rhiannonlynne
commented
7 years ago Regarding the existing DD implementation, it's based on the whitepapers submitted previously. You can see a summary of these (and links to the actual whitepapers) at https://community.lsst.org/t/deep-drilling-whitepapers/732
Please note that in these whitepapers, deep u band coadds were indeed requested! 5 of the fields were actually requested to a deeper u band depth than others; some science collaborations wanted more u band depth, but fewer fields, while others wanted more fields but needed less depth.
I agree that the actual cadence of visits should be examined, as it was based simply on requests from (primarily the SN community) to get all filters possible in each night of DD. It sounds like we're drifting away from that, but clarification of what is actually required would be good. For example - if you got g+r+i on three successive nights, and then i+z+y on two nights, would that be good or bad? If we split it up into more frequent, but less comprehensive filter coverage (i.e. g+r+i, then i+z+y, etc.) do these need to be interleaved? Can they be random groupings or should there be a pattern? If there is a pattern, is that bad?
We took notes on our Tucson 2016 PCW discussion on this hackpad. For your convenience and mine, here they are pasted in:
When can we get a realistic OpSim Rolling Cadence realization?
The new version of OpSim will be released in November, then some testing will occur, then rolling cadence can be run (source: Andy C.) The SOCS/Scheduler functionality development plan is posted on LSST confluence here.
We need to propose rolling cadences for then! We have a mechanism for making suggestions, on a GitHub suggestion board . We can also transcribe these into the white paper
Zeljko showed the OpSim development schedule slide in this morning's plenary.
Things to think about when proposing Rolling Cadences:
Eric B: you don't have to roll in all filters, maybe its enough to increase cadence in 1-2 bands? Michael S thinks in terms of declination stripes. "Swiss Cheese" rolling cadences are defined in Chapter 2 of WP v1, but may not be worth investigating (Steve R) Andy: could we piece together high cadence annual declination stripes into a 10 year simulation? Michael S: probably doing whole sky early will be important, to get a good template/reference image. Keith: "Sandwich" strategy, year 1 and year 10 at universal cadence, and rolling in between. Tony T: wonders whether rolling cadence could help weak lensing, perhaps improving airmass? Jeonghee's proposed strategy could do this.
What are the potential downsides to Rolling Cadence?
Sky uniformity could take longer to build up Could asteroid coverage suffer? ...
Action: Figure out coordination of OpSim proposals. Each collaboration has X sims to spend?
Other Cadence Ideas
Fast transients: pair of visits, followed by a third one several hours later (or maybe the following night). Colors in visit pairs?
Deep Drilling Fields: observed using "deep drill" sequences (~20 visits in r, then ~20 in i, and so on, all in succession on the same field), which are efficient but maybe sub-optimal. Transients group are less bothered about getting all filters in one night. Some consensus that drilling sequences would be better shorter, and perhaps the visits spread over two nights, or separated within one night.
What is the u band depth driven by? it seems that extragalactic does not need it because of redshift, and nobody else seems to need such deep and costly u band coverage.
The location of the four current deep drilling fields is chosen for synergy with other surveys. Not all deep drilling fields need to go to the same magnitude or follow the same strategy design. The survey the field is chosen to meet would drive the decision of how deep it needs to go.
The current deep drilling fields cadences are designed from the white papers, summarized by Lynne Jones into a single document, but it appears that the requirements should be updated in the light of growing expertise and better understanding of the LSST strategy design. Rahul points out that the current OpSim DDF fields do not implement the requirements laid out in the white paper.
White paper: it is proposed that the paper includes a section written by the LSST project.
How complete is the white paper? how biased is the set of phenomena considered? How do we focus on the transient cases that drive the cadence and are unique to LSST?
Other Ideas
Can we define a cadence that samples the phase space of amplitude of variation, time scale, and rarity ?
MAF metric execution times needs thinking about, especially as more and more OpSim runs are produced.
Global metric? Will we need it, or will we settle on an observing strategy before that? Andy C: scheduler will not be a simple greedy algorithm, but will do more optimization than this - so will need an objective function that could be science (ie MAF)-related.