overviewApJreferee.txt

          Dear Editor,
we are submitting a revised version of the paper "LSST: from Science Drivers to Reference Design 
and Anticipated Data Products by Ivezic et al.". We are thankful to the referee for careful reading
and excellent suggestions for how to improve this paper. We have implemented all the suggestions
where it was feasible. Detailed listing of implemented edits is appended below. We hope that the
revised version is now suitable for publication. 
         For the authors, 
           Zeljko Ivezic


> Given the 10 year history of this paper, some of the science discussions are stale and should 
> be updated. I have identified some places where this should be done, but there are no doubt
> other places that could be improved.

The chairs of eleven LSST Science Collaborations have carefully revisited science discussion
and references and updated them in many instances. 

> The "Deep-Wide-Fast" survey is supplemented by a "Very Deep" survey and a "Fast time domain" 
> survey [in the abstract]. This is too reminiscent of Starbucks' coffee sizes, where "grande" isn't big.

"Very" was meant to apply to both "Deep" and "Fast"; fixed. 

> It would be good to have more current seeing data in Figure 1. Seeing always seems to get worse
> after site selection probably due to regression to the mean.

The LSST Project is will install a permanent seeing monitor at the summit in 2019. Since
it will take at least a year to collect an adequate data set, existing Figure 1 cannot be significantly
improved at this time. 

> The deep drilling fields in Figure 18 do not appear to be very much deeper than the main survey: 
> 1.8x on the color bar. This figure might need a logarithmic color bar.

We have improved the design of this figure, and also added more information to its caption. 

> A Figure showing the cadence for 1 or more typical spots would be useful, and it should show which 
> filter is used for each visit. Something like a timeline with different color ticks indicating the visits. 
> This should probably cover one lunation to keep the scale readable.

Figure 19 in the revised version is our response to this excellent suggestion. 

> State how long it takes to change filters. This is relevant for getting an SED of a rapidly changing transient.

We have added this information to section 2.6.2: "The five filters in the camera can be changed in 
90-120 seconds, depending on the initial camera rotator position." 

> The Zwicky Transient Facility is up and running, and the author list of this paper includes many ZTF 
> team members. Based on ZTF experience, you should have a fairly good updated estimate of what 
> the LSST transient alert stream should look like. While you do mention several million alerts per 
> night, you should mention how these will break down into supernovae, fast transients, stellar flares, 
> and asteroids.

Our ZTF colleagues report that they don't have robust results yet and claim that a paper describing
LSST simulations by Ridgway et al. (2014) is still the best reference. We have added a reference to
that paper (not including it was an omission on our part). 

> Standard sirens will not be "found" by LSST [p 36]. They have to found and measured using 
> LIGO/VIRGO/KAGRA/IndIGO and then the EM counterparts need to be found, and LSST will be a 
> great asset for this follow-up activity. So rewrite this to say LSST will followup GWB alerts and is 
> likely to find 70 EM counterparts if the gravitational wave network is working well.

Excellent point! Implemented as suggested. 

> The trailing loss Eqn (8) has weird units. vt/theta should be dimensionless, so express v in 
> arc-sec/sec, t in sec, and theta in arc-sec. This gets rid of the weird 24 in Eqn (9). Eqn (8) is 
> basically noise \propto sqrt(FWHM^2+[trail length]^2) which is pessimistic relative to a search
> tuned for trails. This would have noise \propto sqrt(FWHM*sqrt(FWHM^2+[trail length]^2)) but 
> would be a substantial computational burden to do a 4-D search for trails. You have to either 
> explain your reasoning or give a citation for Eqn (8).

Eq. 8 (now eq. 9) is based on a more detailed model, described in Section 5.1.4 in Jones et al. (2018)
that accounts for both SNR losses and detection losses due to non-optimal extraction profile. 
We have added a reference to Jones et al. (2018). 

> On page 28, H does NOT scale directly with the diameter of an object. That would say that an 
> object with D=1400 m would have H=220 since an object with H=22 has D=140 m for 
> albedo = 14%. H is a logarithmic quantity representing a flux which scales quadratically with 
> diameter. Explain what size distribution is equivalent to your dN/dH~10^{H/3}.

We agree that our phrasing of this well known relation was poor. We have modified
the text as suggested and clarified that adopted magnitude distribution corresponds 
to a size distribution $dN/dD \propto 1/D^\beta$, with $\beta = 5\alpha+1$. 

> The claim in Figure 24 that LSST colors will be twice as accurate as SDSS colors for asteroids 
> depends on being able to phase up very sparse light curves in the different filters, which will 
> be a considerable task, while the SDSS filters were all obtained within about 2 minutes.

The accuracy of SDSS colors is about 0.03 mag (0.02 mag errors per band with uncorrelated errors), 
when the achieved SNR allows it. Assuming that the rms for asteroid light curves is ~0.2 mag (e.g. 
an amplitude of 0.3 mag and sinusoidal variation), and 100 observations per band, the expected 
color accuracy is 0.02 mag (0.2/sqrt(100) with correlated errors between bands). This color accuracy
is achieved without any light curve phasing (and the rizy bands have more than 100 observations).
In addition, a discussion about "Sparse lightcurve inversion" in section 5.7.1 from the LSST Science 
Book (https://www.lsst.org/scientists/scibook) shows that light curve inversion, while "a considerable 
task", will not be a hopeless task. Thus, we believe that our statement is correct. 

> \S 2.1.2: It should be noted in the text that this requirement: "The photometry should be better than 1-2% 
> to measure asteroids' colors and thus determine their types. The different filters should be observed over a 
> short time span to reduce apparent variations in color due to changes in observing geometry, but should 
> be repeated over many lunations in order to determine phase curves and allow shape modeling." is not met 
> by the proposed main survey. The typical asteroid changes its flux by 1% in 100 seconds or so. The filter 
> change cycle is hard to find - I suppose I could look at minion_1016 - but I believe it is on a much longer 
> time scale than 100 seconds. Thus the caption of Figure 24 stating that LSST colors will be twice as good as 
> SDSS colors is not supported. SDSS observed all 5 bands in O(10^2) seconds, while LSST may well take days.

The time to change the filter in the camera is 90-120 seconds, and following your suggestion, 
this information was added to the manuscript. But given the discussion about asteroid color accuracy 
just  above, we believe that the required accuracy will be achieved even without back-to-back observations
in different filters (though we anticipate that some tests based on back-to-back observations will be done
to empirically confirm the claim of ~2% color accuracy). 
 
> Also the Congressional mandate to find 90% of all NEOs 140 m and up will not be met, although finding a 
> large fraction of NEOs brighter the H=22 is feasible. It's those pesky coal black C types that won't be found 
> by LSST: a quarter of the NEOs at 140 m diameter have H > 23. The albedo used in \S 3.2.2 is hard to find, 
> but the statement that H=22 is equivalent to 140 m gives 14%. Most C-types are much darker than this.

It is customary to interpret 140 m size as H=22.  But we agree that there is a substantial fraction of 
asteroid with albedos below the implied median value of pV=0.142. We have added a reference to recent 
papers by Wright et al. (2016) and Grav et al. (2016) who discuss this issue in quantitative detail and find 
that a more realistic treatment of the albedo distribution can lead to a decrease in completeness 
on the order 5% (we note that "C types that won't be found" is too strong a statement because LSST 
completeness doesn't drop to 0 at H=22; indeed, it's still as high as 20% at H=25). 

> Specify the atmospheric dispersion in the filters. You mention that going for wider passbands would make 
> this worse but do not give the magnitude. And call it "atmospheric dispersion", not "chromatic atmospheric 
> refraction", even though these mean the same thing. Some instruments have ADC systems, none have CARC 
> systems. This seems like an important systematic noise term for weak lensing studies.

We have quantitatively specified atmospheric dispersion as a function of wavelength, and followed 
the suggestion to call it "atmospheric dispersion", instead of "chromatic atmospheric refraction"
Regarding the systematic noise for weak lensing studies, we have clarified that the PSF is modeled as 
a function of the color of the object, and given enough information about the object's morphology,
can be handled well enough to enable weak lensing measurements. We have also added references to
recent papers which discuss this effect (Meyers and Burchat 2015; Carlsten et al. 2018).

> How does the effective number density of weak lensing background galaxies compare with Euclid and 
> WFIRST? Euclid is observing a similarly large area with a better PSF, no seeing, no weather, no Moon, no 
> daytime, and a darker sky.

We have added text to the paper to make this comparison. The Euclid galaxy density should be about the 
same as LSST's over a comparable survey area, enabling a weak lensing analysis with similar statistical 
precision but with quite different systematic errors. WFIRST should go deeper, but over a significantly 
smaller sky area. We provide a reference to a white paper comparing the three surveys' expected performance, 
and the potential for combining them.

> The weak lensing results will be competing with Euclid which has a similar start date. There are several 
> surveys that have been "skimming the cream on weak lensing science", including CFHTLenS, PanSTARRS 
> and DES. So far the cream has not produced spectacular results. Figure 22 seems frozen in the DETF days.
> Figure 22 should acknowledge the surveys in the can, and show an "All" that include Euclid WL and then 
> "All minus LSST WL". So you should include a fair assessment of how much the LSST will advance the field 
> over where it will be given current and under construction surveys.

The Figure 22 that the referee is referring to is now Fig 23 in the current version. We have followed your
excellent suggestion and replaced the plot with the most up to date alternative, from the recently released 
LSST DESC Science Requirements Document, and added an additional contour to show what the Dark Energy 
Survey is expected to achieve with its Y6 dataset as a way of showing what the Stage III surveys will contribute.

We have not added cosmological constraint contours that correspond to "All" including Euclid WL and then 
"All minus LSST WL", as the referee suggested. The reason for this is that a key part of the cosmological 
community's Stage IV dark energy program is that there are multiple surveys that are dominated by different
systematic uncertainties and use (at least mostly) independent analysis algorithms. These enable multiple 
cross-checks that provide confidence in our control of astrophysical and observational systematic uncertainties 
when placing tight constraints on dark energy (or modified gravity, neutrino masses, and other measurements 
that will also be made using those surveys). The Weinberg+13 review article on cosmology, Jain+15 white paper, 
and Rhodes+17 ApJ paper more thoroughly explore the value of these tradeoffs in survey strategy, analysis 
methodology, and so on, that will enable us to validate that we have adequate control of astrophysical and 
observational systematic uncertainties. If we provide a FoM plot that appears to emphasize the purely statistical 
constraining power of these surveys, we would be missing this key element of the community's Stage IV program. 
Finally, as a purely practical matter, we do not have Euclid forecasts that are in the same 7-dimensional 
cosmological parameter constraints as the LSST ones, or with reasonably consistent assumptions, nor do we 
think that such constraints provided by the LSST community would necessarily be seen as credible (as compared 
to whatever the Euclid consortium provides based on their best current understanding of the instrument and 
survey).

> In discussing the proper motion and parallax accuracy, it should be pointed out that achieving the required 
> 10 mas 1 sigma 1 axis single visit accuracy requires SNR of 30, so the magnitude limit is 2 mag brighter 
> than the single visit SNR 5 limit. Still very good: r = 22.5. Later on p 29 there is a more conservative calculation 
> showing r = 21 for 0.2 mas/yr. So give a magnitude limit during the first proper motion accuracy discussion, 
> and state that accuracy scales like 1/flux for fainter objects.

We have clarified that the first mention of "10 mas" corresponds to systematic astrometric errors 
(and not random errors which indeed scale as 1/flux, as discussed in Section 3.2.3)

> You should cite actual performance of Gaia in addition to the pre-launch projected performance.

We have added references to recent papers about Gaia Data Release 2 performance on pages 9 and 30
(which demonstrate that their early performance estimates were robust). 

> The telescope latitude and longitude are specified to 0.01" which is 31 cm. Are these values really that good? 
> And if so, what part of the telescope do they correspond too? That should be specified. [p 16]

We have clarified that the listed coordinates correspond to the center of the telescope pier. 

> Footnote 14: there should be an S^2 term as well since there are flux errors proportional to the flux. The 
> coefficient will probably be small but the term will be present.

If there is an S^2 term for some reason (and one would not expect it from the first principles, 
unless more complicated error models with various systematic effects are invoked), it must be 
very small; for example, Sesar et al. (2007) showed that our Eq.5 is supported by the empirical SDSS 
photometric behavior. More details about our SNR computation are available in Section 4 from http://faculty.washington.edu/ivezic/Teaching/Astr511/LSST_SNRdoc.pdf

> Eqn 6 & 7: These are not consistent in their treatment of the exposure time. Eqn(6) is pure sqrt(t) while Eqn(7) 
> is poorly written and not explained well. \Delta C^\infty_m is not the loss due to readout noise when the 
> integration time goes to infinity, because that is zero. Rewrite to explain better.

We have clarified the details behind Eqs. 6 and 7 in the revised version.