JWST Calibration Uncertainties
Estimated uncertainties for calibrated quantities, including photometry, astrometry, and wavelength calibration, organized by instrument and mode, are documented in this article. It describes the calibration uncertainties for each JWST observing mode as they are currently understood. In most cases, updates to the calibration, including reference files and theJWST pipeline, can be expected in the next few months, which will change the values quoted here.
See also: Known Issues with JWST Data Products
Shortwave channel: typical uncertainties for the wide band filters are <=0.1 pixels, on the order of 3–5 mas for F070W decreasing to 1 mas for F200W. For the other filters, no images were available at the time of the analysis, and appropriate wide band filter distortions were used. In these cases, there are systematic residuals on the order of up to 0.2 pixels (6 mas).
Longwave channel: typical uncertainties for the wide band filters are 1–2 mas. For the other filters, no images were available at the time of the analysis, and appropriate wide band filter distortions were used. In these cases, there are systematic residuals on the order of up to 0.1 pixels (6 mas).
Updated astrometric distortion files are expected winter 2023/2024. After the update, the accuracies for all filters are expected to be at or below 1 mas.
An updated flux calibration was released in CRDS on September 14, 2023 (pmap 1126).
- The absolute flux calibration uncertainties are <1% for most filters/detectors, and 1–2% for the remainder. Four LW filters reach 2–4%: F322W2 & F460M on module B and F430M & F460M on module A.
- The absolute flux calibration of the weak lenses (used in time-series observations) is currently based on ground data, and is within about 5% uncertainty. An update based on in-flight observations of standard stars is expected in October 2023.
- Average detector-to-detector offsets are <0.03 mag between the standard stars (often < 0.01 mag), depending on filter.
- Subarray and full frame observations agree to <1%.
Stability is <1% based on repeated imaging measurements in F212N in the shortwave channel and F470N in the longwave channel throughout Cycle 1.
Target acquisition performance
Analysis of the target acquisition performance for time-series imaging is still in progress.
Wide field slitless spectroscopy and grism time series
The absolute flux calibration uncertainties for the grisms are about 2% (all filters/detectors). The grism subarrays and full frame observations agree to within about 2% based on images in those subarrays using standard stars.
Generally ~0.25 pixels, or 2.5 Å, over the whole field of view using the latest calibration data. The wavelength is tied to NIRSpec long slit observations of a planetary nebulae.
Spectral trace accuracy
All positions, offsets, traces, and spectral shapes of the 1st orders are calibrated to within ~0.25 pixel over the entire field of view. That is, given a coordinate x0, y0 for a source, it can be predicted where the spectral trace will fall on the detector in the WFSS observation down to 0.25 pixel on the detector in WFSS mode. The modeling/calibration of this includes field dependence and variation of the shape of the spectral traces as a function of position.
Target acquisition performance
For grism time series, TA uncertainty has been measured to be ±0.3 px (0.02") for the F335M and ±0.15 (~0.01") for the F405N filters, respectively.
Shortwave channel: Uncertainties for both round and bar occulting masks are generally on the order of 5–10 mas in the area covered by the coronagraph optical mount (COM). Outside the COM, the uncertainties are 3–6 mas.
Longwave channel: Uncertainties for both round and bar occulting masks are generally >10 mas both in and outside of the COM area.
Updated astrometric distortion files for coronagraphy are expected winter 2023/2024.
An updated flux calibration for coronagraphy was released in CRDS on November 1, 2023 (pmap 1146).
The absolute flux calibration uncertainties are <1.5% for the round masks and the short wavelength bar mask (SWB). For the long wavelength bar mask (LWB), the uncertainties are <2% for most filters, but are higher for F277W (6%), F250M (13.5%), and F300M (12.5%) because the standard star was partially obscured by the mask. These numbers should improve when Cycle 2 data are incorporated.
Target acquisition performance
Table 1. target acquisition global uncertainties
|Module A||SW as prime (F210M)||LW channel as prime (F335M)|
BRIGHT (ND) TA
SR offset: +0.007, -0.002
|SR offset: +0.000, -0.015|
Global error: < 15 mas (1σ)
|SR offset: -0.010, +0.006|
Global error: ~15 mas (1σ)
|SR offset: +0.017, -0.015|
Global error: ~20 mas (1σ)
|SR offset: +0.000, +0.031|
Global error: > 0.5 pixel
|FAINT TA||SR offset: -0.001, +0.011|
Global error: TBD
|TBD||SR offset: -0.006, +0.002|
Global error: ~15 mas (1σ)
|SR offset: +0.014, -0.019|
Global error: TBD
- SR offsets are the special requirements offsets (in arcsec) currently needed (in APT) to compensate for imperfect distortion compensation and updated mask positions.
- SR offsets will be discontinued when distortion corrections are updated (winter 2023/2024).
- Values in orange need confirmation (further analysis and/or more data).
Multi-object spectroscopy (MOS)
There is an approximately 15% absolute flux accuracy.
Field-dependent variations may be as large as 10%.
- Issues with the current pipeline treatment of the off-center nodding positions can significantly degrade this accuracy, although workarounds are available.
- Achieving the above accuracies may also require rerunning the pipeline with altered parameters, including changes to outlier correction, weights used for resampling, and the source extraction location.
- The emission line fluxes taken with PRISM match the NIRCam photometry; however, the emission line fluxes in the spectra taken with M gratings is ~10% higher (Bunker et al. 2023). The NIRSpec team is investigating the issue and anticipates that the offset is likely due to the mis-centering of the target used for deriving the flux calibration of the M gratings. The improved flux calibration will be available soon.
NIRSpec meets the requirement of wavelength accuracy of 1/8 of a resolution element (Böker et al., 2023). This assumes a definition of the resolution element as that measured using lamp spectra, which typically corresponds to a wavelength accuracy of about a 1/4 pixel.
This translates into typical velocity accuracies for the different dispersions of:
- High resolution gratings (R2700) ~ 15 km/s
- Medium resolution gratings (R1000) ~ 40 km/s
- PRISM (R100) ~ 750 km/s
- The above numbers apply to well-centered point sources or extended sources that uniformly fill the aperture. The accuracy of wavelength calibration for off-centered sources is less well established.
- Variations over the MOS field of view have yet to be well characterized.
For more information, see JWST NIRSpec MOS Pipeline Caveats.
Integral field unit (IFU) spectroscopy
There is better than 5% absolute flux accuracy measured for the combined spectrum from the 4-point nodding pattern.
- The accuracy of delivered pipeline products may be sensitive to the exact aperture size and background subtraction strategy.
- Variation of sensitivity with position in the IFU aperture is not as well characterized.
The wavelength accuracies are slightly worse for the IFU modes, although the details of the residuals may vary and are still being characterized. For the G235H and G395H, the IFU mode has modest global dispersion errors with the shortest wavelengths for these 2 gratings having offsets gratings corresponding to as much as -20 km/s relating both the long wavelength end of the same grating and to the true velocity. Variations over the IFU field of view have yet to be well characterized.
Additional information is available at JWST NIRSpec IFU Pipeline Caveats.
Fixed slit (FS) spectroscopy
It is generally better than 5% absolute for sources well centered in the aperture.
- The lack of a correction in the pipeline for differences between nodding patterns, which subtract some source flux along with the background, limits the accuracy to about 5% for most modes. The current calibration is optimized for the 3-point nodding pattern for the S200A1, S200A2, and S400A1, and is a compromise between the 3-point nod and no nod correction for the S1600A1. Note that the 5-point dither pattern in the S1600A1 is currently not well handled by the pipeline default settings, leading to a 20% flux underestimate for level 2 and 3 products, although this is easily fixed with customized reprocessing.
- PRISM observations in the narrow S200A1, S200A2, and S400A1 slits may show lower absolute accuracy than the above, as the standard star used to calibrate these combinations was close to saturation with the smallest supported subarray.
- The accuracy of flux corrections for off-centered sources is less well established.
- Achieving the above-quoted accuracy may require rerunning the pipeline with altered parameters, including changes to outlier rejection, weights used for resampling, and the source extraction location.
Typical broadband flux repeatability over a time scale of months is approximately 0.5%. Any time-dependent trend appears to be smaller than 2% per year.
Limiting S/N: For a bright source observed at multiple dither positions, 50:1 is achievable with modest modifications to default pipeline parameters. At least 100:1 should be possible with more specialized data reduction techniques.
The wavelength accuracy is similar to the MOS mode. However, like IFU mode, for the G235H and G395H, FS modes may have modest global dispersion errors, with the shortest wavelengths for these 2 grating having offsets corresponding to as much as -20 km/s shift relative to both the long wavelength end of the same grating and to the true velocity.
Bright object time series (BOTS)
The results on photometric calibration for FS also apply to the BOTS mode when considering the net extracted absolute, as opposed to relative, spectra.
For bright stars (K mag ~ 7.7 to 9.5) it has been demonstrated that observations with G395H can measure broadband transit depth variations to an accuracy of 50 to 60 ppm. Similar behavior is expected for other modes.
Typical drifts in the measured count rate are between 30 to 140 ppm/hour for NRS1 and 10 to 40 ppm/hour for NRS2, with the drift rate declining over the course of multi-day observations.
Stability of the trace location on the detector is typically between 1/300th and 1/1000th of a pixel.
For more information, see NIRSpec Time Series Observations Pipeline Caveats.
Relative accuracy is about 3 mas, absolute accuracy about 0.5 arcsec mainly due to issues with the star tracker roll angle measurement. The absolute world coordinate system accuracy can vary from 0.1 to larger than 2 arcsec and is sometimes off due to guide star acquisition issues.
Current uncertainties are of order 2% at short wavelengths, 5% at longer wavelengths, due to variation in measured values from one standard star to another. The direct statistical measurement uncertainties per star are less than 1% at short wavelengths and up to 2% at longer wavelengths. The actual stability appears to be better than 1% over time.
Filter-to-filter image offsets (i.e., boresight offsets)
The "_rate.fits" images in different filters are subject to small "wedge" offsets between the filters. These values are given in Table 2 as the offsets from the position in the F150W filter to the position in another filter. These offsets are propagated to the world coordinate system (WCS) information in the headers of the "_cal.fits" files (which have been flat fielded, but not yet resampled to a distortion-free pixel grid) and the "_i2d.fits" files (which are distortion corrected). However, in the "_cal.fits" files, these filter-to-filter offsets are still present in the science images. To compare images in different filters on a common pixel grid, STScI's NIRISS team recommends running the resample step with a certain absolute pixel scale (i.e., the pixel_scale argument) to create "_i2d.fits" data products.
Table 2. Filter-to-filter offsets in NIRISS "_rate.fits" and "_cal.fits" images
|Filter||x pixel offset||y pixel offset|
NIRISS WFSS mode
STScI currently estimates a 2.5% accuracy of the photometric conversion factors, which is dominated by issues with spectral extraction and contamination. There is a separate issue with the effect of bad pixels on the extraction which produces systematically low values in the spectra at some wavelengths, so the wavelength-to-wavelength errors in the output flux density can be 10% or more depending on where in the cross-dispersion profile the bad pixel appears. This effect is also somewhat dependent on the width of the extraction aperture.
This is generally about 0.3 pixels, with worse uncertainties in the F200W filter; it is limited by the available calibration data.
Spectral trace accuracy
For point sources there can be offsets of 2–3 pixels between the predicted trace positions and the actual trace positions over the detector. The current accuracy is limited by the available calibration data. Additional trace offsets may be caused by issues in the pipeline source catalogue segmentation map, especially for extended sources.
NIRISS SOSS mode
There is an estimated 1% accuracy in order 1 and 2, based on cross-comparison of standard stars. The order 3 uncertainty is larger than 5% because the response is much lower than for the other orders. This is the uncertainty in the conversion from ADU/s to Jy. Due to the effect of bad pixels on the spectral extraction, the wavelength-to-wavelength photometric errors in the spectrum can be of order 5%, with values being systematically low where the bad pixels affect the extraction. In orders 2 and 3 the uncertainties are much larger in the wavelength ranges where the intrinsic signal is low: for order 2, wavelengths between 1.05 and 1.4 µm, and for order 3 wavelengths below 0.8µm.
The estimated accuracy is 4.0 pixels in the raw pipeline results, which can be improved significantly to about 0.5 pixels by allowing for the small grism rotations and small source offsets from one observation to another using the PASTASOSS package (Baines et al. 2023). The raw accuracy varies with wavelength, being larger at the longest wavelengths because the "lever arm" distance from the rotation point is larger for that part of the spectrum. Table 3 shows the accuracy estimates for the SOSS wavelength calibration using measurements of H2 absorption features in the photometric standard BD+60 1753.
Table 3. Accuracy estimates for the SOSS wavelength calibration (Baines et al. 2023)
Number of Fitted H2 Absorption Features
5.72 x 10-4 ± 1.71 × 10-4
2.01 x 10-4 ± 3.70 × 10-5
0.58 ± 0.17
0.21 ± 0.04
2.42 x 10-4 ± 5.11 × 10-5
6.06 x 10-5 ± 2.07 × 10-5
0.52 ± 0.11
0.13 ± 0.04
The accuracy is better for wavelengths less than 1 µm in order 1, likely due to better S/N at these wavelengths due to the combination of the better grism response and the higher stellar flux density in the spectrum of BD+60 1753. The wavelength calibration for SOSS order 3 is not yet supported.
NIRISS AMI mode
This is the same as for imaging.
A range of about 5% in the scaling factors from different standard stars is observed in the F277W/F380M/F430M filters, and the range is 8% for F480M. The uncertainty in the relative calibration of the normal imaging and AMI imaging modes is under 1%.
Calibration of charge migration (CM)
Commissioning characterized the largest CM effect on the order of a ~1% drop in the count rate between the first half of an exposure and the second half (using 3 groups and a peak pixel with 30,000 e– accumulated charge). Improvement in rejecting affected data in an upcoming release of the JWST pipeline is likely to mitigate against CM affecting AMI contrast. Quantification of the improvement is not yet available. Future work measuring and calibrating CM effects using existing data may also enable existing data to yield better contrast than is currently attainable.
An approximately 0.03 pixel standard deviation per axis random error in AMI target placement was measured in the AMI medium band filters during commissioning.
Commissioning data revealed systematic filter-specific offsets from commanded positions (20 mas for F380M, 12 mas for F430M, and 14 mas for F480M). Updated Science Instrument Aperture File (SIAF) definitions were installed onboard to remove these systematic target placement errors. These updates have not been tested yet.
The absolute astrometric accuracy can vary from 0.1 to 0.2 arcsec (pointing without target acquisition) to 2 arcsec (guide star issues). If the absolute astrometry is refined by means of the Gaia catalog and the TweakReg step of the stage 3 pipeline, the geometric distortion uncertainties (standard deviation of the positional residuals) can be 0.01–0.03 arcsec, depending on the filter. The astrometric accuracy and precision get worse from short to long wavelengths.
After a pending update in mid- to late-September: Uncertainties are of order 3% or better at all wavelengths, due to variations in measured values from one standard star to another. The telescope/instrument stability are better than 0.5% over time based on repeated measurements of one star in F770W. The pending update will account for the time dependent sensitivity loss seen for F1280W and longer.
MIRI medium resolution spectrometer (MRS)
Overall astrometric alignment is 0.03" when using target acquisition or 0.45" without target acquisition. Typical data product WCS accuracy is 0.3" to 1.5", driven by guide star uncertainties. See MRS Astrometric Calibration.
Based on repeated observations of individual standard stars, the typical stochastic flux calibration uncertainty is below 0.5% in channels 1–3 and increases to 5% in Channel 4C (where the star becomes faint relative to the thermal background). This is likely a lower limit, as observations using different numbers of groups/integrations in all channels can show systematic differences of 1–2%. Likewise, there are global systematic uncertainties in the overall calibration at the level of 10% due to differences in the observed calibration factors for various standard stars, driven largely by uncertainties in the CALSPEC models. Calibration uncertainties for faint sources can be significantly larger due to cosmic ray showers (at short wavelengths) or systematics in the background subtraction (at long wavelengths). See MRS Photometric Calibration.
"ERR" values provided by the pipeline are unreliable to factors of 10–50 and should not be used pending further investigation.:
MIRI low resolution spectrometer (LRS)
The current calibration is from October 2022 and based on commissioning data. A recalibration for the slit is underway and will be complete in fall 2023.
Slitless: It is generally better than 20 nm, but possibly larger for wavelengths below ~7 µm.
Slit: Work is ongoing based on Cycle 1 calibration data. Reported wavelengths from the current calibration look to be too red, with larger shifts to the blue required at ends of the LRS wavelength range than in the middle: about 40 nm at 6 um, 10 nm at 10–11 um, and 20 nm past 12 um. This result is close to a shift of 50 nm at 6 µm reported by Beiler et al. (2023).
See Beiler et al. 2023 for additional information.
Pointing distribution and consequences
The dispersion in the position of a star in the TA verification images is roughly ~0.10 pixels (radial, 1-σ). A shift of the target from the nominal pointing position will shift the dispersion solution for that spectrum an equal number of wavelength elements. The pipeline does not correct for that shift at this time.
Slit: The absolute flux calibration, as measured at 6–7 um for standard stars, is about 2% or better, but the overall shape of the extracted spectra for these stars show red excesses of 3–8%. The cause is under investigation. The limited quality of the current pathloss correction will add errors to the photometry for sources not centered in the slit (improvements are expected in fall 2023).
Slitless: The default pipeline concentrates on relative spectrophotometry for the study of exoplanetary transits and light curves and does not perform a background subtraction for slitless spectroscopy. Persistence in the slitless subarray (see next section) adds to the background. For data that have not been background subtracted, these issues affect fainter targets more than brighter ones. For standard stars at about 100–200 mJy at 10 um, the calibration can be better than 1% at 7 um. Standard stars at 50–70 mJy at 10 um can be several percent high at 7 um, and the problem grows worse for fainter targets, with excesses of 20–30% seen in the fainter standard stars. Because stellar spectra are dropping rapidly with increasing wavelengths, the lack of a background subtraction can lead to strong red excesses in the extracted spectra. Observers interested in absolute photometry are strongly encouraged to remove the background manually from their slitless spectra integration by integration using apertures on either side of the region of interest in the spectral images.
Spectrophotometric precision (TSOs with LRS)
The spectrophotometric precision achieved during JWST commissioning was ~50 ppm at at a resolving power of ~50 in the core wavelength range for the LRS (7–8 um; Bouwman et al. 2023). This precision is achievable after an initial settling time; the time constant for this initial settling is of order 30 minutes, and its magnitude approximately 1–3%. The shape of this initial settling ramp clearly depends on flux. Optimal spectrophotometric performance is also contingent on background subtraction, which is not yet automatically performed by the JWST calibration pipeline at this time. The background must be removed on a per-integration basis to mitigate for the 390 Hz noise signature in the SLITLESSPRISM subarray.
Users are recommended to add 30 min of additional time at the start of their observation for transits and eclipses, and 1 hour for phase curves, to mitigate for the initial detector settling behavior.
Photometric calibration, after a pending update in middle/late Sep.: Uncertainties are of order 3% or better at all wavelengths, due to variation in measured values from one standard star to another. The telescope/instrument stability are better than 0.5% over time based on repeated measurements of one star in F770W. The pending update will account for the time dependent sensitivity for F1550C and F2300C.
Baines, T., Espinoza, N., Filippazzo, J., Volk, K. 2023, JWST-STScI-008571
Characterization of the visit-to-visit Stability of the GR700XD Wavelength Calibration for NIRISS/SOSS Observations
Beiler, S.A., Cushing, M.J., Kirkpatrick, J.D., et al. 2023, ApJ, 951, L48
The First JWST Spectral Energy Distribution of a Y Dwarf
Böker, T., et al. 2023, PASP, 135, 038001
In-orbit Performance of the Near-infrared Spectrograph NIRSpec on the James Webb Space Telescope
Bouwman, J., Kendrew, S., Greene, T. P., et al. 2023, PASP, 135, 038002
Spectroscopic Time Series Performance of the Mid-infrared Instrument on the JWST
Bunker, A. J., et al. 2023, submitted to A&A (arXiv)
JADES NIRSpec Initial Data Release for the Hubble Ultra Deep Field: Redshifts and Line Fluxes of Distant Galaxies from the Deepest JWST Cycle 1 NIRSpec Multi-Object Spectroscopy
Espinoza, N., et al. 2023, PASP, 135 018002
Spectroscopic Time-series Performance of JWST/NIRSpec from Commissioning Observations
Lustig-Yaeger, J., et al. 2023, submitted to Nature Astronomy (arXiv)
A JWST transmission spectrum of a nearby Earth-sized exoplanet
Mikal-Evans, T., et al. 2023 ApJL 943 L17
A JWST NIRSpec Phase Curve for WASP-121b: Dayside Emission Strongest Eastward of the Substellar Point and Nightside Conditions Conducive to Cloud Formation
Predicting Accurate Spectra Traces in Astrophysical SOSS Spectra (PASTASOSS): https://github.com/spacetelescope/pastasoss