Number of additional featureswhen compared with
Increase in features detectedwhen used as a secondarysource to the TCC (%)
Increase in features detectedwhen used as a secondarysource to the NIR band (%)
JanuaryNIR band
JanuaryTCC
MayNIR band
MayTCC
January May January May
MRESRI 14 22 11 15 49 38 25 24MRENDVI 9 16 11 12 36 30 16 24SIPI 7 15 14 15 33 38 13 31ARI2 7 14 11 14 31 35 13 24RENDVI 6 10 10 12 22 30 11 22ARI1 6 10 10 11 22 28 11 22SRI 6 9 13 16 20 40 11 29ARVI 5 8 8 9 18 23 9 18NDVI 5 8 9 9 18 23 9 20EVI 1 2 0 0 4 0 2 0PSRI 2 2 0 0 4 0 4 0REPI 0 0 4 3 0 8 0 9
R. Bennett et al.
itself to more refined measures such as MRESRI andRENDVI. While it is clear that some indices can beused to detect archaeological features successfully,a lack of understanding of both spectral sensitivityand resolution often prevents the most appropriateindices from being applied.
Conclusions
This study provides the first quantitative analysis ofvegetation indices derived from remotely sensed datain a grass-dominated environment. It was shownthat a number of vegetation indices, particularly thosein the narrow-band greenness category, are effectivein detecting archaeological features. With hundredsof vegetation indices developed for environmentalpurposes, this study cannot be considered exhaustive,but it does indicate that appropriate selection of
Table 4. Ranking of vegetation indices based on number of features a
January numberof features
January TCCcomplementarity
January NIRcomplementarity
Janufinal s
MRESRI 12 12 12 36MRENDVI 11 11 11 33SIPI 9 10 9 28ARI2 7 9 9 25ARI1 9 7 7 23RENDVI 7 7 7 21SRI 5 6 7 18NDVI 5 4 4 13ARVI 4 4 4 12PSRI 3 2 3 8EVI 2 2 2 6REPI 1 0 1 2
Copyright © 2012 John Wiley & Sons, Ltd.
vegetation indices (based on spectral sensitivity ofthe target and resolution of the sensor) can enhancearchaeological prospection from the use of the TCCcomposite or NIR band alone.
Acknowledgements
Archive CASI and ALS data were supplied by theEnvironment Agency. The authors would like to thankthe Ministry of Defence and Defence Estates forfacilitating access and especially Richard Osgood andMartin Brown, Senior Historic Environment Advisors.Access to the archaeological archive was facilitated bythe staff of the Wiltshire HER and we are especiallygrateful for the expertise and encouragement of RoyCanham, former Wiltshire County Archaeologist.Gratitude is also expressed for the helpful commentsof the two anonymous reviewers. The research issupported by a Bournemouth University DoctoralResearch Bursary.
nd complementarity to other visualizations.
arycore
May numberof features
May TCCcomplementarity
May NIRcomplementarity
May finalscore
11 10 9 308 7 9 24
10 10 12 328 9 9 266 6 6 188 7 6 21
11 12 11 344 4 5 135 4 4 131 0 1 21 0 1 23 3 3 9
Archaeol. Prospect. (2012)DOI: 10.1002/arp
Application of Vegetation Indices in Grass-dominated Environments
References
Aqdus A, Hanson WS, Drummond J. 2012. The potentialof hyperspectral and multi-spectral imagery to enhancearchaeological cropmark detection: a comparativestudy. Journal of Archaeological Science 39: 1915–1924.
Asner GP. 2008. Vegetation Indices: ENVI Documentation.ITT Visual Information Solutions.
Beck A, Philip G, Abdulkarim M, Donoghue D. 2007.Evaluation of Corona and Ikonos high resolution satelliteimagery for archaeological prospection in western Syria.Antiquity 81(311): 161–175.
Bennett RA. 2012. Archaeological Remote Sensing: Visual-isation and analysis of grass-dominated environmentsusing airborne laser scanning and digital spectra data.Unpublished PhD thesis, Bournemouth University.
Bennett R, Welham K, Hill RA, Ford A. 2011. Making themost of airborne remote sensing techniques forarchaeological survey and interpretation. In RemoteSensing for Archaeological Heritage Management, CowleyDC (ed.). EAC Occasional Paper, Archaeolingua:Hungary; 99–107.
Brophy K, Cowley D (eds). 2005. From the Air: Under-standing Aerial Archaeology. Tempus.
Challis K, Kincey M, Howard AJ. 2009. Airborneremote sensing of valley floor geoarchaeology usingDaedalus ATM and CASI. Archaeological Prospection16(1): 17–33.
Curran PJ, Windham WR, Gholz HL. 1995. Exploringthe relationship between reflectance red edge andchlorophyll concentration in slash pine leaves. TreePhysiology 15: 203–206.
Datt B. 1999. A new reflectance index for remotesensing of chlorophyll content in higher plants: testsusing Eucalyptus leaves. Journal of Plant Physiology154: 30–36.
Donoghue D, Shennan I. 1988. The application of remotesensing to environmental archaeology. Geoarchaeology3(4): 275–285.
English Heritage. 2006. The National Mapping ProgrammeManual: A Methodology for the Use of Aerial Photographsfor Archaeological Landscape Mapping and Analysis.English Heritage: Swindon.
Gheyle W, Trommelmans R, Bourgeois J, et al. 2004.Evaluating CORONA: A case study in the Altai Republic(South Siberia). Antiquity 78(300): 391–403.
Gitelson AA, Merzlyak MN. 1994. Spectral reflectancechanges associated with autumn senescence of Aesculushippocastanum L. and Acer platanoides L. Leaves. Spectralfeatures and relation to chlorophyll estimation. Journalof Plant Physiology 143: 286–92.
Gitelson AA, Merzlyak MN, Chivkunova OB. 2001.Optical properties and nondestructive estimation ofanthocyanin content in plant leaves. Photochemistryand Photobiology 71: 38–45.
Hejcman M, Ondracek J, Smrz Z. 2011. Ancient waste pitswith wood ash irreversibly increase crop production inCentral Europe. Plant and Soil 339(1–2): 341–350.
Heute AR, Liu H, Batchily K, van Leeuwen W. 1997.A comparison of vegetation indices over a global setof TM images for EOS-MODIS. Remote Sensing ofEnvironment 59(3): 440–451.
Copyright © 2012 John Wiley & Sons, Ltd.
Kaufman YJ, Tanre D. 1996. ‘Strategy for direct andindirect methods for correcting the aerosol effect onremote sensing: from AVHRR to EOS-MODIS.’ RemoteSensing of Environment 55: 65–79.
Lasaponara R, Masini N. 2007. Detection of archaeo-logical crop marks by using satellite QuickBirdmultispectral imagery. Journal of Archaeological Science34(2): 214–221.
Maxwell G (ed.). 1983. The Impact of Aerial Reconnaissance onArchaeology. Council for British Archaeology: London.
Merzlyak JR, Gitelson AA, Chivkunova OB, Rakitin VY.1999. Non-destructive optical detection of pigmentchanges during leaf senescence and fruit ripening.Physiologia Plantarum 106: 135–141.
Mumford G, Parcak S. 2002. Satellite image analysis andarchaeological fieldwork in El-Markha Plain (SouthSinai). Antiquity 76(294): 953–954.
Pascucci S, Cavalli RM, Palombo A, Pignatti S. 2010.Suitability of CASI and ATM airborne remotesensing data for archaeological subsurface structuredetection under different land cover: the Arpi casestudy (Italy). Journal of Geophysics and Engineering7(2): 183–189.
Penuelas J, Baret F, Filella I. 1995. Semi-empirical indicesto assess carotenoids/chlorophyll-a ratio from leafspectral reflectance. Photosynthetica 31: 221–230
Powlesland D, Lyall J, Hopkinson G, et al. 2006. Beneaththe sand: remote sensing, archaeology, aggregates andsustainability: a case study from Heslerton, the Vale ofPickering,North Yorkshire, UK.Archaeological Prospection13(4): 291–299.
Ray TW. 1994. A FAQ on Vegetation in Remote Sensing. http://www.yale.edu/ceo/Documentation/rsvegfaq.html
Rouse JW, Haas RH, Schell JA, Deering DW. 1973.Monitoring vegetation systems in the Great Plains withERTS. Third ERTS Symposium, NASA SP-351(I):309–317.
Rowlands A, Sarris A. 2007. Detection of exposed andsubsurface archaeological remains using multi-sensorremote sensing. Journal of Archaeological Science 34(5):795–803.
Sims DA, Gamon JA. 2002. Relationships Between LeafPigment Content and Spectral Reflectance Across aWide Range of Species, Leaf Structures and Develop-mental Stages. Remote Sensing of Environment 81:337–354.
Traviglia A. 2005. A semi-empirical index for estimating soilmoisture from MIVIS data to identify sub-surface archaeo-logical sites. 9a Conferenza Nazionale ASITA, Federazionedelle Associazioni Scientifiche per le Informazioni Territorialie Ambientali 15–18 novembre 2005, Centro Congressuale‘Le Ciminiere’, Catania 2005 – Proceedings.
Traviglia A. 2008. The combinatorial explosion: definingprocedures to reduce data redundancy and to validate theresults of processed hyperspectral images. In DigitalDiscovery. Exploring New Frontiers in Human Heritage.CAA2006. Computer Applications and QuantitativeMethodsin Archaeology, Rome
Traviglia A, Cottica D. 2011. Remote sensing applica-tions and archaeological research in the NorthernLagoon of Venice: the case of the lost settlement ofConstanciacus. Journal of Archaeological Science 38(9):2040–2050.
Archaeol. Prospect. (2012)DOI: 10.1002/arp
R. Bennett et al.
Tucker CJ. 1979. Red and photographic infrared linearcombinations for monitoring vegetation. Remote Sensingof Environment 8: 127–150.
Verhoeven G. 2012. Near-infrared aerial crop markarchaeology: from its historical use to current digitalimplementations. Journal of Archaeological Method andTheory 19(1): 132–160.
Verhoeven G, Doneus M. 2011. Balancing on the borderline– a low-cost approach to visualise the red-edge shiftfor the benefit of aerial archaeology. ArchaeologicalProspection 18(4): 267–278.
Copyright © 2012 John Wiley & Sons, Ltd.
Verhoeven GJ, Smet PF, Poelman D, Vermeulen F. 2009.Spectral characterization of a digital still camera’s NIRmodification to enhance archaeological observation. IEEETransactions on Geoscience and Remote Sensing 47(10): 3456.
Wilkinson T. 1993. Linear hollows in the Jazira, UpperMesopotamia. Antiquity 67(256): 548–562.
Wilson DR (ed.). 1975. Aerial reconnaissance for archaeology.Council for British Archaeology: London.
Winterbottom SJ, Dawson T. 2005. Airborne multi-spectralprospection for buried archaeology in mobile sand dom-inated systems. Archaeological Prospection 12(4): 205–219.
Archaeol. Prospect. (2012)DOI: 10.1002/arp