FBS 1.6 release (August 2020)

Dear survey strategy enthusiasts - a belated run release update.

The FBS 1.6 release is a somewhat mixed set, part of which could really be considered part of or replacements for runs from the 1.5 release, and part of which should be considered an orthogonal set of simulations (let’s call these the ‘potential scheduler’ runs).

In the bigger picture, the FBS 1.5 and the FBS 1.6 runs which are not ‘potential scheduler’ runs (i.e. the rolling_fpo family and the even_filters family) are simulations where a single survey strategy option is varied across the family allowing investigation of the science impacts of changing that particular option. The FBS 1.6 runs which are in the ‘potential scheduler’ family take a different approach – these simulations vary multiple survey strategy options at once, with the goal of doing better for a particular science goal … thus the runs within this family are fairly different from one another and represent some of our guesses at potential combinations for survey strategy. It’s worth noting that within the FBS 1.6 family there are several runs which would most definitely not be beneficial for all science aims (and some are particularly bad for some science).

From our point of view, how we think survey strategy evaluation should proceed is by looking at the metrics from the FBS 1.5 and 1.6 even_filters and rolling_fpo families, looking at which survey strategy changes make the biggest impacts and summarizing why this occurs and linking the magnitude of the metric changes with the amount of survey strategy change. The FBS 1.6 ‘potential schedulers’ then gives a set of runs to see if those survey strategy options that are beneficial to your science play together as expected … although it’s entirely likely we didn’t combine them exactly as you would wish – the ‘potential schedulers’ family of simulations serves as examples of combined options, but we do not labor under the impression that they are optimal.

The FBS 1.5 runs are described in a previous community post - FBS 1.5 release (May Update - Bonus FBS 1.5 release) (ignore the ‘bonus’ as this just refers to them being more than we’d originally expected to create).

The FBS 1.6 runs
The FBS 1.6 runs use almost the same codebase as FBS 1.5, but adds a new footprint class that enables weighting the target map by time within the season. What this means is that as new parts of the sky become available for observation at the start of their season, they do not automatically rise to the top of the priority list as they would otherwise (since the new fields would always have fewer observations than the parts of the sky that were recently available and visited in the previous weeks). This means that visits are taken closer to the center of their observing season. This is particularly important for rolling cadence simulations, where the ‘focused’ footprint is smaller than the entire sky (and thus ‘new fields’ end up overweighting themselves even more strongly). Generally, FBS 1.6 runs could be compared fairly closely with FBS 1.5 runs, but we still recommend using the baseline simulation in FBS 1.6 (baseline_nexp1_v1.6_10yrs.db) to normalize any changes in metric results for best results.

You can download the FBS 1.6 runs at either:
https://lsst.ncsa.illinois.edu/sim-data/sims_featureScheduler_runs1.6/ or https://epyc.astro.washington.edu/~lynnej/opsim_downloads/fbs_1.6/ (the epyc link includes WFD labelling).

You can find the MAF results for these runs at http://astro-lsst-01.astro.washington.edu:8080/

The links for FBS 1.5 are in the FBS 1.5 release note, but for convenience:
https://epyc.astro.washington.edu/~lynnej/opsim_downloads/fbs_1.5/ (databases)
and http://astro-lsst-01.astro.washington.edu:8081/ (MAF)

Family summaries
In FBS 1.6 there are three families of simulations, a total of 30 simulations. These simulations should generally be used together with the simulations in FBS 1.5 (86 simulations).

  • even_filters The ‘even_filters’ family investigates the effect of changing the filter-choice weighting. In standard runs, the choice of filter is weighted by the difference between the current five-sigma limiting magnitude and the darkest possible five-sigma limiting magnitude in each filter, as well as the number of visits in each filter and a penalty for changing the filter. This tends to result in g band visits being taken near new moon and a heavy preference for z and y near full moon, which leads to an uneven cadence in a particular filter over the lunar cycle. In these simulations, the weighting between the current five-sigma limiting magnitude and the darkest possible five-sigma limiting magnitude is not used for all bandpasses – in ‘even_filtersv1.6_10yrs.db’ and ‘even_filters_altv1.6_10yrs.db’ only u band avoids bright time (using a per-filter delta limiting magnitude basis function); in ‘even_filters_g_v1.6_10yrs.db’ and ‘even_filters_alt_g_v1.6_10yrs.db’ both u and g band avoid bright time. The difference between the simulations with ‘alt’ and without ‘alt’ in the name is the survey footprint; ‘alt’ runs use a variation of the big-sky alternative WFD footprint.
    Investigate these simulations especially if you are worried about the intervals between visits in the same filter throughout a lunar cycle.
  • rolling_fpo These simulations are familiar - they are replacements for the rolling cadence simulations in FBS 1.5, using the new footprint class to make the cadence within the active rolling area more even. These again split the sky into 2, 3 or 6 declination bands (2nslice, 3nslice and 6nslice respectively), with varying levels of background observations outside the active rolling area – the weight on the rolling area varies between 0.8, 0.9 and 1.0 (with corresponding higher to lower background visits outside the active rolling area). Note that all runs have a minimum number of all-sky visits per year (3 per band per pointing) in order to maintain difference imaging subtraction templates and their availability for ToOs; this means that even at a weighting of 1.0, there are still non-negligible visits outside the active rolling area.
    Investigate these simulations to evaluate the effects of rolling cadence on the interval between visits over the lifetime of the survey.
  • potential_schedulers This family is unique in that instead of varying a single survey strategy option (as in all other families), here we vary several options at once in pursuit of a particular science goal. The point here is to illustrate the effect of combinations of survey strategy variations; some are successful and sometimes we may meet technical goals but not science goals. For further details on each simulation, Section 5 in the Survey Strategy report for the SCOC (https://pstn-051.lsst.io/) is recommended.
    Everyone should run their metrics on these simulations, as they serve as a way to highlight entire styles simulations which will fail to meet particular science goals.

As always, please reach out (either here on community.lsst.org in the ‘survey strategy’ area or via email) for questions or help.