Data and code for “Dietary metal risk to the Persian gazelle (Gazella subgutturosa) near a desert highway: two-dimensional Monte Carlo and Bayesian hierarchical exposure models compared”

Description

Raw data and code for the manuscript "Dietary metal risk to the Persian gazelle (Gazella subgutturosa) near a desert highway: two-dimensional Monte Carlo and Bayesian hierarchical exposure models compared" (Belmaseb et al., submitted to Ecological Modelling). QUESTION: Do road-derived metals (Pb, Cd, Ni, Co, Cr) raise the dietary exposure of a wild herbivore to a level of concern, and does the conclusion depend on the model used? Expected: forage metals decline with distance from the road; benchmark uncertainty, not exposure, controls the risk conclusion. SAMPLING: Mouteh Wildlife Refuge, Esfahan Province, central Iran. Two transects (tr1, tr2; ~20 km apart) perpendicular to a highway, each with points at 0, 50 and 250 m. In mid-June, three replicates per point: surface soil (0-5 cm; n=18), forage of three species (Artemisia sieberi, Scariola orientalis, Launaea acanthodes; n=54, unwashed) and fresh gazelle feces (n=18). All matrices share Transect, Distance (m) and Replicate. ANALYSIS: ICP-MS; mg/kg dry weight, reported as measured (no substitution). Values are flagged OK/BDL against conservative reference limits from the laboratory's ISO/IEC 17025 certificate (AAS scopes, not ICP-MS detection limits). QC: soil field duplicates RPD <=15.8% (criterion 20%); standards recovery 97-106%. FILES: Supplementary_Data_EcolModel.xlsx (soil, plant, fecal, LOD and QA sheets); mc_model.py (2-D Monte Carlo, Sobol sensitivity); bayes_model.py (Bayesian hierarchical models); fig_concept.py; qa_qc.py; results/ (outputs, Figures 2-7, S1-S2). MODELS: (1) 2-D Monte Carlo: 3,000 simulated gazelles (variability) x 1,000 draws of intake rate and US EPA Eco-SSL toxicity reference values (uncertainty). Hazard Quotient HQ = dose/TRV; P(concern) = share of draws with 95th-percentile HQ >= 1. (2) Bayesian hierarchical model of ln concentration (species x distance, transect random effect), PyMC/NUTS. KEY FINDINGS: Forage at the road margin carried 1.2-1.9 times the concentrations at 250 m (Bayesian ratio, P>0.99); soil 1.2-1.4 times; fecal ratios ~1. HQ95 stayed far below 1 for Pb, Cd and Co (P(concern) <1.5%). Ni (2-D MC 24% near road, 17% far; Bayesian 32%, 24%) and Cr (12%, 8%; 17%, 11%) were the elements of concern. TRV uncertainty explained 70-96% of HQ variance (Sobol). Feces-implied diet was 0.1-1.0 x modelled forage, so exposure is unlikely underestimated. INTERPRETATION: HQ is a screening index, not a measured effect; HQ < 1 means the estimated dose is below the benchmark. Run times: ~5 min (mc_model.py), ~20 min (bayes_model.py); seeds are fixed. LIMITATIONS: one season, two transects, n=18 per matrix; reference limits only; generic Eco-SSL benchmarks; no field validation. Licence: CC BY 4.0 (data and code). Contact: H. R. Azimzadeh, Yazd University, hazimzadeh@yazd.ac.ir. Article DOI: to be added

Steps to reproduce

Python 3.13 with the package versions in requirements.txt. pip install -r requirements.txt python mc_model.py Supplementary_Data_EcolModel.xlsx results python fig_concept.py results python qa_qc.py Supplementary_Data_EcolModel.xlsx results python bayes_model.py Supplementary_Data_EcolModel.xlsx results Random seeds are fixed. mc_model.py takes about 5 minutes and bayes_model.py about 20 minutes; outputs and Figures 2-7, S1-S2 are written to results/. README.md maps every output file to the manuscript tables and figures.

Institutions

• Yazd University

  Yazd Province, Yazd

Categories

Funders

• Yazd University

  Yazd Province, Yazd

Licence