In-orbit Performance of the Near-infrared Spectrograph NIRSpec on the James Webb Space Telescope

The quality of the image delivered by the OTE is of fundamental importance for all JWST instruments, as it limits the spatial resolution and sensitivity of images. As discussed in other papers in this edition (McElwain et al. 2022; Menzel et al. 2022), the in-orbit optical performance of the OTE is significantly better than expected, with lower wave front errors and therefore a sharper point-spread function (PSF) across the entire FOV (Feinberg et al. 2022). More specifically, the JWST PSF is diffraction limited already at 1.1 μm, and reaches a Strehl ratio of 0.9 at 4 μm across the FOV.

While this is clearly good news for all JWST instruments, the NIRSpec sensitivity stands to gain the most. This is because a narrower PSF minimizes the "path losses" caused by the physical truncation and subsequent diffraction of the PSF by the narrow apertures of the NIRSpec fixed slits and microshutters. As discussed in Giardino et al. (2022), this is the main reason for the fact that the NIRSpec slitlosses are significantly smaller than expected, with peak gains of more than 10% at the shorter wavelengths. The resulting benefit to the NIRSpec throughput is discussed in Section 6.1.

In addition to the low wave front error, the OTE optics are also free of water ice or other molecular contamination, resulting in a very high telescope throughput across the entire NIRSpec wavelength range. Moreover, the good thermal performance of the sunshield, and the resulting nominal temperatures of the OTE result in thermal backgrounds that are at or below the pre-launch expectations (Rigby et al. 2023a), also enhancing the sensitivity of NIRSpec observations. Taken together, the outstanding optical and thermal performance of the JWST OTE provides a much better wave front to the NIRSpec optics than expected, which is reflected in the sensitivity numbers discussed in Section 6.1.

Another important factor for the quality of NIRSpec observations is the pointing accuracy and stability of the JWST Attitude Control System (ACS), which is discussed in detail in Menzel et al. (2022). Accurate "blind pointing" is required to ensure that the science target is indeed imaged onto the rather small detector area used to identify the target by the onboard TA algorithms, and that the corrective small-angle maneuvers (SAMs) are not too large. The ability to accurately execute those SAMs, and the absence of any drifts or low-frequency jitter in the telescope pointing is crucial for stable and precise target placement throughout the exposure, which is particularly important for time series observations of transiting exoplanets, as discussed further in Section 6.4. On all these accounts, the JWST observatory meets or exceeds the pre-launch requirements, which is the foundation for the problem-free execution of the various NIRSpec observing templates, and the outstanding NIRSpec science performance presented in this paper.

NIRSpec is equipped with two wheel mechanisms: (i) the Filter Wheel Assembly (FWA) with five long-pass transmission filters (F070LP, F100LP, F170LP, F290LP, and CLEAR) that define the wavelength ranges for the dispersers, two filters used for the TA exposures (F110W and F140X), and one position (OPAQUE) that serves both as an instrument shutter toward the telescope side and a coupling mirror for illumination from the CAA on the instrument side, and (ii) the Grating Wheel Assembly (GWA) with the seven NIRSpec dispersers, and a mirror for undispersed imaging (see also Table 6). The functionality of both wheels was verified during commissioning, and their torque profiles were characterized and found to be nominal, i.e., similar to those measured during ground testing, and well within their expected tolerances.

As discussed in Jakobsen et al. (2022), the limited mechanical angular reproducibility of the GWA ball bearings causes small, but significant, variations in the position of NIRSpec spectra on the detector, especially in dispersion direction. Since this has an obvious effect on the wavelength calibration as well as the precision of the TA algorithm, it is important that the actual GWA position of any given NIRSpec exposure can be accurately inferred from telemetry. To this end, the GWA design includes two magneto-resistive position sensors (a.k.a."tilt sensors") which were extensively tested before launch (De Marchi et al. 2012) to ensure that NIRSpec can indeed achieve the required accuracy of its wavelength calibration.

The tilt sensors have to be calibrated after each cooldown, and their post-launch calibration was one of the important goals of the NIRSpec commissioning campaign. The accuracy of the in-orbit sensor calibration has been discussed in detail by Alves de Oliveira et al. (2022): the remaining uncertainties are consistently below 1/10 of a pixel, which is fully in line with expectations and sufficient to ensure a reliable wavelength calibration for all NIRSpec spectra, as demonstrated further in Section 6.3. The tilt sensor performance and the stability of the wavelength calibration will be monitored with a roughly semi-annual cadence as part of the overall JWST calibration plan, using a combination of stellar spectra and the internal LINE and REF calibration lamps (see Jakobsen et al. 2022).

The NIRSpec microshutter array (MSA) has performed excellently throughout flight thus far, with no unexpected hardware issues, overall multiplexing remaining very high, and an average success rate of ∼96% for commanding shutters open in science-like patterns (Rawle et al. 2022). Operability at the level of individual shutters is of course crucial for the MOS mode itself, but less obviously also for the IFS mode, where problems with the shutter array can become a source of parasitic light, either contamination through stuck-open shutters or thermal glow ²³ emanating from electrical shorts in the circuitry. Our knowledge of this performance is critical to allow mitigation, such as short-masking, and enable users to generate the optimal target set around the inoperable shutter population.

As described in detail by Rawle et al. (2022), the marginal increase in inoperable shutters seen during commissioning was in-line with pre-launch estimates. The primary consequence of these trends is that the masking required to mitigate electrical shorts is now the dominant factor affecting the usability of unvignetted shutters (Table 2). While comparatively fewer shorts have emerged after launch, with approximately one row/column needing to be masked for every 50 shutter array re-configurations compared to one masked per 20–25 during ground testing, shorts have continued to appear throughout the latter stages of commissioning and another two required masking in the first three months of science observations. An added complication apparent for two of the shorts seen since launch (and never during ground testing) was that they produced glow when applying the all-closed shutter configuration for IFS observations. Although successfully mitigated in the usual manner, the contamination before masking affected both MOS and IFS exposures, increasing the overall impact of the shorts.

In the regime that shorts may still occur every few months, there are two operational issues to tackle. First, to ensure that data still meet scientific objectives, users need to be informed in a timely manner when executed exposures are directly contaminated by a short and also when upcoming observations are affected by newly masked shutters. The former may now apply to both MOS and IFS visits. Second, as the population of shorts continues to grow, each removing several hundred shutters from operation, this will eventually have a noticeable impact on multiplexing. Shorts are believed to be caused by particulate contamination of the complex MSA control electronics, and those particles may shift during reconfiguration, which is the origin of new shorts but also potentially clears the cause of previously detected shorts. Therefore, as discussed by Rawle et al. (2022), re-checking whether older shorts still remain may be an avenue to recover previously unusable shutters and maintain the multiplexing of NIRSpec MOS at the current high level.

The NIR light collected by the JWST OTE and fed through the NIRSpec optical train is registered by two Hawaii-2RG (H2RG) sensor chip assemblies (SCAs), which are described in detail by Rauscher et al. (2007). The SCAs are operated at a temperature of 42.8 K, chosen as the best compromise between pixel operability and total noise of the detector system: a higher temperature leads to an increased number of hot pixels, while a lower temperature leads to higher noise in the signal chain.

As a reminder, ²⁴ the NIRSpec SCAs are non-destructively read "up-the-ramp," using one of two fundamentally different readout modes: the so-called traditional readout mode (TRAD) or the improved reference sampling and subtraction (IRS²; pronounced "IRS-square"; Rauscher et al. 2017) readout mode. A number of detector subarrays with different sizes such as, for example, the ALLSLITS subarray are supported in TRAD mode only, offering faster readouts and using a slightly higher conversion gain to make the full physical well depth of the pixels accessible, which is of particular importance for time series observations of bright targets.

The in-orbit performance of the NIRSpec SCAs and their readout electronics is an important factor for the NIRSpec science performance, because the sensitivity of most NIRSpec observing modes is limited by the total detector noise in a given exposure (see Appendix A in Jakobsen et al. 2022). The analysis of the detector performance data collected during the NIRSpec in-orbit commissioning campaign characterization has been presented in detail by Birkmann et al. (2022b). Here, we briefly summarize the most relevant results.

5.3.1. Cosmetics

5.3.2. Dark Current

5.3.3. Noise Performance

The total noise for the different readout modes of the two NIRSpec SCAs as a function of effective integration time is summarized in Table 5. These numbers include the effects of cosmic rays, i.e., broken ramps due to cosmic ray hits and early saturation for some pixels.

Table 5. Total Noise of the Two NIRSpec Detectors for Different Readout Modes and Effective Integration Times as Measured During Commissioning

	Effective Integration Time
Readout/SCA	∼950 s	∼1700 s	∼3560 s
TRAD/NRS1	6.9	7.4	N/A
TRAD/NRS2	7.3	7.7	N/A
IRS2/NRS1	5.9	6.6	8.5
IRS2/NRS2	7.2	7.6	9.2
SUB/NRS1	7.0	7.8	N/A
SUB/NRS2	7.0	7.5	N/A

Download table as:  ASCIITypeset image

Figure 2 shows the noise behavior as a function of integration time in more detail. For all readout modes, the total noise decreases with exposure length, up to integration times of ≈500 s where it levels out and then slowly increases for longer integrations. Nevertheless, for observations of faint sources that are detector noise limited, it is still beneficial to use longer integration times for optimal signal-to-noise ratios. On the other hand, it is always advisable to have multiple integrations per observation, ideally in the form of dithered exposures to guard against early saturation after a strong cosmic ray hit, as well as to enable a robust rejection of outliers.

Zoom In Zoom Out Reset image size

The IRS² readout mode shows the best total noise performance, in particular for NRS1, and should be used when observing faint targets. The ALLSLITS subarray readout mode has noise levels comparable to those in traditional full frame mode (TRAD), and is best used for observations of bright targets with short integrations, because the effective full well depth is higher due to the higher conversion gain, like for the other subarrays.

5.3.4. Cosmic Ray Environment

5.3.5. Persistence

For NIRSpec, and in fact any optical instrument, a critical performance parameter is its efficiency, i.e., the fraction of photons incident on the primary mirror that are being registered by the detector after passing through the entire optical path. This metric, together with the noise performance of the NIRSpec detectors, drives the ultimate sensitivity for astronomical observations.

Because NIRSpec is a complex instrument that supports many different observing modes, its optical train is rather complicated, as evident from Figure 1. Except for the order-separation filters and the low resolution double-pass prism, the NIRSpec optics are reflective throughout. Photons captured by the JWST primary mirror undergo a total of 19 reflections before reaching the NIRSpec detector array in the MOS, FS, and BOTS observing modes. For IFS mode, there are an additional 8 reflections in the IFU optics (see Böker et al. 2022a). Cleanliness and high reflectivity of the NIRSpec optics therefore was of ultimate importance throughout the construction and test phases.

By necessity, the metric of choice to measure the optical throughput is the Photon Conversion Efficiency (PCE), i.e., the ratio of photons incident on the JWST primary mirror to electrons registered by the NIRSpec detector system. It can be derived using observations of a "flux standard" (typically a well-characterized star) with an accurately known spectral energy distribution (SED). As can be seen from Table 6, NIRSpec has a total of nine spectral configurations, and the throughput must be measured separately for each of them.

As described in more detail by Giardino et al. (2022), the NIRSpec optical efficiency meets or exceeds the pre-flight expectations. In particular, significantly higher than predicted PCEs are achieved for the high-resolution configurations, for the MOS/FS mode, and for all configurations below ∼2.6 μm for the IFS mode. The 10%–20% lower than expected efficiencies above ∼4 μm, apparent in particular for the IFS mode, are (mostly) explained by the more significant path losses at longer wavelengths, which are reflected in the measurements but not the predictions (see Figure 3 in Giardino et al. 2022).

These in-orbit PCE measurements, together with the detector noise performance discussed in Section 5.3, can be used to calculate the in-flight NIRSpec sensitivity for its various science modes and configuration, following the methodology described in Appendix A of Jakobsen et al. (2022). Figure 3 shows the results of the calculations in terms of continuum sensitivity curves for a point source observed in the MOS and IFS modes, using a bench-mark observation in all spectral configurations. The achieved level of sensitivity is extremely impressive when compared to other NIR spectrographs with similar observing modes and spectral resolutions: by at least two orders of magnitude, NIRSpec is the most sensitive NIR spectrograph currently available for astronomical studies (see also Rigby et al. 2023b).

Zoom In Zoom Out Reset image size

The full photometric calibration of NIRSpec requires three major steps in the reduction pipeline, with their associated reference files: (i) the detector flat (D-Flat) to capture the pixel-to-pixel response variations, and derived from component-level ground test data of the two NIRSpec SCAs, (ii) the spectrograph flat (S-Flat) to correct the throughput variations of the spectrograph optics, and measured from exposures using the internal calibration lamps in the CAA, and (iii) the FORE optics flat (F-Flat) to characterize any field- and wavelength-dependent effects caused by the OTE and the NIRSpec FORE Optics. For the last step, observations of spectro-photometric standard stars are necessary to verify the integrity of the entire NIRSpec optical system, in particular the pick-off and FORE optics (including the FWA), neither of which can be illuminated with the CAA lamps (see te Plate et al. 2005, for a detailed description of the light path from the CAA). For a detailed overview of the NIRSpec flat field strategy we refer to Rawle et al. (2016).

The S-Flat step corrects any throughput variations in the spectrograph, i.e., introduced in the optical path after the FWA and before the detector. The S-Flat reference files for all NIRSpec observing modes are derived from internal lamp exposures and after correction for the detector response (D-Flat) and spectral energy distribution of the lamp(s). Because the throughput is strongly dependent on the properties of the chosen disperser and slit aperture, different reference files are created per detector for each combination of prism or grating and NIRSpec observing mode (FS, IFS, or MOS). For the FS and IFS modes, the S-flat consists of two parts: (i) a 2D image capturing the pixel-to-pixel variations for all slits/slices, and (ii) a vector for each slit and the IFU, to correct the fast throughput variations with wavelength. This approach is possible for the FS and IFS mode, because for a given aperture and spectral configuration, each detector pixel is always illuminated by the same wavelength (modulo the non-repeatability of the GWA, which is corrected separately using the position sensors described in Section 5.1).

For MOS mode, in contrast, the same detector pixel can receive light of different wavelengths, depending on the position of the open MSA shutter. Hence, the MOS S-flat reference consists of a data cube, sampling the wavelength variation for each pixel, so that the specific S-flat for any given MSA configuration can be derived on-the-fly by the pipeline. Given the large number of shutters and the finite amount of time available for commissioning, direct measurement of the throughput for every shutter was impossible. Therefore, only a subset of MSA shutters was observed, from which the complete (smoothed) S-flat cube could be generated by interpolation.

Similar to the S-Flat, the F-Flat must be created for each of the filter and grating wheel combinations listed in Table 1 and for each of the three observing modes (FS, IFS, and MOS ²⁵ ), to take into account any wavelength dependence on the throughput of the OTE and FORE optics. Standard star observations were obtained for all but the F100LP/G140M configuration, for which calibration files based on simulated data will remain in place until such observations are taken during Cycle 1.

We used the A3V star 1808347 (2MASS J18083474+6927286) for all gratings, while for the CLEAR/PRISM configuration, we used either the A8III star 1743045 (2MASS J17430448+6655015), or the G0-5 star SNAP-2 (2MASS J16194609+5534178), because 1808347 would saturate the low-resolution spectra. For the fixed slits, we used a 3-nod pattern (2-nod pattern for S1600A1) to obtain the data, and a 4-nod pattern with a 4 spaxel (04) extraction radius for the IFS observations. The MOS program was executed using a 3-shutter nod pattern for each of the filter/grating combinations.

To derive the conversion to absolute flux units (which is part of the F-Flat step), the extracted spectra were divided by the resampled standard star templates available on the CALSPEC website ²⁶ (Bohlin et al. 2014). To create a smooth F-Flat, any outliers such as remaining hot pixels, stellar absorption lines, or the 0th order contamination of the lamp spectra used to create the S-Flat, were masked and a smooth vector was fitted, creating the final F-Flat.

To verify the validity of the NIRSpec calibration approach, we re-extracted the spectrum of the standard stars using the NIRSpec-internal data reduction pipeline, and compared the resulting spectra to the CALSPEC template, as shown in Figure 4 for the case of FS mode with the S1600A1 aperture. Overall, the agreement is very good, with an rms of the residuals well below 2%. For the other NIRSpec modes and apertures, the results are of similar quality. Note that this comparison only represents an estimate of the "best case" calibration accuracy, because it is performed on the same star that was used to derive the pipeline reference files, and thus does not account for any systematic uncertainties.

Zoom In Zoom Out Reset image size

While an evaluation of the systematic errors must wait for additional calibration data obtained during Cycle 1 or later, commissioning spectra of two other standard stars, WD1057 + 719 (a DA1.2 white dwarf, M-gratings) and P177D (a G0V star, H-gratings), were obtained in FS mode with the S1600A1 slit (see Table 1). These observations were executed early in commissioning and before the wide aperture TA procedure (see Section 6.5.1) was available. Therefore, a proper centering is not guaranteed and (small) uncorrected aperture losses are possible. Nevertheless, we measure residuals between the flux template and the extracted spectra of −2.12% ± 2.55%, −3.76% ± 1.00%, −3.67% ±1.57%, −5.09% ± 1.16%, −1.00% ± 1.21%, and −0.91% ± 1.97% for the F070LP/G140H, F170LP/G235M, F170LP/G235H, F290LP/G395M, F290LP/G395H, and CLEAR/PRISM configurations, respectively.

These residuals are well below the pre-launch requirement (10% absolute photometric accuracy of all NIRSpec spectra, i.e., even after correction for "delta" slit losses due to imperfect source centering). While source centering is most critical for the 200 mas wide slits, it is worth mentioning here that after a successful target acquisition procedure (see Section 6.5), the source is typically placed within 10 mas of the slit center. This magnitude of source displacement would result in insignificant "delta" slit losses.

As noted above, the results presented here were derived using the NIRSpec-internal data reduction pipeline. The user pipeline provided by STScI is currently being checked and improved to deliver similar results, and the status is discussed further in Section 7.

The accuracy of the NIRSpec wavelength calibration is determined by the performance of the parametric instrument model, which was a crucial product to be derived from NIRSpec commissioning data. As described in Lützgendorf et al. (2022), the residuals in the final instrument model were well within the requirements for the wavelength and astrometric calibration of NIRSpec data. This is illustrated in Figure 5, which shows wavelength-calibrated spectra of the spectrophotometric standard 1808347, an A3V star with prominent hydrogen absorption features, taken through the 200 mas wide fixed slits.

Zoom In Zoom Out Reset image size

As can be seen, the NIRSpec instrument model allows for a highly accurate wavelength calibration, with residuals for internal lamp spectra well below the requirement (1/8 of a resolution element) for all gratings analyzed so far. Note that the full verification for all pipeline products in the various NIRSpec modes is still ongoing, especially in the case of the IFS mode.

The temporal stability of the instrument response to a constant flux stimulus is a critical parameter for the accuracy of measured light curves of astronomical targets, e.g., in the case of Time Series Observations (TSOs) of transiting exoplanets. In order to quantify the stability of the NIRSpec response over various timescales, we obtained a TSO of the star HAT-P-14 (PID 1118; PI: Proffitt) during the transit of its well-characterized exoplanet HAT-P-14 b using the G395H/F290LP grating/filter combination (see Table 6). The observations, which also served the purpose of verifying the correct execution of the BOTS observing template (see Birkmann et al. 2022a), were analyzed in detail by Espinoza et al. (2022).

To recap, the transmission spectrum of HAT-P-14 b could be measured with a precision of 50–60 ppm at R = 100 (rebinned down from R = 2700), which is in excellent agreement with pre-flight expectations, and close to the photon-noise limit for a J = 9.094, F-type star like HAT-P-14. There were two noteworthy features observed in this analysis. The first was a weak linear trend in the white-light response as a function of time, which was observed to be stronger in NRS1 (−150 ppm hr⁻¹) than in NRS2 (30 ppm hr⁻¹). This was shown to also be slightly wavelength dependent in NRS1, but not in NRS2. These trends can be easily corrected for, and are most likely caused by low-amplitude instabilities in the detector signal chain. The second is the fact that binning pixels in the spectral direction seems to not decrease the noise level in the time-series as $1/\sqrt{{N}_{{\rm{bin}}}}$ where N_bin is the size of the bin. It is likely this is related to unaccounted covariance between pixels due to effects such as, e.g., 1/f noise. This can also be accounted for by either simply working with the added pixels on each spectral channel or considering a more complex spectral extraction and binning scheme that includes this spatial covariance of pixels in the detector (see, e.g., Schlawin et al. 2020).

Given the excellent performance of the BOTS mode, and the absence of any strong variations and systematic trends in the combined response from telescope and instrument over timescales of a few hours to a day, NIRSpec promises to fulfill its key role for cutting-edge transiting exoplanet atmospheric science with JWST during Cycle 1 and beyond.

Depending on their scientific needs, NIRSpec users have a number of options for fine-tuning of the telescope pointing after the completion of the initial slew to the target field, which typically places the source within 100 mas of its intended position. Programs for which this level of accuracy is sufficient do not need a dedicated TA procedure, and should select TA = NONE or TA = VERIFY_ONLY. The difference between these two options is that (for MOS and IFS mode) the latter obtains an undispersed image of the sky after the spectroscopic exposures in order to allow a determination of the pointing post-facto.

Programs that require a more accurate target placement of an individual source in either the IFU, one of the fixed slit apertures, or the MOS single-point field position should use the NIRSpec Wide Aperture Target Acquisition (WATA) procedure. For MOS observations, on the other hand, the added complexity of requiring a roll angle optimization calls for the use of a dedicated method called Micro-Shutter Array Target Acquisition (MSATA). The details and in-orbit performance of both of these TA methods are discussed in the following subsections.

6.5.1. Wide Aperture Target Acquisition

The WATA procedure takes an image of an isolated, point-like target through the 16 × 16 wide fixed slit aperture (S1600A1). Using this image, the onboard software then computes the centroid of the source emission to determine its position after the initial "blind" telescope slew, and autonomously calculates the corrective "delta" slew required to accurately position either this target or another nearby target at the optimal location in the NIRSpec science aperture. The total duration of the WATA procedure can be as short as 5 minutes, and as long as 11 minutes, depending on the size of the detector area being used (subarray or FULL frame), and the depth of the acquisition exposure.

Figure 6(a) shows the WATA image of a star used to verify the WATA process during commissioning, obtained from Proposal ID 1118 (PID1118). This program contained six successful WATA attempts ²⁷ that were used to evaluate the onboard algorithms, the correct computation and execution of the offset slew, and the subsequent science exposures. Figure 6(b) shows the position of the source after the initial slew, i.e., at the beginning of each WATA. Because of the excellent slew and pointing performance of the observatory, the WATA starting position is already quite close to the aperture center: on average, the target is within 50 mas radial distance, or one half of the size of a NIRSpec detector pixel.

Zoom In Zoom Out Reset image size

Figure 6(c) shows the target position at the end of the WATA procedure, i.e., after the corrective slew to the center of the S1600A1 aperture was computed by the onboard algorithm, and executed by the telescope guiding system. For the six successful WATA observations in PID1118, the average target position was within 2.5 mas from the aperture center, which is almost ten times better than the pre-launch requirement of 20 mas.

For observations that require the science target to be placed in apertures other than the S1600 slit itself, further tests executed during commissioning and the early science program have demonstrated that the telescope slew accuracy is sufficient to place the target at, e.g., the intended location in the IFU aperture with an accuracy of better than 10 mas, again well within the requirement.

6.5.2. Microshutter Array Target Acquisition

The NIRSpec MOS mode requires that the images of astrophysical targets are accurately placed onto the MSA over the entire ($3^{\prime} \times 3^{\prime}$) FOV, such that they fall within their dedicated microshutters. The onboard algorithms to achieve this rely on highly precise coordinate transformations between the detector, MSA, and sky planes, which are another crucial output of the NIRSpec parametric instrument model (Lützgendorf et al. 2022).

As explained in detail by Keyes et al. (2018), the NIRSpec MSA target acquisition (MSATA) process uses a set of 5–8 "reference stars" that are imaged onto the detector through the microshutter grid. For MSATA exposures, the MSA can be either in the all-open configuration or in "protected" mode, which closes the shutters around bright stars to prevent persistence in the subsequent science exposures. Two MSATA exposures are acquired, separated by an offset equivalent to half of the microshutter pitch (in both x and y direction), in order to mitigate the effect of vignetting by the MSA bars. The centroid position of each reference star is calculated autonomously by the on-board software, and the set of centroids is analyzed, outliers are clipped, and their mean offset from the intended position is used to correct the initial spacecraft pointing ("pitch" and "yaw") and position angle ("roll").

The duration of the MSATA procedure depends on the number of reference stars used and the depth of the acquisition exposures, ranging from 23 minutes for the minimum number of reference stars (5) and the shortest readout pattern, to 35 minutes for 8 reference stars (the recommended number), and the deepest available readout pattern. In cases where the MSATA algorithm does not converge to a sufficiently accurate solution, it may be repeated once, adding up to 22 minutes to the duration if the maximum of 8 reference stars is used.

The NIRSpec MSATA process is the only operational situation where the roll angle of the Webb telescope is adjusted after the start of an observation.

To provide an impression of the complexities involved, Figure 7 shows the image used to verify the NIRSpec MSATA process during commissioning (PID 1117, Obs. 31). The target field is the astrometric calibration field (Anderson et al. 2021) in the Large Magellanic Cloud (LMC) which had been pre-observed with HST and ground-based telescopes to provide accurate stellar coordinates for the astrometric calibration of the JWST focal plane. In such highly crowded fields, careful planning is necessary to identify the optimal MSATA reference stars, which must not saturate in the TA exposure, and must be suitably isolated for optimal centroid calculation by the on-board algorithm.

Zoom In Zoom Out Reset image size

To allow accurate calibration of MOS spectra, the MSATA process must place the science targets with an accuracy of just 20 mas across the NIRSpec FOV. The most important driver for the MSATA accuracy is the absolute anchoring of the planning catalog in rotation, as it is critical for the proper derivation of the TA roll solution. If the desired accuracy of the target placement in the shutters can be relaxed, the MSATA can also be planned using catalogs with poorer astrometric precision or rotation anchoring, or using galaxies with compact central cores instead of stars.

To evaluate the performance and post-TA target positioning accuracy, we have analyzed the 20 NIRSpec MSATA procedures carried out so far. Figure 8 shows their distribution of corrective TA slews (in V2, V3 and roll). We find all slew offset solutions to be within 100 mas radial offset and at an average −51'' roll offset compared to the optimal pointing. These relatively small corrections are a testament to the excellent 'blind' pointing accuracy of JWST which allows very accurate target placement even without any TA.

Zoom In Zoom Out Reset image size

To illustrate the improvement achieved by the MSATA procedure, Figure 8 shows the target placement after MSATA execution for the same 20 observations. These measurements were derived by running an offset analysis on the reference image that is acquired after the final TA slew is complete. As can be seen, the average radial offset over the ensemble of TA reference targets is 25.0 mas, and the average roll offset from the optimal pointing solution is 14''. This is close to, but not yet fully in line with the requirements, but it should be kept in mind that the more critical displacement in cross-dispersion direction should be a factor of $\sqrt{2}$ smaller. In addition, many of the observations analyzed so far have used catalogs with non-optimal astrometric accuracy, and/or non-stellar objects as reference targets.