Baseline Isotopic Data
Module 3: Foundations for Reliable Dietary Reconstruction
11 April 2026
Before You Run ReSources
Module 3
This module establishes why the quality and completeness of your baseline isotopic dataset determines the reliability of any ReSources output. The most sophisticated Bayesian framework cannot compensate for inadequate baseline data.
What Is Baseline Isotopic Data?
Definition
Baseline isotopic data = the measured isotope values of all potential dietary sources available to the population under study.
Most commonly: δ13C and δ15N — but also δ34S, δ18O, and 87Sr/86Sr where needed.
These measurements form the end-members of the mixing model — the isotopic anchoring points against which consumer tissue values are interpreted.
If end-members are inaccurate, incomplete, or derived from an inappropriate reference population, ReSources estimates will be unreliable — regardless of how carefully the MCMC chain is configured.
The Three Pillars of a Robust Baseline
1 · Completeness
All plausible dietary contributors have been identified and sampled
2 · Contextual Relevance
Measurements reflect the actual isotopic environment of the site, period, and ecological zone
3 · Sufficient Replication
Each end-member is characterised by enough specimens to capture natural biological variability
Failure in any one of these three pillars propagates directly into mixing model output as bias, spurious precision, or outright misinterpretation.
Why Baseline Collection Cannot Be Deferred
The Asymmetry of Model Errors
Errors at the input stage compound through the entire analytical chain.
Statistical sophistication at the modelling stage cannot reverse them.
ReSources propagates uncertainty honestly via MCMC — but only for quantified uncertainties.
Garbage in, Bayesian clothing
No protection against:
- Sources never measured
- Systematic regional offsets in published values
- Wrong tissue type or time period
A Bayesian model run on flawed baseline data produces a beautifully precise incorrect answer.
Sampling Decisions Are Often Irreversible
Archaeological and zooarchaeological sampling opportunities are frequently non-repeatable:
- Faunal assemblage may be fully excavated
- Stratigraphic context may not be accessible again
- Museum collections often limit destructive sampling
Baseline decisions must be made prospectively — before data analysis begins — on the basis of an explicit ecological and archaeological model of what was available and consumed. In reality the use of published data is more commonplace.
Reconstructing a baseline post hoc risks unconscious selection bias: including sources that make results look plausible and overlooking sources that would complicate interpretation.
Completing Your Source Inventory
Archaeological Evidence as a Starting Point
The first step is not an isotope measurement — it is a thorough assessment of the archaeological record.
| Evidence type | What it tells you |
|---|---|
| Faunal assemblages (NISP, MNI) | Exploited taxa; relative abundance |
| Archaeobotanical remains | Plant foods; seasonality |
| Lipid residue analysis | Processed commodities in vessels |
| Historical / documentary sources | Traded or imported foods absent from excavation |
| Previous isotope studies | Regional baselines; potential comparanda |
Good practice: draw a source diagram before measuring — listing evidence, existing literature values, and available specimen counts for each source.
Protein Sources vs. Energy Sources
ReSources explicitly separates these two categories. Your baseline must too.
Protein sources
(→ bone collagen)
- Terrestrial mammals (cattle, sheep, pig, deer)
- Freshwater & anadromous fish (salmon, eel, pike)
- Marine fish (cod, herring, haddock)
- Legumes (peas, beans, lentils)
- Eggs & dairy products
Energy sources
(→ caloric intake)
- C3 cereals (wheat, barley, oats, rye)
- C4 cereals (millet, maize post-Columbian)
- Root & tuber crops
- Dairy products (secondary protein source)
- Alcohol-bearing crops (often overlooked)
⚠️ The Most Commonly Overlooked Energy Source
C4 plants in European contexts
In European prehistoric and historical contexts, C4 dietary input is often assumed negligible. Always test this assumption.
- Broomcorn millet (Panicum miliaceum) and foxtail millet (Setaria italica) were significant across Bronze Age and Iron Age temperate Europe and medieval Eastern Europe
- δ13C ≈ −12 to −14 ‰ vs. −26 to −28 ‰ for C3 cereals
- Even modest millet consumption produces a strong 13C enrichment signal
Omitting millet will cause ReSources to misattribute C4-enriched consumer values to marine protein.
Don’t Sample Only the Common Taxa
A common weakness: measuring only large, well-represented domesticates while neglecting rarer taxa.
Priority should go to taxa that are:
- Isotopically extreme (marine fish; freshwater fish from unusual catchments; C4 plants)
- Underrepresented in the faunal record but potentially consumed (birds, fish, molluscs)
A small dietary contribution from an isotopically extreme source has a disproportionately large effect on mixing model results. Zooarchaeological rarity ≠ dietary insignificance.
Ensuring Contextual Relevance
The Risk of Published Reference Values
Published baselines from other regions or periods carry risks that are frequently underappreciated.
δ¹⁵N — substrate effects
- Aridity → enriched δ15N throughout the food web
- Manuring → plant δ15N elevated +3 to +6 ‰
- Waterlogging → depressed δ15N in plant macrofossils
- Cattle from southern England vs. northern Scotland: 3–4 ‰ difference from substrate alone
δ¹³C — aquatic variability
- Freshwater fish: δ13C −35 to −18 ‰ depending on catchment geology
- Marine fish: varies with latitude, depth, and productivity regime
- Anadromous species integrate both signals — never approximate from one end-member
Principle: Measure locally. Supplement regionally only with explicit acknowledgement of the offset and sensitivity analyses at ±1–2 ‰.
Temporal Matching
The isotopic baseline shifts through time as climate, land use, and ecology change.
| Factor | Effect | Period of concern |
|---|---|---|
| Agricultural intensification & manuring | Step-change in herbivore δ15N (+2–4 ‰) | Neolithic → Bronze Age; medieval expansion |
| The Suess effect | Modern specimens ~1.5 ‰ depleted in 13C vs. pre-industrial | Post-AD ~1850 reference material |
| Holocene aridity cycles | Ecosystem δ15N baseline shifts | Long-duration sites in arid/semi-arid regions |
For assemblages spanning several centuries, consider whether separate baseline measurements are needed for different phases.
Replication: How Many Specimens?
Replicate to Characterise Variability
Isotope values in animals are not fixed constants. They reflect individual dietary history, physiological state, age, sex, seasonal diet, and habitat use.
The purpose of replication is not just to reduce measurement error (typically ±0.1–0.2 ‰ for good collagen) — it is to characterise the natural biological variability within each source taxon.
Under-replicated source end-members produce artificially narrow standard deviations, causing ReSources to express false confidence in dietary proportion estimates. This makes the output look more precise than the data warrant.
Analytical replicates ≠ biological replicates. Running the same sample twice tells you nothing about inter-individual variability.
Recommended Minimum Sample Sizes
| Source category | Min n | Preferred n | Notes |
|---|---|---|---|
| Large domesticates (cattle, horse) | 5 | 10–15 | Low diet variability if managed |
| Medium domesticates (sheep, pig, goat) | 5 | 10–15 | Pigs: higher variability (omnivory) |
| Wild terrestrial mammals (deer) | 8 | 15–20 | Higher habitat-driven variability |
| Freshwater fish | 10 | 20–30 | Very high inter-individual variability |
| Marine fish (single species) | 8 | 15–20 | Ontogenetic & migratory variability |
| Wild birds | 5 | 10 | Multiple individuals, not one carcass |
| Legumes and cereals | 8 | 15–20 | Soil N heterogeneity; multiple contexts |
| Wild plants | 5 | 10 | Sample representative microhabitats |
Treat these as minima, not targets.
When Sufficient Specimens Are Unavailable
Four appropriate responses — in order of preference:
Pool closely related taxa with similar ecology into a single end-member (e.g., “small ruminants”; “domesticated grain”)
Supplement with contemporaneous material from closely comparable sites — document the offset explicitly
Widen the source SD to reflect insufficient replication:
δ13C → ±1.0–1.5 ‰ · δ15N → ±1.5–2.0 ‰ (where n < 5)Report limitations transparently — note the potential effect on posterior precision
Isotopic Separation: Is Your Baseline Informative?
The Pre-Analysis Check You Must Perform
Before running ReSources, always plot source end-members and consumer values together in δ13C–δ15N bivariate space.
The key questions:
- Do consumer values fall within the source mixing polygon (after applying TEFs)?
- Are the source end-members sufficiently distinct to be resolvable by the model?
- Which sources have high leverage (extreme values) vs. which are poorly resolved?
Well-Structured vs. Poorly-Structured Baseline
Figure 1
Open circles = consumers; coloured ellipses = source end-members (68% confidence)
Four-Step Pre-Analysis Checklist
Step 1 — Draw the δ13C–δ15N bivariate plot for all sources and consumers
Step 2 — Check consumers fall within the source mixing polygon
(after applying trophic enrichment factors)
Step 3 — Identify overlapping sources → pool, add δ34S, or note unresolvable uncertainty
Step 4 — Assess the leverage of each source — extreme values (marine fish; C4 plants) will dominate model inference
If consumers fall outside the mixing polygon: this is a problem with the baseline, not the model. Return to the source inventory — do not adjust TEF priors or λ to force consumers inside.
Trophic Enrichment Factors
TEFs Are Not Universal Constants
Trophic enrichment factors (TEFs) vary as a function of:
- Dietary protein quantity & quality — protein-restricted diets → higher Δ15N (increased N recycling)
- Physiological stress — malnutrition, pregnancy, chronic infection → atypical Δ15N
- Tissue type — bone collagen ≠ muscle ≠ hair ≠ enamel carbonate
- Species — measurable inter-species variation in controlled feeding experiments
For ReSources, the routing parameter λ is effectively part of the enrichment framework — it must also be selected with care.
TEF Selection Hierarchy
| Priority | Source | Notes |
|---|---|---|
| 1 | Controlled feeding experiments — same species, tissue, dietary regime | Rare for humans |
| 2 | Controlled experiments — closely related species or comparable regime | Widen SD explicitly |
| 3 | Empirically calibrated TEFs from well-constrained archaeological contexts | Risk of circularity |
| 4 | Meta-analytic literature compilations | Last resort; always expand SD |
When TEF uncertainty is high, sensitivity analyses across a plausible range of values are an essential component of responsible reporting — not an optional extra.
Practical Workflow
Six Stages Before Running ReSources
Stage 1 · Preliminary Source Inventory
Compile all plausible protein and energy sources from zooarchaeological, archaeobotanical, and historical evidence. Assign protein/energy category.
Stage 2 · Literature Search
Identify existing measurements for listed taxa. Note geographic and temporal offsets. Flag taxa requiring new primary measurement.
Stage 3 · Specimen Selection
Prioritise: (a) taxa with no local measurements; (b) isotopically extreme taxa; (c) high-abundance dietary contributors; (d) high-variability taxa (fish, omnivores). Confirm contemporaneity.
Stage 4 · Analytical Preparation & QC
Apply appropriate extraction protocols. Check collagen quality indicators: C:N ratio 2.9–3.6; yield > 1%; %C > 13%.
Stage 5 · Bivariate Visualisation
Plot sources + consumers. Run four-step diagnostic check. Address overlap before proceeding.
Stage 6 · ReSources Input Files
Calculate means and biologically informed SDs.
Summary: Seven Key Principles
Key Principles
No baseline, no inference. Mixing model outputs are only as reliable as the source end-members. Statistical sophistication cannot remedy inadequate inputs.
The baseline must precede analysis. Post-hoc baseline construction risks selection bias. Treat it as a prerequisite.
Local and contemporaneous measurements are always preferable. Regional δ15N variability is especially large and cannot be assumed away.
Replicate to characterise variability, not just to reduce error. Biological variability must be honestly propagated into ReSources source SDs.
All plausible sources must be included, including rare or isotopically extreme ones. A minor C4 or marine component omitted from the baseline will corrupt the entire dietary reconstruction.
Inspect source separation before running ReSources. If consumers fall outside the mixing polygon, the fault is in the baseline.
TEFs carry uncertainty. Select them with the same rigour as source end-members, and report sensitivity analyses.
Recommended Reading
Baseline data collection and quality
Britton, K., Müldner, G. & Bell, M. (2008) Stable isotope evidence for salt-marsh grazing in the Bronze Age Severn Estuary. J. Archaeol. Sci. 35, 2382–2393. [Multi-isotope baseline; δ34S for source discrimination]
Chenery et al. (2010) Strontium and stable isotope evidence for diet and mobility in Roman Gloucester. J. Archaeol. Sci. 37, 150–163. [Integrated baseline construction]
TEF values and variability
Hedges, R.E.M. & Reynard, L.M. (2007) Nitrogen isotopes and the trophic level of humans in archaeology. J. Archaeol. Sci. 34, 1240–1251. [Critical review — essential before applying Δ15N uncritically]
Best practice in mixing models
Phillips et al. (2014) Best practices for use of stable isotope mixing models in food-web studies. Can. J. Zoology 92, 823–835. [Definitive methodological reference]
Local vs. published baseline values
Styring, A.K., Sealy, J.C. & Evershed, R.P. (2010) Resolving bulk δ15N values via compound-specific amino acid analysis. Geochim. Cosmochim. Acta 74, 241–251. [Scale of local δ15N variability]
Module 3 complete
Proceed to Module 4: From measured collagen to dietary source valuess
Module 3: Baseline Isotopic Data