FBS 1.4 release (January 2020 update - FBS 1.4 Runs)

FBS 1.4 runs - released in January 2020.

There are 75 simulations in this release (although we have recently added extra runs as part of a “1.4.1” set, upon requests from the community - so as of April 2020, closer to 120 runs). Most previous sets of simulations have been recreated with a consistent set of software (updates/bugfixes), however some runs have been dropped. Please be sure to note which version of runs you are comparing (and generally try to compare runs in the same release). If a run that was useful to you was dropped, please let us know so that we can add it back in. The version information is generally included in the simulation name, it is also contained inside the simulation database file (in the info table).

Release notes

These runs are the first to use a more realistic distribution of seeing, which varies more strongly from season to season and year to year than our previous simulations. For more information on this new seeing distribution and why it was adopted, please see the discussion in https://github.com/lsst/sims_seeingModel/issues/2 along with the documentation linked there. In short, the new seeing distribution has a slightly worse median value, significantly worse tails toward ‘worse seeing’, and more variation from year to year and season to season … the new inputs are based on a much longer set of DIMM measurements than was previously available.

In response to evaluations of the u band filter swap/mount simulations in FBS 1.3, we have implemented a filter swap at 15% lunar illumination, together with a strategy with attempts to take u band visits in pairs – either u paired with u, u paired with g or u paired with r. This results in a more uniform u band coverage across the sky and when paired with another filter, allows for the potential for u band colors to be obtained within a night. To continue the pairing of visits within a night in all filters, we have also added y band pairs (instead of singles, as was the previous strategy) – now y band will be paired with y or with z in most simulations. This means all visits are now paired in the ‘blob’ surveys, which operate under stable (not near twilight) conditions.

As in FBS 1.2 and 1.3, 1x30s visits are used as the defaults for the standard simulations (although 2x15s visits must remain the official baseline and will be tested for all final runs).

Observations for the Deep Drilling fields continue to be refined here (and will be updated further in future simulations - we anticipate that less time will be spent in DDFs than is currently used in the standard simulations, once something closer to the AGN/DESC requirements is implemented). In these runs, DDF observations are now taken at lower airmasses than in previous simulations. The DDF dithering strategies in FBS 1.3 are continued, imprinting a 0.7 degree dithering pattern.

The rotation angles are slightly more restricted than previously, in order to maintain the camera within the rotator boundaries. The rotator angle dithering introduced in FBS 1.3 is continued.

The rolling cadence simulations have improved basis functions implemented, which result in a more uniform final survey. Feedback on the rolling cadences will be greatly appreciated.

In the code itself, some changes have been made to improve cross-platform repeatability, and the configuration scripts have been updated to help improve documentation and readability (and thus maintainability as well). This release was created with the sims_featureScheduler v1.4 together with the configuration and slurm scripts in the directories available at NCSA.

Availability
The simulation output databases are available at https://lsst-web.ncsa.illinois.edu/sim-data/sims_featureScheduler_runs1.4/ : follow the directories in that location (mirrored in the descriptions below) to find specific runs. If you are familiar with the general python configuration scripts used to set up these surveys, you may find the configurations at https://github.com/lsst-sims/sims_featureScheduler_runs1.4 useful – the READMEs in the github repo can be useful for understanding the output contents as well.

A variety of MAF outputs (and additional links to the databases) for these runs are also available at http://astro-lsst-01.astro.washington.edu:8082 – more outputs will come online here over the next few weeks. The MAF outputs are identified by the same run names as the DB links.
You can also download (proposal labelled) versions of the databases at https://epyc.astro.washington.edu/~lynnej/opsim_downloads/fbs_1.4/

FBS 1.4 runs

  • baseline This now contains a comparison of 1x30s visits vs. 2x15s visits (inter-night variation of survey strategy have gone to the ‘pair_strat’ section). For other sets of simulations, we have taken the 1x30s + mixed pairs of filters configuration as the standard intranight cadence … however please take this as a working premise, but NOT an official constraint or choice. The final decision to choose 1x30s vs. 2x15s visits will not be made until on-sky observations with the survey camera are acquired, and 2x15s visits must be carried as the official baseline until that point.

    • baseline_v1.4 Each visit is 1x30s long. All visits in ugrizy are obtained in pairs, with filters mixed between pairs of visits (u will primarily be paired with u but occasionally g and r, and y primarily paired with y and occasionally z, but all filters are paired with other filters at least some of the time). Specifically: u-u, u-g and u-r are paired, as are g-r, r-i, i-z, then z-y and y-y.
    • baseline_2snapsv1.4 The same filter and pair strategy as above, but each visit is 2 snaps of 15s long.
    • The expected increase in efficiency due to taking 1x30s visits instead of 2x15s visits is about 7% – this is consistent with the increase in number of visits between this baseline_v1.4 vs. baseline_2snapsv1.4.
  • Pair strategy These runs look at variations in the intra-night strategy. In FBS 1.3 runs, the visits were paired as r-g , i-r , z-i , with some z visits unpaired, and u and y visits always unpaired. One idea that came out of the filter_load simulations in FBS1.3 was that u band visits would perhaps be better paired with another filter, at least some of the time, in order to obtain u band colors for transients. Variations on the filter pairing are evaluated here.

    • pair_strat_0 This is the previous approach, with pairs of visits in r-g , i-r , z-i , with some z visits unpaired, and u and y visits always unpaired. This would be generally equivalent to the FBS 1.3 baseline, with the updated seeing used in FBS 1.4.
    • pair_strat_1 This runs pairs u-u, g-r, r-i, i-z, z-y, and some y visits unpaired (so adds u visit pairing and some y band pairing).
    • pair_strat_2 This run pairs u-u but also u-g, then g-r, r-i, i-z, with some z visits unpaired while y visits are always unpaired (so adds u visit pairing with u and g, but the rest of the visits have the standard pairing).
    • pair_strat_3 This run pairs u-u, u-g and also u-r, then g-r, r-i, i-z, with some z visits unpaired while y visits are always unpaired (so adds u visit pairing with u and g and u and r, but the rest of the visits have the standard pairing).
    • pair_strat_4 This run pairs u-u, g-r, r-i, i-z, z-y, y-y (so adds a simple u-u visit pairing, but adds z and y visit pairs, so that all visits are now in pairs).
    • Not listed in this section, but related, is the baseline_v1.4 simulation. From these simulations, we decided to adopt u-u, u-g and u-r pairs as the defaults, as they didn’t significantly decrease any obvious metrics (but improved u band coverage and improved the opportunity for u band transient colors). The remaining visit pairings are g-r, r-i, i-z, z-y, and y-y – thus combining pair_strat_4 with pair_strat_3 above.
  • footprints Variations on the WFD footprint and overall survey footprint. For now, we’ve continued to take the ‘traditional WFD and minisurvey’ footprint to be the basis for most of the other experiments, but this may change in a future set of simulations.

    • gp smooth continue traditional WFD coverage throughout the galactic plane (low galactic latitude regions + bulge)
    • bluer footprint This is a standard survey footprint, but with more observations in bluer bands (shifted from z and y bands).
    • big sky a footprint roughly matching that suggested by the “Big sky” whitepaper - extends WFD north (+12) and south (-72), limited by |b|<15 from low galactic latitude regions. This drops any concept of the North Ecliptic Spur (NES) or other minisurveys as separate designs.
    • big sky no uiy just like above, but no observations in u , i or y bands outside the WFD footprint.
    • big sky with dust cut The big sky concept, but defining the galactic plane cutout by a dust extinction cut, rather than a galactic latitude.
    • newA This is based on a similar concept to the big sky simulations, but takes more observations from WFD to extend coverage in the NES and galactic plane regions. This simulation runs WFD from -72 to +12 degrees declination.
    • newB Similar to newA, but moves observations from the galactic anti-center to the galactic center to increase coverage over the galactic plane. This simulation runs WFD from -72 to +12 degrees declination (a bugfix updated this from FBS 1.2).
    • no gp north This is a standard footprint, but without the small spur of the galactic plane that has traditionally stuck up north of the usual WFD footprint.
    • add magellanic clouds This is a standard footprint that adds extended (WFD style) coverage over the Magellanic clouds.
    • ‘stuck’ rolling This is a test simulation, an extreme case of what could possibly happen if there was a simple rolling cadence that always had bad weather in the even (or odd) years. Basically, this is an example simulation that should show that your metrics break when sky coverage is bad.
    • Here is a visualization of these footprints, including their filter coverage. (If github fails to load the notebook repeatedly, try it on nbviewer).
  • alt_roll_dust These two (now three) simulations investigate some rolling cadence variations on an extended N/S dust-limit footprint.

    • roll_mod2_dust_sdf_0.2 A seasonal N/S dec band rolling cadence.
    • alt_roll_mod2_dust_sdf_0.2 A hybrid of the alt scheduler N/S nightly emphasis and a seasonal rolling cadence, where bands of declination are emphasized in alternating years. Here, the footprint is essentially sliced in 4 parts, two in northern declination bands and two in southern declination bands; like in the 2-dec band delayed rolling cadences below, half of the sky is chosen to roll in each of the alternating years (with a season that corresponds to their RA range), but the ‘bands’ are actually one each of a N and S section, and these are further alternated from night to night.
    • alt_dust this is like the above, but without the two-band rolling cadence (i.e. just the footprint).
  • Rolling_cadence A range of variations on the rolling cadence. There is more to do here, but we need better metrics to distinguish between the effects of these difference cadences. The rolling cadences tested so far are all declination band-based, using between 2-6 bands (mod2, mod3, or mod6). All of these rolling cadences feature a ‘delay’ or modification, so that the dec bands switch at an appropriate point in the season for each field. There are also variations on the weight for the ‘background’ (non-rolling / non-emphasized) portions of the WFD, from 10-20% (_sdf_0.10 or _sdf_0.20). For now, we’ve taken a non-rolling cadence as the basis for other experiments, but this may change in a future set of simulations.

  • DDF These simulations experiment with the cadence and number of visits per filter (per night) requested by the DESC and AGN white papers, as well as adding a fifth DDF at the Euclid location. The filter swap time is the same as in the baseline (15% lunar illumination), and translational dithers are applied to the DD visits.

    • DESC_DDF
    • AGN_DDF
    • Euclid DDF Moving the default fifth field, to the location desired for Euclid ground-based observations, using traditional DD sequences.
  • bulge These experiments vary the number and strategy of observations within the galactic bulge region. The “background” survey footprint is the “big sky” footprint - i.e. an extended WFD region, but minimal coverage north or south of this region, or beyond a galactic latitude of +/-15 degrees unless otherwise modified in the simulation. The ‘low galactic latitude’ region in question as specified by the SAC is then |b| < 10 degrees, with a further specification of galactic ‘bulge’ |l| < 20 degrees and |b| < 10 degrees.

    • bulge big sky This is approximately the baseline ‘big sky’ footprint, with an enhancement throughout the low galactic latitude region to increase the number of visits to approximately 250.
    • bulge wfd This further increases the enhancement in the bulge region within the galactic low-latitude region.
    • bulge i heavy Similar to the bulge enhancement above, but shifts visits into i band in the bulge.
    • bulge cadence big sky This is the same overall footprint and number of visit target goal as the ‘bulge big sky’ survey above, but with additional constraints on the cadence of observations within the bulge region, tightening the season to 2.5 months and intra-night revisit time of < 2.5 days.
    • bulge cadence wfd This combines the WFD coverage in the bulge region with the cadence constraints described above (tighten season to 2.5 months and a revisit time < 2.5 days).
    • bulge cadence i-heavy Adds cadence constraints onto the i-heavy bulge footprint.
  • twilight_neo The previous twilight survey indicated that twilight time was not necessary to meet the SRD requirements; with the degraded weather added in FBS 1.2, this is no longer true. These experiments tested adding an NEO-specific survey during twilight. The NEO survey included fields with |ecliptic latitude| < 40 degrees and Dec < 30 degrees, then attempted to take 3 observations per pointing with about 3 minutes spacing, in one of riz filters, up to airmass of 2 and with a preference for pointing as near as possible toward the Sun. This twilight survey was repeated on various intervals - either every night (mod_1) or every other night (mod_2), every third night (mod_3) or every fourth night (`mod_4).

  • twilight_filters These runs investigated if there was a benefit to restricting the filters available for use in twilight (i.e. forcing the simulator to use a particular filter or filters during twilight). The number (_1 through _5) indicates how many filters were available for use during twilight, dropping filters from the blue end until _1 is equivalent to allowing only y band during twilight.

  • DCR Add high airmass observations in u and g to enhance estimates of colors/spectra via DCR. The number of high airmass observations per filter per season was varied in the different simulations, from 1 (nham1) to 5 (nham5).

  • Short exposures These simulations add a mini-survey which takes short exposures – either 1 or 5 seconds long (_1expt or _5exp) – over the entire sky either 2 or 5 times per year (_2ns or _5ns). In FBS 1.2, we saw these had a minimal impact on the overall survey.

  • Spiders This run investigated the effect of setting the rotator angle such that the expected diffraction spikes from stars would run along columns/rows (i.e. match the rotTelPos to the angle of the spiders, +/- 45 degrees, for each visit).

  • Variable exposure times This run allows the exposure time to vary between 20s to 100s to maintain a more uniform depth per image. This mode may not be supported by Data Management.

  • WFD depth This set of runs experiments with weighting the WFD portion of the footprint to different depths. The survey footprint is the standard traditional footprint. The WFD % in each case represents an estimate of the requested number of visits in WFD. The runs with _noddf in the name do not have DD fields and hold exactly to the requested scaling, the runs with DDF can have slight deviations due to small variations of the time spent on DDFs. The WFD fraction was varied from 65 - 99%, in steps of 5%.

  • Weather These runs attempt to shed light on over/under subscription effects, by varying the limit for stopping observing due to clouds. This should preferentially reduce the time available during bad weather periods of the year, increasing the effect of oversubscription in areas which are correlated with bad weather. We have noticed that the area around RA=0 (near most of the DD fields, nearby but not directly next to the galactic plane, and under the edge of the NES) appears oversubscribed with regard to the amount of time available at that time of year. These runs vary the cloud limit from 0.3 (_0.3) to 1.2 (_1.2), where 0.3 would indicate less time available (shutting down at a lower cloud limit) while 1.2 would indicate never closing due to clouds, and 0.7 (_0.7) is close to the earlier (OpsimV3/FBS1.1) values. The current standard value is 0.3, which results in about 32% downtime due to clouds (see release notes in the FBS 1.2 runs release announcement). The other runs in this simulation should be viewed as instructive about oversubscription, but not realistic.

Added later in FBS 1.4 (April 2020)

Due to requests for updates in u band coverage, we did re-run the filter load sequence, but also extended these by evaluating the effect of ‘flexible’ DD sequences (i.e. dropping the separation between dark time in u band and non-dark time in grizy, so that the sequences are continuous in whatever filters are available), and then evaluating the effect of varying the filter load time while also attempting to pair u band visits with either g or r band (including varying the amount of u band visits requested, so as to add more pairs).
These sets of runs are available in

  • Filter_load
  • DDF_experiment
  • u pairs
  • We also ran an series of experiments where we added a priority on obtaining ‘good seeing’ images in r and i bands – good_seeing
  • And a further series of runs where we attempted to add a third visit in at the end of some percent of nights, in the third_obs series.
  • We extended the WFD variation runs by varying the area in the WFD, not just the fraction of time devoted to WFD, in the wfd_vary series.

Dropped in FBS 1.4

If these sets of simulations are of interest to you, please get in touch and we can recreate them with the updated seeing inputs and code used for the FBS 1.4 runs. In the meantime, we’ve dropped these sets of runs to simplify analysis or because we think we learned what we need for the moment.

  • Filter_load The SAC recommended confining u band observations to within +/- 2 days of new moon, and understanding when to change the u band filter (for either z or y band) is useful. These sets of runs complete an (updated and extended) evaluation of the impact of changing the u band filter loading time. The time of filter swap is varied from 3% lunar illumination - 60% lunar illumination. 10% is the default carried to other simulations. We evaluated this for FBS 1.2 and FBS 1.3, and came to the conclusion that below about 7-10% lunar illumination resulted in extremely uneven sky coverage in u band, due to oversubscription at some points in the year (due to a combination of weather and sky footprint). Pairing the u band filter resulted in more even coverage in general, and as was pointed out in the previous community thread - obtaining u band images in combination with visits in g or r allows color determination for very interesting transients such as TDEs. Based on an evaluation of this in FBS 1.3, we are carrying forward a strategy of swapping the filter at 15% lunar illumination and requesting u band visits in pairs – most of the time the visits are paired with other u band images, but some percent of the time they are instead paired with g or r.

  • alt_sched
    The previous simulations were an attempt to come closer to recreating the alt scheduler outputs, without some of the drawbacks to the alt scheduler. A footprint that looks more like the ‘standard’ footprint is included in the ‘roll_alt’ set. These simulations illustrate what a basic North-South daily interleaving of observations gives you, over some simple footprints. This is somewhat contained in the alt_roll/alt_roll_mod2_dust_v1.4 simulation. The general alt_sched footprint did not meet SRD (too much area, not enough depth) and the bluer filters were obtained at bad times (even with the FBS algorithms).

  • roll_alt These previous simulations investigated imposing the alt_scheduler daily N/S variation on top of a season N/S dec band rolling cadence. This has been replaced by a simulation in alt_roll_dust, removing the attempt to impose this on a standard footprint.

  • Rolling_cadence Many of these runs are recreated here, but we dropped the “simple” runs, where the declination bands swapped based on time rather than at a given point in their (individual RA) seasons.

  • templates Difference imaging will require a usable template image. This set of simulations experimented with adding weighting to parts of the sky which had not received at least 3 visits within the last ~300 days. The run parameters are similar, just with different weighting for this feature, the weight ranging from 0 to 10. The default, 6, has been carried out as a baseline for the other simulations. Note that we need better information from Data Management regarding what is required for a template! These simulations are dropped for now, until we have more information about the requirements; the default is continued through these simulations.

  • WFD only These simulations contained only WFD visits, on a traditional WFD footprint. They are somewhat included in the wfd_depth series above, as there are now versions without DD fields.

1 Like

Hi, can you point me to the web interface to see the MAFs results on these OPSIMS?

Of course! It’s similar to the old location, but I moved it up a port so that we could continue to host the FBS 1.3 runs in the previous location for DESC while they’re working on their journal article. Please let me know if that causes problems for anyone!

http://astro-lsst-01.astro.washington.edu:8082/

First a quick question: I followed the url but couldn’t find any science metrics yet.

Second question is about the range of possible cadences explored in the current runs. I believe this is not diverse enough to allow a fair comparison of the different science programs.

I have mentioned in other posts and via Slack that pairing of u-band and a wider temporal coverage (outside of the new moon) is important for photometric classification, in particular to discriminate blue transients (SLSN, TDEs) from rest (ie, SN Ia and most core-collapse SN). This request is also formulated in the white paper of Gezari et al.. It seems none of the 1.4 simulations approximate the request of this white paper. Finally, some of the runs from 1.3 that preformed best in our TDE metric (eg, illum60) are removed in 1.4.

Within TVS, we have also formulated a metric for “Discovery of the Unknown”, this metric would also benefit from a broader, less conservative set of simulations.

I understand optimizing the cadence of different science cases is non-trivial and I think that “voting” via different metrics for each science case is a beautiful solution. But with the current set of runs, TDE science has almost no voting power.

I hope we can still expand the range of the simulations before picking the final cadence.

3 Likes

Yes, sorry - the “Metadata+” links have metrics for all of the metadata plus some of the more basic kind of science metrics (like coadded depth, number of visits, astrometry proper motion + parallax, the gaps between nights and season length information), but I don’t have the larger science metric outputs up yet. They should be up by Monday.

Sorry - some of the links above apparently are not correct … but I can no longer edit the post.
So, please take the name / description as accurate but don’t try to follow the links exactly if they don’t work.
Instead, you can download ALL of the runs from
https://epyc.astro.washington.edu/~lynnej/opsim_downloads/fbs_1.4/
or from
https://lsst-web.ncsa.illinois.edu/sim-data/sims_featureScheduler_runs1.4/

The difference between these two locations are:
epyc hosts the databases which were in the original fbs 1.4 release, and others as they’re added … but they go through an extra process to label the visits with a proposal ID, so may lag behind.
ncsa hosts the same databases, and maybe more, but they have typically been synced over before they received the proposal labelling.
Please use whichever you prefer … if proposal labelling is important to you, use epyc.

1 Like

Announcement at top updated with some additional series of runs - good seeing, u pairs, filter load, wfd footprint variation, and DDF variations.

1 Like