Causes of floristic variation in the Haute Provence weed flora
Correspondence analysis of all 60 crop field transects in Haute Provence (Fig. 3a) on the basis of 78 weed species occurring in at least 5 % of fields presents a clear contrast on axis 1 (horizontal) between the Sault region (right) and the Lubéron (left). Axis 2 (vertical) distinguishes between the majority of fields (including both cereals and pulses) concentrated at the negative (bottom) end from four intensively worked pulse fields towards the positive end. These were fields of spring-sown lentils and chickpeas on loamy stone-free soils along the river Nesque near Sault (Farm 2, Table 1) ‘weeded’ mechanically using a vibroculteur once the early spring weeds had germinated. The weed species associated with the intensively worked pulse fields belonged in phytosociological terms to the Chenopodietea, garden or row-crop weeds (Fig. 3b). Although some Chenopodietea species are associated with low intensity cereal (and pulse) cultivation, the vast majority of species associated with low intensity cultivation belong to the class Secalinetea. The small number of fields did not permit full ecological characterisation of the more intensive form of pulse cultivation in Haute Provence, but the evident contrast with less intensively managed fields is relevant to comparison with intensive cultivation in Asturias (see below).
In a second correspondence analysis, excluding the intensively worked pulse fields, the geographical distinction between the Sault area (right) and the Lubéron (left) remains dominant on axis 1 (Fig. 4a). Axis 2 distinguishes the Cavalon and Chaffère drainages in the Lubéron (Fig. 4b). Figure 4a, coded by farm, reveals subregional clustering of farms in the Sault region and Cavalon drainage; the Chaffère drainage (upper left) is represented by a single farm. The dominance of geography as a correlate with floristic variation in Haute Provence indicates a contrast with the study of ‘garden’ and ‘field’ cultivation in Evvia, Greece, where any geographical variation in the weed flora was subordinate to the gradient in agricultural intensity (Jones et al. 1999).
While the contrast between the Cfb climate of the Sault region and the more mediterranean climate of the Lubéron likely contributes to the geographical trend on axis 1, the contrast also involves different soil types. This is reflected in a contrast in weed species between those preferring acidic soils around Sault and those preferring alkaline soils in the Lubéron (Fig. 5a). These different soil types also encourage different agricultural practices: spring harrowing of cereals is associated with the soil types of the Sault/Albion plateau but is not practiced on the more clay-rich soils in the Lubéron (Fig. 4c). This explains the tendency for long-flowering annuals, able to recover from disturbance, to be located towards the positive (right) end of axis 1 (Fig. 5b). It is clear, however, that it is the location of fields (which determines soil type), rather than harrowing, that governs the distribution of fields in the correspondence analysis, because the few unharrowed fields in the Sault area cluster with the other (harrowed) Sault fields rather than the unharrowed fields of the Lubéron (Fig. 4), and the association of weed species preferring acidic soils with the Sault area is stronger than the association of long-flowering species with harrowing (Fig. 5).
As noted above, manuring of cereals in the Lubéron with chicken dung is a means of reducing the soil pH; manuring rates are very low, however, and no differences in weed functional attributes relating to soil productivity are apparent (plots not shown). Finally, neither crop sowing time nor crop taxon show any clear patterning (plots not shown). For example, the analysis did not detect any differences in the weed flora of einkorn fields due to their later harvesting time.
Overall, the weed flora of the Provence fields primarily vary according to their geographical location (in response to soil type and perhaps climate), and only secondarily in response to agricultural practice (notably harrowing).
Ecological comparison of Haute Provence and Asturias regimes on the basis of weed functional attribute values
Discriminant analysis was used to distinguish between the weed flora of large-scale/low-intensity production in Haute Provence and small-scale/intensive cereal farming in Asturias, on the basis of the five functional attributes previously shown to distinguish between high- and low-intensity pulse cultivation regimes in Evvia, Greece as discriminating variables (Table 3). Though geographical differences between the two areas may affect the discrimination (Jones et al. 2010), our hypothesis was that the strong agronomic contrasts (Table 2) would enable the regimes to be distinguished along predictable lines using these attributes. All of the Asturias fields surveyed (Charles et al. 2002), and 56 of the Haute Provence field transects, were included in the discrimination; the four intensively worked pulse fields in Haute Provence (Fig. 3a) were entered into the classification phase of the discriminant analysis.
Figure 6 shows that the Haute Provence and Asturias fields are clearly distinguished on the basis of fully quantitative (98 % correctly reclassified, 119 of 121) and semi-quantitative (presence/absence) data (100 % correctly reclassified; Fig. 6). Functional attributes characterise the two regimes in accordance with ecological predictions (Fig. 7). In both analyses, weed species with tall, broad canopies, high specific leaf area (SLA) and a high leaf area:thickness ratio are associated with intensive cereal farming in Asturias, and vice versa for the less fertile Provence fields; a long flowering period is also associated with greater disturbance in Asturias (cf. Jones et al. 2000). It is worth noting that three of the four more intensively worked pulse fields in Haute Provence—those that emerged as distinct in the correspondence analysis (Fig. 3a)—were classified as intensively cultivated by both discriminant functions (Fig. 6).
The functional attributes discriminate between the Haute Provence and Asturias regimes as predicted based on contrasts in agronomic practice, despite geographical differences between the two areas (Table 2; cf. Jones et al. 2010). Moreover, pronounced geographic variation within Haute Provence (above) does not obscure the predicted ecological contrast with Asturias.
Comparison of Haute Provence and Asturias regimes on the basis of cereal stable isotope values
Table 4 and Fig. 8 show that the production regimes in Haute Provence and Asturias exhibit largely distinct ranges of δ15N values (a difference of 2.36 ‰, 95 % CI 0.71, 4.01 ‰, p = 0.002). The Asturias fields exhibit a range of values equivalent to ‘medium’ and ‘high’ levels of manure application in long-term agricultural experiments (Fraser et al. 2011). Agronomic observations in Asturias show that manure is applied at rates equivalent to c. 15–40 tons per year for 1 ha (Table 2; Bogaard 2012), spanning experimental medium (c. 10–15 t/ha) to high (30+ t/ha) rates (Fraser et al. 2011). With two exceptions, the Haute Provence fields exhibit δ15N values lower than 3 ‰, as predicted by long-term agricultural experiments where there is continuous cultivation with little to no manuring (Fraser et al. 2011). The exceptions are two einkorn fields (Farm 3, Table 1) located near Sault along the river Nesque. Manuring can be excluded as a cause of the high δ15N values for these two fields, which were managed in the same way as the other fields on Farm 3), so natural organic matter or anaerobic denitrification are more likely explanations (see also below).
Higher δ15N values coincide with higher mean discriminant scores in the weed functional attribute analysis for Asturias, and vice versa for Haute Provence (Table 4). This relationship reflects the contribution of manuring to the high cultivation intensity ‘signature’ of the weed species in Asturias.
Carbon discrimination (∆13C) values in Asturias and Haute Provence suggest that cereals in both regions generally had access to sufficient water (∆13C > 17 ‰ for wheat, Wallace et al. 2013). Values in Haute Provence tend to be higher (1.24 ‰, 95 % CI 0.67, 1.82 ‰, p < 0.001) despite lower annual rainfall than those from Asturias (Table 2), potentially due to an exceptionally wet spring/summer when the Haute Provence study was conducted in 2013, and/or the [stress-tolerant] crop varieties grown in Haute Provence.
Comparison with ‘traditional’ cereal production regimes in Romania and northern Turkey
In order to test the accuracy of the new ‘weed plus isotope’ intensity model, we first used the discriminant function extracted to separate the Haute Provence and Asturias fields on the basis of semi-quantitative data (i.e. the form in which it would be applied archaeobotanically) to classify intensively managed crop fields in a study area in Transylvania, near Sighisoara, Romania (Fig. 1), in a region of temperate-continental climate (Köppen–Geiger zone Dfb). Here, einkorn and other cereals are grown at c. 350–600 m altitude in small fields (c. 50–1,500 m2) with variable levels of manuring (often applied to maize or potatoes, grown in rotation with the cereals) and hand-weeding (Hajnalová and Dreslerová 2010; Fraser et al. 2011). The weeds in a total of 17 cereal fields in this region were surveyed in 2008 using the same methods as in Haute Provence and Asturias; 14 fields with ripe crops (including eight surveyed for weeds) were sampled for crop stable isotope analysis. Table 5 and Fig. 9 show that the Sighisoara fields were all (correctly) classified as ‘intensive’. It is notable that most discriminant scores for the Sighisoara fields are located at the ‘low-intensity’ end of the Asturias spectrum; this result corresponds well with agronomic observations, which indicate that the Sighisoara fields were managed rather less intensively than most of the Asturias fields [e.g. less consistent manuring and hand-weeding: see Charles et al. (2002), Hajnalová and Dreslerová (2010)].
Next we compared the Sighisoara fields with those in Asturias and Haute Provence in terms of their crop isotope values (Table 4; Figs. 8c, 10). The majority of Sighisoara fields (11 of 14) have δ15N values consistent with ‘medium’ rates of manuring (Fraser et al. 2011; Bogaard et al. 2013), while a small number have lower values and had not been manured in recent years (only one of the latter was included in the weed surveys). In terms of ∆13C values, those of the Sighisoara einkorn tend to be higher (‘wetter’) than those of bread wheat by an average of 1.1 ‰ (Fig. 8c; t(12) = 2.60, p = 0.023). Since the einkorn fields were managed in the same way as the bread wheat crops, and on similar soils, it appears that the relationship between carbon discrimination and water status is different in Romanian einkorn and bread wheat; a similar ‘offset’ has previously been noted between barley and free-threshing wheat (Araus et al. 1997; Wallace et al. 2013). Such an offset is not apparent between einkorn and other wheats in Haute Provence (Fig. 8a), and may reflect landrace-specific variation.
In order to test the ability of the new model to correctly identify low-intensity cereal production, we applied it to a second case study area, centred around the town of İhsangazi in the (Cfb) Kastamonu province of northern Turkey (Fig. 1), where einkorn and emmer were grown on a moderately large scale (fields ranging from c. 0.2 ha to 4.5 ha) with little to no manuring and no weeding, in rotation regimes incorporating forage legumes and bare (ploughed) fallow (Karagöz 1996; Ertuğ2004). The weeds in a total of 13 cereal fields in this region were surveyed in 2008 using the same methods as in Haute Provence and Asturias; eight fields with ripe crops (including seven surveyed for weeds) were sampled for crop stable isotope analysis. Table 5 and Fig. 9 show that the Kastamonu fields are all classified correctly, on the basis of their weed functional attributes, as being subject to low-intensity cultivation, in one case with a much lower discriminant score than the Haute Provence fields, suggesting poorer growing conditions than in the Haute Provence study.
We then compared the Kastamonu fields with those in the other studies in terms of their δ15N values (see Fraser et al. 2011 for the isotope methodology pertaining to these samples); ∆13C values are not available for these samples (Table 4; Fig. 10). Seven of the eight Kastamonu fields have δ15N values below 3 ‰, consistent with their reported history of little to no manuring; the cause(s) of the single higher value (4.1 ‰), from a field not manured in recent years, is unclear.
The relationship between weed ecology and cereal isotope values on a field-by-field basis
Figure 11 shows the relationship between discriminant scores based on (semi-quantitative) weed ecological data and δ15N values for cereals in the Haute Provence, Sighisoara and Kastamonu studies (n = 34), where the two variables can be compared on a field-by-field basis; there is a weak but significant positive correlation for all fields (R2 = 0.19, p = 0.009). This is consistent with the agronomic observation that manuring contributes to the contrast in intensity between Sighisoara, on the one hand, and Haute Provence and Kastamonu, on the other. Field discriminant scores and isotope values for the same fields are not available for Asturias, but Fig. 11 includes the mean ±1σ for all Asturias weed survey transects and cereal isotope samples. The Asturias data fit well within the overall pattern.
The two einkorn fields from Provence (Farm 3) with unusually high cereal δ15N values emerge as outliers in Fig. 11; they do not have a high cultivation intensity signature based on their weed functional ecology. In fact, their weed ecological signature places them at the ‘low-intensity’ end of the Provence range. We hypothesise that these elevated δ15N values are due to seasonal waterlogging along the river, rather than to high organic matter content (through manuring or natural abundance). This example illustrates the value of combining crop isotope analysis with weed ecology: interpretation of high crop δ15N values is informed by ecological analysis of the associated weed flora. If these two fields are excluded, the correlation between the weed ecology discriminant scores and cereal δ15N values is stronger (R2 = 0.45, p < 0.001).
There was no correlation between cereal ∆13C values and weed ecology discriminant scores; this is to be expected since artificial watering did not form part of management intensity in these studies. Topographical analysis of the fields using GIS, however, did reveal a relationship between slope and cereal ∆13C values (Fig. 12), such that steeply sloping fields are associated with lower (‘drier’) values than flat fields (R2 = 0.45, p = 0.003). This relationship presumably reflects greater water run-off on slopes. No other topographical variables (aspect, solar radiation, distance from stream/river) exhibited a clear relationship with cereal ∆13C or δ15N values.
Application of the Asturias/Haute Provence model to archaeobotanical data from Neolithic central Europe
The discriminant function extracted to distinguish the high-intensity cultivation in Asturias and low-intensity cultivation in Haute Provence on the basis of semi-quantitative weed ecological data was used to classify 141 Neolithic archaeobotanical samples from 30 sites across central Europe (Supplementary Table 2). These samples were selected because each derives from a single ‘deposit’, contains a minimum of 30 potential weed seeds identified to species and contains crop and weed material consistent with a single crop processing stage; this set of samples includes the 126 discussed in Bogaard (2004), another 14 samples subsequently analysed from Vaihingen/Enz (Bogaard 2012) and a weed-rich pit fill from Ecsegfalva (Bogaard et al. 2007). Most of these samples belong to the early Neolithic Linearbandkeramik (LBK) complex; half come from Vaihingen (Supplementary Table 2) (Bogaard 2012).
Table 5 and Fig. 9 show that most samples (71 %, 100 out of 141) are grouped with the Asturias fields, and the remainder with Haute Provence transects. A significant proportion of the archaeobotanical samples (41 %, 57 out of 140) are classified with low (<0.90) probability, and are positioned between the two groups on the discriminant function in Fig. 9d. Most of the archaeobotanical samples appear to derive from conditions akin to the ‘low-intensity’ end of the Asturias spectrum, in a manner reminiscent of the Sighisoara fields. While a significant minority of samples are grouped with the Provence transects, most are placed at the ‘high-intensity’ end of the Haute Provence spectrum, with only a few samples around or below the Provence group centroid.
Overall, the archaeobotanical results suggest a range of agricultural conditions centred around the medium to high intensity part of the spectrum. While the problem-oriented approach of classifying archaeobotanical samples as unknown cases in discriminant analysis has a limited ability to detect regimes without a present-day analogue (Jones et al. 2010), it is evident that the vast majority of the archaeobotanical samples point to a consistent range of conditions between the Asturias and Haute Provence groups, suggesting a distinctive set of intermediate conditions. Variation in growing conditions is to be expected, particularly in small-scale subsistence production where labour is limiting and is invested strategically depending on its availability (Halstead 2014).
Previous application of the Evvia pulse intensive ‘garden’ versus extensive ‘field’ cultivation model to the weeds of Neolithic central European (Bogaard 2004, 2012) mostly identified the samples as deriving from intensive cultivation, with few samples classified as extensively cultivated. The Evvia model identified attributes relating to both soil fertility (e.g. specific leaf area) and disturbance (flowering period) as important to the discrimination between intensive and extensive cultivation (Jones et al. 2000; Charles et al. 2002, Fig. 3) whereas soil disturbance (flowering period) was less important in the discrimination between intensive cultivation in Asturias and low-intensity cultivation in Haute Provence. Nevertheless, a consistent outcome from application of both the Evvia and the Haute Provence/Asturias models has been that the Asturias fields appear on the whole more intensively managed than the regime(s) represented by the central European archaeobotanical samples (cf. Bogaard 2004, 110). Taking the classifications of the Neolithic samples by the Evvia and Asturias/Provence models together, it is plausible that the Neolithic cultivation plots tended to be more disturbed but less fertile than the Asturias plots.
Full application of the new ‘multi-stranded’ model requires measurement of crop δ15N and ∆13C values in a large number of archaeobotanical samples. At present, only a pilot study in archaeobotanical crop isotope determination has been undertaken from a few Neolithic sites in central Europe, including Vaihingen (Bogaard et al. 2013). Cereal δ15N values from this work are comparable with those from Asturias and Sighisoara (Fig. 10). Combined consideration of weed ecology and crop isotope based inferences on crop growing conditions at Neolithic sites in central Europe suggests that cultivation was of an intensive type but that management intensity was variable.