juliensimon/galah-dr4-stellar-abundances

---
license: cc-by-4.0
pretty_name: "GALAH DR4 — Stellar Abundances for 917k Stars"
language:
- en
description: "The fourth data release of the GALactic Archaeology with HERMES (GALAH) survey, providing radial velocities, stellar parameters, and up to 31 elemental abundances for 917,588 stars observed with the H"
task_categories:
- tabular-classification
- tabular-regression
tags:
- space
- stars
- spectroscopy
- galah
- abundances
- astronomy
- open-data
- tabular-data
- parquet
size_categories:
- 100K<n<1M
configs:
- config_name: default
data_files:
- split: train
path: data/galah_dr4_allstar.parquet
default: true
---

# GALAH DR4 — Stellar Abundances for 917k Stars

<div align="center">
<img src="banner.jpg" alt="A youthful globular star cluster observed by the Hubble Space Telescope" width="400">
<p><em>Credit: NASA/ESA/Hubble</em></p>
</div>

*Part of a [dataset collection](https://huggingface.co/collections/juliensimon/astronomy-datasets-69c24caf2f17e36128946743) on Hugging Face.*

## Dataset description

The fourth data release of the GALactic Archaeology with HERMES (GALAH) survey, providing radial velocities, stellar parameters, and up to 31 elemental abundances for 917,588 stars observed with the HERMES spectrograph on the Anglo-Australian Telescope.

GALAH DR4 is one of the largest stellar spectroscopic surveys, designed to unravel the formation and evolution of the Milky Way through chemical tagging. Each star has high-resolution spectra decomposed into fundamental stellar parameters and individual elemental abundances spanning light elements, alpha-elements, iron-peak elements, and neutron-capture elements.

GALAH was specifically designed for chemical tagging — the idea that stars born in the same molecular cloud retain a unique multi-dimensional chemical fingerprint that persists long after the birth cluster has dispersed. The HERMES spectrograph delivers four non-contiguous optical wavelength channels at R ~ 28,000, capturing lines of light elements (Li, C, N, O), alpha-elements (Mg, Si, Ca, Ti), iron-peak elements (Sc, V, Cr, Mn, Fe, Co, Ni, Cu, Zn), and neutron-capture elements (Rb, Sr, Y, Zr, Mo, Ba, La, Ce, Nd, Ru, Sm, Eu) — up to 31 distinct abundance dimensions per star.

DR4 represents a major advance over DR3, incorporating improved spectral analysis techniques, better treatment of non-LTE effects for critical elements, and cross-matching with Gaia DR3 for precise astrometric information. The inclusion of both s-process elements (Ba, La, Ce from AGB nucleosynthesis) and r-process elements (Eu from neutron star mergers) makes GALAH uniquely powerful for constraining the sites and timescales of heavy element production in the Milky Way.

## Schema

| Column | Type | Description | Sample | Null % |
|--------|------|-------------|--------|--------|
| `sobject_id` | object | GALAH spectroscopic observation identifier (unique per exposure); format encodes field and fiber number | 170910004101274 | 0.0% |
| `tmass_id` | object | 2MASS photometric catalog cross-identifier (e.g. 'J12345678+1234567'); null if no 2MASS match | 00000011+0522500 | 0.0% |
| `gaiadr3_source_id` | object | Gaia DR3 astrometric source identifier; enables cross-match for precise positions, proper motions, and parallaxes; null if unmatched | 2745049530295263232 | 0.0% |
| `ra` | float64 | Right ascension, ICRS J2000.0, in decimal degrees (0-360) | 0.0005000000237487257 | 0.0% |
| `dec` | float64 | Declination, ICRS J2000.0, in decimal degrees (-90 to +90) | 5.380555629730225 | 0.0% |
| `teff_k` | float32 | Effective temperature in Kelvin from spectral synthesis; GALAH targets FGK stars, typical range 4000-7500 K; uncertainty ~100 K; null if spectral pipeline failed (flag_sp > 0) | 4486.2554 | 1.2% |
| `logg` | float32 | Log surface gravity in cgs (log cm/s²); main sequence dwarfs: 4.0-5.0, subgiants: 3.5-4.5, red giants: 1.5-3.5; null if flag_sp > 0 | 4.657445 | 1.2% |
| `fe_h_dex` | float32 | [Fe/H] iron abundance in dex relative to solar; GALAH surveys -2.5 to +0.5 dex; typical uncertainty ~0.1 dex; null if flag_sp > 0 | -0.3963167 | 1.2% |
| `vmic` | float32 | Microturbulence velocity in km/s; internal parameter of the spectral model capturing small-scale turbulent broadening; typical range 0.5-2.0 km/s | 0.7832511 | 1.2% |
| `vsini` | float32 | Projected rotational velocity v sin i in km/s; slow rotators (FGK dwarfs) typically < 10 km/s; null for stars where rotation is unresolved at R~28,000 | 6.9111066 | 1.2% |
| `radial_velocity_kms` | float32 | Barycentric radial velocity in km/s from cross-correlation; precision ~0.1 km/s; null for very low S/N spectra | 10.483474 | 1.2% |
| `radial_velocity_comp2_kms` | float32 | Barycentric radial velocity of a detected binary companion in km/s; non-null only for double-lined spectroscopic binaries (SB2) | 186.25742 | 96.5% |
| `snr_px_ccd1` | float32 | Signal-to-noise ratio per pixel for HERMES CCD 1 (blue channel, ~4713-4903 Å); drives which light-element abundances can be measured | 26.964375 | 1.9% |
| `snr_px_ccd2` | float32 | Signal-to-noise ratio per pixel for HERMES CCD 2 (green channel, ~5648-5873 Å); drives which iron-peak abundances can be measured | 45.697865 | 0.0% |
| `snr_px_ccd3` | float32 | Signal-to-noise ratio per pixel for HERMES CCD 3 (red channel, ~6478-6737 Å); drives which alpha-element abundances can be measured | 75.889725 | 0.6% |
| `snr_px_ccd4` | float32 | Signal-to-noise ratio per pixel for HERMES CCD 4 (IR channel, ~7585-7887 Å); drives which neutron-capture abundances can be measured | 72.90598 | 2.9% |
| `flag_sp` | Int64 | Spectroscopic analysis quality flag; 0 = good stellar parameters; >0 encodes specific problems (binary contamination, emission, grid edge); use flag_sp == 0 for clean samples | 0 | 0.0% |
| `flag_red` | Int64 | Reduction pipeline quality flag; 0 = successful reduction; >0 indicates issues with sky subtraction, cross-talk, or cosmic rays | 0 | 0.0% |
| `c_fe` | float32 | [C/Fe] carbon abundance ratio in dex; elevated in carbon-enhanced metal-poor (CEMP) stars | 0.05938914 | 1.4% |
| `n_fe` | float32 | [N/Fe] nitrogen abundance ratio in dex; a tracer of CNO cycling and AGB dredge-up | -0.3913465 | 24.8% |
| `o_fe` | float32 | [O/Fe] oxygen abundance ratio in dex; key alpha-element tracing core-collapse supernova enrichment | -0.19825687 | 4.7% |
| `na_fe` | float32 | [Na/Fe] sodium abundance ratio in dex; anti-correlates with O in globular cluster stars | -0.16347441 | 2.6% |
| `al_fe` | float32 | [Al/Fe] aluminium abundance ratio in dex; traces Mg-Al chain proton captures in massive stars | -0.048628658 | 7.4% |
| `k_fe` | float32 | [K/Fe] potassium abundance ratio in dex; sensitive to non-LTE effects; limited by spectral coverage | -0.03559644 | 6.2% |
| `mg_fe` | float32 | [Mg/Fe] magnesium abundance ratio in dex; primary alpha-element; high in old, metal-poor disk stars; decreases with increasing [Fe/H] due to Type Ia SNe iron contribution | 0.3087402 | 2.1% |
| `si_fe` | float32 | [Si/Fe] silicon abundance ratio in dex; alpha-element; co-produced with Mg in core-collapse supernovae | 0.19373856 | 3.5% |
| `ca_fe` | float32 | [Ca/Fe] calcium abundance ratio in dex; alpha-element; traces both core-collapse and Type Ia supernova nucleosynthesis | 0.12320362 | 1.9% |
| `ti_fe` | float32 | [Ti/Fe] titanium abundance ratio in dex; odd alpha-element; useful for separating thin disk, thick disk, and halo populations | 0.3395407 | 1.6% |
| `sc_fe` | float32 | [Sc/Fe] scandium abundance ratio in dex; iron-peak element; produced mainly in core-collapse supernovae | 0.01370742 | 2.0% |
| `v_fe` | float32 | [V/Fe] vanadium abundance ratio in dex; iron-peak element; constrains explosive nucleosynthesis models | 0.32754487 | 10.6% |
| `cr_fe` | float32 | [Cr/Fe] chromium abundance ratio in dex; iron-peak element with known non-LTE corrections required | 0.1857635 | 1.6% |
| `mn_fe` | float32 | [Mn/Fe] manganese abundance ratio in dex; traces Type Ia supernova contribution (Mn is overproduced in Chandrasekhar-mass SNe Ia) | -0.023918502 | 1.9% |
| `co_fe` | float32 | [Co/Fe] cobalt abundance ratio in dex; iron-peak element sensitive to neutron excess in the explosive burning region | 0.13685165 | 8.6% |
| `ni_fe` | float32 | [Ni/Fe] nickel abundance ratio in dex; closely follows Fe; used to distinguish thick-disk from halo stars | -0.123810135 | 1.7% |
| `cu_fe` | float32 | [Cu/Fe] copper abundance ratio in dex; iron-peak element with significant s-process contribution | 0.16626377 | 9.5% |
| `zn_fe` | float32 | [Zn/Fe] zinc abundance ratio in dex; bridges iron-peak and neutron-capture elements; useful metallicity probe | 0.087473795 | 4.9% |
| `rb_fe` | float32 | [Rb/Fe] rubidium abundance ratio in dex; s-process element; traces AGB stellar nucleosynthesis | -0.008337121 | 66.0% |
| `sr_fe` | float32 | [Sr/Fe] strontium abundance ratio in dex; light s-process element; also has r-process and charged-particle process contributions | 0.035072897 | 69.8% |
| `y_fe` | float32 | [Y/Fe] yttrium abundance ratio in dex; s-process element with Ba/Y ratio used to age-date stellar populations | 0.11842539 | 1.7% |
| `zr_fe` | float32 | [Zr/Fe] zirconium abundance ratio in dex; s-process element co-produced with Y and Sr | 0.11218737 | 29.0% |
| `mo_fe` | float32 | [Mo/Fe] molybdenum abundance ratio in dex; neutron-capture element with both s- and r-process origin | -0.14827245 | 83.9% |
| `ba_fe` | float32 | [Ba/Fe] barium abundance ratio in dex; dominant s-process tracer; high in AGB-enriched stars and young thin-disk stars | -0.090060145 | 2.0% |
| `la_fe` | float32 | [La/Fe] lanthanum abundance ratio in dex; s-process element; La/Eu ratio distinguishes s- from r-process enrichment | 0.061767366 | 26.8% |
| `ce_fe` | float32 | [Ce/Fe] cerium abundance ratio in dex; s-process element produced in low-mass AGB stars | -0.025505573 | 28.7% |
| `nd_fe` | float32 | [Nd/Fe] neodymium abundance ratio in dex; mixed s- and r-process origin | -0.1502057 | 12.2% |
| `ru_fe` | float32 | [Ru/Fe] ruthenium abundance ratio in dex; primarily r-process origin; rare to measure in stellar spectra | -0.30789047 | 76.8% |
| `sm_fe` | float32 | [Sm/Fe] samarium abundance ratio in dex; r-process dominated element; traces neutron star merger enrichment | -0.017710395 | 49.8% |
| `eu_fe` | float32 | [Eu/Fe] europium abundance ratio in dex; the cleanest r-process tracer; high in metal-poor halo stars; r-process enrichment from neutron star mergers | -0.08262927 | 69.8% |
| `n_abundances` | Int64 | Count of non-null [X/Fe] abundance measurements for this star; ranges 0-31; derived column useful for selecting well-characterised stars | 27 | 0.0% |
| `snr_mean` | float32 | Mean S/N per pixel averaged across all four HERMES CCDs; derived column; stars with snr_mean < 30 have fewer reliable abundance measurements | 55.364487 | 0.0% |

## Quick stats

- **917,588** stars observed with HERMES spectrograph
- **906,689** stars with radial velocity measurements
- **906,432** stars with at least one elemental abundance (99%)
- **30** elemental abundance columns ([X/Fe]), median **24** per star
- **663,075** stars with clean spectroscopic flags (flag_sp == 0)
- Median SNR across 4 HERMES CCDs: **48.1** per pixel

## Usage

```python
from datasets import load_dataset
import matplotlib.pyplot as plt

ds = load_dataset("juliensimon/galah-dr4-stellar-abundances", split="train")
df = ds.to_pandas()

# Kiel diagram (Teff vs logg) coloured by [Fe/H] — shows stellar populations
best = df[(df["flag_sp"] == 0) & df["teff_k"].notna() & df["logg"].notna()]
sample = best.sample(min(50_000, len(best)), random_state=42)

sc = plt.scatter(sample["teff_k"], sample["logg"],
c=sample["fe_h_dex"], s=0.1, cmap="coolwarm",
vmin=-1.5, vmax=0.5, alpha=0.6)
plt.gca().invert_xaxis()
plt.gca().invert_yaxis()
plt.xlabel("Effective Temperature (K)")
plt.ylabel("log g (dex)")
plt.title("GALAH DR4 Kiel Diagram")
plt.colorbar(sc, label="[Fe/H] (dex)")
plt.tight_layout()
plt.show()

# Abundance pattern: alpha-element enhancement vs metallicity
alpha_cols = ["mg_fe", "si_fe", "ca_fe", "ti_fe"]
best["alpha_fe"] = best[alpha_cols].mean(axis=1)
sub = best.dropna(subset=["fe_h_dex", "alpha_fe"]).sample(30_000, random_state=0)
plt.figure()
plt.scatter(sub["fe_h_dex"], sub["alpha_fe"], s=0.1, alpha=0.3, c="steelblue")
plt.axhline(0, color="gray", lw=0.5, ls="--")
plt.xlabel("[Fe/H] (dex)")
plt.ylabel("[alpha/Fe] (dex)")
plt.title("Alpha-element Enhancement vs Metallicity")
plt.tight_layout()
plt.show()
```

## Data source

https://www.galah-survey.org/dr4/

## Update schedule

Static dataset — uploaded once from the DR4 release catalog

## Related datasets

- [juliensimon/apogee-dr17-stellar-abundances](https://huggingface.co/datasets/juliensimon/apogee-dr17-stellar-abundances)

- [juliensimon/hipparcos-catalog](https://huggingface.co/datasets/juliensimon/hipparcos-catalog)

- [juliensimon/pulsar-catalog](https://huggingface.co/datasets/juliensimon/pulsar-catalog)

## Citation

```bibtex
@dataset{galah_dr4_stellar_abundances,
title = {GALAH DR4 — Stellar Abundances for 917k Stars},
author = {juliensimon},
year = {2026},
url = {https://huggingface.co/datasets/juliensimon/galah-dr4-stellar-abundances},
publisher = {Hugging Face}
}
```

## License

[CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/)