Source-linked AI summary
Pandeia: A Multi-mission Exposure Time Calculator for JWST and WFIRST
Klaus M. Pontoppidan, Timothy E. Pickering, Victoria G. Laidler, Karoline Gilbert, Christopher D. Sontag, Christine Slocum, Mark J. Sienkiewicz, Christopher Hanley, Nicholas M. Earl, Laurent Pueyo, Swara Ravindranath, Diane M. Karakla, Massimo Robberto, Alberto Noriega-Crespo, Elizabeth A. Barker
TL;DR
Exposure-time calculators support users as they explore observatory capabilities and prepare observing proposals, but JWST’s data-rich and high-contrast modes call for detailed modeling. Pandeia uses a pixel-based engine to model signal and noise across observing modes; benchmark comparisons were expected to match instrument-team signal-to-noise estimates to about 10%, while reference-data verification remained ongoing.
Problem
JWST’s information-rich observing modes and high-signal-to-noise or high-contrast observations require accurate calculations, while exposure-time calculators also need to help users explore capabilities and prepare proposals.
Source: §1 · p. 1Method
Pandeia uses a pixel-based approach to model detector signal and noise for complex scenes and supports JWST observing modes, including MULTIACCUM readouts and correlated-noise effects.
Source: §1 · pp. 1–2Results
Benchmark comparisons were generally required to match instrument-team sensitivity calculations to about 10% in derived signal-to-noise ratios, while verification of reference data remained ongoing.
Source: §8 · p. 13Takeaways & Limitations
Pandeia’s data-driven design supports JWST modes and has been extended to WFIRST wide-field imaging and integral-field spectroscopy.
Source: §1 · p. 2Pandeia is not a complete instrument simulator: it uses small postage-stamp fields, and reference-data verification was still ongoing.
Source: §2.2 · p. 3, §8 · p. 13
Abstract
from arXiv · showhide
Pandeia is the exposure time calculator (ETC) system developed for the James Webb Space Telescope (JWST) that will be used for creating JWST proposals. It includes a simulation-hybrid Python engine that calculates the two-dimensional pixel-by-pixel signal and noise properties of the JWST instruments. This allows for appropriate handling of realistic point spread functions, MULTIACCUM detector readouts, correlated detector readnoise, and multiple photometric and spectral extraction strategies. Pandeia includes support for all the JWST observing modes, including imaging, slitted/slitless spectroscopy, integral field spectroscopy, and coronagraphy. Its highly modular, data-driven design makes it easily adaptable to other observatories. An implementation for use with WFIRST is also available.
1. INTRODUCTION
Pandeia is designed to help users explore JWST observing concepts and prepare proposals with accurate, usable exposure-time calculations. Its pixel-based, modular engine supports demanding JWST modes and has also been adapted for selected WFIRST instruments.
1. INTRODUCTION
- Pandeia aims to combine accurate instrument modeling with an intuitive, powerful interface for exploring observing concepts and preparing proposals.Source: §1 · p. 1
- The ETC is intended to support information-rich modes and high-signal-to-noise or high-contrast observations, including IFU, NIRSpec MSA, exoplanet spectroscopy, coronagraphy, and non-redundant aperture masking.Source: §1 · p. 1
- Pandeia models two-dimensional detector pixels to support complex scenes and reproduce steps in JWST calibration and reduction, improving accuracy and interpretability.Calculations focus on small postage-stamp regions and can take as little as a few seconds, depending on observing mode.Source: §1 · pp. 1–2
- Pandeia supports a wide range of JWST imaging, coronagraphic, and spectroscopic modes, including MULTIACCUM readouts and correlated noise and inter-pixel capacitance effects.Source: §1 · p. 2
- Pandeia’s feature set includes advanced astronomical scene creation and interfaces to the full JWST background model.Source: §1 · p. 2
- Its modular, data-driven design supports adaptation beyond JWST, and an implementation is available for WFIRST wide-field imaging and integral-field spectroscopy.Source: §1 · p. 2
2. BASIC PHILOSOPHY AND CONCEPT
Pandeia’s pixel-based approach addresses detector and observing-strategy effects that simpler models may not capture, while its speed and realism are balanced against explicit scope limits.
2.1 Features of a Pixel-based Approach
- Pixel-level modeling matters because source flux, MULTIACCUM exposure time, correlated noise, and inter-pixel coupling can vary across pixels and affect signal-to-noise estimates.Source: §2.1 · p. 2
- Pixel-based calculations let users examine how background-subtraction strategy and contaminating sources in complex scenes affect measured signal-to-noise.Source: §2.1 · p. 2
2.2 Scope of Pandeia
- Pandeia prioritizes adequate signal-to-noise accuracy and calculation speed over complete instrument simulation, using postage-stamp fields rather than full instrument fields of view.The scope reflects the data volume of MULTIACCUM exposures and the need for users to explore parameter space efficiently.Source: §2.2 · p. 3
- Pandeia assumes linear pixel arithmetic and does not explicitly model optical field distortion or exact instrument-aperture boundaries, which are typically accessible from APT.Source: §2.2 · p. 3
- Pandeia uses precomputed PSFs and does not model field-dependent PSF variation except in coronagraphic modes.Source: §2.2 · p. 3
- Because field distortion is ignored, Pandeia does not generally determine individual pixel positions relative to sources and assumes dithers mitigate bad pixels; stochastic effects are represented on average.Source: §2.2 · p. 3
3. IMPLEMENTATION
Pandeia represents astronomical scenes as spatial and spectral brightness distributions, then realizes them on an observing-mode pixel grid. Source construction and wavelength-dependent PSFs carry that scene into the detector-oriented calculation.
3.1 Astronomical Scenes
- Scenes combine point or extended sources with spectra; each extended source has one normalized spectrum, while multiple sources can represent richer spatio-spectral structure.A calculation defaults to a single on-axis point source, while more complex scene configurations are available when needed.Source: §3.1 · p. 3
- Scene stamps are typically 3–20 arcseconds across, limiting computation; modeling the full instrument field may require a full-fledged simulator.The stamp size balances computational time and typical use cases.Source: §3.1 · pp. 3–4
- The abstract scene is instrument-independent and shareable across modes; realization samples it on the pixel grid relevant to the observing mode.Realized scenes are spatially subsampled to avoid undersampling and aliasing while allowing efficient resampling to detector scale.Source: §3.1 · p. 4
3.2 The Scene Coordinate System
- The realized scene is rotated to the aperture angle, with ξ along dispersion and η cross-dispersion for spectroscopy; Pandeia does not report the sky rotation angle.The pixel grid is assumed aligned with the aperture angle, generally a good approximation.Source: §3.2 · p. 4
3.3 Source Creation
- Sources specify spatial extent and offset plus a spectrum; available spectra include templates and user inputs, renormalizable to a reference flux density or magnitude.The scene’s location relative to the aperture is implicit through the PSF and configuration, and is generally assumed centered.Source: §3.3 · pp. 4–5
3.4 Spectral Lines and other Spectral Modifications
- Spectral lines use Gaussian profiles at idealized infinite resolution; the ETC applies observing-mode broadening, allowing sensitivity modeling of under-resolved lines amid strong continuum.Spectra can also be shifted by intrinsic Doppler or cosmological redshift, and modified with standard infrared extinction laws.Source: §3.4 · p. 5
3.5 The PSF Library
- Mode-specific monochromatic PSFs convolve each realized-scene plane, with wavelength interpolation and usually field-constant profiles; coronagraphic sources use position-relevant PSFs.The library PSFs are normalized to one incident photon before throughput corrections, while including losses from diffractive optical elements.Source: §3.5 · p. 5
4. SCENE CUBE PROJECTIONS
Pandeia projects realized scene cubes onto virtual detector planes to calculate electron rates, using different mappings for imaging and spectroscopic modes. The projections cover imaging, slit-based, IFU, slitless, and multi-order slitless observations.
4. SCENE CUBE PROJECTIONS
- Pandeia maps realized scene cubes onto virtual detector planes, with the projection determined by the observing mode.The five projection types are imaging, single slit/slitlet, IFU, slitless, and multi-order slitless.Source: §4 · p. 5
4.1 Imaging Projections
- Imaging projections integrate the scene rate along wavelength using Simpson’s Rule, including for coronagraphic and non-redundant aperture mask images.Source: §4.1 · p. 6
4.2 Fast Projection for Narrow-Slit Spectroscopy
- Narrow-slit spectroscopy masks the PSF-convolved scene and integrates along the dispersion axis, then interpolates rates to detector wavelength and pixel scales.Because the PSF is already sub-pixel sampled and rebinning conserves flux, integration can use a simple sum; the fast method assumes slit width is similar to or smaller than the PSF width, so spatial offsets within the slit do not shift spectra.Source: §4.2 · pp. 6–7
4.3 Fast Projection for Integral Field Spectroscopy
- For IFU observations, Pandeia treats each image-slicer slice as a narrow slit and stacks the resulting two-dimensional spectra onto one detector plane.The resulting plane approximates detector data without field distortion.Source: §4.3 · p. 7
4.4 Projection for Slitless or Wide-Slit Spectroscopy
- Slitless and wide-slit projections integrate along an oblique path through the scene cube, assuming the dispersion function does not vary across the field.With no dispersion, the expression reduces to imaging; sources outside the scene postage stamp are excluded, though spatially invariant background can be added.Source: §4.4 · p. 7
4.5 Projection for NIRISS Single-Object Slitless Mode
- NIRISS SOSS projects the scene once for each supported order and applies an order-specific trace function to model spatial distortion.Including all orders accounts for overlapping background and contaminating signal.Source: §4.5 · p. 8, Figure 3 · p. 8
5. PROPAGATION THROUGH TELESCOPE AND INSTRUMENTS
This section describes instrument and exposure configuration, throughput conversion, and detector saturation handling. Pandeia uses forward calculations for MULTIACCUM exposures and applies mode-specific optical efficiencies before modeling detector effects.
5.3 Throughputs and Optical Path
- Throughput conversion applies the multiplicative efficiencies of optical elements to the scene cube before detection.Elements include the telescope optics, mirrors, filters, gratings, and detector quantum efficiency; diffractive components are generally represented in the PSF model.Source: §5 · p. 8, §5.3 · p. 9
5.1 Instrument Configuration
- Pandeia configures calculations by instrument and observing mode, including their associated parameters such as filters or dispersing elements.Source: §5.1 · p. 8
5.2 Exposure Configuration
- Because MULTIACCUM SNR versus exposure time is neither analytic nor unique, Pandeia performs forward calculations from a specified exposure configuration.Exposure settings include sub-array, readout pattern, groups, integrations, and exposures; dithering is currently supported only for IFU observations.Source: §5.2 · p. 8
5.4 Dynamic Range and Saturation
- Soft-saturated pixels remain usable with an effective count of unsaturated groups, while hard-saturated pixels cannot receive an SNR calculation.A single-read noise model can replace ramp fitting in some cases, but yields significantly higher noise; implementation was planned for at least NIRCam and NIRSpec.Source: §5.4 · p. 9
6. NOISE PROPAGATION
Pandeia propagates noise through pixel-level covariance rather than assuming independent pixel errors, accounting for correlated detector noise and observing strategies. It models read-noise correlations, inter-pixel capacitance, and cosmic-ray effects on usable integration time.
6.1 MULTIACCUM Ramp Noise Model and Correlated Noise
- Correlated read noise means pixel variances do not simply add in quadrature, so the extraction variance must include covariance between pixels.Read noise is strongly correlated in HAWAII 2RG detectors, especially along the fast-read direction; MIRI may also have weaker correlated noise.Source: §6.1 · pp. 9–10
6. NOISE PROPAGATION
- A covariance matrix combined with pixel-by-pixel extraction weights yields the total variance for a source measurement.The framework supports realistic observing strategies and evaluates their effect on final data quality.Source: §6 · p. 9, §6.1 · p. 10
6.2 The Correlated Read Noise Term
- The read-noise covariance matrix is constructed from detector-coordinate correlations and depends on detector and readout configuration.Extraction weights select aperture and background pixels; Figure 4 illustrates a HAWAII 2RG correlation matrix.Source: §6.2 · p. 10, Figure 4 · p. 11
6.3 Inter-Pixel Capacitance (IPC)
- Inter-pixel capacitance deterministically spreads charge into adjacent pixels, correlating their Poisson noise and requiring distinct signal and noise images in the ETC.Pandeia calculates strategy-product noise from an unconvolved detector image while calculating signal from an image convolved with the IPC kernel.Source: §6.3 · pp. 10–11
6.4 Cosmic Rays
- For cosmic-ray-affected ramps, Pandeia conservatively counts only the ramp portion before the event as usable.The average usable integration time is derived using the ramp duration, number of readouts, and minimum reads needed for a valid slope.Source: §6.4 · p. 11
7. EXTRACTION STRATEGIES
Pandeia represents an observing strategy as a pixel-weight map and read-noise covariance matrix that transform detector rates into scientific products. It adapts the extraction and background treatment to imaging, dispersed spectroscopy, and integral-field observations.
7. EXTRACTION STRATEGIES
- An observing strategy maps detector pixel rates and noise to products such as a signal, noise, and signal-to-noise ratio using a covariance matrix and pixel weights.The weights define how detector pixels contribute to an extracted spectrum or photometric measurement.Source: §7 · p. 12
7.1 Imaging Aperture Photometry
- Imaging aperture photometry sums pixels in a source aperture, subtracts an estimated background, and applies an aperture correction when source flux falls outside the aperture.The resulting weights are propagated to obtain the photometric variance and signal-to-noise ratio.Source: §7.1 · p. 12
7.2 Spectroscopic Aperture Photometry
- For dispersed modes, aperture photometry is performed at each wavelength step, producing an extracted spectrum and a signal-to-noise array.Source and background aperture definitions vary among dispersed observing modes.Source: §7.2 · p. 12
7.2.1 Fixed Slit
- Fixed-slit extraction uses one-dimensional source and optional background apertures, with NIRSpec MSA aperture sizes and locations constrained by shutter geometry.Source: §7.2.1 · p. 12
7.2.2 Slitless
- Slitless extraction uses one-dimensional apertures for wavelength-dependent source flux and two-dimensional regions for a shared, wavelength-independent background estimate.Source: §7.2.2 · p. 12
7.2.3 Integral Field
- Integral-field extraction maps two-dimensional spatial apertures into detector pixels and offers aperture, dither-off-scene, and dither-on-scene background methods.Off-scene dithering more than doubles observation time and is likely for sources that significantly fill the IFU; on-scene dithering suits sources much smaller than the field of view.Source: §7.2.3 · p. 13
8. PERFORMANCE AND BENCHMARKS
Pandeia was benchmarked against JWST instrument-team sensitivity calculations using the same or similar reference data. The comparisons generally matched derived signal-to-noise ratios to about 10%, while verification of reference data remained ongoing.
8. PERFORMANCE AND BENCHMARKS
- About 10% agreement in derived signal-to-noise ratios was generally required in comparisons with JWST instrument-team sensitivity calculations.The comparisons used the same or similar reference data, including throughputs and background.Source: §8 · p. 13
- Benchmarking was a two-step process: compare algorithms and implementation with instrument-team calculations, then check that the ETC uses the best-known reference data.Source: §8 · p. 13
- Verification that the ETC uses the best-known reference data was still ongoing.Source: §8 · p. 13
9. PANDEIA FOR WFIRST AND OTHER MISSIONS
Pandeia's data-driven design allows new observatories and instrument suites to be implemented largely through configuration files and reference data. JWST modes and current WFIRST WFI and integral-field designs are supported; WFIRST coronagraph support was planned pending a finalized baseline design.
9. PANDEIA FOR WFIRST AND OTHER MISSIONS
- New observatories and instrument suites can be implemented largely by adding configuration files and appropriate reference data.Source: §9 · p. 13
- All JWST instruments and observing modes were implemented in preparation for the first call for proposals.Source: §9 · p. 13
- Support covered current WFIRST Wide-field Imager and integral-field channel designs, while coronagraph support awaited a finalized baseline design.Source: §9 · p. 13
10. ACCESSING AND USING PANDEIA
Pandeia access was planned through a web interface with workbook sharing and richer visual comparisons, while a limited WFIRST-only notebook route was already available. The paper describes release timing and access pathways for the interface and compute engine.
10. ACCESSING AND USING PANDEIA
- The planned JWST web interface supports workbook grouping, sharing, and richer 2D/3D comparisons; Figure 5 shows a representative screenshot.Users could also superimpose results from multiple calculations for comparison.Source: §10 · p. 13, Figure 5 · p. 14
- A limited-support JWST interface pre-release was expected by late summer 2016, with its official release for Cycle 1 preparations scheduled for January 2017.Source: §10 · p. 13
- The Python compute engine was also slated for public release through AstroConda, and the paper points readers to the JWST ETC site for current access information.The release information is stated as of summer 2016.Source: §10 · p. 13
- A limited WFIRST-only engine had been available to team members on request via an STScI-hosted JupyterHub since January 2016; notebooks support reproducible, shareable exploration.The paper describes the notebook route as less user-friendly than the web interface, but useful for experienced users.Source: §10 · p. 13