The TESS Mission Target Selection Procedure

The first challenge of the target selection process is to select targets from each CTL in a way that fairly allocates the limited number of target slots while maximizing the scientific return of the mission. In order to do this, the POC adopted the following strategy.

For any number of input CTLs, each CTL is assigned a maximum number of target slots. The CTLs are then searched in a specific order for targets. Once the maximum number of targets allocated to a CTL has been selected, the remaining targets from that CTL are ignored, and selection from the next CTL begins. The number of targets for each CTL and the order in which CTLs are selected are adjustable parameters, allowing the POC to distribute the 20,000 target slots across different CTLs in a flexible way.

For the TESS primary mission, targets were selected from the CTLs in the order shown in Table 2: ppa, bright, exo, astero, GI, DDT. The maximum number of target slots that were assigned to each CTL are also given in Table 2: 1920 for ppa, 13,400 for exo, 751 for astero, 1501 for GI, and 1501 for DDT. No limits were assigned to the bright CTL; in practice, the number of observable bright stars was between about 450 and 900 targets, all of which were assigned pixel subarrays. These limits were initially based on predefined resource allocations. However, the limits were refined over the first five TESS observing sectors, and we report the final choices here.

The order in which the CTLs are searched for targets is important because a single target is often requested at high priority in multiple CTLs. If a target appears in multiple CTLs, it is only counted against the total allotment in the first CTL that requested it. For example, a GI target that was first selected by the exo CTL is not counted against the 1501 slots allotted to the GI CTL, nor would an exo target that was first selected as a PPA star be counted against the 13,400 slots for the exo CTL. In this way, the POC algorithm does not need to compare the priorities between different CTLs. However, there is an advantage for a CTL to come up later in the target selection process, because a high priority target selected by an earlier CTL will not be counted against a later CTL's total allocation of targets.

The POC balances this advantage against the total number of assigned target slots—because the exo CTL is assigned nearly ten times as many target slots as each of the the astero, GI, and DDT CTLs, it obtains an overwhelming number of high priority targets regardless of its order in the target selection process. Furthermore, the exo CTL would benefit less from the small number of targets selected from the astero, GI, and DDT CTLs than those CTLs benefit from a large number of selected exo CTL targets. Selecting from the exo CTL first therefore increases the total number of targets observed from each CTL, and accordingly bolsters the total scientific return of the mission. Similarly, the ppa and bright star lists represent required targets, and so are selected first to the benefit of the exo CTL.

Figure 1 shows an example of the target selection results for Sector 20, highlighting the extra targets observed from each CTL beyond the fiducial limits in Table 2. Observable stars from the CTL ("on silicon") are shown with the black histograms. "Selected" stars are stars that the CTL explicitly requested based on their priorities (see Section 3.3) and are added to the observed target list. However, each star is only "billed," that is counted against the target allotments in Table 2, if it has not already been selected by a previous CTL. Because targets are selected from the exo CTL first, there are very few examples of this kind in the top panel of Figure 1 (i.e., the difference between the cyan and red histograms is very small). The difference between the billed and selected histograms is more pronounced for the astero and GI CTLs because they have a high degree of overlap with the exo CTL.

Zoom In Zoom Out Reset image size

Stars that are listed at low priority in a given CTL are also often selected at high priority in a different CTL. The blue histograms in Figure 1 show the total "observed" stars, meaning all stars added to the target list that appear in the given CTL but were not necessarily added based on priority. The lists of billed and selected stars are always a subset of the observed target list, but a substantial number of lower priority targets are also added. These targets are freely observed for a given CTL in the sense that they are billed to a different CTL.

Figure 2 compares the number of these additional targets to the baseline allocation. On average, the exo CTL receives about 25% additional targets compared to the allocated slots, while the astero and GI receive a factor of about 2.1 and 3.4 times as many targets as allocated slots, respectively. These extra targets suggest an improved scientific return from the target selection algorithm compared to the baseline resource allocation given in Table 2. In the Appendix, we provide a table with the number of billed and observed targets for each sector from each CTL.

Zoom In Zoom Out Reset image size

Finally, not all target star slots are filled for every CTL in every sector—this depends on the overall size of the CTLs and the distribution of CTL targets on the sky. For example, the DDT CTL rarely reaches the full allotment of 1501 targets, especially for the first half of the primary mission when relatively few DDT proposals were submitted (see the Appendix). Thus, after selecting targets from all CTLs, any remaining target slots (up to 20,000) are filled by re-selecting stars from the exo CTL. Typically, about 1 500 additional exo targets are added in this way, so that the exo CTL usually reaches a total of ∼15 000 observed targets per sector.

In addition to the constraint of 20,000 targets observed per sector, no more than 2000 targets can be selected per TESS CCD. This requirement is usually in tension with the priorities of the CTLs, which often assign higher priorities to targets near the ecliptic poles where TESS can observe the targets over multiple sectors. Figure 3 shows the sky distribution of the top 200,000 targets by priority in the year 1 and year 2 exo CTLs—roughly 55% of these targets are within 36 degrees of the ecliptic pole (the area of the sky observed by Cameras 3 and 4). Of the highest priority 20,000 observable targets per sector, on average about 60% would be assigned to Camera 4 and 90% would be assigned to Cameras 3 and 4 combined.

Zoom In Zoom Out Reset image size

A more even distribution of targets over the 16 TESS CCDs results in better systematic error correction by PDC, and the POC makes a concerted effort to more evenly distribute targets over the full TESS field of view. To do this, the POC imposes a maximum number of targets from each CTL selected on each CCD. The particular number is an adjustable parameter, and can be used to accommodate a range of target distributions across the CCDs. For example, a perfectly even distribution of targets could be obtained by dividing the limits assigned per CTL (Table 2) by 16 and setting the parameter to this value for all CCDs. On the other hand, the scientific priorities can completely determine the target selection by setting the limits on each CCD to 2000 targets (the maximum allowed by ground systems). Intermediate cases that balance these two options can be obtained by setting the limits for each CCD between these two extremes. The limits for individual CCDs also do not need to have the same value for a given CTL, and so different CCDs/cameras can be more highly weighted with respect to the scientific priorities versus technical constraints.

No CCD limits were applied to the bright, astero, and GI CTLs—owing to their small size, it is more important to select as many high priority targets as possible rather than enforce an even distribution across CCDs. The exo CTL, owing to its large number of target slots, can absorb the work of dividing targets across the many CCDs for systematic error correction. However, we give extra weight to the scientific priorities of exo CTL targets in Camera 4, because these targets will have long baseline light curves and therefore additional scientific value. We assign 1400 exo targets to each of the 4 CCDs in Camera 4 from the exo CTL, with 650 assigned to each of the other 12 CCDs in Cameras 1, 2, and 3. The DDT list is assigned a maximum of 500 targets per CCD. Although this limit suggests no particular preference per CCD for the DDT CTL, the total number is well beyond the allotment in Table 2 and allows DDT targets to be distributed mainly based on scientific priority.

The available slots for Camera 4 have usually filled up by the time the second pass of the exo CTL is implemented. Thus, many higher priority targets in this region of the sky are skipped in order to add targets to the other three cameras. In some cases, the slots in Camera 4 are exhausted during the GI target selection—this depends specifically on the observing sector, patch of sky, and available targets and their priorities in other cameras.

TESS is usually pointed at +54° ecliptic latitude so that the center of the field of view of Camera 4 is near the ecliptic pole. However, in six sectors of year 2, TESS was pointed at +85° in ecliptic latitude to avoid high backgrounds caused by earthshine in Camera 1 (Sectors 14–16 and 24–26). In these cases, Cameras 2 and 3 straddle the north ecliptic pole, and we adjusted the weighting of targets across CCDs/cameras and CTLs to account for this. All CCDs in Camera 3 were allotted 1150 exo CTL slots, along with the two CCDs in Camera 2 nearest the ecliptic pole. The other CCDs were assigned 650 exo CTL slots. No change was made for the number of targets per CCD for the other CTLs.

For each CTL, the POC adds targets with the following procedure. First, we filter each CTL to identify which targets land on the imaging regions of the CCDs (see 3.4). Next, we sort the targets by decreasing priority. For the highest priority target, we then check if adding its associated pixels violates any technical requirements—see Section 3.4 for a description of how pixel subarrays are assigned for each target and Table 3 for a full list of tests. If the target passes all tests, we add it to the target list, update the pixel table to track the associated subarray pixels, and write files that command the instrument to save these pixels at 2 minute cadence. We then repeat this process for the next-highest priority target and iterate until reaching the maximum number of targets per CTL, or exhausting the observable targets from the CTL. If a target lands on a CCD which is already full for a given CTL (Section 3.2), that target is skipped.

Storage space in the SSR onboard the spacecraft also implies a limit on the total number of pixels that can be collected at 2 minute cadence, and a parameter is available to limit the total number of pixels associated with 2 minute targets. However, limiting the total number of pixels for 2 minute targets would lead to a variable number of selected targets per sector. Instead, we use the total number of targets to limits the 2 minute target selection—20,000 targets have been found to produce a data volume close to (but always less than) the total storage space reserved for 2 minute cadence data.

The total number of targets per CCD is tracked. Once the maximum number of targets on a given CCD is reached, candidate targets that land on that CCD are skipped.

The total number of targets per CTL is tracked. Once the slots allotted to that CTL are filled, the remaining targets from that CTL are ignored and the selection process moves on to the next CTL.

The total number of targets per sector is tracked. Once the maximum number of targets is reached, the target selection loop exits, logs of the targets and their CTLs of origin are saved, and files are produced for tracking the target list/pixel table and commanding the spacecraft. If there are still slots remaining after searching all six CTLs, we return to the exo CTL to fill out any remaining space.

After completing target selection, the final target list is checked against each CTL to find the total number of targets observed per CTL. This number is always larger than the allocated slots per CTL, because targets that are listed at low priority in one CTL are often selected at high priority by another CTL.

The subarray pixels associated with each selected target must contain the optimal photometric aperture and enough background pixels to robustly estimate the sky level. For the majority of targets, an 11 × 11 pixel box is sufficient. For saturated stars, additional rows are collected to fully capture the bleed trails.

The locations of stars in the image are calculated with the TESS focal plane geometry (FPG) model—a full description of the FPG model will be presented in future work. The FPG model is known to be accurate to a few hundredths of a TESS pixel. When calculating the expected position of targets in the TESS field of view, celestial coordinates are corrected for proper motion and velocity aberration. Once the location of a target is identified, it is trivial to define the corresponding 11 × 11 pixel subarray.

For saturated stars, we collect additional rows proportional to the expected flux beyond a "saturation magnitude." The saturation magnitude is defined as the magnitude of a star whose integrated flux would saturate a single TESS pixel in a single two-second exposure. The TESS pixel full-well depth is approximately 200,000 photoelectrons per pixel, with the exact number varying depending on the CCD output node (see Vanderspek et al. 2018, Figure 4.2). Adopting 200,000 photoelectrons per two seconds as the saturation flux results in a saturation magnitude of T_mag = 7.5. We calculate the total number of additional rows by scaling the expected flux of the source relative to the 7.5 magnitude saturation magnitude. As an example, a 2.5 magnitude target has 100 times the flux of a 7.5 magnitude target, and so we expect to require 100 extra rows to collect all the saturated pixels for the brighter source. We allocate half of the extra rows above and half below the original 11 × 11 pixel subarray. If the calculation results in an odd number of extra rows, we always add one additional row so that the extra rows can be distributed evenly above and below the original pixel subarray.

This procedure results in subarrays that grow somewhat faster than the observed bleed trails of bright stars. The faster growth is caused by pointing jitter and the finite width of the instrument's PSF, both of which spread light from a point source across multiple pixels and effectively diminish the amount of flux individual pixels receive relative to the saturation magnitude. It is preferable to have a buffer of pixels around the bleed trails in the subarrays of saturated stars for users interested in photometry of these stars, and so we have not found it necessary to tune the bleed trail growth calculation further. The procedure works well unless the magnitudes of the bright stars are underestimated in the TIC, or multiple bright stars land close enough to each other (within a few pixels) that the combined light saturates the central pixel faster than expected. Although uncommon, several examples have been collected—see the TESS Data Release Notes for specific examples. ³⁴

Apart from extra rows collected for bleed trails, all pixels associated with a target must land in the science imaging region of the target CCD. If a target identified as "on silicon" is too close to the edge of a CCD image (within 5 pixels) it is not added to the final target list and the target selection algorithm continues to the next-highest priority target. The target is also skipped over if any of its pixels are within three columns of a saturated star and the associated bleed rows, where the bleed rows are calculated in the same way as above.

There is a small amount of overlap, about 10 pixels, between the fields of view of each camera. In cases where a target lands in this region, the target's pixel subarray is assigned to the camera farthest from the ecliptic pole in order to save target slots for higher priority targets. Because we require the full 11 × 11 subarray to be observable, this only happens to targets that land on rows 6–10 of a given CCD.

Each CTL may also request a different shape of the subarray using a keyword provided in the CTL. In practice, some GI programs have requested 25 × 25 pixel apertures. In cases where different CTLs request different pixels (e.g., for larger apertures), the superset of pixels is selected for download.

The POC generates a list of PPA stars for each sector, which are used to monitor the instrument performance and derive WCSs for each subarray and FFI. As a reminder, PPA stars must be bright but unsaturated, show little astrophysical variability, be reasonably well isolated in the TESS images (to avoid crowding and possible bias in their position measurements), and be evenly spread out across the TESS fields of view. These requirements are often at odds with each other, in particular when a particularly crowded region of the sky (such as an open cluster or the Galactic plane) would cause a large gap in the coverage of PPA stars across a TESS image.

The POC developed an algorithm to perform the PPA selection in a way that balances these requirements, and includes adjustable parameters that can be used to override the default PPA selection for especially difficult fields. The procedure starts by generating a list of stars between 8th and 10th magnitude that land in the TESS image. The stars must be 5 pixels away from the image edges, as well as 5 pixels away from the edges of CCD output nodes (see Vanderspek et al. 2018, Figure 4.2). We then remove stars that are known variables based on the Variable Star Index (Watson et al. 2006), and we remove stars that land near the bleed trails of bright saturated stars (see Section 3.4). Next, we identify and remove stars with near neighbors using the following criteria: there must be no star brighter than T_mag = T_PPA + 2 within 5 pixels of the PPA star candidate, where T_PPA is the T_mag of the PPA star candidate, and no star brighter than T_mag = T_PPA + 5 within 3 pixels. Stars within 16 pixels of a 6th magnitude star or brighter are also removed, because light from the nearby bright companion can affect the background and centroid estimates of the PPA star. These pixel and magnitude thresholds are adjustable parameters, and can be relaxed if the field of view is exceptionally crowded and few or no viable PPA stars are identified (as occurs near the Galactic center, for example). There is a minimum of 100 PPA stars required for every sector—if fewer stars than this are found, target selection is rerun with adjusted crowding parameters.

Usually several hundred suitable PPA stars per CCD can be found with this method. However, a maximum of 120 PPA stars per CCD is imposed because only 100 PPA star centroids are needed to obtain accurate astrometry and WCSs. The final step of the POC algorithm is to select the subset of 120 stars that most evenly cover the CCD image. First, if fewer than 120 stars are available (but more than 100), a better set of stars cannot be found for this field of view and the stars are selected as PPAs. If more than 120 stars are available, the POC finds pairs of candidate PPA stars and orders the pairs by increasing separation. For each pair in order of increasing separation, the faintest star is removed until 120 PPA stars remain. This results in the maximally spaced set of good PPA stars available on each CCD. We always check the spatial distribution of PPA stars and their magnitudes before uploading the target list tables to the spacecraft—if there are problems (for example, an exceptionally large gap between PPA stars in some portion of the image), we rerun the target selection with adjusted crowding parameters until a suitable distribution of PPA stars is found.

In the first TESS extended mission, starting 2020 July 4, TESS began selecting a limited number of targets for observations at 20 s cadence. The POC uses the same target selection algorithm described above for the 20 s cadence targets, but with parameter changes driven by tighter data margins in the extended mission (caused by 20 s cadence data and 10-minute cadence FFIs). These changes are described below and summarized in Table 4.

• 1.  
  Only 1000 total targets per sector are observed at 20 s cadence, with no more than 400 targets observed per CCD.
• 2.  
  For 20 s cadence observations, 25 PPA stars per CCD are selected instead of 120 PPA stars for 2 minute observations. The smaller number of stars degrades the accuracy and precision of the WCSs for 20 s target data, but the trade-off is acceptable given the limited number of 20 s target slots.
• 3.  
  20 s cadence CTLs are produced by the GI office and DDT program. The POC selects targets from the GI CTL first, followed by the DDT CTL. If any remaining targets slots are available, the GI CTL is searched a second time.
• 4.  
  On the first pass of the 20 s GI CTL target selection, 540 targets are selected. A maximum of 40 targets per CCD is imposed on Cameras 1 and 2, a maximum of 80 targets per CCD on Camera 3, and a maximum of 160 targets per CCD on Camera 4. The DDT targets are limited to 60 total, with a maximum of 10 targets per CCD. These parameters will be adjusted in Year 4 observations for ecliptic pointings and pointing adjustments that avoid scattered light in Camera 1.

The 2 minute target selection changed in the extended mission in the following ways:

• 1.  
  The POC adds a CTL listing the 20 s cadence targets to the 2 minute target selection, so that all targets observed at 20 s cadence are also observed at 2 minute cadence. This is done in order to monitor the pipeline performance, particularly with respect to cosmic ray removal in 20 s cadence data and noise properties of the light curves.
• 2.  
  The POC selects 1501 targets from the exo CTL and 12,001 targets from the GI CTL. The order of selection also changes: ppa, bright, 20 s cadence, GI, DDT, exo. There is no astero CTL in the extended mission separate from targets proposed through the GI program. If any targets slots remain, the GI CTL is searched a second time, and if there are still remaining target slots, the exo CTL is searched a second time.
• 3.  
  The GI targets are subjected to the same process described above for of evenly distributing targets over CCDs for systematic error correction. There is a maximum of 660 targets per CCD in Cameras 1–3, and 1060 targets per CCD in Camera 4. No other CTL has per CCD limits applied.
• 4.  
  The exo CTL is reprioritized in the extended mission, favoring stars promoted to TESS Objects of Interest during the primary mission (Guerrero et al. 2021). Stars that landed in gaps between the TESS CCDs in the primary mission observations are also added, but at lower priority. After these two lists, stars with potential signs of transits in the primary mission, but at low signal-to-noise ratio, are added at the lowest priority. The majority of these targets are likely false positives due to measurement noise, but the targets are added because additional observations in the extended mission might improve the transit search for very small planets.

The mission imposed a requirement that over 200,000 targets be observed at 2 minute cadence in the two-year primary mission. For year 1, 128,292 unique targets were observed, and, for year 2, 104,473 unique targets were observed.

Figure 4 shows the histogram of priorities of selected targets for the exo CTL. The targets from year 1 (Sectors 1–13) and year 2 (Sectors 14–26) are shown separately because the version of the TIC and CTL changed in Sector 14, along with associated priorities in the CTLs. This change corresponds to TIC 8, which is based on Gaia DR2 rather than 2MASS—details are given in Stassun et al. (2019). Figure 4 also shows the subsets of "billed" and "observed" targets compared to the parent population of observable targets ("on silicon"). Almost all of the higher priority targets were selected for 2 minute observations. There is little difference between the billed and observed targets because the exo CTL targets are selected early in the target selection process to the benefit of the astero and GI CTLs (see Section 3.1).

Zoom In Zoom Out Reset image size

Figure 5 shows the histograms of priorities of targets for the astero and GI CTLs. As for Figure 4, the year 1 and year 2 targets are shown separately, as well as the on silicon, billed, and observed subsets of targets. The astero and GI CTLs are allocated a small number of targets relative to the exo CTL, but many more targets were observed because of the CTL merging procedure described in Section 3.1. The overlap between the various CTLs also adds additional targets at lower priority. The differences between the red and blue histograms in Figure 5 serves as a simple metric for the success of the target selection algorithm, because a larger difference indicates additional targets from each CTL that bolster the scientific return of the mission.

Zoom In Zoom Out Reset image size

The mission also imposed a requirement that >5000 targets be observed for more than 250 days in each hemisphere. A total of 6010 such targets were observed in year 1, and 7162 such targets in year 2. Prior to launch, we experimented with forcing stars observed in previous sectors to be reselected in subsequent sectors, regardless of priority. This resulted in relatively small changes in the number of targets observed for more than 250 days, only about 300 more targets than were obtained in a single hemisphere. We, therefore, did not force stars selected in early sectors to be re-observed in subsequent sectors.

Figure 6 shows the sky distributions of targets from the exo CTLs and those actually observed on the sky. The left panel shows the density of the highest priority 200,000 targets in each hemisphere. The right panel shows the density of exo targets selected for observation. Note that different versions of the TIC and CTLs are displayed for the southern (v7) and northern (v8) ecliptic hemispheres.

Zoom In Zoom Out Reset image size

While there is a smooth distribution with decreasing density from the poles to the ecliptic plane in the CTL histogram (left panel), the dominant features of the selected target histogram (right panel) are the discrete boundaries caused by gaps between the CCDs. The density of targets on individual CCDs is mostly flat, and any gradient between the poles and ecliptic is much weaker than in the parent CTL population. The main exceptions are for CCDs that contain the plane of the Galaxy. Crowding due to the high density of stars in the Galactic plane causes targets in this area of the sky to have significantly lower priorities than the rest of the CTL. A single CCD that contains the Galaxy will therefore preferentially add targets away from the Galactic plane, resulting in a clumpy distribution of targets around the edges of these CCDs.

The differences between the sky distributions of CTL targets and selected targets in Figure 6 are caused by the technical requirements of the mission, particularly the requirement that targets be roughly evenly distributed over all TESS CCDs. This requirement is applied to improve the light curve systematic error correction in the TESS mission pipeline (PDC). However, the complicated interaction of the technical requirements and CTL priorities results in an inhomogeneous sample of target stars. In general, the properties of the CTL catalog cannot be used to infer the properties of the sample of stars actually observed by TESS. Therefore, we suggest that researchers use carefully tailored subsamples of TESS targets when studying the statistical properties of TESS planet hosts. A detailed comparison of the stellar properties of the CTL and the stars actually observed for 2 minute observations by TESS will be presented in future work (J. Pepper et al., 2021 in preparation).