- Email: [email protected]
Establishing discovery probabilities of lithic artefacts in Palaeolithic and Mesolithic sites with core sampling
Journal of Archaeological Science 40 (2013) 240e247
Contents lists available at SciVerse ScienceDirect
Journal of Archaeological Science journal homepage: http://www.elsevier.com/locate/jas
Establishing discovery probabilities of lithic artefacts in Palaeolithic and Mesolithic sites with core sampling Philip Verhagen a, *, Eelco Rensink b, Machteld Bats c, Philippe Crombé c a
Research Institute for the Heritage and History of the Cultural Landscape and Urban Environment (CLUE), Faculty of Arts, VU University, De Boelelaan 1105, 1081 HV, Amsterdam, The Netherlands b Cultural Heritage Agency (RCE), Amersfoort, The Netherlands c Department of Archaeology, Ghent University, Ghent, Belgium
a r t i c l e i n f o
a b s t r a c t
Article history: Received 17 January 2012 Received in revised form 22 April 2012 Accepted 28 May 2012
This paper reports the results of a study into the effectiveness of core sampling for discovering Palaeolithic and Mesolithic hunter-gatherer sites in the Netherlands and northwestern Belgium. Earlier work established optimal sampling strategies for use in archaeological heritage management survey in the Netherlands. However, the statistical model used for this was based on a limited amount of data on the distribution of lithic artefacts in Palaeolithic and Mesolithic sites. For the current study we have analyzed the distribution of artefacts in a selected number of excavated sites, and estimated discovery probabilities of these sites through simulation. The simulation results indicate that discovery probabilities are lower than expected due to the effect of clustering of finds. Furthermore, the density of flints in Palaeolithic and Mesolithic sites is generally lower than the estimates that were used for setting up the optimal sampling strategies, and a substantial number of sites is very small. This means that, in order to discover Palaeolithic and Mesolithic sites with sufficient reliability, we will have to apply more intensive survey strategies than have been recommended up to now. Ó 2012 Elsevier Ltd. All rights reserved.
Keywords: Archaeological survey Core sampling Lithic artefacts Archaeological heritage management Simulation Palaeolithic and Mesolithic sites
1. Introduction Over the past ten years, procedures for fieldwork in archaeological heritage management (AHM) in the Netherlands have become codified in a system of regulations set up by the archaeological sector itself. This set of quality norms (the Quality Norm for Dutch Archaeology or KNA version 3.2; SIKB, 2010) describes the procedures to be followed in AHM research, moving from deskbased assessment through survey to excavation. The quality norms specify what needs to be done, but do not prescribe how things should be done. However, in some cases, it was felt that additional guidance was needed on the ‘how’ as well. One of these issues is the establishment of the most effective and efficient strategies for detecting archaeological sites. Accompanying guidelines have therefore been developed concerning the use of core sampling (Tol et al., 2006) and trial trenching (Borsboom and Verhagen, 2009) e these being the most frequently used survey methods in the Netherlands. The guidelines are based on * Corresponding author. Tel.: þ31 20 5982848. E-mail addresses: [email protected] (P. Verhagen), e.rensink@ cultureelerfgoed.nl (E. Rensink), [email protected] (M. Bats), [email protected] (P. Crombé). 0305-4403/$ e see front matter Ó 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.jas.2012.05.041
theoretical statistical models that specify the probability of detecting archaeological sites of a certain dimension and find density (see Tol et al., 2004; Verhagen, 2005; Verhagen and Borsboom, 2009). They are used to help design survey project briefs and to evaluate survey results. The guidelines provide preferred survey strategies that will result in a 75% chance of discovery of archaeological sites that are classified according to such prospection characteristics as size and artefact density (Table 1). However, it is still difficult to assess the actual effectiveness of these strategies, since the prospection characteristics of many archaeological site types are only known in general terms. More empirical data are needed to compare the situation in the field to the theoretical assumptions used for the guidelines. Unfortunately, this type of data is still in small supply, and no mechanisms are available in Dutch AHM that would enable us to increase our knowledge on these aspects. The excavation of sites is usually considered to be the closing chapter of AHM, and is not systematically used to collect information on, for example, the spatial distribution and density of artefacts or features that can be of use in the earlier phases of fieldwork in the future. This paper reports the results of an investigation that has tried to do just that. We have used excavation data of Stone Age sites, especially from the Late Palaeolithic (ca. 13,000e8700 cal BC) and Mesolithic (ca.
P. Verhagen et al. / Journal of Archaeological Science 40 (2013) 240e247
241
Table 1 Overview of standard strategies for core sampling survey in the Netherlands for different site types. Site type
Lithology
Stone Age Medium size, 200e1000 m2 Base camp House plan Large size, >2000 m2 Large base camp Aggregated settlement Multiple house plans
Flint scatter Sand Clay/loess Clay/loess Sand Clay/loess Clay/loess
20 17 13 40 30 20
25 20 15 50 35 25
m m m m m m
15 12 12 15 12 12
cm cm cm cm cm cm
3 mm 3 mm e 3 mm 3 mm e
Bronze Age e Middle Ages House plan(s), 500e2000 m2
Ceramic scatter Sand Clay/loess Clay/loess Sand Clay/loess Clay/loess
30 20 17 80 60 40
35 25 20 90 70 50
m m m m m m
15 12 12 15 12 12
cm cm cm cm cm cm
4 mm 4 mm e 4 mm 4 mm e
Sand Clay/loess
20 25 m 13 15 m
15 cm 12 cm
4 mm e
‘Village’, >8000 m2
Grid spacing
Core diameter
Sieving mesh
Grid spacing
Core diameter
Sieving mesh
Cultural layer 20 25 m
3 cm
e
40 50 m
3 cm
e
Cultural layer 30 35 m
3 cm
e
80 90 m
3 cm
e
Unspecified
Flint and ceramic scatters can only be effectively detected when artefact densities are >80 per m2, for lower densities core sampling is not recommended. Sieving is used to increase the detection probability of artefacts. In clay or loess soils however, sieving may be too difficult, and an alternative strategy is given using a larger number of samples. Cultural layers are distinct lithostratigraphical units that can be recognized directly as archaeological relics, and hence have a detection probability of 1. Source: Tol et al., 2006: p. 38.
Table 2 Example of calculating discovery probabilities from core samples using the binomial distribution. Given a detection probability of artefacts ranging from 0.1 to 0.9, the probability of discovering the site is given when 5 cores are placed inside the site. For example, when the detection probability is 0.1, there is a probability of 0.59 that no cores will recover an artefact; a probability of 0.33 that 1 core will recover an artefact, 0.07 that 2 cores will recover one, and 0.01 that 3 cores will. The discovery probability is then 0.41. Number of hits
8700e4500/4000 cal BC) in the Netherlands and northwestern Belgium (Flanders) to better understand and describe their prospection characteristics. This is especially relevant since many of these sites are difficult to discover by means of core sampling because of their relatively small size, low density of artefacts and often deep stratigraphic position. 2. Background: core sampling and site detection
Detection probability 0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0 1 2 3 4 5
0.59 0.33 0.07 0.01 0.00 0.00
0.33 0.41 0.20 0.05 0.01 0.00
0.17 0.36 0.31 0.13 0.03 0.00
0.08 0.26 0.35 0.23 0.08 0.01
0.03 0.16 0.31 0.31 0.16 0.03
0.01 0.08 0.23 0.35 0.26 0.08
0.00 0.03 0.13 0.31 0.36 0.17
0.00 0.01 0.05 0.20 0.41 0.33
0.00 0.00 0.01 0.07 0.33 0.59
Discovery probability
0.41
0.67
0.83
0.92
0.97
0.99
1.00
1.00
1.00
The Netherlands and northwestern Belgium are located in the highly dynamic Rhine-Meuse-Scheldt delta, where fluvial and marine sedimentation as well as erosion have had a substantial effect on the three-dimensional distribution of archaeological sites. In many places, the archaeological remains found at the surface only form a small portion of the actual archaeological record, and subsurface survey methods are needed to detect archaeological sites. From the late 1980s on, core sampling has been an important
Table 3 List of analyzed excavations. Site name
Code
Landscape setting
Period
Excavated area
Grid size
Sieving strategy
Eyserheide
EY
Loess
Late Palaeolithic (Magdalenian) Late Palaeolithic (Ahrensburg culture) Late Mesolithic/Early Neolithic Late Mesolithic/Early Neolithic Late Mesolithic/Early Neolithic Neolithic (Single Grave culture) Early Mesolithic
158 m2
11m
2
305 m
22m
4 mm wet and dry (only in centre of the site) 2 mm wet
345 m2
50 50 cm
4 mm wet (only sandy soil)
50 50 cm
3 mm wet
50 50 cm
2 mm wet, only counted in selected transects Unknown
Geldrop-Aalsterhut
GA
Aeolian sand
Hardinxveld-De Bruin
HA
Hempens
HE
River dune covered with fluvial clay Aeolian sand covered with marine clay Aeolian sand covered with marine clay Tidal marsh deposits covered with marine clay Point bar deposits covered with fluvial clay Valley side covered with aeolian sand Loess
A27-Hoge Vaart
HV
Keinsmerbrug
KB
Oudenaarde-Donk
OD
Stroe
ST
Sweikhuizen-Groene Paal Verrebroek-Aven Ackers (2 sites) Verrebroek-Dok Zutphen-Ooijerhoek
SW VA VD ZO
Aeolian sand covered with peat and fluvial clay Aeolian sand covered with peat and fluvial clay Aeolian sand
Late Palaeolithic (Hamburg cultuur) Late Palaeolithic (Magdalenian) Early, Middle and Late Mesolithic Early Mesolithic Early Mesolithic
443 m2 2
1342.75 m 2
145 m2
1 1 m; some squares 2 1 and 2 2 m 50 50 cm
29.5 m2
50 50 cm
1 mm wet
625 m
22m
3 mm wet
321.5 (2007) þ 43.75 m2 (2006) 2091 m2
50 50 cm
2 mm wet
50 50 cm
2 mm wet
50 50 cm
3 mm wet
432 m
2
294.75 m
2
1 mm wet
242
P. Verhagen et al. / Journal of Archaeological Science 40 (2013) 240e247
Fig. 1. Location of analyzed excavations in the Netherlands and Belgium. Site abbreviations can be found in Table 3.
survey technique in the Netherlands, since it is a relatively cheap method that can be applied in larger areas and allows archaeologists to reach depths of up to 7 m using only manual labour. Since the mid 1990s, core sampling has also become important within northwestern Belgium, mainly for the survey of river floodplains (Bats, 2007, in preparation). However, this technique has the drawback of taking only a small sediment sample, in which archaeological features are seldom recognized and artefacts may be missed altogether. Tol et al. (2004) applied the concepts of intersection probability and detection probability (Krakker et al., 1983) to predict the probability that archaeological sites of a certain dimension and artefact density will be discovered using core sampling. Intersection probability refers to the probability that a site will be intersected by a grid of core samples, test pits or trial trenches (see Drew, 1967, 1979; Singer, 1975; Krakker et al., 1983; Banning, 2002; Verhagen and Borsboom, 2009). It was proved by Drew (1979) that an equilateral triangular grid layout is the most effective layout for intersecting circular targets. Intersection probability then depends on the
actual spacing that is chosen between sample points. For differently shaped targets, other configurations may be more effective. However, since core samples only take very small sediment samples, the discovery probability of an archaeological site through core sampling depends to a large degree on the probability that a core taken inside an archaeological site actually contains artefacts. The detection probability of an individual artefact is determined using the following equation (Koopman, 1980; Stone, 1981; Verhagen, 2005), based on the Poisson distribution:
D ¼ 1 eAdW where D ¼ detection probability; e ¼ the base of natural logarithms (2.711828); A ¼ the area of the sampling unit; d ¼ the density of artefacts per area unit; and W ¼ the observation probability, i.e. the probability that an artefact will be recognized as such when it is recovered.
P. Verhagen et al. / Journal of Archaeological Science 40 (2013) 240e247 400
405
243
410
415
420
410
415
420
-290
-290
Verrebroek-Aven Ackers 2007 pieces of flint per excavared square 0 1 - 10 11 - 20
-295
51 - 100
-300
101 - 330
-320
-320
-315
-315
-310
-310
-305
-305
-300
-295
21 - 50
0 0,5 1
2
3
4
5 m
400
405
Fig. 2. Distribution of lithic artefacts in the Mesolithic site of Verrebroek-Aven Ackers excavation campaign 2007.
The number of samples taken within a site with a certain artefact density then determines its actual discovery probability and can be calculated using the binomial distribution (Table 2). A problem with this model is that it assumes artefact distribution to
be random within the site. In practice, artefacts are often found clustered, and calculations based on a random distribution of artefacts will not be very realistic (see e.g., Nance and Ball, 1986). Kintigh (1988) therefore recommended calculating artefact
Table 4 Discovery probabilities for the site of Verrebroek-Aven Ackers 2007. For different sampling distances and core diameters, discovery probabilities have been simulated using the flint counts per excavated square. For purposes of comparison, discovery probabilities were also obtained by simulation based on mean flint density, and calculated on the basis of the detection probability equation given in Section 2, as well as the k-parameter. Distance between sampling points
Auger diameter
Simulation based on counts per square
Simulation based on mean find density
Calculation based on Poisson distribution
Calculation based on k-parameter
25 m 20 m 15 m 10 m 5m
12 12 12 12 12
cm cm cm cm cm
0.232 0.350 0.622 0.920 1.000
0.310 0.466 0.740 0.959 1.000
0.310 0.466 0.739 0.959 1.000
0.214 0.334 0.526 0.847 1.000
25 m 20 m 15 m 10 m 5m
15 15 15 15 15
cm cm cm cm cm
0.275 0.415 0.724 0.970 1.000
0.379 0.553 0.852 0.989 1.000
0.377 0.553 0.852 0.990 1.000
0.214 0.390 0.608 0.905 1.000
244
P. Verhagen et al. / Journal of Archaeological Science 40 (2013) 240e247
1,0
A27-Hoge Vaart Eyserheide
0,9
Zutphen-Ooijerhoek Stroe
0,8
Geldrop-Aalsterhut
discovery probability
0,7
Sweikhuizen-Groene Paal Hempens
0,6
Oudenaarde-Donk Verrebroek-Aven Ackers 2006
0,5
Verrebroek-Aven Ackers 2007 Verrebroek-Dok
0,4
Hardinxveld-De Bruin
0,3
Keinsmerbrug
0,2 0,1 0,0 50 m
25 m
20 m
15 m
10 m
5m
sample point spacing Fig. 3. Overview of discovery probabilities with an auger diameter of 15 cm and different sample point spacings for all analyzed sites.
detection probabilities using the k-parameter (Pielou, 1977: pp. 124e134; Nance, 1983: pp. 300e310; McManamon, 1984: pp. 269e273). This is a measure for clustering that can be used on the assumption that the distribution of artefacts in a site follows the negative binomial distribution. However, the k-parameter can only be calculated on the basis of known artefact distributions, and therefore needs good empirical data in order to be applied to the problem of establishing appropriate survey strategies. Furthermore, it is not clear whether the negative binomial distribution is really a suitable model for the distribution of artefacts in a site. This is why we decided to use a different approach, and apply statistical simulations on the available data to obtain discovery probabilities independently of any theoretical assumptions of artefact distribution.
excavations in the Netherlands and northwestern Belgium (Flanders), (Table 3; Fig. 1). These were partly taken from EASY, the digital archiving system for archaeological data maintained by DANS (https://easy.dans.knaw.nl). Additional datasets were made available by the Cultural Heritage Agency (RCE), Ghent University and Groningen University. The data mainly concern buried sites and a few surface sites. Unfortunately, different excavation practices and soil conditions imply that sieving was done at different mesh sizes of 1, 2, 3 or 4 mm; furthermore, excavation square sizes were not always equal (ranging from 50 50 cm to 2 2 m), and not all sites had been completely excavated. All in all, this means that the sites cannot be compared very well; however, the data can be used to simulate discovery probabilities for the registered site sizes and lithic artefact densities.
3. Dataset
4. Simulations
The current study has aimed to collect datasets that allowed us to investigate the actual density of artefacts in Palaeolithic and Mesolithic sites. For this, it was necessary to analyze excavations where lithic artefacts were sieved and counted per square. Surprisingly, the number of suitable excavations available in digital form is still rather small. In total, we collected data from 12
In order to establish the discovery probabilities for each excavated site, simulations were set up using different core sampling grid spacings, covering the range from the recommended strategies in Tol et al. (2006) to a very dense spacing of 5 m. An equilateral triangular grid layout was applied throughout, as this is the most efficient grid configuration possible (Krakker et al., 1983). In
Table 5 Overview of analysis results. Find density and site size were determined on the basis of squares containing flints.
Eyserheide Geldrop-Aalsterhut Hardinxveld-De Bruin Hempens A27-Hoge Vaart Keinsmerbrug Oudenaarde-Donk Stroe Sweikhuizen-Groene Paal Verrebroek-Aven Ackers 2006 Verrebroek-Aven Ackers 2007 Verrebroek-Dok Zutphen-Ooijerhoek
Lithology
Mesh size in mm
Find density per m2
Estimated find density at 3 mm
k-parameter
Effect clustering (average)
Area in m2
Loess Sand Sand Sand Sand Clay Clay Sand Loess Sand Sand Sand Sand
4 2 4 3 10 ? 1 1 3 2 2 2 3
20.3 19.5 50.6 167.9 31.6 1.8 21.7 13.6 53.0 41.2 95.4 200.2 66.1
27.1 13.0 67.5 167.9 105.3 ? 7.2 4.5 53.0 27.5 63.6 133.5 66.1
0.25 0.12 0.78 0.50 0.29 0.39 0.41 0.35 0.06 0.64 0.51 0.17 0.34
3.43% 2.05% 3.64% 9.34% 5.64% 0.04% 2.90% 0.57% 12.35% 2.81% 7.40% 11.58% 10.80%
104.75 82.00 246.50 443.00 916.00 190.00 87.25 17.25 160.00 43.75 305.75 1428.50 246.75
P. Verhagen et al. / Journal of Archaeological Science 40 (2013) 240e247
practice, somewhat less efficient triangular layouts are used (like 40 50 m or 20 25 m), since a perfect equilateral triangular layout is often thought to be more difficult to set up in the field using measuring tape e although in practice this should not be a problem, certainly not with modern GPS equipment. The simulations were carried out through a Python-script. In each simulation, a grid was randomly placed on top of the site 100,000 times, and for each ‘virtual’ core sample hitting the site, detection probabilities were calculated on the basis of the counted flint fragments per square of 50 50 cm. The detection probability per square was established using the artefact detection equation based on the Poisson distribution; it is assumed that artefacts are distributed randomly within each square. The observation probability was set to 1. A core sample was then considered successful if a randomly generated number between 0.0 and 1.0 was smaller than the detection probability at that particular spot. A single positive core was considered to constitute a site discovery. The total number of ‘successful’ core samples was then used to calculate the discovery probability of the site analyzed. For purposes of comparison, separate calculations were performed using the kparameter, as well as the detection probability equation given in Section 2. 5. Results The simulations resulted in discovery probabilities for each analyzed site, and for the selected grid spacings. As an example, the case of the Mesolithic site of Verrebroek-Aven Ackers, consisting of several very small sand dunes each yielding evidence of Mesolithic occupation (Bats et al., 2004), is presented here; the remaining cases are discussed in Verhagen et al. (2011). The simulations were carried out on the data from the excavations conducted in 2007 (Sergant et al., 2007; Crombé et al., 2009), which cover almost an entire dune top, limited by local coordinates 399.5/320.5 and 419.5/291.5 (1286 50 50 cm squares, measuring 321.5 m2; Fig. 2; of these, 1223 squares were actually excavated). The other dunes were only excavated partially due to time limitations. Sediment was sieved at a 2 mm mesh size, and the average find density within the excavated zone is 95.4 per m2. The distribution of artefacts is left-skewed (Pearson’s skewness 3.21), and the kparameter for this site is 0.51, pointing to relatively strong clustering. Simulation results (Table 4) indicate that this site can be discovered with relative ease using core sampling, even when a 10 m spacing is necessary to achieve the required 75% discovery probability. The strategy recommended by Tol et al. (2006) for this type of site (20 25 m grid with 15 cm diameter core) will only result in a 41.5% chance of discovery. An overview of the simulation results for all analyzed sites is given in Fig. 3 and Table 5.
245
Table 6 Calculation of values of Df based on the data from Tol et al. (2004) and additional data from analyzed excavations. Site
Df
Size class boundary
Gennep site F Haelen Broekweg Ittersumerbroek West
1.21 1.03 0.92 1.00 0.39 2.43 0.58 1.10 1.73 0.70 1.96 0.45 1.16
4 3 4 3 2 2 4 2 4 3 4 2 4
Merselo Mienakker Site 0 Voetakker vindplaats 28 Zutphen-Ooijerhoek Schipluiden-Harnaschpolder A27-Hoge Vaart Eyserheide
mm mm mm mm mm mm mm mm mm mm mm mm mm
6. The effect of sieving It is difficult to directly compare the excavations analyzed and draw general conclusions about stone artefact densities in prehistoric sites because of the different sieving strategies applied, ranging from 1 to 4 mm mesh sizes in various soil types. Sieving is a major factor influencing the detection of flints. Tol et al. (2004: p. 46, Tables 10 and 11) show for a selected number of 7 cases that the fraction between 3 or 4 mm and 1 cm took up more than 50% of the total number of stone artefacts recovered. In 3 cases where smaller mesh sizes were applied, the fraction between 1 and 2 mm again took up more than 50% of total number of artefacts recovered. Brown (2001) suggested that the size distribution of flints in archaeological sites can be modelled using a fractal relation:
Nð>rÞ ¼ r Df where: N(>r) ¼ the number of fragments larger than r; r ¼ fragment size; Df ¼ the fractal dimension, or:
lnðNð>rÞÞ Df ¼ lnðrÞ The value of Df however can only be calculated if we know the actual fragment size distribution. For this purpose, Brown (2001: p.
Recommended A4 A5 A1 A6 A2 A3 strategy (Tol 50 m 15 cm 35 m 12 cm 25 m 15 cm 25 m 12 cm 20 m 15 cm 20 m 12 cm 15 m 15 cm 15 m 12 cm 10 m 15 cm 10 m 12 cm 5 m 15 cm 5 m 12 cm et al., 2006) e e A1 A1 A2 e e e e e A1 A1 A1
0.138
0.437
0.205
0.626
0.041 0.041 0.218 0.480 0.486 0.011 0.044 0.006 0.113 0.031 0.275 0.972 0.196
0.037 0.029 0.166 0.420 0.376 0.007 0.031 0.005 0.088 0.024 0.232 0.830 0.157
0.072 0.057 0.344 0.701 0.665 0.017 0.067 0.009 0.176 0.052 0.415 0.982 0.298
0.051 0.042 0.266 0.623 0.534 0.011 0.047 0.007 0.140 0.038 0.350 0.958 0.245
0.131 0.096 0.569 0.923 0.871 0.029 0.112 0.016 0.289 0.095 0.724 1.000 0.483
0.094 0.069 0.450 0.861 0.754 0.019 0.081 0.011 0.228 0.073 0.622 0.998 0.403
0.269 0.218 0.864 0.999
0.196 0.157 0.748 0.996
0.757 0.683 1.000 1.000
0.617 0.534 0.997 1.000
0.068 0.258 0.039 0.600 0.202 0.970
0.044 0.189 0.027 0.491 0.156 0.920
0.245 0.715 0.140 0.989 0.649 1.000
0.165 0.575 0.099 0.960 0.521 1.000
0.797
0.707
1.000
1.000
246
P. Verhagen et al. / Journal of Archaeological Science 40 (2013) 240e247
Table 7 Estimation of flint densities per m2 for different mesh sizes of the analyzed excavations, based on Df ¼ 1. In bold the originally registered flint densities per m2. Site
Mesh size
1 cm
4 mm
3 mm
2 mm
1 mm
Eyserheide Geldrop-Aalsterhut Hardinxveld-De Bruin Hempens A27-Hoge Vaart Keinsmerbrug Oudenaarde-Donk Stroe Sweikhuizen-Groene Paal Verrebroek-Aven Ackers 2006 Verrebroek-Aven Ackers 2007 Verrebroek-Dok Zutphen-Ooijerhoek
4 mm 2 mm 4 mm 3 mm 10 mm Unknown 1 mm 1 mm 3 mm
8.1 3.9 20.2 50.4 31.6 e 2.2 1.4 15.9
20.3 9.8 50.6 125.9 79.0 e 5.4 3.4 39.8
27.1 13.0 67.5 167.9 105.3 e 7.2 4.5 53.0
40.6 19.5 101.2 251.9 158.0 e 10.9 6.8 79.5
81.2 39.0 202.4 503.7 316.0 e 21.7 13.6 159.0
2 mm
8.2
20.6
27.5
41.2
82.4
2 mm
19.1
47.7
63.6
95.4
190.8
2 mm 3 mm
40.0 19.8
100.1 49.6
133.5 66.1
200.2 99.2
400.4 198.3
263) cites a number of (American) sources describing the size distribution of flint, quartz and obsidian fragments obtained from stone-working experiments. The calculated values of Df range from 0.38 to 3.33 (mean 2.10; standard deviation 0.62). Tol et al. (2004: pp. 45e46) provide some figures that can be used to calculate values for Df as well (Tables 6 and 7). The accuracy of these calculations is probably not very high, since the original data only differentiate between two size classes (smaller/larger than 2, 3 or 4 mm). The calculated values of Df (mean 1.11; standard deviation 0.60) are substantially lower than those mentioned by Brown. Data provided by Ghent University from 29 excavated artefact clusters from Verrebroek-Dok sieved over 2 mm give a similar outcome (mean 1.13; standard deviation 0.20; Crombé, 1998, 2005). A reason for this may be that the fine fraction in archaeological sites is vulnerable to erosion, and may have been blown or washed away, whereas the material from flint-working experiments will still contain this finer material. The figures obtained, while not extremely accurate, are nevertheless useful to make an educated guess of the effect of sieving on detection probabilities. If we take a (probably unrealistically) conservative estimate of Df ¼ 1, the registered flint densities can be recalculated to provide estimates for each mesh size (Table 6). Since Table 8 Recommended new survey strategies for Stone Age sites. Type Very small (<50 m2) Low find density (40e80 per m2) Very low find density (<40 per m2) Small (50e200 m2) Low find density (40e80 per m2) Very low find density (<40 per m2) Medium-sized: 200e1000 Medium-high find density (>80 per m2) Low find density (40e80 per m2) Very low find density (<40 per m2) Large: >1000 m2 Medium-high find density (>80 per m2) Low find density (40e80 per m2)
Lithology
Core sampling Auger Observation grid diameter technique
Irrelevant e
e
e
Irrelevant e
e
e
Irrelevant 4 5 m
15 cm
Irrelevant 4 5 m þ test pits m2 Irrelevant 13 15 m
e
3 mm sieving mesh e
12 cm
Irrelevant 8 10 m
15 cm
Irrelevant 4 5 m þ test pits
e
Irrelevant 20 25 m
12 cm
Irrelevant 13 15 m
12 cm
3 mm sieving mesh 3 mm sieving mesh e
3 mm sieving mesh 3 mm sieving mesh
the original guidelines written by Tol et al. (2006) specified that sieving should always be done at 3 mm, we can in this way obtain a baseline to compare our empirical data with the recommendations in the guidelines. From this we can conclude that a density of 80 artefacts per m2, which was thought necessary to guarantee an artefact detection probability of >75%, is not met by 9 of the 11 analyzed sites. 7. Conclusions The model presented by Tol et al. (2006) departs from the assumption of a random distribution of artefacts within an archaeological site. In practice, artefacts always display a certain amount of clustering. This is clearly illustrated when we calculate the k-parameter for the sites analyzed (mean 0.37, standard deviation 0.20). Clustering is always there, even when it can vary per site. The effect of clustering on the discovery probability of an archaeological site through core sampling was analyzed using simulation. This shows that clustering diminishes the probabilities of finding artefacts in a core. The calculation of discovery probabilities on the basis of the k-parameter shows similar results, but generally provides a lower estimate. However, the effect of clustering is relatively limited, and only becomes a major issue when high artefact densities are combined with a less intensive survey strategy. On the basis of the simulations it is unfortunately impossible to specify a clear relationship between clustering, artefact density, site size and core sampling strategy. The (conservative) estimate of the fractal dimensions of artefact size distributions lead us to conclude that a significant proportion of Palaeolithic and Mesolithic sites in the Netherlands and northwestern Belgium have a considerably lower artefact density (<80 per m2) than the cases considered in the survey guidelines (Tol et al., 2006). Using smaller mesh sizes for sieving will substantially increase the detection probabilities of these sites (Bats, 2007). Furthermore, it seems probable that a substantial number of sites are smaller than the dimensions specified in the guidelines (<200 m diameter). From the simulation analysis (Fig. 3, Table 5) it can be concluded that, using a 15 cm auger and 3 mm meshes, for the majority of sites a 10 m grid is sufficient to reach a 75% discovery probability. A coarser grid spacing will only yield this probability if the find density is high to very high, as is the case of Hempens, A27-Hoge Vaart, and Verrebroek-Dok. On the other hand, it seems that sites with very low (20e40 per m2) to extremely low (<20 per m2) artefact densities, which in addition are often smaller in size (e.g., Stroe and Geldrop-Aalsterhut), demand a more intensive core sampling strategy using a 5 m grid combined with test pits. The latter is unfortunately difficult to apply at larger depths. Another solution is to reduce the sieve meshes from 3 to 1 mm. Recent tests (Bats, 2007) have clearly demonstrated that smaller mesh sizes considerably increase the discovery rate of low-density sites. In any case, finding sites that combine a very small surface area and a very low artefact density will largely depend on chance, certainly if these are situated at great depth. This means that the guidelines will have to be revised to take this new information into account (Table 8). A further consideration is the question how we want to deal with these ‘difficult’ sites in archaeological heritage management. The effort needed to find the small and unobtrusive kind of sites is of course much larger than for other site types. Until now, Dutch archaeologists have taken the stance that survey efforts should be customized to provide the 75% probability of finding these. However, in cases when this would imply an enormous investment in survey, it is legitimate to ask whether the 75% discovery probability is the target that should always be aimed for. The results of the current study make this
P. Verhagen et al. / Journal of Archaeological Science 40 (2013) 240e247
question more urgent. If small prehistoric sites with low artefact densities are relatively common occurrences, it would imply a substantial increase in survey effort to meet the demands of current AHM policies. If this effort should be devoted is of course a policy issue, but it is advisable to give them some priority in future research, if only to establish how much they contribute to our knowledge of Stone Age archaeology. Furthermore, it should be questioned whether it is wise to adapt the core sampling strategy to the types of sites that are expected within a study region, Tol et al. (2006) propose. Is it possible to determine the occurrence of these types of sites a priori (in terms of size, density of artefacts) on the basis of desk-based research, i.e. before fieldwork has started? How can specific categories of sites that are not expected ever be discovered? And how can we be sure about our predictions? One solution to this problem is to apply a common core sampling and sieving strategy for all regions; based on our findings a core-sampling grid of 10 m seems the best option. Alternatively, we may choose to apply intensive core sampling (4 5 m grid) and/or test pits in targeted locations with a high expectation for the presence of Stone Age sites, irrespective of their (expected) size or artefact density. Acknowledgements The research done for this paper was made possible by the Rijksdienst voor het Cultureel Erfgoed (RCE), as part of the programme “What is Heritage?”. The authors would like to thank the colleagues who provided datasets for the analysis e in particular Jos Deeben (RCE) and Gary Nobles (Groningen University) e as well as the project advisory committee: Jan-Willem de Kort, Dr. Bjørn Smit, José Schreurs, Jos Deeben (all RCE), Dr. Hans Peeters, Marcel Niekus (both Groningen University), Willem-Jan Hogestijn (Municipality of Almere), Antoine Wilbers (IDDS Archeologie), Jaap Beuker (Province of Drenthe), Dr. Jos de Moor (EARTH Integrated Archaeology), Axel Müller (ADC Archeoprojecten), Johan Jelsma (De Steekproef), and Esther Wieringa (SIKB). References Banning, E.B., 2002. Archaeological Survey. Kluwer Academic/Plenum Publishers, New York (Manuals in Archaeological Method, Theory, and Technique). Bats, M., 2007. The Flemish Wetlands: an archaeological survey of the valley of the river Scheldt. In: Barber, J., Clark, C., Cressey, M., Crone, A., Hale, A., Henderson, J., Housley, R., Sands, R., Sheridan, A. (Eds.), Archaeology from the Wetlands: Recent Perspectives. Proceedings of the 11th WARP Conference, Edinburgh 2005. Society of Antiquaries of Scotland, Edinburgh, pp. 93e100 (WARP Occasional Paper, 18). Bats, M., Klinck, B., Meersschaert, L., Sergant, J., 2004. Verkennend en waarderend booronderzoek in het alluvium van de Schelde. Notae Praehistoricae 24,175e179. Bats, M., in preparation. De Vlaamse Wetlands, een archeologische verkenning van de Scheldevallei. PhD-thesis Ghent University.
247
Borsboom, A.J., Verhagen, J.W.H.P., 2009. KNA Leidraad Inventariserend Veldonderzoek. Deel: Proefsleuvenonderzoek (IVO-P). SIKB, Gouda. Brown, C.T., 2001. The fractal dimensions of lithic reduction. Journal of Archaeological Science 28, 619e631. Crombé, P., 1998. The Mesolithic in Northwestern Belgium, Recent Excavations and Surveys. Archaeopress, Oxford (British Archaeological Reports, International Series 716). Crombé, P. (Ed.), 2005. The Last Hunter-gatherer-fishermen in Sandy Flanders (NW Belgium); the Verrebroek and Doel Excavation Projects. Part 1: Palaeoenvironment, Chronology and Features. Academia Press, Ghent (Archaeological Reports Ghent University 3). Crombé, P., Sergant, J., Lombaert, L., Van Strydonck, M., Boudin, M., 2009. The Mesolithic and Neolithic site of Verrebroek e Aven Ackers (East Flanders, Belgium): the radiocarbon evidence. Notae Praehistoricae 29, 15e21. Drew, L.J., 1967. Grid-drilling exploration and its application to the search for petroleum. Geology 62, 698e710. Drew, L.J., 1979. Pattern drilling exploration: optimum pattern types and hole spaces when searching for elliptical shaped targets. Mathematical Geology 11, 223e254. Kintigh, K.W., 1988. The effectiveness of subsurface testing: a simulation approach. American Antiquity 53, 686e707. Koopman, B.O., 1980. Search and Screening: General Principles with Historical Applications. Pergamon Press. Krakker, J.J., Shott, M.J., Welch, P.D., 1983. Design and evaluation of shovel-test sampling in regional archaeological survey. Journal of Field Archaeology 10, 469e480. McManamon, F.P., 1984. Discovering sites unseen. In: Schiffer, M.B. (Ed.), 1984. Advances in Archaeological Method and Theory, vol. 7. Academic Press, New York, pp. 223e292. Nance, J.D., 1983. Regional sampling in archaeological survey: the statistical perspective. In: Schiffer, M.B. (Ed.), 1983. Advances in Archaeological Method and Theory, vol. 6. Academic Press, New York, pp. 289e356. Nance, J.D., Ball, F.B., 1986. No surprises? The reliability and validity of test pit sampling. American Antiquity 51, 457e483. Pielou, E.C., 1977. Mathematical Ecology. Wiley, New York. Sergant, J., Bats, M., Noens, G., Lombaert, L., D’Hollander, D., 2007. Voorlopige resultaten van noodopgravingen in het afgedekte dekzandlandschap van Verrebroek Aven Ackers (Mesolithicum, Neolithicum). Notae Praehistoricae 27, 101e107. SIKB, 2010. KNA Landbodems Versie 3.2. SIKB, Gouda. Singer, D.A., 1975. Relative efficiencies of square and triangular grids in the search for elliptically shaped resource targets. Journal of Research of the U.S. Geological Survey 3, 163e167. Stone, G.D., 1981. On artifact densities and shovel probes. Current Anthropology 22, 182e183. Tol, A.J., Verhagen, J.W.H.P., Borsboom, A.J., Verbruggen, M., 2004. Prospectief boren. Een studie naar de betrouwbaarheid en toepasbaarheid van booronderzoek in de prospectiearcheologie. RAAP Archeologisch Adviesbureau, Amsterdam (RAAP-rapport 1000). Tol, A.J., Verhagen, J.W.H.P., Verbruggen, M., 2006. Leidraad inventariserend veldonderzoek; Deel: karterend booronderzoek. SIKB, Gouda. Verhagen, P., 2005. Prospection strategies and archaeological predictive modelling. In: Leusen, M. van, Kamermans, H. (Eds.), Predictive Modelling for Archaeological Heritage Management: a Research Agenda. Rijksdienst voor het Oudheidkundig Bodemonderzoek, Amersfoort, pp. 109e121 (Nederlandse Archeologische Rapporten 29). Verhagen, P., Borsboom, A., 2009. The design of effective and efficient trial trenching strategies for discovering archaeological sites. Journal of Archaeological Science 36, 1807e1816. Verhagen, P., Rensink, E., Bats, M., Crombé, P., 2011. Optimale strategieën voor het opsporen van Steentijdvindplaatsen met behulp van booronderzoek. Een statistisch perspectief. Rijksdienst voor het Cultureel Erfgoed, Amersfoort (Rapportages Archeologische Monumentenzorg 197).
Contents lists available at SciVerse ScienceDirect
Journal of Archaeological Science journal homepage: http://www.elsevier.com/locate/jas
Establishing discovery probabilities of lithic artefacts in Palaeolithic and Mesolithic sites with core sampling Philip Verhagen a, *, Eelco Rensink b, Machteld Bats c, Philippe Crombé c a
Research Institute for the Heritage and History of the Cultural Landscape and Urban Environment (CLUE), Faculty of Arts, VU University, De Boelelaan 1105, 1081 HV, Amsterdam, The Netherlands b Cultural Heritage Agency (RCE), Amersfoort, The Netherlands c Department of Archaeology, Ghent University, Ghent, Belgium
a r t i c l e i n f o
a b s t r a c t
Article history: Received 17 January 2012 Received in revised form 22 April 2012 Accepted 28 May 2012
This paper reports the results of a study into the effectiveness of core sampling for discovering Palaeolithic and Mesolithic hunter-gatherer sites in the Netherlands and northwestern Belgium. Earlier work established optimal sampling strategies for use in archaeological heritage management survey in the Netherlands. However, the statistical model used for this was based on a limited amount of data on the distribution of lithic artefacts in Palaeolithic and Mesolithic sites. For the current study we have analyzed the distribution of artefacts in a selected number of excavated sites, and estimated discovery probabilities of these sites through simulation. The simulation results indicate that discovery probabilities are lower than expected due to the effect of clustering of finds. Furthermore, the density of flints in Palaeolithic and Mesolithic sites is generally lower than the estimates that were used for setting up the optimal sampling strategies, and a substantial number of sites is very small. This means that, in order to discover Palaeolithic and Mesolithic sites with sufficient reliability, we will have to apply more intensive survey strategies than have been recommended up to now. Ó 2012 Elsevier Ltd. All rights reserved.
Keywords: Archaeological survey Core sampling Lithic artefacts Archaeological heritage management Simulation Palaeolithic and Mesolithic sites
1. Introduction Over the past ten years, procedures for fieldwork in archaeological heritage management (AHM) in the Netherlands have become codified in a system of regulations set up by the archaeological sector itself. This set of quality norms (the Quality Norm for Dutch Archaeology or KNA version 3.2; SIKB, 2010) describes the procedures to be followed in AHM research, moving from deskbased assessment through survey to excavation. The quality norms specify what needs to be done, but do not prescribe how things should be done. However, in some cases, it was felt that additional guidance was needed on the ‘how’ as well. One of these issues is the establishment of the most effective and efficient strategies for detecting archaeological sites. Accompanying guidelines have therefore been developed concerning the use of core sampling (Tol et al., 2006) and trial trenching (Borsboom and Verhagen, 2009) e these being the most frequently used survey methods in the Netherlands. The guidelines are based on * Corresponding author. Tel.: þ31 20 5982848. E-mail addresses: [email protected] (P. Verhagen), e.rensink@ cultureelerfgoed.nl (E. Rensink), [email protected] (M. Bats), [email protected] (P. Crombé). 0305-4403/$ e see front matter Ó 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.jas.2012.05.041
theoretical statistical models that specify the probability of detecting archaeological sites of a certain dimension and find density (see Tol et al., 2004; Verhagen, 2005; Verhagen and Borsboom, 2009). They are used to help design survey project briefs and to evaluate survey results. The guidelines provide preferred survey strategies that will result in a 75% chance of discovery of archaeological sites that are classified according to such prospection characteristics as size and artefact density (Table 1). However, it is still difficult to assess the actual effectiveness of these strategies, since the prospection characteristics of many archaeological site types are only known in general terms. More empirical data are needed to compare the situation in the field to the theoretical assumptions used for the guidelines. Unfortunately, this type of data is still in small supply, and no mechanisms are available in Dutch AHM that would enable us to increase our knowledge on these aspects. The excavation of sites is usually considered to be the closing chapter of AHM, and is not systematically used to collect information on, for example, the spatial distribution and density of artefacts or features that can be of use in the earlier phases of fieldwork in the future. This paper reports the results of an investigation that has tried to do just that. We have used excavation data of Stone Age sites, especially from the Late Palaeolithic (ca. 13,000e8700 cal BC) and Mesolithic (ca.
P. Verhagen et al. / Journal of Archaeological Science 40 (2013) 240e247
241
Table 1 Overview of standard strategies for core sampling survey in the Netherlands for different site types. Site type
Lithology
Stone Age Medium size, 200e1000 m2 Base camp House plan Large size, >2000 m2 Large base camp Aggregated settlement Multiple house plans
Flint scatter Sand Clay/loess Clay/loess Sand Clay/loess Clay/loess
20 17 13 40 30 20
25 20 15 50 35 25
m m m m m m
15 12 12 15 12 12
cm cm cm cm cm cm
3 mm 3 mm e 3 mm 3 mm e
Bronze Age e Middle Ages House plan(s), 500e2000 m2
Ceramic scatter Sand Clay/loess Clay/loess Sand Clay/loess Clay/loess
30 20 17 80 60 40
35 25 20 90 70 50
m m m m m m
15 12 12 15 12 12
cm cm cm cm cm cm
4 mm 4 mm e 4 mm 4 mm e
Sand Clay/loess
20 25 m 13 15 m
15 cm 12 cm
4 mm e
‘Village’, >8000 m2
Grid spacing
Core diameter
Sieving mesh
Grid spacing
Core diameter
Sieving mesh
Cultural layer 20 25 m
3 cm
e
40 50 m
3 cm
e
Cultural layer 30 35 m
3 cm
e
80 90 m
3 cm
e
Unspecified
Flint and ceramic scatters can only be effectively detected when artefact densities are >80 per m2, for lower densities core sampling is not recommended. Sieving is used to increase the detection probability of artefacts. In clay or loess soils however, sieving may be too difficult, and an alternative strategy is given using a larger number of samples. Cultural layers are distinct lithostratigraphical units that can be recognized directly as archaeological relics, and hence have a detection probability of 1. Source: Tol et al., 2006: p. 38.
Table 2 Example of calculating discovery probabilities from core samples using the binomial distribution. Given a detection probability of artefacts ranging from 0.1 to 0.9, the probability of discovering the site is given when 5 cores are placed inside the site. For example, when the detection probability is 0.1, there is a probability of 0.59 that no cores will recover an artefact; a probability of 0.33 that 1 core will recover an artefact, 0.07 that 2 cores will recover one, and 0.01 that 3 cores will. The discovery probability is then 0.41. Number of hits
8700e4500/4000 cal BC) in the Netherlands and northwestern Belgium (Flanders) to better understand and describe their prospection characteristics. This is especially relevant since many of these sites are difficult to discover by means of core sampling because of their relatively small size, low density of artefacts and often deep stratigraphic position. 2. Background: core sampling and site detection
Detection probability 0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0 1 2 3 4 5
0.59 0.33 0.07 0.01 0.00 0.00
0.33 0.41 0.20 0.05 0.01 0.00
0.17 0.36 0.31 0.13 0.03 0.00
0.08 0.26 0.35 0.23 0.08 0.01
0.03 0.16 0.31 0.31 0.16 0.03
0.01 0.08 0.23 0.35 0.26 0.08
0.00 0.03 0.13 0.31 0.36 0.17
0.00 0.01 0.05 0.20 0.41 0.33
0.00 0.00 0.01 0.07 0.33 0.59
Discovery probability
0.41
0.67
0.83
0.92
0.97
0.99
1.00
1.00
1.00
The Netherlands and northwestern Belgium are located in the highly dynamic Rhine-Meuse-Scheldt delta, where fluvial and marine sedimentation as well as erosion have had a substantial effect on the three-dimensional distribution of archaeological sites. In many places, the archaeological remains found at the surface only form a small portion of the actual archaeological record, and subsurface survey methods are needed to detect archaeological sites. From the late 1980s on, core sampling has been an important
Table 3 List of analyzed excavations. Site name
Code
Landscape setting
Period
Excavated area
Grid size
Sieving strategy
Eyserheide
EY
Loess
Late Palaeolithic (Magdalenian) Late Palaeolithic (Ahrensburg culture) Late Mesolithic/Early Neolithic Late Mesolithic/Early Neolithic Late Mesolithic/Early Neolithic Neolithic (Single Grave culture) Early Mesolithic
158 m2
11m
2
305 m
22m
4 mm wet and dry (only in centre of the site) 2 mm wet
345 m2
50 50 cm
4 mm wet (only sandy soil)
50 50 cm
3 mm wet
50 50 cm
2 mm wet, only counted in selected transects Unknown
Geldrop-Aalsterhut
GA
Aeolian sand
Hardinxveld-De Bruin
HA
Hempens
HE
River dune covered with fluvial clay Aeolian sand covered with marine clay Aeolian sand covered with marine clay Tidal marsh deposits covered with marine clay Point bar deposits covered with fluvial clay Valley side covered with aeolian sand Loess
A27-Hoge Vaart
HV
Keinsmerbrug
KB
Oudenaarde-Donk
OD
Stroe
ST
Sweikhuizen-Groene Paal Verrebroek-Aven Ackers (2 sites) Verrebroek-Dok Zutphen-Ooijerhoek
SW VA VD ZO
Aeolian sand covered with peat and fluvial clay Aeolian sand covered with peat and fluvial clay Aeolian sand
Late Palaeolithic (Hamburg cultuur) Late Palaeolithic (Magdalenian) Early, Middle and Late Mesolithic Early Mesolithic Early Mesolithic
443 m2 2
1342.75 m 2
145 m2
1 1 m; some squares 2 1 and 2 2 m 50 50 cm
29.5 m2
50 50 cm
1 mm wet
625 m
22m
3 mm wet
321.5 (2007) þ 43.75 m2 (2006) 2091 m2
50 50 cm
2 mm wet
50 50 cm
2 mm wet
50 50 cm
3 mm wet
432 m
2
294.75 m
2
1 mm wet
242
P. Verhagen et al. / Journal of Archaeological Science 40 (2013) 240e247
Fig. 1. Location of analyzed excavations in the Netherlands and Belgium. Site abbreviations can be found in Table 3.
survey technique in the Netherlands, since it is a relatively cheap method that can be applied in larger areas and allows archaeologists to reach depths of up to 7 m using only manual labour. Since the mid 1990s, core sampling has also become important within northwestern Belgium, mainly for the survey of river floodplains (Bats, 2007, in preparation). However, this technique has the drawback of taking only a small sediment sample, in which archaeological features are seldom recognized and artefacts may be missed altogether. Tol et al. (2004) applied the concepts of intersection probability and detection probability (Krakker et al., 1983) to predict the probability that archaeological sites of a certain dimension and artefact density will be discovered using core sampling. Intersection probability refers to the probability that a site will be intersected by a grid of core samples, test pits or trial trenches (see Drew, 1967, 1979; Singer, 1975; Krakker et al., 1983; Banning, 2002; Verhagen and Borsboom, 2009). It was proved by Drew (1979) that an equilateral triangular grid layout is the most effective layout for intersecting circular targets. Intersection probability then depends on the
actual spacing that is chosen between sample points. For differently shaped targets, other configurations may be more effective. However, since core samples only take very small sediment samples, the discovery probability of an archaeological site through core sampling depends to a large degree on the probability that a core taken inside an archaeological site actually contains artefacts. The detection probability of an individual artefact is determined using the following equation (Koopman, 1980; Stone, 1981; Verhagen, 2005), based on the Poisson distribution:
D ¼ 1 eAdW where D ¼ detection probability; e ¼ the base of natural logarithms (2.711828); A ¼ the area of the sampling unit; d ¼ the density of artefacts per area unit; and W ¼ the observation probability, i.e. the probability that an artefact will be recognized as such when it is recovered.
P. Verhagen et al. / Journal of Archaeological Science 40 (2013) 240e247 400
405
243
410
415
420
410
415
420
-290
-290
Verrebroek-Aven Ackers 2007 pieces of flint per excavared square 0 1 - 10 11 - 20
-295
51 - 100
-300
101 - 330
-320
-320
-315
-315
-310
-310
-305
-305
-300
-295
21 - 50
0 0,5 1
2
3
4
5 m
400
405
Fig. 2. Distribution of lithic artefacts in the Mesolithic site of Verrebroek-Aven Ackers excavation campaign 2007.
The number of samples taken within a site with a certain artefact density then determines its actual discovery probability and can be calculated using the binomial distribution (Table 2). A problem with this model is that it assumes artefact distribution to
be random within the site. In practice, artefacts are often found clustered, and calculations based on a random distribution of artefacts will not be very realistic (see e.g., Nance and Ball, 1986). Kintigh (1988) therefore recommended calculating artefact
Table 4 Discovery probabilities for the site of Verrebroek-Aven Ackers 2007. For different sampling distances and core diameters, discovery probabilities have been simulated using the flint counts per excavated square. For purposes of comparison, discovery probabilities were also obtained by simulation based on mean flint density, and calculated on the basis of the detection probability equation given in Section 2, as well as the k-parameter. Distance between sampling points
Auger diameter
Simulation based on counts per square
Simulation based on mean find density
Calculation based on Poisson distribution
Calculation based on k-parameter
25 m 20 m 15 m 10 m 5m
12 12 12 12 12
cm cm cm cm cm
0.232 0.350 0.622 0.920 1.000
0.310 0.466 0.740 0.959 1.000
0.310 0.466 0.739 0.959 1.000
0.214 0.334 0.526 0.847 1.000
25 m 20 m 15 m 10 m 5m
15 15 15 15 15
cm cm cm cm cm
0.275 0.415 0.724 0.970 1.000
0.379 0.553 0.852 0.989 1.000
0.377 0.553 0.852 0.990 1.000
0.214 0.390 0.608 0.905 1.000
244
P. Verhagen et al. / Journal of Archaeological Science 40 (2013) 240e247
1,0
A27-Hoge Vaart Eyserheide
0,9
Zutphen-Ooijerhoek Stroe
0,8
Geldrop-Aalsterhut
discovery probability
0,7
Sweikhuizen-Groene Paal Hempens
0,6
Oudenaarde-Donk Verrebroek-Aven Ackers 2006
0,5
Verrebroek-Aven Ackers 2007 Verrebroek-Dok
0,4
Hardinxveld-De Bruin
0,3
Keinsmerbrug
0,2 0,1 0,0 50 m
25 m
20 m
15 m
10 m
5m
sample point spacing Fig. 3. Overview of discovery probabilities with an auger diameter of 15 cm and different sample point spacings for all analyzed sites.
detection probabilities using the k-parameter (Pielou, 1977: pp. 124e134; Nance, 1983: pp. 300e310; McManamon, 1984: pp. 269e273). This is a measure for clustering that can be used on the assumption that the distribution of artefacts in a site follows the negative binomial distribution. However, the k-parameter can only be calculated on the basis of known artefact distributions, and therefore needs good empirical data in order to be applied to the problem of establishing appropriate survey strategies. Furthermore, it is not clear whether the negative binomial distribution is really a suitable model for the distribution of artefacts in a site. This is why we decided to use a different approach, and apply statistical simulations on the available data to obtain discovery probabilities independently of any theoretical assumptions of artefact distribution.
excavations in the Netherlands and northwestern Belgium (Flanders), (Table 3; Fig. 1). These were partly taken from EASY, the digital archiving system for archaeological data maintained by DANS (https://easy.dans.knaw.nl). Additional datasets were made available by the Cultural Heritage Agency (RCE), Ghent University and Groningen University. The data mainly concern buried sites and a few surface sites. Unfortunately, different excavation practices and soil conditions imply that sieving was done at different mesh sizes of 1, 2, 3 or 4 mm; furthermore, excavation square sizes were not always equal (ranging from 50 50 cm to 2 2 m), and not all sites had been completely excavated. All in all, this means that the sites cannot be compared very well; however, the data can be used to simulate discovery probabilities for the registered site sizes and lithic artefact densities.
3. Dataset
4. Simulations
The current study has aimed to collect datasets that allowed us to investigate the actual density of artefacts in Palaeolithic and Mesolithic sites. For this, it was necessary to analyze excavations where lithic artefacts were sieved and counted per square. Surprisingly, the number of suitable excavations available in digital form is still rather small. In total, we collected data from 12
In order to establish the discovery probabilities for each excavated site, simulations were set up using different core sampling grid spacings, covering the range from the recommended strategies in Tol et al. (2006) to a very dense spacing of 5 m. An equilateral triangular grid layout was applied throughout, as this is the most efficient grid configuration possible (Krakker et al., 1983). In
Table 5 Overview of analysis results. Find density and site size were determined on the basis of squares containing flints.
Eyserheide Geldrop-Aalsterhut Hardinxveld-De Bruin Hempens A27-Hoge Vaart Keinsmerbrug Oudenaarde-Donk Stroe Sweikhuizen-Groene Paal Verrebroek-Aven Ackers 2006 Verrebroek-Aven Ackers 2007 Verrebroek-Dok Zutphen-Ooijerhoek
Lithology
Mesh size in mm
Find density per m2
Estimated find density at 3 mm
k-parameter
Effect clustering (average)
Area in m2
Loess Sand Sand Sand Sand Clay Clay Sand Loess Sand Sand Sand Sand
4 2 4 3 10 ? 1 1 3 2 2 2 3
20.3 19.5 50.6 167.9 31.6 1.8 21.7 13.6 53.0 41.2 95.4 200.2 66.1
27.1 13.0 67.5 167.9 105.3 ? 7.2 4.5 53.0 27.5 63.6 133.5 66.1
0.25 0.12 0.78 0.50 0.29 0.39 0.41 0.35 0.06 0.64 0.51 0.17 0.34
3.43% 2.05% 3.64% 9.34% 5.64% 0.04% 2.90% 0.57% 12.35% 2.81% 7.40% 11.58% 10.80%
104.75 82.00 246.50 443.00 916.00 190.00 87.25 17.25 160.00 43.75 305.75 1428.50 246.75
P. Verhagen et al. / Journal of Archaeological Science 40 (2013) 240e247
practice, somewhat less efficient triangular layouts are used (like 40 50 m or 20 25 m), since a perfect equilateral triangular layout is often thought to be more difficult to set up in the field using measuring tape e although in practice this should not be a problem, certainly not with modern GPS equipment. The simulations were carried out through a Python-script. In each simulation, a grid was randomly placed on top of the site 100,000 times, and for each ‘virtual’ core sample hitting the site, detection probabilities were calculated on the basis of the counted flint fragments per square of 50 50 cm. The detection probability per square was established using the artefact detection equation based on the Poisson distribution; it is assumed that artefacts are distributed randomly within each square. The observation probability was set to 1. A core sample was then considered successful if a randomly generated number between 0.0 and 1.0 was smaller than the detection probability at that particular spot. A single positive core was considered to constitute a site discovery. The total number of ‘successful’ core samples was then used to calculate the discovery probability of the site analyzed. For purposes of comparison, separate calculations were performed using the kparameter, as well as the detection probability equation given in Section 2. 5. Results The simulations resulted in discovery probabilities for each analyzed site, and for the selected grid spacings. As an example, the case of the Mesolithic site of Verrebroek-Aven Ackers, consisting of several very small sand dunes each yielding evidence of Mesolithic occupation (Bats et al., 2004), is presented here; the remaining cases are discussed in Verhagen et al. (2011). The simulations were carried out on the data from the excavations conducted in 2007 (Sergant et al., 2007; Crombé et al., 2009), which cover almost an entire dune top, limited by local coordinates 399.5/320.5 and 419.5/291.5 (1286 50 50 cm squares, measuring 321.5 m2; Fig. 2; of these, 1223 squares were actually excavated). The other dunes were only excavated partially due to time limitations. Sediment was sieved at a 2 mm mesh size, and the average find density within the excavated zone is 95.4 per m2. The distribution of artefacts is left-skewed (Pearson’s skewness 3.21), and the kparameter for this site is 0.51, pointing to relatively strong clustering. Simulation results (Table 4) indicate that this site can be discovered with relative ease using core sampling, even when a 10 m spacing is necessary to achieve the required 75% discovery probability. The strategy recommended by Tol et al. (2006) for this type of site (20 25 m grid with 15 cm diameter core) will only result in a 41.5% chance of discovery. An overview of the simulation results for all analyzed sites is given in Fig. 3 and Table 5.
245
Table 6 Calculation of values of Df based on the data from Tol et al. (2004) and additional data from analyzed excavations. Site
Df
Size class boundary
Gennep site F Haelen Broekweg Ittersumerbroek West
1.21 1.03 0.92 1.00 0.39 2.43 0.58 1.10 1.73 0.70 1.96 0.45 1.16
4 3 4 3 2 2 4 2 4 3 4 2 4
Merselo Mienakker Site 0 Voetakker vindplaats 28 Zutphen-Ooijerhoek Schipluiden-Harnaschpolder A27-Hoge Vaart Eyserheide
mm mm mm mm mm mm mm mm mm mm mm mm mm
6. The effect of sieving It is difficult to directly compare the excavations analyzed and draw general conclusions about stone artefact densities in prehistoric sites because of the different sieving strategies applied, ranging from 1 to 4 mm mesh sizes in various soil types. Sieving is a major factor influencing the detection of flints. Tol et al. (2004: p. 46, Tables 10 and 11) show for a selected number of 7 cases that the fraction between 3 or 4 mm and 1 cm took up more than 50% of the total number of stone artefacts recovered. In 3 cases where smaller mesh sizes were applied, the fraction between 1 and 2 mm again took up more than 50% of total number of artefacts recovered. Brown (2001) suggested that the size distribution of flints in archaeological sites can be modelled using a fractal relation:
Nð>rÞ ¼ r Df where: N(>r) ¼ the number of fragments larger than r; r ¼ fragment size; Df ¼ the fractal dimension, or:
lnðNð>rÞÞ Df ¼ lnðrÞ The value of Df however can only be calculated if we know the actual fragment size distribution. For this purpose, Brown (2001: p.
Recommended A4 A5 A1 A6 A2 A3 strategy (Tol 50 m 15 cm 35 m 12 cm 25 m 15 cm 25 m 12 cm 20 m 15 cm 20 m 12 cm 15 m 15 cm 15 m 12 cm 10 m 15 cm 10 m 12 cm 5 m 15 cm 5 m 12 cm et al., 2006) e e A1 A1 A2 e e e e e A1 A1 A1
0.138
0.437
0.205
0.626
0.041 0.041 0.218 0.480 0.486 0.011 0.044 0.006 0.113 0.031 0.275 0.972 0.196
0.037 0.029 0.166 0.420 0.376 0.007 0.031 0.005 0.088 0.024 0.232 0.830 0.157
0.072 0.057 0.344 0.701 0.665 0.017 0.067 0.009 0.176 0.052 0.415 0.982 0.298
0.051 0.042 0.266 0.623 0.534 0.011 0.047 0.007 0.140 0.038 0.350 0.958 0.245
0.131 0.096 0.569 0.923 0.871 0.029 0.112 0.016 0.289 0.095 0.724 1.000 0.483
0.094 0.069 0.450 0.861 0.754 0.019 0.081 0.011 0.228 0.073 0.622 0.998 0.403
0.269 0.218 0.864 0.999
0.196 0.157 0.748 0.996
0.757 0.683 1.000 1.000
0.617 0.534 0.997 1.000
0.068 0.258 0.039 0.600 0.202 0.970
0.044 0.189 0.027 0.491 0.156 0.920
0.245 0.715 0.140 0.989 0.649 1.000
0.165 0.575 0.099 0.960 0.521 1.000
0.797
0.707
1.000
1.000
246
P. Verhagen et al. / Journal of Archaeological Science 40 (2013) 240e247
Table 7 Estimation of flint densities per m2 for different mesh sizes of the analyzed excavations, based on Df ¼ 1. In bold the originally registered flint densities per m2. Site
Mesh size
1 cm
4 mm
3 mm
2 mm
1 mm
Eyserheide Geldrop-Aalsterhut Hardinxveld-De Bruin Hempens A27-Hoge Vaart Keinsmerbrug Oudenaarde-Donk Stroe Sweikhuizen-Groene Paal Verrebroek-Aven Ackers 2006 Verrebroek-Aven Ackers 2007 Verrebroek-Dok Zutphen-Ooijerhoek
4 mm 2 mm 4 mm 3 mm 10 mm Unknown 1 mm 1 mm 3 mm
8.1 3.9 20.2 50.4 31.6 e 2.2 1.4 15.9
20.3 9.8 50.6 125.9 79.0 e 5.4 3.4 39.8
27.1 13.0 67.5 167.9 105.3 e 7.2 4.5 53.0
40.6 19.5 101.2 251.9 158.0 e 10.9 6.8 79.5
81.2 39.0 202.4 503.7 316.0 e 21.7 13.6 159.0
2 mm
8.2
20.6
27.5
41.2
82.4
2 mm
19.1
47.7
63.6
95.4
190.8
2 mm 3 mm
40.0 19.8
100.1 49.6
133.5 66.1
200.2 99.2
400.4 198.3
263) cites a number of (American) sources describing the size distribution of flint, quartz and obsidian fragments obtained from stone-working experiments. The calculated values of Df range from 0.38 to 3.33 (mean 2.10; standard deviation 0.62). Tol et al. (2004: pp. 45e46) provide some figures that can be used to calculate values for Df as well (Tables 6 and 7). The accuracy of these calculations is probably not very high, since the original data only differentiate between two size classes (smaller/larger than 2, 3 or 4 mm). The calculated values of Df (mean 1.11; standard deviation 0.60) are substantially lower than those mentioned by Brown. Data provided by Ghent University from 29 excavated artefact clusters from Verrebroek-Dok sieved over 2 mm give a similar outcome (mean 1.13; standard deviation 0.20; Crombé, 1998, 2005). A reason for this may be that the fine fraction in archaeological sites is vulnerable to erosion, and may have been blown or washed away, whereas the material from flint-working experiments will still contain this finer material. The figures obtained, while not extremely accurate, are nevertheless useful to make an educated guess of the effect of sieving on detection probabilities. If we take a (probably unrealistically) conservative estimate of Df ¼ 1, the registered flint densities can be recalculated to provide estimates for each mesh size (Table 6). Since Table 8 Recommended new survey strategies for Stone Age sites. Type Very small (<50 m2) Low find density (40e80 per m2) Very low find density (<40 per m2) Small (50e200 m2) Low find density (40e80 per m2) Very low find density (<40 per m2) Medium-sized: 200e1000 Medium-high find density (>80 per m2) Low find density (40e80 per m2) Very low find density (<40 per m2) Large: >1000 m2 Medium-high find density (>80 per m2) Low find density (40e80 per m2)
Lithology
Core sampling Auger Observation grid diameter technique
Irrelevant e
e
e
Irrelevant e
e
e
Irrelevant 4 5 m
15 cm
Irrelevant 4 5 m þ test pits m2 Irrelevant 13 15 m
e
3 mm sieving mesh e
12 cm
Irrelevant 8 10 m
15 cm
Irrelevant 4 5 m þ test pits
e
Irrelevant 20 25 m
12 cm
Irrelevant 13 15 m
12 cm
3 mm sieving mesh 3 mm sieving mesh e
3 mm sieving mesh 3 mm sieving mesh
the original guidelines written by Tol et al. (2006) specified that sieving should always be done at 3 mm, we can in this way obtain a baseline to compare our empirical data with the recommendations in the guidelines. From this we can conclude that a density of 80 artefacts per m2, which was thought necessary to guarantee an artefact detection probability of >75%, is not met by 9 of the 11 analyzed sites. 7. Conclusions The model presented by Tol et al. (2006) departs from the assumption of a random distribution of artefacts within an archaeological site. In practice, artefacts always display a certain amount of clustering. This is clearly illustrated when we calculate the k-parameter for the sites analyzed (mean 0.37, standard deviation 0.20). Clustering is always there, even when it can vary per site. The effect of clustering on the discovery probability of an archaeological site through core sampling was analyzed using simulation. This shows that clustering diminishes the probabilities of finding artefacts in a core. The calculation of discovery probabilities on the basis of the k-parameter shows similar results, but generally provides a lower estimate. However, the effect of clustering is relatively limited, and only becomes a major issue when high artefact densities are combined with a less intensive survey strategy. On the basis of the simulations it is unfortunately impossible to specify a clear relationship between clustering, artefact density, site size and core sampling strategy. The (conservative) estimate of the fractal dimensions of artefact size distributions lead us to conclude that a significant proportion of Palaeolithic and Mesolithic sites in the Netherlands and northwestern Belgium have a considerably lower artefact density (<80 per m2) than the cases considered in the survey guidelines (Tol et al., 2006). Using smaller mesh sizes for sieving will substantially increase the detection probabilities of these sites (Bats, 2007). Furthermore, it seems probable that a substantial number of sites are smaller than the dimensions specified in the guidelines (<200 m diameter). From the simulation analysis (Fig. 3, Table 5) it can be concluded that, using a 15 cm auger and 3 mm meshes, for the majority of sites a 10 m grid is sufficient to reach a 75% discovery probability. A coarser grid spacing will only yield this probability if the find density is high to very high, as is the case of Hempens, A27-Hoge Vaart, and Verrebroek-Dok. On the other hand, it seems that sites with very low (20e40 per m2) to extremely low (<20 per m2) artefact densities, which in addition are often smaller in size (e.g., Stroe and Geldrop-Aalsterhut), demand a more intensive core sampling strategy using a 5 m grid combined with test pits. The latter is unfortunately difficult to apply at larger depths. Another solution is to reduce the sieve meshes from 3 to 1 mm. Recent tests (Bats, 2007) have clearly demonstrated that smaller mesh sizes considerably increase the discovery rate of low-density sites. In any case, finding sites that combine a very small surface area and a very low artefact density will largely depend on chance, certainly if these are situated at great depth. This means that the guidelines will have to be revised to take this new information into account (Table 8). A further consideration is the question how we want to deal with these ‘difficult’ sites in archaeological heritage management. The effort needed to find the small and unobtrusive kind of sites is of course much larger than for other site types. Until now, Dutch archaeologists have taken the stance that survey efforts should be customized to provide the 75% probability of finding these. However, in cases when this would imply an enormous investment in survey, it is legitimate to ask whether the 75% discovery probability is the target that should always be aimed for. The results of the current study make this
P. Verhagen et al. / Journal of Archaeological Science 40 (2013) 240e247
question more urgent. If small prehistoric sites with low artefact densities are relatively common occurrences, it would imply a substantial increase in survey effort to meet the demands of current AHM policies. If this effort should be devoted is of course a policy issue, but it is advisable to give them some priority in future research, if only to establish how much they contribute to our knowledge of Stone Age archaeology. Furthermore, it should be questioned whether it is wise to adapt the core sampling strategy to the types of sites that are expected within a study region, Tol et al. (2006) propose. Is it possible to determine the occurrence of these types of sites a priori (in terms of size, density of artefacts) on the basis of desk-based research, i.e. before fieldwork has started? How can specific categories of sites that are not expected ever be discovered? And how can we be sure about our predictions? One solution to this problem is to apply a common core sampling and sieving strategy for all regions; based on our findings a core-sampling grid of 10 m seems the best option. Alternatively, we may choose to apply intensive core sampling (4 5 m grid) and/or test pits in targeted locations with a high expectation for the presence of Stone Age sites, irrespective of their (expected) size or artefact density. Acknowledgements The research done for this paper was made possible by the Rijksdienst voor het Cultureel Erfgoed (RCE), as part of the programme “What is Heritage?”. The authors would like to thank the colleagues who provided datasets for the analysis e in particular Jos Deeben (RCE) and Gary Nobles (Groningen University) e as well as the project advisory committee: Jan-Willem de Kort, Dr. Bjørn Smit, José Schreurs, Jos Deeben (all RCE), Dr. Hans Peeters, Marcel Niekus (both Groningen University), Willem-Jan Hogestijn (Municipality of Almere), Antoine Wilbers (IDDS Archeologie), Jaap Beuker (Province of Drenthe), Dr. Jos de Moor (EARTH Integrated Archaeology), Axel Müller (ADC Archeoprojecten), Johan Jelsma (De Steekproef), and Esther Wieringa (SIKB). References Banning, E.B., 2002. Archaeological Survey. Kluwer Academic/Plenum Publishers, New York (Manuals in Archaeological Method, Theory, and Technique). Bats, M., 2007. The Flemish Wetlands: an archaeological survey of the valley of the river Scheldt. In: Barber, J., Clark, C., Cressey, M., Crone, A., Hale, A., Henderson, J., Housley, R., Sands, R., Sheridan, A. (Eds.), Archaeology from the Wetlands: Recent Perspectives. Proceedings of the 11th WARP Conference, Edinburgh 2005. Society of Antiquaries of Scotland, Edinburgh, pp. 93e100 (WARP Occasional Paper, 18). Bats, M., Klinck, B., Meersschaert, L., Sergant, J., 2004. Verkennend en waarderend booronderzoek in het alluvium van de Schelde. Notae Praehistoricae 24,175e179. Bats, M., in preparation. De Vlaamse Wetlands, een archeologische verkenning van de Scheldevallei. PhD-thesis Ghent University.
247
Borsboom, A.J., Verhagen, J.W.H.P., 2009. KNA Leidraad Inventariserend Veldonderzoek. Deel: Proefsleuvenonderzoek (IVO-P). SIKB, Gouda. Brown, C.T., 2001. The fractal dimensions of lithic reduction. Journal of Archaeological Science 28, 619e631. Crombé, P., 1998. The Mesolithic in Northwestern Belgium, Recent Excavations and Surveys. Archaeopress, Oxford (British Archaeological Reports, International Series 716). Crombé, P. (Ed.), 2005. The Last Hunter-gatherer-fishermen in Sandy Flanders (NW Belgium); the Verrebroek and Doel Excavation Projects. Part 1: Palaeoenvironment, Chronology and Features. Academia Press, Ghent (Archaeological Reports Ghent University 3). Crombé, P., Sergant, J., Lombaert, L., Van Strydonck, M., Boudin, M., 2009. The Mesolithic and Neolithic site of Verrebroek e Aven Ackers (East Flanders, Belgium): the radiocarbon evidence. Notae Praehistoricae 29, 15e21. Drew, L.J., 1967. Grid-drilling exploration and its application to the search for petroleum. Geology 62, 698e710. Drew, L.J., 1979. Pattern drilling exploration: optimum pattern types and hole spaces when searching for elliptical shaped targets. Mathematical Geology 11, 223e254. Kintigh, K.W., 1988. The effectiveness of subsurface testing: a simulation approach. American Antiquity 53, 686e707. Koopman, B.O., 1980. Search and Screening: General Principles with Historical Applications. Pergamon Press. Krakker, J.J., Shott, M.J., Welch, P.D., 1983. Design and evaluation of shovel-test sampling in regional archaeological survey. Journal of Field Archaeology 10, 469e480. McManamon, F.P., 1984. Discovering sites unseen. In: Schiffer, M.B. (Ed.), 1984. Advances in Archaeological Method and Theory, vol. 7. Academic Press, New York, pp. 223e292. Nance, J.D., 1983. Regional sampling in archaeological survey: the statistical perspective. In: Schiffer, M.B. (Ed.), 1983. Advances in Archaeological Method and Theory, vol. 6. Academic Press, New York, pp. 289e356. Nance, J.D., Ball, F.B., 1986. No surprises? The reliability and validity of test pit sampling. American Antiquity 51, 457e483. Pielou, E.C., 1977. Mathematical Ecology. Wiley, New York. Sergant, J., Bats, M., Noens, G., Lombaert, L., D’Hollander, D., 2007. Voorlopige resultaten van noodopgravingen in het afgedekte dekzandlandschap van Verrebroek Aven Ackers (Mesolithicum, Neolithicum). Notae Praehistoricae 27, 101e107. SIKB, 2010. KNA Landbodems Versie 3.2. SIKB, Gouda. Singer, D.A., 1975. Relative efficiencies of square and triangular grids in the search for elliptically shaped resource targets. Journal of Research of the U.S. Geological Survey 3, 163e167. Stone, G.D., 1981. On artifact densities and shovel probes. Current Anthropology 22, 182e183. Tol, A.J., Verhagen, J.W.H.P., Borsboom, A.J., Verbruggen, M., 2004. Prospectief boren. Een studie naar de betrouwbaarheid en toepasbaarheid van booronderzoek in de prospectiearcheologie. RAAP Archeologisch Adviesbureau, Amsterdam (RAAP-rapport 1000). Tol, A.J., Verhagen, J.W.H.P., Verbruggen, M., 2006. Leidraad inventariserend veldonderzoek; Deel: karterend booronderzoek. SIKB, Gouda. Verhagen, P., 2005. Prospection strategies and archaeological predictive modelling. In: Leusen, M. van, Kamermans, H. (Eds.), Predictive Modelling for Archaeological Heritage Management: a Research Agenda. Rijksdienst voor het Oudheidkundig Bodemonderzoek, Amersfoort, pp. 109e121 (Nederlandse Archeologische Rapporten 29). Verhagen, P., Borsboom, A., 2009. The design of effective and efficient trial trenching strategies for discovering archaeological sites. Journal of Archaeological Science 36, 1807e1816. Verhagen, P., Rensink, E., Bats, M., Crombé, P., 2011. Optimale strategieën voor het opsporen van Steentijdvindplaatsen met behulp van booronderzoek. Een statistisch perspectief. Rijksdienst voor het Cultureel Erfgoed, Amersfoort (Rapportages Archeologische Monumentenzorg 197).