ariel-data-preprocessing 1.5

Signal correction module for Ariel Data Challenge 2025.

Ariel Data Preprocessing

This module contains the complete FGS1 and AIRS-CH0 signal data preprocessing pipeline for the Ariel Data Challenge.

Overview

The DataProcessor class provides an integrated pipeline that combines:

1. Signal correction - Complete 6-step calibration pipeline for raw telescope data
2. Signal extraction - Intelligent data reduction and spectral signal extraction

This unified approach transforms raw Ariel telescope data directly into extracted, science-ready spectral time series in a single processing step.

1. Signal correction

Implements the complete six-step signal correction pipeline outlined in the Calibrating and Binning Ariel Data notebook shared by the contest organizers. This module transforms raw Ariel telescope data into science-ready corrected signals.

Processing Pipeline:

1. ADC Conversion - Convert raw counts to physical units
2. Hot/Dead Pixel Masking - Remove problematic detector pixels
3. Linearity Correction - Account for non-linear detector response
4. Dark Current Subtraction - Remove thermal background noise
5. Correlated Double Sampling (CDS) - Reduce read noise via paired exposures
6. Flat Field Correction - Normalize pixel-to-pixel sensitivity variations

Key Features:

• Multiprocessing support for parallel planet processing
• Optional FGS1 downsampling to match AIRS-CH0 cadence (83% data reduction)
• Configurable processing steps (enable/disable individual corrections)
• HDF5 output for efficient large dataset storage

See the following notebooks for implementation details and performance analysis:

1. Signal correction
2. Signal correction optimization

Example use:

from ariel_data_preprocessing.data_preprocessing import DataProcessor

data_processor = DataProcessor(
    input_data_path='data/raw',
    output_data_path='data/corrected',
    n_cpus=4,
    downsample_fgs=True,
    n_planets=100
)

data_processor.run()

The signal correction pipeline will write the corrected frames and hot/dead pixel masks as an HDF5 archive called train.h5 by default with the following structure:

    train.h5:
    │
    ├── planet_id_1/
    │   ├── signal  # Combined corrected/extracted spectral time series
    │   └── mask    # Dead/hot pixel mask for spectra
    |
    ├── planet_id_2/
    │   ├── signal  
    │   └── mask    
    |
    └── planet_id_n/

2. Signal extraction

Complete extraction pipeline integrated with DataProcessor

The signal extraction functionality is integrated within the DataProcessor class, which handles both signal correction and extraction in a unified pipeline. This approach transforms 3D detector arrays into focused time series suitable for exoplanet atmospheric analysis.

Processing Features:

• AIRS-CH0 Extraction: Selects brightest detector rows containing spectral traces, sums to create 1D spectra per frame
• FGS1 Extraction: Uses 2D block extraction to identify signal regions, collapses to single brightness value per frame
• Combined Output: Merges FGS1 and AIRS-CH0 signals (FGS1 as first column for transit detection)
• Adaptive Thresholding: Automatically selects signal-bearing pixels based on configurable intensity thresholds
• Optional Smoothing: Applies moving average filtering across wavelengths to reduce noise
• Massive Data Reduction: Achieves ~97-98% volume reduction while preserving transit signals

Key Benefits:

• Dramatically faster downstream processing due to reduced data volume
• Improved signal-to-noise ratio by focusing on high-signal detector regions
• Preserved exoplanet transit signatures with cleaner temporal structure
• Unified processing for both instrument types

See the following notebooks for implementation details and analysis:

1. AIRS-CH0 signal extraction
2. FGS1 signal extraction
3. Wavelength smoothing

Example usage:

The signal extraction is performed as part of the integrated DataProcessor pipeline:

from ariel_data_preprocessing.data_preprocessing import DataProcessor

data_processor = DataProcessor(
    input_data_path='data/raw',
    output_data_path='data/processed',
    inclusion_threshold=0.75,
    smooth=True,
    smoothing_window=200
)

data_processor.run()

Output data will be written to train.h5 by default in the directory passed to output_data_path. The structure of the HDF5 archive combines both AIRS-CH0 and FGS1 signals:

    train.h5
    |
    ├── planet_1/
    │   ├── signal  # Shape: (n_frames, n_wavelengths + 1) - FGS1 + AIRS-CH0
    │   └── mask    # Shape: (n_wavelengths + 1,) - combined mask
    │
    ├── planet_2/
    │   ├── signal  # Shape: (n_frames, n_wavelengths + 1) - FGS1 + AIRS-CH0  
    │   └── mask    # Shape: (n_wavelengths + 1,) - combined mask
    │
    └── planet_n/

Note: The first column of the signal dataset contains the extracted FGS1 brightness time series, followed by the AIRS-CH0 spectral channels. This structure facilitates easy transit detection using FGS1 data while providing wavelength-dependent atmospheric information from AIRS-CH0.