Estimating grain weight in archaeological cereal crops: a quantitative approach for comparison with current conditions

Journal of Archaeological Science 31 (2004) 1635e1642 http://www.elsevier.com/locate/jas Estimating grain weight in archaeological cereal crops: a quantitative approach for comparison with current conditions Juan P. Ferrioa, Natàlia Alonsob, Jordi Voltasa, José Luis Arausc,) a Departament de Producció Vegetal i Cie`ncia Forestal, Universitat de Lleida, Rovira Roure 191, E-25198, Lleida, Spain b Departament d’Història, Universitat de Lleida, Victor Siurana 1, E-25003, Lleida, Spain c Unitat de Fisiologia Vegetal, Facultat de Biologia, Universitat de Barcelona, Diagonal 645, E-08028, Barcelona, Spain Received 4 August 2003; received in revised form 2 April 2004 Abstract Data relating to grain weight of cereal crops cultivated in the past could be useful to track early genetic and/or agronomic improvements, facilitating the comparison of archaeological data with current agronomic studies. However, archaeological grains are usually preserved by a process of carbonization, during which there is a considerable reduction in mass and dimension changes. The aim of this study was to develop a model for the estimation of original grain weight from the dimensions of charred grains, taking into account the effect of carbonization. For this purpose, extant grains of wheat and barley were experimentally carbonized at three temperature levels (200, 250, 300 (C) and under two atmospheric conditions (oxidant, reducing). Weight, length (L), breadth (B) and thickness (T) were measured before and after carbonization. The products L!B and L!T provided the best estimation of grain weight (r2 ¼ 0:80e0:86) and were relatively stable across treatments. As a case study, we applied grain weight models to wheat and barley grains gathered from a range of archaeological sites in the Segre and Cinca Valley (Catalonia, NE Spain, ca. 3900e2200 cal. BP). In this region, we observed a relatively constant increment in grain weight during the first half of I millennium BCE, probably explained by the increased water availability indicated by palaeoenvironmental studies. In contrast, the increased grain weight of current samples, compared with that of archaeological grains, is largely attributable to more recent genetic improvements. Ó 2004 Elsevier Ltd. All rights reserved. Keywords: Agriculture; Charred seeds; Dimensions; Morphology; Carbonization; Wheat; Barley 1. Introduction Grain weight is one of the three main agronomic components of grain yield in cereals, along with the number of spikes per unit grown area and the number of grains per spike; moreover, it has direct implications for grain quality [16,24]. Therefore, data relating to grain weight of cereal crops cultivated in the past would be of interest as a way to track genetic improvement in this trait, but also as an indicator of the potential quality of the food products that could be delivered from them. ) Corresponding author. E-mail address: josel@bio.ub.es (J.L. Araus). 0305-4403/$ - see front matter Ó 2004 Elsevier Ltd. All rights reserved. doi:10.1016/j.jas.2004.04.006 Although strongly genetically determined, grain weight also depends on environmental constraints, such as water availability or temperature stress [26,35]. For example, when crops are initially grown under favourable conditions, but become stressed during grain filling (which is a common occurrence in rain-fed Mediterranean environments), the time span for grain growth is reduced and grain weight is limited [17,29]. Thus, estimates of grain weight in ancient cereal crops could also provide helpful information about the agronomic conditions under which the plants were grown. From an archaeological perspective, current excavation techniques often involve the systematic recovery of carbonized plant remains, including grains from cereals and other crops, and grain dimensions are routinely 1636 J.P. Ferrio et al. / Journal of Archaeological Science 31 (2004) 1635e1642 measured. However, grain proportions change considerably after charring and, consequently, are not directly comparable with those of intact grains [21,23]. Similarly, although archaeological grains can be weighed, their weight is not directly comparable with the original weight, as there can be considerable loss during carbonization [31]. Moreover, grains are often found partially broken and/or with mineral inclusions, still allowing reliable measurements of principal dimensions, but not of the total weight. In contrast, current agronomic studies are conducted with non-carbonized grains and whereas grain dimensions are rarely considered, grain weight is measured routinely [2,12,20,29]. Consequently, being able to estimate grain weight in archaeological grains would allow us to compare them with the large body of modern agronomic data currently available. Hence, the aim of this work was to find a way to link archaeobotanic and agronomic studies, as the comparison between such extensive sources of data might help to answer some unsolved questions. Our specific objective was to develop a model to estimate original grain weight from charred grain dimensions that would be applicable to archaeological cereal seeds. We analyzed the effect of carbonization on principal grain dimensions (length, breadth and thickness), morphological indices and grain weight. A wide range of carbonization levels, from slightly toasted to fully carbonized or nearly combusted, under either aerobic or anaerobic conditions, was assayed. By including such a range of charring conditions in model calibration, we expected to cover the variability found in archaeological assemblages. A case study on archaeological seeds recovered from several sites in the Segre and Cinca Valley (Catalonia, NE Spain) is presented to illustrate the application of these models in an archaeological context. 2. Experimental procedure 2.1. Plant material and charring conditions The relationship between charred grain dimensions and grain weight was modelled using extant grains of barley (Hordeum vulgare L.) and wheat (Triticum aestivum L.), harvested in the experimental fields of the UdL-IRTA Centre in Lleida (Catalonia, NE Spain). We took 144 grains for each species, covering a wide range of grain weight, dimension and morphology (Table 1). We assigned grains to six separate subsets, selected to include a similar range of grain weight and dimensions. We ensured that there were no significant differences in grain weight and dimensions among the subsets. All samples were dried at 60 (C for 24 h, and submitted to a pre-heating ramp of 2 h to minimize grain explosion Table 1 Mean and range for grain weight, dimensions and morphological indices before and after experimental carbonization Grain Intact grains Charred grains Variable Range Mean Range Mean Wheat Grain weight (mg) Length (L) (mm) Breadth (B) (mm) Thickness (T) (mm) L/B L/T B/T N ¼ 144 7.3e61.4 4.1e7.4 1.5e3.6 1.4e3.4 1.8e3.7 1.9e4.5 0.7e1.7 29.5 5.7 2.6 2.4 2.2 2.4 1.1 N ¼ 95 3.5e55.0 3.4e7.1 1.9e5.4 1.6e4.8 0.9e2.8 1.0e3.4 1.0e1.5 20.9 5.1 3.3 2.7 1.5 1.8 1.2 Barley Grain weight (mg) Length (L) (mm) Breadth (B) (mm) Thickness (T) (mm) L/B L/T B/T N ¼ 144 3.4e56.5 3.8e7.7 1.6e3.7 0.7e3.0 1.7e3.0 2.2e7.4 1.2e2.5 28.0 6.3 2.9 2.1 2.2 3.1 1.4 N ¼ 92 3.4e48.5 3.8e7.9 1.9e4.7 1.1e4.1 1.2e2.7 1.6e4.8 1.0e2.1 20.0 6.1 3.3 2.6 1.9 2.5 1.3 The data shown here correspond to global values, i.e. across the six treatments assayed. N, number of samples. (i.e. ‘pop-corn’ effect) due to violent output of water and combustible gases. The subsets were carbonized under three levels of maximum temperature (200, 250, 300 (C) and two atmospheric conditions (oxidant, reducing). Samples were kept at these temperatures for 45 min. To obtain a reducing atmosphere, three of the subsets were buried in high crucibles under 3 cm of sand, whereas the other three (oxidant) were placed in flat crucibles without burying. 2.2. Grain measurements Weight and dimensions were determined before and after the experimental carbonization of extant grains. We measured length (L), breadth (B) and thickness (T) under a stereoscopic microscope with an LCD camera coupled to a PC, assisted by image analysis software (Leika Qwin Lite 2.1, 1997). After carbonization, some grains were too distorted and/or broken to be measured and were not included in the analysis. 2.3. Statistical analyses Data were subjected to analysis of variance (ANOVA) to determine the effect of carbonization treatments (temperature and atmospheric conditions). Linear regressions were calculated to assess the relationships between original grain weight and dimension variables. Unless otherwise stated, differences were considered statistically significant when P ! 0:05. All analyses were carried out using standard SAS-STAT procedures [28]. J.P. Ferrio et al. / Journal of Archaeological Science 31 (2004) 1635e1642 Model fitness was assessed using root mean square error (RMSE): sP ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 ðYref
Ypred Þ RMSE ¼ ðn
pÞ where Yref and Ypred are observed and predicted values for each sample, n stands for the number of samples and p is the number of estimated parameters (p ¼ 2). From this value we could obtain the standard error (SEp) for each predicted value, which is a function of all errors associated with the regression equation [10]: sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 ðX
Xm Þ 1 P 2 C SEp ¼ RMSE n X where RMSE and n are as described above, X is the value of the independent variable for a given sample and Xm stands for the mean value of X over all samples. Grain weight of archaeological seeds was estimated from the products length!breadth (L!B) and length!thickness (L!T), taking the mean value of both estimates. RMSE could not be calculated for archaeological samples, as Yref (i.e. real) values were unknown. Instead, we took root mean square difference (RMSD) between the values estimated from the two variables (L!B and L!T) as a rough indicator of the prediction error in archaeological samples: sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi P 2 ðYLB
YLT Þ RMSD ¼ n where YLB and YLT stand for the predicted values using either the L!B or L!T model, and n is the number of samples. RMSD was also calculated within calibration samples as a reference value for archaeological results. The overall standard error for a given archaeological sample (SEsample) and site (SEsite) was calculated following the general rules of error propagation [22]: ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi vffiq u ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi u 2 2 t ðSEpLB Þ CðSEpLT Þ CðSELB;LT Þ2 SEsample ¼ 2 SEsite ¼ qffiP ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðSEsample Þ2 N where SEpLB and SEpLT are the SEp associated with the estimation of grain weight from L!B and L!T, respectively, SELB;LT represents the standard error of the mean between the two estimates (from L!B and L!T) and N stands for the number of samples within each site. 1637 3. Results and discussion 3.1. Changes in grain size and morphology during carbonization Overall, both grain weight and grain dimensions changed considerably after charring, thus confirming that they are not directly comparable with those of intact seeds (Table 1). We observed an average depletion in grain weight of about 30%, and the grains became more rounded, especially in wheat. There was a general decrease in L (11% in wheat, 5% in barley), whereas B and T increased (for B: 27% in wheat, 14% in barley; for T: 13% in wheat and 23% in barley). Similar changes in grain dimensions during carbonization have been reported previously for several species of wheat and barley, under a variety of carbonization conditions [13,21,23,25,36]. In all cases, charred grains tended to shrink in length, whilst increasing in breadth and thickness. In spite of these general trends, the quantitative effect of charring varied considerably according to the species and experimental conditions. This is partly confirmed by our work, as grain weight and dimensions after charring differed significantly between temperatures (Table 2). However, we did not find any significant effect of atmospheric conditions on the variables studied. Comparing the two species, we observed that the degree of distortion was somewhat greater in wheat than in barley. The greater density (i.e. tighter packing) of wheat (75e80 kg/hl) than barley grains (60e65 kg/hl) probably enhanced the distortion during charring. Nevertheless, despite showing greater deformations, they were less dependent on the carbonization environment for wheat than for barley (Table 2). In wheat, L was the most variable dimension, whereas for barley T was the least consistent. As our aim was to develop a model sufficiently robust to be applied over a wide range of seed morphologies, the best variables would be those that are relatively stable regardless of carbonization conditions. The products L!B and L!T were the only two variables not showing significant differences among treatments in both species. Apparently, the increment in either B or T brought about by higher temperatures was compensated by the reduction in L, thus giving a more stable variable than each dimension taken separately. In addition, L!B and L!T showed the strongest relationships with grain weight across all treatments (Table 2). Consequently, the variables of choice for model calibration were L!B and L!T. As expected, although the weight of charred grains showed a relatively good correlation with original grain weight across treatments, it was highly dependent on the charring conditions. Indeed, grain weight showed little variation when carbonization was performed at 200 (C (the lowest temperature assayed), but was reduced 1638 J.P. Ferrio et al. / Journal of Archaeological Science 31 (2004) 1635e1642 Table 2 Effect of carbonization conditions on weight and grain dimensions, and determination coefficients (r2) of the linear relationship (across treatments) between initial grain weight and different variables measured after carbonization Treatment Measurable samples (%) Intact grains Wheat Barley Charred grains Wheat Temperature Atmosphere Temp.!Atm. Mean values 200 (C 250 (C 300 (C r2 (across treatments) Barley Temperature Atmosphere Temp.!Atm. Mean values 200 (C 250 (C 300 (C r2 (across treatments) 100 69 29 100 73 19 Grain weight (mg) L (mm) B (mm) T (mm) L!B!T (mm3) B!T (mm2) L!B (mm2) L!T (mm2) 29.5 28.0 5.7 6.3 2.6 2.9 2.4 2.1 36 38 06.2 06.1 14.8 18.3 13.7 13.2 *** n.s. n.s. *** n.s. n.s. n.s. n.s. n.s. * n.s. n.s. n.s. n.s. n.s. * n.s. n.s. n.s. n.s. n.s. n.s. n.s. n.s. 27.7a 15.4ab 10.5ab 0.67 5.5a 4.6ab 4.6ab 0.43 3.3a 3.4a 3.5a 0.50 2.6a 2.9ab 2.9ab 0.49 48a 47a 50a 0.77 08.6a 10.0ab 10.5ab 0.52 18.2a 16.0a 16.0a 0.80 14.6a 13.4a 13.7a 0.82 *** n.s. n.s. * n.s. n.s. ** n.s. n.s. *** n.s. n.s. ** n.s. n.s. *** n.s. n.s. n.s. n.s. n.s. n.s. n.s. n.s. 25.3a 14.8ab 12.2ab 0.69 6.3a 5.8ab 6.3ab 0.64 3.2a 3.4a 3.8ab 0.67 2.4a 2.8ab 3.3abc 0.61 51a 56ab 85ab 0.77 07.8a 09.5ab 12.8abc 0.65 20.5a 19.8a 24.7a 0.84 15.2a 15.9a 21.0a 0.82 Mean values for intact grains (i.e. before carbonization) are included for reference (N ¼ 144). The significance of the effect of temperature, atmosphere and their interaction is indicated for each variable (*P ! 0:05, **P ! 0:01, ***P ! 0:001). Mean values with the same letter are nonsignificantly different according to the LSD test. Linear regressions are all significant at P ! 0:001. Measurable samples (%), percentage of the original samples still measurable after carbonization; L, length; B, breadth; T, thickness. drastically (about 50%) at temperatures above 250 (C (Table 2). This, added to the aforementioned considerations (grain fragmentation, presence of mineral inclusions and other taphonomic alterations), confirms that this variable cannot be used as an estimator of original weight in archaeological cereal seeds. 3.2. Sample preservation: does initial grain weight affect sample survival during charring? As the degree of carbonization increased, a greater number of samples were either destroyed or too distorted to be measured (Table 2). In order to account for potential drifts due to sample loss, we checked whether or not the ‘surviving’ seeds differed in their original grain weight when compared with the rest of the samples. There were no significant differences (data not shown) and, consequently, no evidence of a differential preservation of seeds related to their initial grain weight. Moreover, we did not find any significant relationship between original grain weight and the changes in grain dimensions during carbonization. This suggests that the grain size of a carbonized cereal sample is likely to be representative of the original sample, provided that there is no further drift during burial and recovery. Wilson [37] and, more recently, Wright [38] assayed the effect of different carbonization environments on the relative survivability of weed seeds, both concluding that it was strongly dependent on the heating conditions. In a series of field experiments, including experimental fires, sampling and flotation, Gustafsson [18] also found that weed seeds showed contrasting levels of carbonization, even under the same conditions, and suggested that this could be related to substantial differences in seed composition (e.g. oil content). However, in the same experiment, grains from different cereal species were carbonized and recovered in similar proportions, probably due to their relatively uniform chemical composition. Considering the great morphological variability found among cereal species, the work of Gustafsson [18] supports the absence of any differential preservation of cereal grains according to their size and/or morphology. Nevertheless, random drifts should not be discounted, although they can be minimized by increasing sample size, as well as including samples from several carbonization events [9,11,18]. 3.3. Grain weight models: performance and range of application Once the most suitable variables were selected, we used regression models to predict grain weight, pooling samples from all treatments together, with the aim of covering the widest range of experimental situations. 1639 J.P. Ferrio et al. / Journal of Archaeological Science 31 (2004) 1635e1642 In both species, L!B and L!T were strongly related to grain size (r2 ¼ 0:80e0:86), although they showed different trends in wheat and barley (Fig. 1). In wheat, both L!B and L!T were linearly related to grain size. However, in barley the relationship between these variables and grain size was improved using a power model (Fig. 1c,d). This could be explained by the lower morphological distortion of barley grains, as the relationship between these variables and weight in intact grains of both species fitted power curves (data not shown). RMSE values of 5.1e5.4 mg (Fig. 1) suggest a good predictive accuracy of the models. Nevertheless, taking the mean between the values estimated from L!B and L!T gave the most reliable estimations of grain weight, and RMSE was lower (4.9 and 5.0 mg for wheat and barley, respectively) than for each of them taken separately. Within the whole calibration range, SEp was always less than 2 mg, and it was less than 1 mg in the most common ranges of 9e44 and 10e46 mg of predicted values for wheat and barley, respectively. It should be noted that the range of dimensions and morphology included in calibration was similar to that usually reported in archaeological grains for both species, which can be 3.0e7.5 mm in L, 1.7e4.6 mm in B and 1.5e4.0 mm in T [3e5,11,14,23,36]. Moreover, mean site values reported in the literature would always be included within the aforementioned optimal range (SEp ! 1 mg). This suggests that current models can be 60 Grain weight (mg) a) Wheat b) Wheat 50 40 30 20 Y=-15.1+2.98x N=95 r2=0.82*** RMSE=5.1mg Y=-15.4+2.47x N=95 r2=0.80*** RMSE=5.3mg 10 d) Barley Grain weight (mg) 50 To further assess the suitability of the models, we applied them to a set of archaeological samples. Charred grains of naked wheat (Triticum aestivum/durum) and hulled barley (Hordeum vulgare) were gathered from six archaeological sites in the Segre and Cinca Valley (Catalonia, NE Spain, see Table 3), ranging from Bronze Age (ca. 3900 cal. BP) to Second Iron Age (Late Iberian period: ca. 2200 cal. BP). Currently, all these sites show similar climatic and soil conditions, and are characterized by a semiarid Mediterranean climate. The samples were gathered from various archaeological contexts, such as domestic fires, cooking ovens, storage jars, room floors and levels of rubble from housing structures and pits (for further details about the sites and sampling procedures, see Refs. [3e5]). We determined grain dimensions in these samples as described in Section 2.2. Seeds too distorted and/or broken were not measured. Firstly, we confirmed that the range of values within the set of archaeological samples was suitable for the application of the models. Thus, 96% of the samples fell within the calibration range for grain dimensions and morphological indices described in Table 1, with about 95% of the samples falling within the optimal range (SEp ! 1 mg) of the models. Removing the samples that fell outside of these limits did not significantly affect the results, so we decided to keep them in the analysis. Moreover, calculated RMSD for archaeological samples (3.9 and 3.4 mg, for wheat and barley, respectively) was similar to that found within calibration samples (3.6 mg Site 40 30 Minferri 20 Y=0.78x1.28 N=92 r2=0.85*** RMSE=5.4mg Y=0.21x1.60 N=92 r2=0.86*** RMSE=5.1mg 10 0 4. A case study with archaeological samples Table 3 Site description and number of archaeological grains used in the case study 0 c) Barley applied to most archaeological cereal seeds, provided that they are sufficiently well preserved to allow reliable measurements. 5 10 15 20 25 30 35 L*B (mm2) 0 5 10 15 20 25 30 35 L*T (mm2) Fig. 1. Regression models across treatments to estimate initial grain weight of wheat (a,b) and barley (c,d) from the products length! breadth (L!B; a,c) and length!thickness (L!T; b,d) of carbonized grains. Circles, triangles and squares indicate temperature levels of 200, 250 and 300 (C, respectively. Open and closed symbols indicate reduction and oxidation treatments, respectively. All regressions were significant at P ! 0:001. Lat. Long. Cultural period 41(30# 0(45# Bronze Age El Vilot 41(38# 0(30# Bronze III Age El Vilot 41(38# 0(30# Bronze II Age Els 41(34# 0(56# First Vilars II Iron Age Margalef 41(34# 0(47# Second Iron Age Tossal 41(37# 0(47# Second Tenalles Iron Age Calibrated Wheat Barley age (BP) samples samples 4100e3650 118 29 3113e2794 13 17 2923e2393 14 70 2550e2425 92 170 2250e2150 100 100 2250e2150 100 100 Calibrated age represents the approximate age in years BP estimated from a combination of archaeological and calibrated 14C dating using the program CALIB 3.03 [30]. 1640 J.P. Ferrio et al. / Journal of Archaeological Science 31 (2004) 1635e1642 for both species), suggesting that the accuracy of the estimates in archaeological seeds was close to those reported for current samples (Section 3.3). Fig. 2 plots average grain weight for the six archaeological sites, taking the mean between the estimates from L!B and L!T models, as described in Section 2.3. We also included a reference value for present-day samples of wheat [2,8] and barley [32,33] grown in rainfed conditions in the same area. We observed a large increase in grain weight of present-day samples as compared with that of archaeological grains. Indeed, even under the harsher conditions in which barley and wheat can grow, values of grain weight are usually over 30 mg, and rarely fall below 25 mg [1,6,17]. Such results contrast with the estimated grain size among the archaeological sites studied, which were always below 25 mg. Moreover, when we applied our models to other archaeological data acquired from the literature [11,23], estimated grain weight was never greater than 30 mg. Data from geomorphology [19], antracology [5,27] and palynology [5] indicate that the current climate in the area is more arid than in the period studied. In addition, the analysis of carbon isotope composition of barley and wheat grains recovered from other archaeological sites in Catalonia also suggested that they were grown under much wetter conditions between the Neolithic period and the Middle Ages than currently [7]. Thus, such a large increase in grain weight from archaeological to present times is mostly attributable to more recent advances in plant breeding and agronomic practices, including the extensive application of nitrogen fertilizers 40 a) Wheat b) Barley Grain weight (mg) 35 30 25 20 15 10 4.0 3.5 3.0 2.5 2.0 Ky BP 0.0 4.0 3.5 3.0 2.5 2.0 0.0 Ky BP Fig. 2. Evolution of grain weight of barley and wheat in the Segre and Cinca Valley (Catalonia, NE Spain) from Bronze Age (ca. 3900 cal. BP) to Second Iron Age (ca. 2200 cal. BP), compared with present-day samples (see Table 3 for a description of the archaeological sites). Archaeological values were estimated from the products length! breadth (L!B) and length!thickness (L!T), taking the mean value of both estimates. Present-day data correspond to barley [32,33] and wheat [2,8] grown under rain-fed conditions around the studied area. Horizontal error bars indicate the range of calibrated age. Vertical error bars of the archaeological samples indicate the standard error of predicted means, including within-site variability, regression error and the difference between predicted values using the two models (see text for details). For present-day data, vertical error bars indicate the standard error of the mean across environments (N ¼ 6). [1,17,26,34]. The differences observed among archaeological samples, however, appear to be related to environmental changes in the studied area. At the second half of the II millennium BCE, vegetation was characterized by evergreen, schlerophyllous species, mostly forming associations of Rhamno-Quercetum cocciferae and Rosmarino-Ericion, typically Mediterranean [5,27]. During the first half of the I millennium BCE, however, there was a progressive change in vegetation, with the formation of meso/submediterranean forests, composed of a mixture of oaks (evergreen and deciduous) and pines, together with shrub woodlands in the driest places [5,27]. This is in agreement with the existence of a period of wetter climate, according to geomorphological studies [19]. Consequently, the increase in grain weight observed from 3000 to 2000 BP, particularly apparent in barley, may be due to changes in water availability. Nevertheless, we cannot rule out the possibility of genetic improvement in grain size (possibly unintentional in agronomic terms) during the period studied, as increased grain size implies better end products. Indeed, grain weight is directly associated with flour yield [20] and baking properties of wheat [16], and also determines the malting quality of barley [24]. Moreover, the differences between archaeological and current data were not limited to modern cultivars, but also applicable to traditional landraces [1,6,12], supporting the idea that genetic improvement in grain size had already started before the implementation of modern breeding programs. Indeed, grain size is an easily identifiable trait, and, for example, it has been documented that Spanish farmers have traditionally reserved the best seeds (i.e. the biggest) for sowing [12]: it is probable that such practice began during the early phases of agriculture. Nevertheless, we should bear in mind that carbonized plant remains are not necessarily a random sub-sample of the overall harvest. Indeed, carbonized grains might have been those discarded by the farmers, and thrown onto fires along with other plant debris [9,15]. Exceptionally, for two of the sites studied (Margalef and Tossal Tenalles), there was evidence that the samples analyzed were representative of harvested grains, as thousands of them were found within storage jars. For the remaining sites, however, we are unaware of the origin of the grains and, although a large number of samples (and carbonization events) would reduce any biasing, these results should be interpreted with caution. Nevertheless, the difference in grain weight between stored grains and those recovered from other sources was much smaller than that between archaeological and present-day grains. Therefore, even considering only stored material (i.e. undoubtedly representative of harvested grains), we can conclude that a significant genetic improvement in grain weight occurred in the area, at least during the last two millennia. J.P. Ferrio et al. / Journal of Archaeological Science 31 (2004) 1635e1642 5. Conclusions We have confirmed that although there is considerable change in dimensions and morphology of wheat and barley grains during carbonization, some measurements are relatively constant over a wide range of carbonization levels. The products L!B and L!T were the best correlated with grain weight, and also the most stable across carbonization treatments. The range of dimensions and morphology present in the sample set used for calibrations suggested that it was representative of the variety of (unknown) conditions in which archaeological grains were carbonized. Indeed, this range was wide enough to allow accurate predictions in about 95% of the archaeological seeds in which the models were tested. In our case study, we concluded that the increase in grain weight between archaeological samples and present-day crops was principally due to genetic improvements. Differences among archaeological samples, however, were apparently related to environmental change, although genetic differences could not be dismissed. Further work will be necessary to clarify the origin of such variations, as well as to determine the timing of the change in grain weight to its current values. Acknowledgements This work was partly supported by the CICYT grant BTE2001-3421-C02-01 and the INCO-MED project MENMED (ICA3-CT-2002-10022). J.P. Ferrio has a PhD Fellowship from Generalitat de Catalunya. We thank the extensive crops department from UdL-IRTA for providing extant seeds. We also acknowledge the useful comments of the referees, which helped us to significantly improve the original manuscript. References [1] E. Acevedo, P.Q. Craufurd, R.B. Austin, P. Pérez-Marco, Traits associated with high yield in barley in low-rainfall environments, Journal of Agricultural Science 116 (1991) 23e36. [2] A.E.T.C., Encuesta de Calidad de los Trigos Españoles: Cosecha 2002, Ministerio de Agricultura, Pesca y Alimentación, Madrid, 2003. [3] N. Alonso, Conreus i Agricultura a la Plana Occidental Catalana en Època Ibèrica. Estudi Arqueobotànic de Margalef (Torregrossa, el Pla d’Urgell) i Tossal de les Tenalles (Sidamon, el Pla d’Urgell), Agrupació Cultural La Femosa, Lleida, 1992. [4] N. Alonso, De la Llavor a la Farina: els Processos Agrı́coles Protohistòrics a la Catalunya Occidental, Monographies d’Archéologie Méditerranéenne, 4, CNRS, Lattes, 1999. [5] N. Alonso, M. Gené, E. Junyent, A. Lafuente, J.B. López, A. Moya, E. Tartera, Recuperant el Passat a la Lı́nia del Tren d’Alta Velocitat. L’Assentament Protohistòric, Medieval i d’Època Moderna de El Vilot de Montagut (Alcarràs, Lleida), GIF, Generalitat de Catalunya, Lleida, 2002. 1641 [6] J.L. Araus, T. Amaro, J. Casadesus, A. Asbati, M.M. Nachit, Relationships between ash content, carbon isotope discrimination and yield in durum wheat, Australian Journal of Plant Physiology 25 (1998) 835e842. [7] J.L. Araus, R. Buxó, Changes in carbon isotope discrimination in grain cereals from the north-western Mediterranean basin during the past seven millennia, Australian Journal of Plant Physiology 20 (1993) 117e128. [8] J.L. Araus, D. Villegas, N. Aparicio, L.F. Garcı́a-del-Moral, S. Elhani, Y. Rharrabti, J.P. Ferrio, C. Royo, Environmental factors determining carbon isotope discrimination and yield in durum wheat under Mediterranean conditions, Crop Science 43 (2003) 170e180. [9] S. Boardman, G. Jones, Experiments on the effects of charring on cereal plant components, Journal of Archaeological Science 17 (1990) 1e11. [10] S.R. Bowley, A Hitchhiker’s Guide to Statistics in Plant Biology, Plants et al., Guelph, Ontario, 1999. [11] R. Buxó, Arqueologı́a de las Plantas: la Explotación Económica de las Semillas y los Frutos en el Marco Mediterráneo de la Penı́nsula Ibérica, Crı́tica, Barcelona, 1997. [12] J. Cascón, Agricultura Española: Antologı́a de Artı́culos, Monografı́as y Conferencias, Dirección General de Agricultura, Madrid, 1934. [13] S.M. Colledge, Plant exploitation on Epipalaeolithic and Early Neolithic sites in the Levant, PhD Thesis, University of Sheffield, 1994. [14] S.M. Colledge, Plant Exploitation on Epipalaeolithic and Early Neolithic Sites in the Levant, British Archaeological Reports, Oxford, 2001. [15] R.W. Dennell, The economic importance of plant resources represented on archaeological sites, Journal of Archaeological Science 3 (1976) 229e247. [16] T. Evers, J. Flintham, K. Kotecha, a-Amylase and grain size in wheat, Journal of Cereal Science 21 (1995) 1e3. [17] M.J. Gooding, R.H. Ellis, P.R. Shewry, J.D. Schofield, Effects of restricted water availability and increased temperature on the grain filling, drying and quality of winter wheat, Journal of Cereal Science 37 (2003) 295e309. [18] S. Gustafsson, Carbonized cereal grains and weed seeds in prehistoric housesdan experimental perspective, Journal of Archaeological Science 27 (2000) 65e70. [19] M. Gutiérrez, J.L. Peña, Evolución climática y geomorfológica del Holoceno Superior (Cordillera Ibérica, Depresión del Ebro y PrePirineo), in: A. Cearreta, F.M. Ugarte (Eds.), The Quaternary in the Western Pyrenean Region, Bilbao, 1992, pp. 109e122. [20] R.J. Henry, P.S. Kettlewell, Cereal Grain Quality, Chapman and Hall, London, 1996. [21] M. Hopf, Form veranderung von Getreide-Kornern beim Verkohlen, Berichte der Deutschen Botanischen Gesellschaft 68 (1955). [22] H. Ku, Notes on the use of propagation of error formulas, Journal of Research of the National Bureau of Standards 70C (4) (1966) 263e273. [23] F.A. Lone, M. Khan, G.M. Buth, Palaeoethnobotany: Plants and Ancient Man in Kashmir, A.A. Balkema, Rotterdam, 1993. [24] J.L. Molina-Cano, M. Francesch, A.M. Perez-Vendrell, T. Ramo, J. Voltas, J. Brufau, Genetic and environmental variation in malting and feed quality of barley, Journal of Cereal Science 25 (1997) 37e47. [25] J. Renfrew, Palaeoethnobotany, Methuen, London, 1973. [26] Y. Rharrabti, C. Royo, D. Villegas, N. Aparicio, L.F. Garcı́a-delMoral, Durum wheat quality in Mediterranean environments I. Quality expression under different zones, latitudes and water regimes across Spain, Field Crops Research 80 (2003) 123e131. [27] M.T. Ros, La vegetació de la Catalunya Meridional i territoris propers de la Depressió de l’Ebre, en la Prehistòria recent 1642 [28] [29] [30] [31] [32] J.P. Ferrio et al. / Journal of Archaeological Science 31 (2004) 1635e1642 i Protohistòria, a partir dels estudis antracològics, Gala 3e5 (1997) 19e32. SAS, SAS/STAT User’s Guide, release 6.03, SAS Institute, Cary, NC, 1988. R. Savin, J.L. Molina-Cano, Changes in malting quality and its determinants in response to abiotic stresses, in: G.A. Slafer, J.L. Molina-Cano, R. Savin, J.L. Araus, I. Romagosa (Eds.), Barley Science: Recent Advances from Molecular Biology to Agronomy of Yield and Quality, Food Product Press, New York, 2002, pp. 523e550. M. Stuiver, P.J. Reimer, Extended 14C data base and revised CALIB 3.0 14C age calibration program, Radiocarbon 35 (1993) 215e230. J. Threadgold, T.A. Brown, Degradation of DNA in artificially charred wheat seeds, Journal of Archaeological Science 30 (2003) 1067e1076. J. Voltas, F.v. Eeuwijk, A. Sombrero, A. Lafarga, E. Igartua, I. Romagosa, Integrating statistical and ecophysiological analyses of genotype by environment interaction for grain filling of barley I. Individual grain weight, Field Crops Research 62 (1999) 63e74. [33] J. Voltas, I. Romagosa, J.L. Araus, Grain size and nitrogen accumulation in sink-reduced barley under Mediterranean conditions, Field Crops Research 52 (1997) 117e126. [34] J. Voltas, I. Romagosa, J.L. Araus, Growth and final weight of central and lateral barley grains under Mediterranean conditions as influenced by sink strength, Crop Science 38 (1998) 84e89. [35] J. Voltas, I. Romagosa, A. Lafarga, A.P. Armesto, A. Sombrero, J.L. Araus, Genotype by environment interaction for grain yield and carbon isotope discrimination of barley in Mediterranean Spain, Australian Journal of Agricultural Research 50 (1999) 1263e1271. [36] G. Willcox, Measuring grain size and identifying Near Eastern cereal domestication: evidence from the Euphrates valley, Journal of Archaeological Science 31 (2004) 145e150. [37] D.G. Wilson, The carbonisation of weed seeds and their representation in macrofossil assemblages, in: W. Van Zeist, W.A. Casparie, A.A. Balkema, Rotterdam, 1984, pp. 201e206. [38] P. Wright, Preservation or destruction of plant remains by carbonization?, Journal of Archaeological Science 30 (2003) 577e583.