Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
David Stott, School of Computing, University of Leeds, UK Anthony Beck, School of Computing, University of Leeds, UK Doreen Boyd, Department of Geography, University of Nottingham, UK Gary Llewelyn, NERC ARSF, Gloucester, UK Quinton Carroll, County Archaeology Office, Cambridgeshire County Council, UK
Abstract
The aim of this study was to understand how hyperspectral remote sensing can be optimally utilised for the detection of archaeological residues. In June 2011 the UK NERC ARSF acquired Eagle/Hawk hyperspectral data over an area of arable farmland comprising of heavy and well draining soils (Diddington, Cambridgeshire, UK). Concurrent field spectra were collected and in addition further spectra were collected at dates earlier and later in the crop growing season. Both the field and airborne hyperspectral data were pre-processed and continuum removed metrics were derived. These metrics are used to determine the nature of the archaeological electromagnetic response. Preliminary results illustrate the value of the hyperspectral approach which does allow for more nuanced archaeological prospection. This is the first study of this kind to undertake systematic aerial and ground based analyses and once results are transferred has the potential to reveal new approaches for heritage detection. Since this study occurred in non-normal conditions (after the driest spring in 100 years) it provides a benchmark for future multi-temporal studies in the area (which took place during for 2012), as well as similar areas and conditions worldwide. Keywords: archaeology, heritage, vegetation marks, spectroradiometry, hyperspectral, detection
Introduction
Vegetation marks produced by localised changes in stress or vigour can be indicative of sub-surface archaeological remains. For over 100 years these vegetation marks have been observed and recorded by archaeologists in aerial photographs (APs). The interpretation of these images has substantially increased our understanding of the nature and distribution of the archaeological record. However, aerial photography is over-reliant on the visual component of the electromagnetic spectrum. The collection technique and the technology utilised militate against using any other sensor (i.e., peripatetic surveys are directed by visual observation from a plane and collected using an optical system: a camera out of a window). This is further exacerbated by the policy frameworks that define the application of aerial detection: it is predominantly used on specific, free draining soils where vegetation marks, particularly those in arable crop, more regularly occur. Such biases mean that the large proportion of the UK under clay, pasture and other difficult conditions are rarely flown (Cowley 2002; Mills 2007). This was dramatically detailed in the 1996 flights over the Cambridgeshire clays where a serendipitous vertical survey undertaken by the County Council provided unprecedented information on the buried landscape (Mills and Palmer 2007). Landscape, as opposed to feature, surveys using nonvisual sensors will help address many of the biases associated with traditional archaeological aerial photography. In this respect the National Heritage Science Strategy (NHSS) reports (Williams 2009a; Williams 2009b) recognised that hyperspectral remote sensing (RS) techniques have a huge potential for use in archaeological prospection but are currently underutilised. The current resistance in using hyperspectral RS results from the fact that the science underpinning how archaeological contrasts form and their sensor detection characteristics is not well understood and so there are difficulties in their optimal deployment. Which is a significant issue given the relatively cost of data collection. Archaeological activity creates localised physical, chemical and biological variations in the soil matrix (Schiffer 2002) which can be detected directly or by proxy (normally through the detection of crop-marks: localised crop stress/vigour variations, but also as temperature, topographic and soil structure anomalies). The subtleties associated with archaeological formation means that detectable contrast alters as environmental conditions change. Hence, there is a seasonal and diurnal dimension to detection. The challenge is to detect both the ephemeral expression of contrast and the environmental process that can accentuate or diminish this response. This paper presents preliminary results from an investigation using hyperspectral data acquired by the UKs Natural Environment Research Councils (NERC) Airborne Research and Survey Facilitys (ARSF) Eagle/Hawk sensors, and in-field spectroradiometry, to understand archaeological detection dynamics, particularly with respect to the extrapolation of local detection to the landscape. The research complements that of previous work supported by NERC ARSF (e.g., Bennett et al. 2011) and extends it by focussing on arable landscapes.
This investigation forms part of a wider research project supported by the UKs Arts and Humanities Research Council/Engineering and Physical Sciences Research Council funded Detection of Archaeological Residues using remote sensing Techniques (DART) project (http://dartproject.info/) which has the overall aim of developing analytical methods for identifying and quantifying gradual changes and dynamics associated with near-surface archaeological features under different environmental and land-management conditions. An understanding of how remote sensing technologies detect contrast caused by different underlying factors under different environmental conditions, can lead to the deployment of sensors and survey techniques that are likely to detect a greater range of ARs (Beck 2011; Verhoeven and Doneus 2011). This is particularly important for those anomalies that express their contrast outside the visible spectrum. This understanding will allow different sensors to be deployed at the times when the ARs have the greatest likelihood of being detected.
phenological process. Spectroradiometry readings are taken on transects across known archaeological features. Traditional archaeological aerial photographs are being collected at regular intervals, these are supplemented by a number of aerial hyperspectral and laser scanning surveys from NERC ARSF and the Environment Agency.
Figure 2 DART probe installation and sensor surveys The combination of multi-temporal spectroradiometry and aerial imaging surveys will also allow the examination of scaling issues (particularly, pixel mixing and leaf area index) and environmental dynamics. The different spatial resolutions of the aerial sensors will be key for examining issues of scale. The data from the insitu probes and weather stations will highlight key environmental events. The results from the spectroradiometry transects will be extrapolated to the local area through the hyperspectral images. This paper focuses on airborne hyperspectral data and ground spectroradiometry data associated with the pasture field. The airborne data were collected as part of a pilot study supported by NERC ARSF award (GB11/01) in which Eagle/Hawk hyperspectral data were collected on 14th June 2011. The size of the archaeological features (1m or less) requires that fine spatial resolution data are acquired, thus the Eagle/Hawk sensors were deployed at an altitude of 600m. The ground data were collected along a RTK GNSS located transect 15m long, centred over the known archaeological feature. Spectroradiometry readings were taken with an ASD FieldSpec Pro taken using at least 1m intervals. This paper will use the readings taken at the following dates; 14/06/11; 28/06/11 and 15/07/11 which correspond to mature, senescing and senesced vegetation respectively.
Methodology
It is understood that subsurface archaeological features are different from the surrounding natural soil matrix. These changes in the soil mean that the feature influences its surroundings, for example, in terms of water and nutrient movement. This influence can sometimes be observed as localised changes in the overlying vegetation growth, commonly referred to as crop or vegetation marks. The current understanding is that these marks tend to be most strongly expressed during periods of low rainfall, where there is a Soil Moisture Deficit (SMD) and cereal crops are at or near maturity (Evans and Jones 1977; Jones and Evans 1975). However, the causes of these changes in vegetation growth are likely to be more complex than a simple water stress/vigour relationship. For example, localised variations between archaeological sediments and the surrounding soil matrix exhibit differences in particle size distribution, soil density, pore size, drainage rates, thermal capacity, which all potentially influence the growth of the plant from its very earliest stages of development. Preliminary growth experiments conducted show that plants grown on archaeological sediments develop more quickly than those from the surrounding areas. These plants have more tillers and a larger leaf area, and are thus structurally different and represent significantly increased photosynthetic biomass. Evans and Jones (1977)argue that variations in biomass are the primary means by which archaeological vegetation marks are detectable in aerial photography. The contrasts which are expressed are also likely to change as the plant goes through different phases of the growth cycle from germination to senescence.
Hence, there is a temporal dimension which will be influenced by environmental processes. Models need to be developed that clearly identify these and other contrast determinants, the conditions under which they occur and which spectral wavelengths and hence sensors, can be used to detect them. For vegetation, in the visible/near-infrared we are looking at vegetation response (red-edge etc.) and in the short-wave infrared we are looking at leaf-water content. Boyd et al 2012 described the use of different Vegetation Indices in the visible/near-infrared on the data described in this paper. This paper will look at the use of bandnormalised continuum removed metrics. The image and field spectroscopy data were preprocessed (i.e., image mosaicing and atmospheric correction) using both bespoke (e.g., APL) and proprietary software (e.g., ENVI)) to derive spectral reflectance. Band-normalised continuum removal metrics were calculated on the ground based spectroscopy readings using the techniques proposed by Kokaly and Clark (1999)and Curran et al (2001). This technique enables the normalisation of the effects of illumination geometry and biomass to provide metrics that enable investigation of specific absorption features. This enables the separation between the effects of stress and vigour variations in foliar biochemistry of the crop canopy, and the effects of greater biomass. For this paper metrics were calculated on the diagnostic absorption features associated with chlorophyll a+b (670nm), water (1200nm) and lignin (1730nm).
Figure 3 Raw spectro-radiometry reflectance spectra from June & July 2011 Figure 3 shows how the reflectance of the crop and the contrast between the vegetation mark changes over time in the data from the spectroradiometry transect. On the 14th of June the crop over the archaeology shows greater absorption in the visible spectrum and greater reflectance in the NIR when compared to the background. This is consistent with variations in the density and health of photosynthetic vegetation. On the 29 th of June this relationship changes, with the crop over the archaeology showing greater reflectance in the visible spectrum compared to the background. This may indicate that the crop over the archaeology has reached maturity and started to senesce sooner than the background. On the 15th of July the crop has senesced and ripened, with the archaeology demonstrating greater reflectance over most of the visible and NIR portions of the spectrum. This contrast is indicative of the greater mass of senesced vegetation over the archaeology compared to the background.
Figure 4 Continuum removed spectra for the 670nm absorption feature The band-normalised continuum removed spectra from the 670nm (chlorophyll a+b) absorption feature show little contrast in between the archaeology and the background (see Figure 4) on the 14th and 29th of June. This indicates that on these dates there is comparatively little difference in the photosynthetic pigments in the foliage, and that the contrasts visible in the raw spectra are more likely the result of variations in biomass. The data from the 15 th of July shows greater contrast, but given that the senesced vegetation is no longer photosynthesising these data are less reliable.
Figure 5 Continuum removed spectra for the 1200nm absorption feature In the 1200nm absorbtion feature (foliar water) there is again little contrast between the archaeology and the background. There is a greater contrast on the 29th of June than on the 14th of June, but this remains subtle. The data from the 15th of July shows greater contrast, but the depth and breadth of the absorption feature is much narrower compared to the data from June. This is a result of the dry, senesced vegetation
The 1730nm absorption feature (lignin) shows significant variation for the 14 th of June. This is probably a result of the lower leaf area index (LAI) the crop growing on the background compared to the archaeology, meaning that there is a greater return from the dry mass of the plant. The data from the 29 th of June shows less contrast a change in the shape of the absorption feature, indicative of ripening and senescence. The data from the 15th of July shows a lateral shift resultant from the difference between active and senesced vegetation.
References
Beck, A. R. 2011. Archaeological Applications of Multi/hyper-spectral Data Challenges and Potential. In Remote Sensing for Archaeological Heritage Management., ed. D. C Cowley, 8798. EAC Occasional Paper 5. Europae Archaeologia Consilium. Bennett, R., K. Welham, R. A. Hill, and A. L. J. Ford. 2011. Making the Most of Airborne Remote Sensing Techniques for Archaeological Survey and Interpretation. In Remote Sensing for Archaeological Heritage
Management., ed. D. C. Cowley, 99106. EAC Occasional Paper 5. Europae Archaeologia Consilium. Boyd, D.S., A. R. Beck, D. Stott, G. Llewellyn and Q. Carrol, 2012, The driest spring in 100 years: hyperspectral imaging and spectro-radiometry of vegetation over known archaeological features, Proceedings of the Remote Sensing and Photogrammetry Society Conference 2012 Cowley, D. C. 2002. A Case Study in the Analysis of Patterns of Aerial Reconnaissance in a Lowland Area of Southwest Scotland. Archaeological Prospection 9 (4) (December 1): 255265. doi:10.1002/arp.199. Curran, Paul J, Jennifer L Dungan, and David L Peterson. 2001. Estimating the Foliar Biochemical Concentration of Leaves with Reflectance Spectrometry: Testing the Kokaly and Clark Methodologies. Remote Sensing of Environment 76 (3) (June): 349359. doi:10.1016/S0034-4257(01)00182-1. Evans, R., and R. J. A. Jones. 1977. Crop Marks and Soil Marks at Two Archaeological Sites in Britain. Journal of Archaeological Science (4): 63 76. Jones, R. J. A., and R. Evans. 1975. Soil and Crop Marks in the Recognition of Archaeological Sites by Air Photography. In Aerial Reconnaissance for Archaeology, ed. D.R. Wilson, 111. London: Council for British Archaeology. Kokaly, R. F., and R. N. Clark. 1999. Spectroscopic Determination of Leaf Biochemistry Using Band-depth Analysis of Absorption Features and Stepwise Multiple Linear Regression. Remote Sensing of Environment 67: 267287. Mills, J. 2007. Bias and the World of the Vertical Aerial Photograph. In From the Air: Understanding Aerial Archaeology, ed. K. Brophy and D. C. Cowley, 291299. Stroud: Tempus. Mills, J., and R. Palmer. 2007. Populating Clay Landscapes. First ed. Stroud: Tempus. Schiffer, M. B. 2002. Formation Processes of the Archaeological Record. University of Utah Press. Verhoeven, G., and M. Doneus. 2011. Balancing on the Borderline a Lowcost Approach to Visualize the Rededge Shift for the Benefit of Aerial Archaeology. Archaeological Prospection. doi:10.1002/arp.420. http://onlinelibrary.wiley.com/doi/10.1002/arp.420/abstract. Williams, J. 2009a. Use of Science to Enhance Our Understanding of the Past. NHSS Report. http://nhss.englishheritage.org.uk/upload/pdf/nhss_report_2_web.pdf?1317030601. . 2009b. The Role of Science in the Management of the UKs Heritage. NHSS Report. http://nhss.englishheritage.org.uk/upload/pdf/nhss_report_1_web.pdf?1317030601.