Detector Calibrations
Detector level calibrations for Roman provide information about how the Wide Field Instrument's detectors and their electronics capture light signals. Many effects are calibrated in the Exposure Pipeline.
Scheduling to be Finalized
Overview
The calibration of the Wide-Field Instrument (WFI) detectors of the Nancy Grace Roman Space Telescope (Roman) includes several components. Some are fairly standard, e.g., correction for an additive ("dark”) and multiplicative ("flat”) terms, persistence, or the correction for non-linearity of the detector response. Others are less commonly considered and include, for instance, count-rate-dependent non-linearity, correlation in the read noise, and inter-pixel capacitance. The characterization of these effects are based on a combination of WFI Ground Testing Campaigns, Commissioning Campaigns, and regular in-flight calibration activities that are described in this article. The On-Orbit calibration plan was developed through a community working group and the recommendations are generally conservative. The On-Orbit Calibration Plan adopts a nominal cadence of activities based on the experience obtained from other space missions (primarily HST and JWST, which has flown several detectors similar to those in the WFI), but changes in detector manufacturing and specifications, orbital environment, and required precision for the Roman mission may require calibrations taken at a different frequency or using different methodologies. The Roman Science Requirement Document provides a detailed flow-down from science requirements to calibration requirements and the required accuracy in the calibration is the strongest driver for the observation sequences and observation cadence.
A summary of the detector calibrations are given in the Table of Detector Calibrations and individual sections provide a more detailed description (including links to the WFI Detectors articles), summarize the requirements on the calibration (if they exist), briefly describe the nominal activities planned for the calibrations and their cadence, and connect to the calibration files that are provided for users. Links to other relevant content related to the Roman Community-defined Surveys and Science Data Pipelines are provided when applicable.
Table of Detector Calibrations
| Calibration Name | Short Name | Short Description | Observation Cadence | SOC Products |
|---|---|---|---|---|
| Dark Signal | darks | monitors dark signal | Weekly | dark reference file |
| Pixel-level Flat | p-flat | pixel-level variation in response to uniform illumination | Monthly | component of the flat reference file |
| Classical NonLinearity | CNL | deviation from linear accumulation of counts | Yearly | linearity reference file |
| Count Rate NonLinearity | CRNL | non-linear effect with a dependence on count rate | Quarterly Monitoring One Direct Measurement | n/a |
| Read Noise | noise in the conversion from electrons to data number | Weekly monitoring, updates as required | read noise reference file | |
| Persistence | residual signal present after high levels of illumination | Annual Monitor | ||
| Burn In | response of a pixel depends on if the pixel was previously exposed | Monthly | ||
| Gain | conversion factor between electrons and data numbers | Monthly monitoring, updates as required | gain reference file | |
| Inter-Pixel Capacitance | IPC | coupling factor between neighboring pixels | TBD | IPC reference file |
| Pixel Behavior Families | pixels with atypical behavior | Weekly and Monthly monitoring | mask reference file |
Dark for the imaging mode (darks)
Description
Dark-current signal is an additive component that is measured at the detector but does not correspond to photons impinging on the telescope. This signal is caused by a combination of internal detector dark current, residual bias, and thermal background from the instrument itself. A small amount of this signal will also come from thermal emission from the Dark element on the Element Wheel. Dark signal does not include the additive component coming from using other optical elements, but this term is included in the Roman Background Visualization Tool. Roman WFI science data have the dark subtracted in the Exposure Pipeline.
Based on data acquired during WFI Ground Testing Campaigns, the dark signal for the imaging mode of WFI is well described by a dark rate (Betti et al. 2025), though this will be tested again during commissioning.
Requirements
The most stringent requirement on dark calibration derives from the weak lensing additive shear error budget, with a top-line error of 2.7 × 10−4 root mean square (RMS) per component, of which 4% (in quadrature) is allocated to dark current residuals. "Darks" are taken with the dark element in place and directly measures the dark signal. The dark signal is known to be 0.0114 electrons per second per pixel in the absence of correlations.
Calibration Activity
The total dark sequence consists of 50 integrations of 140 seconds each, and 24 integrations of 300 seconds each, to be repeated weekly. During the Galactic Bulge Time Domain Survey campaigns, the detectors will be monitored for overall health through a shortened observational program of no more than 2 hours.
Cadence
Dark calibration observations will occur on a weekly basis.
Calibration Files
The dark signal is well-described by a constant dark rate and a single calibration file is provided per detector for WFI imaging and is hosted on the Roman Calibration Reference Data System (CRDS).
Pixel-Level Flat field (p-flat)
Description
The pixel-level flat field (P-Flat) quantifies the response of individual pixels to a common, spatially flat stimulus signal and corrects for this source of pixel-to-pixel variations. The flat field is defined so that a uniform illumination level (constant flux per unit area on the sky) produces a uniform calibrated response (constant data number per pixel per second). The Exposure Pipeline uses flat-field files that combine pixel flats (p-flat) with large-scale variation flats (L-flats). The calibration plan is to obtain flat fields with near-uniform illumination using the Relative Calibration System (RCS). The methodology for flat field correction at large scales is described in Imaging Calibrations. The combined flat is applied in the Exposure Pipeline.
Requirements
The pixel response to a flat illumination needs to be calibrated to an accuracy of 0.1% (RMS) for pixel-to-pixel small-scale variation.
Calibration Activities
For each wavelength, this requirement implies collecting about 106 electrons. A total of 20 exposures, each 50 seconds long, with an illumination level of 1000 electrons per second per pixel can achieve this total signal.
Cadence
P-Flat calibration activities will occur on a monthly cadence.
Calibration Files
A pixel-level flat field is generated for each detector and combined with the L-flat. The combined flat files are hosted on the Roman Calibration Reference Data System (CRDS).
Classic Non-Linearity (CNL)
Description
Classic Non-Linearity (CNL) is the deviation from linear accumulation of counts on the detector. A CNL correction will make a ramp with a constant input illumination linear, up to read and photon noise. The CNL correction can be derived from internal flats, as they are very effective in measuring the pixel-to-pixel variations of the non-linearity. The CNL correction is applied in the Exposure Pipeline.
Requirements
Weak lensing and supernova cosmology measurements strongly depend on the correct calibration of the PSF, the shape of which is affected by a non linearity of the detector response. The resulting requirement on non linearity is for the effect to be corrected to an accuracy of 0.3%. This is not a per-pixel requirement, but more as an upper limit on systematics.
Calibration Activities
Estimates suggest that a CNL accuracy of 0.3% can be achieved with 100 ramps and a total well capacity of about 105 electrons. The CNL calibration will rely on the analysis of internal flat fields and does not require specific observations.
Cadence
The CNL will be validated yearly.
Calibration Files
CNL calibration files storing the correction of each pixel of the 18 WFI detectors will be generated and hosted on the Roman Calibration Reference Data System (CRDS).
Count-Rate Non-Linearity
Description
The count rate-dependent non-linearity (CRNL) is a non-linear effect in which the measured signal level depends on the count rate (electrons per second per pixel) in addition to the total collected charge (electrons per pixel). CRNL results in a residual systematic error in photometry after the CNL correction is applied.
CRNL can be parametrized by a power law exponent, CR_{obs}}\,\propto \,F^{\alpha_{CRNL}}, where CR_{obs} is the observed count rate and F is the flux.
Requirements
There are two science requirements on CRNL:
- The Supernova Survey demands the error induced over 11 magnitudes to be less than 0.3%, implying a knowledge of \alpha_\textrm{CRNL} to \pm 3 x 10-4.
- Weak Lensing needs to characterize the second moment of the PSF to better than 7.2 x 10-4, which also implies knowing \alpha_\textrm{CRNL} to \pm 3 x 10-4.
Calibration Activities
Calibrating CRNL is a challenging task and, because of the broad impact of CRNL effects, two different calibration activities are in the baseline plan: (1) the direct measurement using combanatorial flux addition (CFA) that provides a correction at the pixel-level and the (2) Lamp-On Lamp-Off (LOLO) method that provides a correction at the catalog level. These measurements are extremely difficult and were major design drivers of the RCS (see the description of the RCS for more information.
Cadence
The LOLO method should be executed once per year in all optical elements and repeated quarterly (every 3 months) for one optical element (likely F146). The direct measurement method involving all optical elements is planned to be executed only once in the primary-mission lifetime.
Calibration Files
CRNL corrections are not applied in the current SOC pipeline.
Read Noise
Description
Read noise originates from the electron to data number conversion in the digitized images and can be due to several contributions in the electronics (e.g., from from readout electronics to the multiplexer). Generally, only the read noise impact on rate images (e.g., Level-2 data) is considered. Read noise and its correlations can be measured from dark calibration frames.
A pixel-level read noise map is used in the Exposure Pipeline during the ramp-fit step to measure the count-rate and populate the error array.
Requirements
The most stringent requirement comes from the additive shear error budget for weak lensing, where 0.32% of the error (in an residual sum of squares, RSS, sense) is allocated to uncertainty from read noise correlations. The total value of allowed additive shear error is 1.5 x 10-5.
Calibration Activities
Read noise is measured from a large quantity of dark calibration frames and is performed using the same frames as the Dark Signal measurements.
Cadence
The read noise will be monitored with the regular Dark Signal data collects and updated as required.
Calibration Files
A pixel-level read noise file is available on the Roman Calibration Reference Data System (CRDS).
Persistence
Description
Persistence refers to the additional signal present in a detector pixel as a consequence of the pixel's previous exposure to high levels of illumination. Persistence can arise from multiple sources: science exposures, calibration procedures, cosmic rays, and even bright stars trailing in the field of view during telescope slews.
Based on ground testing, the impact of persistence is expected to be less than 1 electron per second per pixels for a signal of 5 × 105 electrons per pixel. It is estimated that fewer than 2% of sources are likely to be affected by persistence for crowded-area observations (e.g., Galactic Bulge Time Domain Survey).
Requirements
There are two persistence requirements that flow down from weak-lensing error budgets:
- Out of the total shear error budget of 2.7 x 10-4, 2% of it (in an RSS sense) is allocated to slew persistence, 2% of it (in an RSS sense) is allocated to science persistence, and 1% of it (in an RSS sense) is allocated to previous flat exposures.
- No more than 1% of the pixels can be lost to slew persistence flagging and 0.25% to science persistence flagging.
Calibration Activities
The intensity and duration of persistence depends on multiple factors, including the intensity of the original source, exposure time, and time elapsed since the stimulus. As a consequence, most measurement techniques rely on observations of known sources as calibrators.
Cadence
Observations suitable to measure the persistence decay timescale will occur annually (see e.g., Figure of Measured WFI Persistence).
Calibration Files
The SOC will monitor rate-images (e.g., Level 2 data) observations for signatures of persistence.
Burn-in
Description
Burn-in (or response hysteresis) is an effect in which the response of a pixel to light depends on whether or not the pixel has been exposed to light before the reset. A typical consequence of burn-in happens, e.g., during repeated observations of a bright star with no dithering: the star will appear brighter in exposures after the first one. Here the term “burn-in” is used to denote an effect beyond the additive effect of persistence from a previous exposure. The standard measurement technique to determine burn-in is to take a series of flat-field exposures and measure the pixel response as a function of exposure number. Typical burn-in effects in modern H4RG detectors are of the order of 0.5%.
Requirements
There are no specific requirements for the understanding of burn-in.
Calibration Activities
Burn-in measurements will occur when two or more exposures are taken in a sequence with the internal lamps on.
Cadence
Burn-in can be estimated from flat-field campaigns.
Calibration Files
No correction is made for burn-in at this time.
Gain
Description
The gain is the conversion factor between data numbers (DNs) and electrons, and it is measured in units of electrons per data number. The exact relationship between gain and the statistics of an individual pixel ramp is complicated by interactions between neighboring pixels, e.g., inter-pixel capacitance (IPC) and the brighter-fatter effect. The gain is used in the Exposure Pipeline during the ramp fitting step and is used for two purposes, as follows: (1) to determine the relative weights of the individual reads as they are combined and (2) to compute Poisson-noise contribution to the resulting slope.
Requirements
There are no direct requirements on gain. The gain does have a sub-allocation of the shear error for weak lensing and for the High Latitude Time Domain Survey. The current requirement is for super-pixel regions of 200 pixels by 200 pixels over which knowledge of the average photon transfer curve needs to be to the 0.5% level. There is no direct requirement on the per-pixel gain measurement.
Calibration Activities
The standard method for measuring gain on each pixel is to use the photon-transfer curve, i.e., the mapping between signal (DN) and variance (DN2), measured in a flat with different signal levels. Inter-pixel capacitance reduces the apparent photon noise, resulting in an higher inferred gain from the photon-transfer curve. If the inter-pixel capacitance is known, the gain can be corrected for this effect, but this correction is not applied here. By using the gain directly measured from the PTC, we will adopt the gain that most accurately represents the statistical properties of the ramp itself.
Cadence
The gain can be monitored in the weekly dark signal and monthly pixel-level flat calibration campaigns. The gain is derived from a series of high- and low-illumination flats and requires data collected for the pixel-level flat and for classical non-linearity.
Calibration Files
A pixel-level gain reference file is hosted on the Roman Calibration Reference Data System (CRDS). These gain measurements are not corrected for inter-pixel capacitance.
Inter-Pixel Capacitance (IPC)
Description
Inter-Pixel Capacitance (IPC) refers to the electrostatic coupling of neighboring pixels that causes the observed signal on each pixel to depend on the charge not only in that pixel, but in the neighboring pixels as well. IPC is usually expressed as 3 x 3 pixel smoothing kernel:
| \textrm{K}= \left(\begin{matrix}\alpha' & \alpha-\alpha_+ & \alpha'\\\alpha+\alpha_+& 1-4\alpha-4\alpha' & \alpha+\alpha_+\\ \alpha' & \alpha-\alpha_+&\alpha'\end{matrix}\right) |
where \alpha is the coupling to neighboring pixels, \alpha_+ describes the horizontal-vertical asymmetry, and \alpha' is the diagonal coupling. WFI Ground Testing Campaigns found that the IPC varies in the WFI detectors.
Requirements
A secondary effect of the IPC is to bias the gain measurement based on the photon-transfer curve, where the raw gain is reduced by a factor of \sim 1-8\alpha. Thus, the requirement on gain knowledge flows down to a requirement on the knowledge of |\alpha| to better than 2 x 10-4.
Calibration Activities
On-orbit IPC calibration could occur in two ways:
- Flat-Field Auto-Correlations: A Poisson-dominated flat field has a correlation coefficient between neighboring pixels of 2\alpha. Thus, a sequence of flat fields can be use to measure the IPC \alpha.
- Hot Pixels: In dark exposures, the signal of isolated hot pixels sloshes to their neighboring pixels and can be used to model the IPC smoothing kernel. For typical values of \alpha of 0.015, a signal of 104 electrons n a pixel should produce a signal of about 150 electrons in each of the side-adjacent pixels.
Cadence
IPC characterization can use both dark and flat calibration exposures obtained for the weekly dark signal and monthly pixel-level flat calibration campaigns.
Calibration Files
One IPC kernel is provided for each detector and hosted on the Roman Calibration Reference Data System (CRDS). IPC is not used in the Exposure Pipeline, but the effects will be captured in the effective point-spread function (ePSF) hosted on the Roman Calibration Reference Data System (CRDS). The IPC reference file is used by the Roman WFI Exposure Time Calculator and can be used in STPSF outputs (but are not applied by default).
Pixel Behavior Families
Description
Pixel behavior families classify different types of unstable pixels. These pixels can have very low or unstable response to light, elevated dark current, or other anomalies. WFI Ground Testing Campaigns provided extensive cataloging of these pixels, but behaviors can change over time and the algorithms can be repeated using in-flight data. Hot, warm, and unstable pixels add uncertainty to the measurements. Knowing the variability of the hot pixel signal is at least as important as knowing the typical values.
Requirements
The WFI has a pixel operability requirement that more than 95% of the pixels have stable properties.
Calibration Activities
Pixel behaviors can be identified from a wide-range of data types. The activities are largely opportunistic with other calibration activities.
Cadence
Many pixel behaviors can be catalogued by exposures obtained for the weekly dark signal and monthly pixel-level flat calibration campaigns, while others will show anomalous behaviors in other calibration types.
Calibration Files
Static bad pixel locations are encoded in the data-quality (DQ) array of all Level-2 products, with values originating from the mask reference file and from the DQ arrays of other reference files. More information on DQ in different data-product levels can be found in the Data Quality Information article of the Data Handbook.
For additional questions not answered in this article, please contact the Roman Help Desk.
References
- Roman Science Requirement Document, revision D: https://roman.gsfc.nasa.gov/science/docs/RST-SYS-REQ-0020D_DOORs_Export.pdf
- Roman On-Orbit Calibration Plan, 2026 edition
- CRDS: https://roman-crds.stsci.edu/
- "The Statistical Properties of Dark Ramps for the Roman-WFI Detectors", Betti et al. 2025