CERES3

A Python 3 port of CERES — a collection of fully automated pipelines for the reduction, extraction and analysis of echelle spectrograph data. CERES3 produces wavelength-calibrated spectra, precise radial velocities, bisector spans, and chromospheric activity indicators (S-index, H-alpha, HeI, NaI D1/D2).

If you use this code, please cite Brahm et al 2017.

Installation

System prerequisites

# Debian/Ubuntu
sudo apt install gfortran libgsl-dev libopenblas-dev build-essential pkg-config

# macOS (Homebrew)
brew install gcc gsl openblas

Install CERES3

pip install ceres3

Or from source:

git clone https://github.com/jvines/ceres3.git
cd ceres3
pip install .

This compiles the C/Fortran extensions (Marsh optimal extraction, CCF cross-correlation) automatically.

Quick start

# Step 1: Reduce calibration frames (bias, flat, ThAr)
ceres3 feros /path/to/raw/calibrations --is-calib

# Step 2: Reduce science frames
ceres3 feros /path/to/raw/science --calib-dir /path/to/calibrations_red/

Output goes to {input_dir}_red/ by default. Override with -o /custom/path.

Supported instruments

Instrument	Telescope	Pipeline
ARCES	APO 3.5m	ceres3 arces
CAFE	CAHA 2.2m	ceres3 cafe
Coralie	Euler 1.2m	ceres3 coralie
Echelle	DuPont 2.5m	ceres3 dupont
ESPaDOnS	CFHT 3.6m	ceres3 espadons
FEROS	MPG/ESO 2.2m	ceres3 feros
FIDEOS	ESO 1.0m	ceres3 fideos
FIES	NOT 2.5m	ceres3 fies
HARPS	ESO 3.6m	ceres3 harps
HIRES	Keck 10m	ceres3 hires
MIKE	Magellan 6.5m	ceres3 mike
PFS	Magellan 6.5m	ceres3 pfs
PUCHEROS	PUC 0.5m	ceres3 pucheros
VBT Echelle	VBT 2.3m	ceres3 vbt

CLI options

ceres3 <instrument> <input_dir> [options]

Options:
  -o, --output-dir DIR     Output directory (default: {input_dir}_red/)
  --npools N               CPU cores to use (default: 4)
  --nsigmas N              Order detection sigma threshold (default: 5.0)
  --is-calib               Process as calibration data
  --calib-dir DIR          Path to pre-computed calibration directory
  --do-class               Enable spectral classification (Teff, logg, [Fe/H])
  --target NAME            Process only this target (default: all)

Python API

# Import and run a pipeline directly
from ceres3.instruments import get_pipeline_module
feros = get_pipeline_module('feros')
# Pipeline runs on import (module-level code)

Data products

All final products are in the proc/ subdirectory of the output directory.

FITS spectrum cube (*_sp.fits)

Shape: (11, n_orders, n_pixels) — 11 data layers per echelle order:

Layer	Contents
0	Wavelength (Angstroms)
1	Extracted flux
2	Flux error (1/sqrt(variance))
3	Blaze-corrected flux
4	Blaze-corrected flux error
5	Continuum-normalized flux
6	Continuum-normalized flux error
7	Estimated continuum
8	Signal-to-noise ratio per pixel
9	Continuum-normalized flux × d(wavelength)/d(pixel)
10	Error of layer 9

Merged 1D spectrum (*_1d.fits)

A single FITS binary table with columns WAVELENGTH, FLUX, ERROR — the echelle orders merged into one continuous spectrum using S/N-weighted crossover. This is the rest-frame, continuum-normalized spectrum suitable for input to spectral analysis tools (e.g., SPECIES).

FITS header keywords

Keyword	Description
RV	Radial velocity (km/s)
RV_E	RV error (km/s)
BS	Bisector span (km/s)
BS_E	BS error (km/s)
FWHM	CCF full width at half maximum (km/s)
XC_MIN	CCF contrast (depth of minimum)
BJD_OUT	Barycentric Julian Date
SNR	Signal-to-noise at ~5130 A
S_INDEX	Ca H+K S-index
HALPHA	H-alpha activity index
HEI	HeI 5876 activity index
NAI_D1D2	NaI D1+D2 activity index
SPEC1D	Path to merged 1D spectrum
INST	Instrument name
PIPELINE	CERES

results.txt

Tab-separated summary with columns: object name, BJD, RV, RV error, bisector span, BS error, FWHM, instrument, pipeline, resolving power, Teff, log(g), [Fe/H], v*sin(i), XC_min, CCF dispersion, exposure time, SNR, S-index, H-alpha, HeI, NaI D1/D2, CCF plot path.

CCF plots (*.pdf)

PDF files showing the cross-correlation function and Gaussian fit for each target.

Auxiliary files

Place these in the raw data directory to customize the reduction:

reffile.txt

Comma-separated file with target-specific parameters:

HD157347,17:22:51.28809,-02:23:17.4297,49.39,-107.16,1,G2,4.0
HD32147,05:00:48.99977,-05:45:13.2303,550.12,-1109.23,1,G2,4.0

Columns: name, RA (J2000), Dec (J2000), PM_RA (mas/yr), PM_Dec (mas/yr), use_coords (0/1), CCF mask (G2/K5/M2), velocity width (km/s).

bad_files.txt

List filenames (one per line) to exclude from processing.

moon_corr.txt

List filenames for which the CCF fit includes a double Gaussian to correct for scattered moonlight.

Spectral classification

To estimate atmospheric parameters (Teff, log(g), [Fe/H], v*sin(i)), use --do-class. This requires the Coelho et al. (2005) model grid:

mkdir -p ~/.ceres3/COELHO_MODELS
cd ~/.ceres3/COELHO_MODELS
wget http://www.astro.puc.cl/~rbrahm/coelho_05_red4_R40.tar.gz
tar -xf coelho_05_red4_R40.tar.gz

What changed from the original CERES

• Python 3.10+ (was Python 2.7)
• No R dependency (rpy2 was never used)
• No SWIG / SOFA / manual ephemeris updates — astropy handles barycentric corrections with DE440 ephemeris (auto-downloaded)
• Inline activity indicators — S-index, H-alpha, HeI, NaI D1/D2 computed during reduction (no separate post-processing step)
• Merged 1D spectrum output — rest-frame continuum-normalized spectrum as a FITS binary table
• Batch CCF — Fortran subroutine processes all velocity steps in one call (3-4x faster)
• pip installable — pip install ceres3 compiles extensions automatically

Authors

Rafael Brahm, Andres Jordan, Nestor Espinoza, Jose Vines.

License

MIT