MixSIAR_Thesis

Goals

The purpose of this analysis is to use the MixSIAR package to implement a Bayesian hierarchical mixing model to estimate the mixing proportion of two species of co-dominant marsh plants, Schoenoplectus americanus and Spartina patens. As it is currently written, we estimate the mixing proportions using one fixed effect, environmental treatment, with four levels: high_sal/high_elev, high_sal/low_elev, low_sal/high_elev, low_sal/low_elev. We also include a continuous effect of seed year determined by ²¹⁰Pb dating of sediment layers.

Workflow

Files are organized and named in the order to be run sequentially (0-4).

• 0_data_assembly.R: This script cleans the raw biomass and isotope data and formats each
  • Inputs:
    • Belowground_Biomass.csv Biomass spreadsheet from HSK Thesis
    • BlueGenes2023_Belowground_Biomass - BG2023.csv Biomass spreadsheet from summer 2023 weighing effort
    • Compiled_Traits.csv Trait data from HSK Thesis
    • 220729_JPM.xls - CN sum.csv Smithsonian lab isotope data
    • 220801_JPM.xls - CN sum.csv Smithsonian lab isotope data
    • 220519_JPM.xls - CN sum.csv Smithsonian lab isotope data
    • CEST_Sample_Runs.csv CEST lab isotope data - all runs, not including replicate testing
    • Trial_Carbon_Data_unedited.csv CEST lab isotope data - replicate trials
    • Corr_Trial_COMP.csv CEST lab isotope data - replicate trials
    • Milled_replicates.csv CEST lab isotope data - replicate trials
  • Outputs:
    • marsh_source_all.csv raw source isotopes formatted to MixSIAR specs, 151 observations of 2 variables
    • Compiled_Traits_appended.csv traits with added weights and root-shoot-ratios, 672 observations of 35 variables
    • training.csv mixture distribution for training MixSIAR model, 274 observations of 17 variables
    • bgb_biomass.csv all weights across each layer, 1692 observations of 11 variables
    • holdout_DONOTTOUCH.csv holdout data, 49 observations of 36 variables
• 1_run_model.R: Describing model inputs and running model, adjust "run" in run_model for MCMC parameters
  • Inputs:
    • training.csv
    • marsh_source_all.csv
    • marsh-discr null trophic enrichment factors
  • Output
    • jags.Env.rds save full jags model run as RDS object
    • figs/isospace_env_plot d13C plot
    • prior_env_plot uniform uninformative distribution
    • diagnostics.pdf pdf for trace, density, and running means plot for each parameter
    • diagnostics.txt text file containing Gelman-Rubin, Geweke, and Heidelberger-Welch diagnostics
    • summary_statistics.txt abbreviated file for main model output, group estimates, and DIC
    • MixSIAR_model text file with model description
    • posterior_density_high_sal folder containing posterior density plots for group characteristics
    • posterior_density_low_sal folder containing posterior density plots for group characteristics
• 2_biomass_calc.R: Calculate mixing proportion and weight densities, some plotting
  • Inputs:
    • training.csv
    • jags.1.rds
    • bgb_biomass.csv
  • Outputs:
    • bgb_biomass_mod.rds modified to include mixing proportions and weight density
• 3_model_validation temporary file holding space until model validation is pursued
  • Outputs: none
• 4_graphing.R: Basic plotting functions for factors
  • Inputs:
    • bgb_biomass.rds
    • Compiled_Traits_appended.csv
  • Outputs: none

File Structure

├── data                                  # raw biomass and isotope data
│   ├── Biomass                           # Biomass weights from two efforts
│   ├── Holdout                           # Folder to place holdout
│   └── Isotope                           # Isotope data from CEST and ND
├── figs                                  # folder for produced figures
├── output                                # intermediary files created during analysis
│   ├── data                              # modified data products
│   ├── posterior_density_high_salinity   # Empty folder to hold mixsiar outputs 
│   ├── posterior_density_high_salinity   # Empty folder to hold mixsiar outputs
│   └── jags.normal.rds                   # JAGS product used in analysis, takes a decently long time to run so provided here
├── src                                   # supporting functions
└── README.md

R Version

All code was run in R version 4.4.0 (2024-04-24). Necessary packages to run script are identified at the top of each script.

• dplyr v1.1.4
• ggplot2 v3.5.1
• tidyverse v2.0.0
• MixSIAR v3.1.12