BayesianDepage (Python / PyMC)

A reproducible BEA-like (Bayesian eruption age) → BAD-like (Bayesian age–depth) workflow for single-zircon U–Pb datasets, implemented in Python with PyMC.

• BEA-like: infer eruption age from a population of zircon ages with analytical uncertainty (allowing older inherited/crystallization ages).
• BAD-like: build a monotonic age–depth model using eruption-age tie points (with optional depth uncertainty) and return credible intervals.

Goal: reproduce the workflow and result behavior (e.g., age–depth curve with credible intervals) used in Dai et al., not necessarily a bit-for-bit match to any author-specific/private implementation.

---

At a glance

• ✅ Works with CSV inputs (zircon.csv, tiepoints.csv)
• ✅ Outputs summary tables + a publication-ready age–depth plot
• ✅ Can be packaged as a Windows GUI .exe (PyInstaller)

---

Updates & release notes

Last updated: 2026-01-06
Current version: v0.1.0

Changelog

• v0.1.0 (2026-01-06)
  • Initial BEA→BAD pipeline (PyMC/NUTS)
  • Bootstrapped prior option for BEA
  • Monotonic BAD with optional depth uncertainty (default 0.03 m)
  • CSV outputs + age–depth figure

Tip: When you publish on GitHub, copy this section into your Release notes and keep the changelog here in sync.

---

Table of contents

• Project layout
• Quickstart
• Inputs
• Depth sign convention
• Outputs
• Windows GUI app (PyInstaller)
• Troubleshooting
• Model notes
• Citation / acknowledgement

---

Project layout

bea_bad_project/
  bea_bad/
    __init__.py
    __main__.py        # enables: python -m bea_bad ...
    cli.py             # pipeline orchestration
    dataio.py          # input reading & validation
    utils.py           # helpers (MAD, bootstrap prior, filtering)
    bea.py             # BEA-like eruption-age model per ash bed
    bad.py             # BAD-like age–depth model
    plot.py            # plotting/export
  zircon_depthdown.csv
  tiepoints_depthdown.csv
  out/                 # created after running

---

Quickstart

1) Install dependencies

Recommended (conda):

conda create -n bea python=3.11
conda activate bea
pip install numpy pandas pymc arviz matplotlib scipy

2) Run the pipeline

From the project root (the directory that contains the bea_bad/ folder):

python -m bea_bad --zircon zircon_depthdown.csv --tiepoints tiepoints_depthdown.csv --outdir out

Optional: query ages at specific depths

Create query_depths.csv:

depth_m
10
20
30

Run:

python -m bea_bad --zircon zircon_depthdown.csv --tiepoints tiepoints_depthdown.csv --query query_depths.csv --outdir out

---

Inputs

(1) zircon.csv (single-zircon U–Pb ages; one row per grain)

Required columns:

• ash_id : ash bed identifier (same for grains from the same bed)
• age_ma : age in Ma
• sigma_ma : 1σ uncertainty in Ma

Recommended columns:

• zircon_id : grain identifier (useful for traceability)
• is_discordant / omit_flag : if you already filtered discordant grains

Example:

ash_id,zircon_id,age_ma,sigma_ma
Gujiao:GJ-ASH-10,GJ10_z4s,251.710,0.0465
...

If your uncertainties are given as 2σ, convert before running:

• sigma_ma = age_2sigma_ma / 2

(2) tiepoints.csv (one row per ash bed)

Required columns:

• ash_id
• depth_m : stratigraphic depth (meters)

Optional:

• depth_sigma_m : depth uncertainty (default used by this workflow is 0.03 m ≈ 3 cm)

Example:

ash_id,depth_m,depth_sigma_m
Gujiao:GJ-ASH-10,50.8,0.03
...

---

Depth sign convention

The current BAD implementation assumes:

• Depth increases downward (larger depth_m = deeper)
• Age increases with depth (deeper = older)

If your raw depths are negative downward (e.g., -50 m is deeper than -10 m), convert to downward-positive before modeling:

• depth_down_m = -depth_m

The example files zircon_depthdown.csv and tiepoints_depthdown.csv are already converted to downward-positive depths.

---

Outputs

After a successful run, out/ will contain:

• bea_eruption_age_summary.csv
  BEA-like eruption-age estimates per ash bed (mean, sd, 95% interval, n grains used)

• bad_posterior.nc
  BAD-like posterior samples (ArviZ NetCDF)

• tiepoint_summary.csv
  Observed eruption ages vs. model-implied tie ages (with 95% intervals)

• query_age_summary.csv
  Age estimates for query depths (or an auto-generated grid)

• age_depth_model.png
  Age–depth curve with 95% credible band + tie points error bars

---

Windows GUI app (PyInstaller)

If you want a double-clickable Windows GUI (.exe):

1. Create a packaging environment (recommended):

conda create -n bea_pack python=3.11
conda activate bea_pack
pip install numpy pandas pymc arviz matplotlib scipy
pip install pyinstaller

2. Add a GUI entry file app_gui.py (Tkinter) that calls the pipeline.

3. Build the executable from the project root:

python -m PyInstaller --noconfirm --onefile --windowed --name BEA_BAD app_gui.py

Output:

• dist\BEA_BAD.exe

Notes:

• Scientific stacks can produce large executables (hundreds of MB). This is normal.
• --onefile is convenient, but a folder build (no --onefile) is sometimes more stable for distribution.

---

Troubleshooting

1) No module named bea_bad

Run from the directory that contains the bea_bad/ folder and ensure it has __init__.py and __main__.py.

Correct command pattern:

python -m bea_bad --zircon ... --tiepoints ... --outdir out

2) Age–depth curve goes the “wrong way” (deeper becomes younger)

This is almost always the depth sign convention. Ensure your depths are downward-positive (see above).

3) Sampling is extremely slow (PyTensor compiler warnings)

If you see warnings about missing g++, the model will still run but can be slower. On Windows + conda, you can install a toolchain:

conda install -c conda-forge m2w64-toolchain

---

Model notes

• BEA-like: eruption age E is inferred from zircon ages with analytical uncertainty, allowing grains to be older than eruption (a positive offset term). A bootstrapped prior (from resampling “youngest plausible ages”) is supported to stabilize inference.
• BAD-like: tie-point eruption ages (from BEA) are combined with stratigraphic depths (with optional depth uncertainty) under a monotonic, positive-rate age–depth model to produce an age–depth curve with credible intervals.

---

Citation / acknowledgement

If you use this workflow in your work, cite the relevant methodological foundations (e.g., BEA and BAD literature) and the study you are reproducing. This code is intended as a transparent, editable reference implementation for research workflows.