Materials and methods

Sediment samples for extraction of archaeobotanical material from trench 23 were initially selected according to ‘judgment’ sampling (M. Jones 1991) i.e. where dark organic remains were visible, or charred plant remains were expected. From 2019, additionally, an ‘interval’ (M. Jones 1991) sampling grid-system was implemented in an area surrounding kiln 3. Sample volume varied from 5 to 70 L.

Plant material was recovered using an Ankara-style flotation device (French 1971). Estimated richness, plant taxa diversity and context information for the samples were used to select samples to analyse. Altogether, including 15 samples from the gridded area, 55 samples from the malting complex have been analysed.

Samples were sorted using a stereomicroscope in the School of Archaeology, University of Oxford, with plant material identified using the lab’s reference collection and relevant literature (Jacomet 2006; Stace 2010; Cappers et al. 2013). Latin nomenclature follows Stace (2010). All plant material in the assemblage is charred.

The minimum number of individual plant items (MNI) was determined by counting specific diagnostic zones (G. Jones 1991). For cereal grains the frequency of both apical and embryo ends per sample was counted, and the higher figure recorded. Most weed seed taxa occurred infrequently in each sample and were scored even when fragmented. Where several fragments co-occurred, the MNI was estimated.

Functional weed ecology

Two sets of FWE analyses were conducted. The first pertains to intensity of agricultural regime, where intensity relates to both fertility (which may be associated with the level of manuring) and, to a lesser extent, disturbance caused by weeding/tillage. The second analysis relates to disturbance (level of weeding/tillage) alone. First, the weed spectrum from Sedgeford was classified by a discriminant analysis function using an existing ‘extensive/intensive’ FWE model. The model contrasts two sets of weed flora from modern surveys in known extensive and intensive agricultural regimes from Haute Provence, France and Asturias in Spain, respectively (Charles et al. 2002; Bogaard et al. 2016). Secondly, the Sedgeford weed set was classified by a different discriminant analysis function within a model contrasting floras from high- and low-disturbance modern regimes, based on modern weed surveys at Laxton, (Nottinghamshire, UK) and Highgrove Home Farm (Gloucestershire, UK) (Hamerow et al. 2020; Bogaard et al. 2022). The disturbance model contrasts low disturbance unploughed ‘sykes’ at Laxton with high disturbance regimes: fallow fields in use as part of a rotation scheme at Laxton and arable fields at both Laxton and Highgrove (Hamerow et al. 2020; Bogaard et al. 2022). ‘Syke’ is the name given to an occasionally grazed, cut once-yearly, untilled patch of hay meadow between or on the edges of other (‘open’) fields at Laxton, which is not treated with herbicide (Hamerow et al. 2020 p.13).

The analysis is based on recording functional traits in modern weed plants which correlate with species’ potential under variable growing conditions (Charles et al. 1997; Bogaard et al. 2001; Jones 2002). The two FWE models applied here use different sets of functional traits to achieve the best separation of the modern regimes. Traits which best distinguish more- and less-fertile regimes (included in the intensity model) are plant canopy height, canopy diameter, specific leaf area and the ratio of leaf area per node to fresh leaf thickness (Bogaard et al. 2016) (Fig. 3e). The length of the flowering period and – for perennial plants – capacity for vegetative regeneration best distinguish between high- and low-disturbance regimes (Bogaard et al. 2022) (Fig. 4f). In each case, in the ‘discrimination phase’, a discriminant analysis model was created, using a linear equation which optimally separated the two sets of known regime attribute data. In the subsequent ‘classification phase’, functional trait data based on modern plant analogues of weed species represented as seeds in the Sedgeford samples were used to classify each sample into one of the two contrasting known groups, according to the linear discriminant equation previously extracted to maximise separation between the contrasting modern regimes.

Fig. 3: ‘Intensity’ discriminant analysis plots. a Scores of modern arable plots along the discriminant function extracted to separate low- (white) and high-intensity (black) crop cultivation regimes. Larger symbols indicate group centroids (average scores). b Relationship between units from the Sedgeford assemblage and the discriminant function used to distinguish high- and low-intensity modern regimes (centroids as before). c Replica of plot b) with episodes coded by malting complex feature. d Replica of plot b) with episodes coded by date range. e Correlations between functional trait scores used as discriminating variables and the discriminant function. Both plots (a) and (e) are reproduced with permission from Bogaard et al. (2016, p 66, Figs. 6b and 7b)

Fig. 4: Disturbance discriminant analysis plots. a Relationship of low-disturbance regime (white squares) and of high-disturbance regimes (other symbols) with the discriminant function. Larger symbols are group centroids (average scores). b Relationship between units from the Sedgeford assemblage and the discriminant function extracted to distinguish low-disturbance from high-disturbance regimes. Larger symbols are group centroids. c Relationship of experimental mouldboard-ploughed fields from Lorsch with the extracted discriminant function, centroids as before (shared with permission from Bogaard et al. (2022) Figs. 13, 14, 15). d Same plot as b) with units coded by malting complex feature. e Same plot as b) with units coded by date range. In plots b) and c), the ‘Lorsch baseline’ is shown as a dashed line. f Correlations between functional trait scores used as discriminating variables and the discriminant function. Both plots (a) and (f) are reproduced with kind permission from Hamerow et al. (2020, p 599, Fig. 8a and f).

With reference to the disturbance model, a reconstructed ox-drawn mouldboard plough has recently been used as part of a three-field rotation system at the Lauresham Open-Air Laboratory for Experimental Archaeology in Lorsch, Germany (Bogaard et al. 2022; Kropp 2022; Stroud et al. 2024). FWE was conducted on the weed flora from this field study. Bogaard et al. (2022) devised a ‘Lorsch baseline’ representing the lowest expected discriminant score for fields tilled with a mouldboard plough (Fig. 4c), and this is compared to Sedgeford data here.

Only Sedgeford samples with at least 10 weed seeds identified to species level were included in the analyses (on this basis, sample 17018 was excluded). Further, spatially proximate samples with similar composition were grouped, to give 26 distinct ‘behavioural episode’ units. Grouping samples into ‘behavioural episodes’ was undertaken for the purposes of FWE alone. Stable isotope and correspondence analyses were conducted using ungrouped samples. This analysis is equivalent to the workflow now formalised in the ‘R’ package, WeedEco (Models 1 and 3) (Stroud et al. 2024) but was conducted before this package was available. Analysis was conducted using IBM SPSS version 27.

Stable isotope analysis

FWE results are complemented by additional independent lines of evidence for revealing crop husbandry methods, e.g. stable isotopic analysis (Bogaard et al. 2016; Stroud et al. 2021). Both stable carbon and nitrogen isotope analyses were conducted. The ratio of stable carbon isotopes, 13C to 12C (δ13C), in preserved grains can indicate levels of soil water availability for crops (e.g. Wallace et al. 2013; Styring et al. 2017; Stroud 2022), whilst archaeological grains’ ratio of stable nitrogen isotopes, 15N to 14N (δ15N), can be used to investigate soil nitrogen isotopic ratio, and thus the possibility of manuring (e.g. Bogaard et al. 2007; Styring et al. 2017; Stroud 2022).

Reduced enzymatic discrimination against the heavier 13C isotope during plant photosynthesis when stomatal pores are closed at times of water stress, means a plant’s carbon stable isotope ratio correlates with its water status (e.g. Farquhar et al. 1989; O’Leary 1993). δ13C values can aid elucidation of cultivation methods even in temperate climates such as the UK’s, where plant water stress is rare, e.g. by revealing whether sets of crops were grown in similar water availability conditions, potentially consistent with crop rotation (Hamerow et al. 2025, p 129). Where crop δ13C values differ significantly and consistently between species, indicating different water availability conditions, crop rotation is unlikely to have been in use. There are various possible explanations for crops having been cultivated in similar water availability conditions – one amongst these is crop rotation.

Barley grains have a δ13C value of ~ 1–2‰ lower (six-row barley grains ~ 2‰ lower) than free-threshing wheat grains from crops grown with equal water availability (Anyia et al. 2007; Wallace et al. 2013). δ13C values are also influenced, inter alia, by soil type, canopy cover, light, temperature, and topography (Heaton 1999; Bogaard et al. 2016).

Turning to stable nitrogen isotopic analysis, factors influencing soil and thus plant tissue δ13N values include aridity, salinity, waterlogging and crop manuring (Heaton 1987; Bogaard et al. 2007; Larsson et al. 2019). Preferential volatisation of the lighter 14N isotope in manure means that plants grown in manured soil themselves become enriched in 15N (Bol et al. 2005; Bogaard et al. 2007). Detecting manuring of fields requires knowledge of the δ13N values of plants grown on local unenriched soils (e.g. Stroud et al. 2021); this may be estimated based on δ13N values of archaeological wild herbivore bone collagen (Hedges and Reynard 2007; Bogaard et al. 2013) assuming herbivores were foraging on wild rather than crop plants. Nitrogen stable isotope values are available for collagen from two wild herbivores from Anglo-Saxon era East Anglia (two deer, both in Suffolk) (ESM 1 Table S1) (Leggett 2021, p 93). The mean δ15N value for the herbivore collagen is 5.9‰. To compare these with cereal grain δ15N values, 4‰ was subtracted to compensate approximately for the trophic shift between vegetation and herbivore, and 2.4‰ added to account for the offset between grains (equivalent to seeds or fruits) and rachis (equivalent to leaves or stems), the plant parts generally consumed by herbivores (Fraser et al. 2011; Bogaard et al. 2013). The mean calculated ‘wild herbivore baseline’ thus derived for unmanured grains is 4.3‰. These wild herbivore-baseline derived δ15N values should be applied with some caution: the samples originate > 34 miles from Sedgeford; further, red deer can browse in woodlands, (i.e. are not always grazers), potentially reducing their bone collagen δ15N values (Sykut et al. 2021).

Other instances where a wild herbivore baseline has been applied to assess δ15N values include at the medieval site of Lyminge in Kent (Hamerow et al. 2025, p 88, Fig. 3.31). A particularly robust use of such baselines is Styring et al. 2017 (see especially p 366 Fig. 3).

Single-grain samples of rye, free-threshing wheat and hulled barley were selected from kilns 1, 2, 3 and the steeping cistern. Additional criteria informing grain selection strategy were as follows. First, visibly germinated grains were excluded, to limit potential confounding effects. Secondly, charring can affect δ13C and δ15N values (Styring et al. 2013; Nitsch et al. 2015; Stroud et al. 2023a). Grain interiors give the best indication of charring temperature (Stroud et al. 2023a; Vaiglova et al. 2023); grains were halved and those judged to fall outside an optimum charring window (215–300 °C) (Stroud et al. 2023a) excluded. In total 112 single-grain samples were analysed; numbers of grains belonging to each species selected, and their contexts, are listed in ESM 1 Table S2.

Fifteen representative single-grain samples were screened using Attenuated Total Reflection Fourier Transform Infrared Spectroscopy (ATR-FTIR) to identify possible contamination post-deposition. No evidence of contamination with carbonates, nitrates or humic acids was found (ESM 1 Fig.S1), thus pre-treatment was deemed unnecessary.

The samples were analysed using a Sercon 20–22 EA-GSL isotope mass spectrometer at the Oxford University Research Laboratory for Archaeology and the History of Art. In-house standards of cow collagen (COW), seal collagen (SEAL) and Alanine were used to calibrate the data according to the internationally determined scales: VPDB (Vienna Pee Dee Belemnite) for carbon and AIR (Atmospheric nitrogen) for nitrogen (Hoefs 2018, p 34). All standards used have well-characterised isotopic compositions; their means and standard deviations (SDs) are listed in ESM 1 Table S3.

To understand accuracy, precision and uncertainty, an internal leucine standard and EMA P2 were used as check standards, and every 10th sample was duplicated. Following Szpak et al. (2017), the accuracy (u(bias) of the carbon runs was calculated as ± 0.28‰, and the precision (u(Rw) as ± 0.08‰. Combined total analytical uncertainty for δ13C values was ± 0.29‰. For nitrogen, accuracy was ± 0.40‰ and precision ± 0.20‰. Overall uncertainty for δ15N values was calculated as ± 0.45‰ (Szpak et al. 2017). Data were calibrated using the statistical programme R (version 4.1.2), and accuracy, precision and uncertainty calculated using Excel, version 16.58.

Following Stroud et al. (2023b), ‘charring’ offsets were applied to the calibrated values: +0.33‰ for δ15N values, and + 0.12‰ for δ13C values. Following FeedSax practice (e.g. Stroud 2022), calibrated δ13C values were not converted to Δ13C, since δ13C values for archaeological grains are not here compared with those for modern grains (see e.g. Farquhar et al. 1989).

Correspondence analysis and seasonality

Correspondence analysis is a means of exploring variability in the composition of samples from an assemblage, by representing these graphically. Correspondence analysis was used to investigate trends in the sowing season of crops supplying the malting complex. Species occurring frequently in the assemblage are represented in the correspondence analysis scatterplots featured here. Those whose datapoints fall close to the graph’s origin are common or ubiquitous in samples from the assemblage. Where species are clustered (spatially proximate) in the plot, this indicates that these regularly co-occur in the samples, consistent with a tendency to grow together. Where species are spatially dispersed in the plot, the reverse is true (Bogaard 2004, pp 92–94).

Correspondence analysis plots were created using CANOCO version 5.0 (ter Braak and Smilauer 2012). CANOCO generates for each analysis four axes, with axis 1 accounting for the greatest variation. In all plots shown here the horizontal axis is axis 1, and the vertical, axis 2. It is possible, using CANOCO, to apply mathematical ‘transformations’ to the correspondence analysis values. A square-root transformation, applied in one plot here, can help data to fit the assumptions of correspondence analysis better by reducing skew and ‘stabilising’ variance in the data (Bartlett 1936; Greenacre 2009). Table 1 summarises the short names allotted to each taxon included in the correspondence analyses, as displayed in the scatterplots.

Type | Flowering onset | Flowering duration (months) | Competitive advantage in…
Early/short | Jan.–Jun. | 1–3 | Autumn-sown fields
Late | Jul.–Dec. | 1–5 | Spring-sown fields
Long | Jan.–Jun. | > 5 | Spring-sown fields
Intermediate | Apr.–Jun. | 4–5 | Autumn and spring-sown fields

Data in the correspondence analyses conducted here are additionally coded by seasonality. We here use correspondence analysis to infer the season of sowing for crops based on associations with weeds whose ecological traits make them more likely to co-occur with spring or autumn-sown cereals (cf. McKerracher 2019, pp 96–127). Weed ecological traits which best reveal crop seasonality are timing of flowering onset and flowering duration (Bogaard et al. 2001). Annual weed species whose germination and flowering time is late in the year or of long duration (setting seed after ploughing for spring sowing) are advantaged among spring sown crops. Species which germinate and flower early and for a short time flourish (being undisturbed) among autumn-sown fields (Bogaard et al. 2001; McKerracher 2019, p 97).

Our assumption that weed species featured in the correspondence analyses grew amongst the crops as genuine crop weeds is based on these surmises: firstly, featured samples were grain-rich and relatively-speaking weed-species poor. Secondly, these were recovered from in situ contexts (rich concentrations of charred grains surrounding malting kilns) not suggestive of secondary mixing. Finally, only weed species occurring above a threshold frequency were incorporated into the analyses.

Samples with < 10 weed seeds were excluded, as were weed species occurring in fewer than 10% of samples. Weed taxa identified only to the family or genus level were also excluded. Eligible Sedgeford weed seeds included two perennials: Plantago lanceolata and Phleum pratense. Each of these regularly regenerates by seed (as well as through vegetative propagation) hence these were treated as annuals (cf. Bogaard et al. 2001). A total of 53 samples, and 12 weed species (in addition to the cereal taxa) were included in the analyses. The seasonality category assigned to each species is shown in Table 2.

Species | Class | Resultant seasonality | Short name
Agrostemma githago L. | Early/short | Autumn | A_gith
Anthemis cotula L. | Late | Spring | A_cot
Atriplex hastata L. / patula L. / prostrata Boucher ex. D.C. | Intermediate | N/a | Atri_p
Bromus arvensis L. / hordeaceus L. / secalinus L. | Early/short | Autumn | Bro
Chenopodium album L. | Late | Spring | C_alb
Fallopia convolvulus (L.) Á.Löve | Late | Spring | F_conv
Phleum pratense L. | Early/short | Autumn | P_prat
Plantago lanceolata L. | Long | Spring | P_lanc
Raphanus raphanistrum L. | Intermediate | N/a | R_raph
Urtica urens L. | Intermediate | N/a | U_uren
Vicia hirsuta L. (Gray) / tetrasperma L. (Schreb.) | Intermediate | N/a | V_h_l