LSST Systematics Caused by Observing Strategy and Milky Way Dust

(Poster will be presented by Eric Gawiser.) Starting in 2022, the Large Synoptic Survey Telescope (LSST) will survey the southern sky with unprecedented detail. To understand the potential sources of artifacts induced in the observed data, we investigate the effects of depth variations resulting from survey strategy, photometric calibration errors, and Milky Way dust. Using the LSST Operations Simulator and the Metric Analysis Framework, we implement observing strategies with large telescope pointing offsets (dithers) and account for dust extinction using the Schlegel-Finkbeiner-Davis dust map. Our results illustrate how the observing strategy, calibration errors, and dust extinction produce fluctuations in the co-added survey depth, with dust dominating the largest angular scales. We find that visually, dust serves to wash out the strong honeycomb pattern of depth variations in an undithered survey, while in the angular power spectrum, all sources of variance add. Furthermore, while dust extinction creates much larger systematics than photometric calibration errors, survey strategy is the dominant source of artificial fluctuations in galaxy number density on ~ degree scales relevant for Baryon Acoustic Oscillations. These artificial fluctuations correspond to the survey window function, which can be corrected for in correlation function analyses for weak lensing and large-scale structure studies. We note that a limiting systematic will be generated by uncertainties in this window function, making uncertainties in the dust extinction on all relevant scales a crucial systematic for LSST. Our work illustrates an approach to quantitatively analyze the effects of Milky Way dust and LSST observing strategy upon weak lensing studies.

Milky Way Reddening from Stellar Photometry

I present a three-dimensional map of dust reddening, covering the three quarters of the sky north of a declination of -30 degrees. Dust reddening is determined from Pan-STARRS 1 and 2MASS optical and NIR stellar photometry. In each sightline (approximately 7' in scale), we combine stellar photometry with a smooth model of the distribution of stars and dust to infer dust reddening in each distance bin, as well as stellar distances and types. Beyond its use in Galactic astronomy, where distance-dependent reddening is important, our 3D dust map has a number of useful properties for extragalactic astronomy, and studies of large-scale structure in particular. Unlike maps based on far-infrared dust emission, which convert from an inferred dust opacity to dust reddening, we use stellar colors to more directly measure reddening. Because far infrared maps trace dust emission not just from the Milky Way, but also from extragalactic dust, they have residuals that are correlated with large-scale structure. Dust maps based on stellar photometry are free of this effect, and therefore may be preferable in applications to large-scale structure. LSST itself will allow great improvement in photometric dust maps, but even before then, Gaia parallaxes, by providing fixed distance measurements to a large sample of stars, will aid in the creation of better 3D dust maps.

Open question: We currently determine dust reddening along each sightline separately. How can the inclusion of priors that dust density is correlated in nearby volume elements improve our model of Galactic dust reddening?

Reddening Law in the DES Footprint

DES Year 1 and Year 2 data have been calibrated using APASS/2MASS (top of the atmosphere) and the SLR method (top of the Milky Way). The difference between the two measures dust extinction as a function of wavelength. The reddening law clearly varies within the SDSS footprint and is most ill-behaved on Stripe 82, where many photo-Z calibration fields are located.

The Optical-Infrared Extinction Curve and its Variation

The dust extinction curve is a critical component of many observational programs, as well as an important diagnostic of the physics of the interstellar medium. We present new measurements of the dust extinction curve and its variation towards tens of thousands of stars, a hundred-fold larger sample compared to existing detailed studies. We use data from the APOGEE spectroscopic survey in combination with ten-band photometry from Pan-STARRS1, 2MASS, and WISE. We find that the extinction curve in the optical through infrared is well characterized by a one-parameter family of curves described by R(V), with little need for further parameters. We show that the extinction curve in the infrared is not "universal," in contrast to several widely-used extinction curve parameterizations. We find that the extinction curve varies more in the optical than in the infrared, but is more uniform than suggested in past works, with sigma(R(V)) = 0.2, and with less than two percent of sight lines having R(V) great than 4. Our data show new, large, spatially coherent structures in R(V) throughout the Galactic plane. The R(V) variations are on scales much larger than individual molecular clouds, indicating that grain growth in dense molecular cloud environments is not the primary driver of R(V) variations in dust at large. Indeed, we find no correlation between R(V) and dust column density up to E(B-V) ~ 2.
Open questions:
- our technique can measure reddenings very well, but is insensitive to total extinctions, or the "gray component" of the extinction. How does the gray component of the extinction vary with the reddenings?
- what physical process gives rise to the large scale (greater than ~500 pc) variations in the extinction curve we observe?
- how variable is the extinction curve in the high latitude, nearby dust, most relevant for extragalactic work?

Can photo-z gain spectral information from atmospheric transmission variations?

The LSST calibration plan is to correct all observations to a "standard" bandpass. However, in order to do this estimations of the "true" bandpass transmission at each visit will have to be made. There will be O(100) visits to each pointing on the sky, therefore O(100) slightly different bandpass transmissions per filter, each probing slightly different spectral regions of an object. So an open question is: can these individual observations, with more information but lower signal-to-noise, improve photometric redshift estimation? This poster presents initial results and future plans to investigate this question, along with plans to test the fidelity of the LSST calibration plan for precision galaxy photometry.

Galactic Dust and Photometric Redshifts

To investigate the impact of imperfect removal of Galactic reddening on LSST photometric redshifts, I create a simulation of galaxy photometry in a 4 square degree patch on the sky. I apply Galactic extinction following a continuous version of the SFD dust map, then deredden the photometry in a patch using only the center point of a pixelized version of the footprint, simulating a low resolution dust map. In addition, I create a second dataset where the Galactic extinction applied is slightly different from that used to deredden, simulating the effects of a mischaracterization of the wavelength dependent absorption of Milky Way dust. I present comparisons of the photo-z performance for these two cases relative to results with perfect knowledge of Galactic extinction.

Shear measurement bias due to color gradients in galaxies in the presence of wavelength-dependent point spread functions

Cosmic shear studies require highly accurate measurements of galaxy shapes in the presence of the point spread function (PSF) that smears the image. The size and shape of the PSF can depend on wavelength due to physical effects that include atmospheric seeing and refraction, sensor effects, and diffraction. The impact of using the PSF measured with stars to determine the shapes of galaxies that have different spectral energy distributions (SED) than the stars has been studied for atmospheric chromatic effects in future ground-based surveys and for diffraction in future space-based missions. Additional biases due to a chromatic PSF applied to a galaxy with a spatially varying SED (so-called color gradients) have not been studied in detail for ground-based surveys.

We present a study of the impact of chromatic gradients in parametric bulge + disk galaxies, on shear estimators due to the wavelength dependence of seeing for a ground-based telescope such as the LSST, and due to diffraction for a space-based mission such as Euclid . We compare the latter results to those of previous studies of the implications for the Euclid mission (Semboloni et al. 2013, Voigt et al. 2012 ). We study the dependence of the shear bias on properties of the PSF and the galaxies. We measure the relative sensitivity to color gradients of moment-based and model-fitting algorithms for measuring the galaxy shape in the presence of the PSF.

Impact of chromatic point spread functions on cosmic shear

Wavelength dependence in the LSST point spread function (PSF) introduces a source of systematic uncertainty in cosmic shear, since the PSF measured using stars is used to determine the shapes of galaxies with different spectral energy distributions (SEDs) than the stars. We review the current understanding of and proposed corrections for PSF effects on cosmic shear due to chromaticity in the atmosphere, optics, and sensors. We then attempt to enumerate currently unquantified LSST chromatic PSF effects and make suggestions on how to constrain these remaining unknowns. In particular, one such unknown is the impact of galaxies with spatially varying SEDs. To address this unknown, we introduce a method to simulate observations of realistic chromatic surface brightness profiles convolved with realistic chromatic PSFs using multi-band imaging from the Hubble Space Telescope as input.

Modeling the LSST pixel grid for systematics tests

Open question: Can we empirically model the science impact of all sensor effects simultaneously?

We propose an empirical maximum-likelihood framework for jointly modeling the effects of transverse electric fields in CCD pixel grids, and apply the method to LSST prototype data. Previous analyses have considered of the impact of individual sensor effects (tree rings, edge effects, etc.). To simultaneously determine their impact, we use a gradient descent algorithm to fit a model of the underlying pixel grid to coadded flat fields. We then deposit simulated PSFs and galaxy profiles onto our model pixel grid to assess the impact of pixel area variation on photometry, astrometry, and shape measurement. Finally, we estimate the impact of these biases on LSST weak lensing science.

Pixelization Bias

Pixel detectors provide the astronomer with a uniform grid of brightness measurements which are surface brightnesses integrated over the collecting area of each pixel. For very well sampled images, this pixelization affects the PSF in a minor way, and thus the locations and shapes of detected objects is not greatly affected. However, as the size of an object approaches the pixel scale, the pixel binning of it's brightness can create a measurable systematic bias in the calculation of some parameters. Here I present an investigation of the biasing effects of pixelization on the PSF using both simulations and data from the laboratory, and I discuss ways to correct for the bias using proper modeling.

Brighter/fatter effect from lateral fields and charge-dependent diffusion

The interaction of incoming charge with charges already accumulated in CCD pixels causes a broadening of the observed point-spread function (PSF) with increasing flux, the so-called brighter/fatter effect. Since a lot of the information on the PSF is contained in relatively bright stars, this is an issue for weak lensing and model fitting photometry.

Two non-exclusive physical causes for the brighter/fatter effect have been proposed: (1) lateral electric fields due to accumulated charges that shift the effective borders of pixels and (2) the dependence of diffusion width on the amount of accumulated charge. The first effect can be characterized from correlation measurements of the Poissonian noise in flat fields (cf. Antilogus et al. 2014; Guyonnet et al. 2015) and corrected on a pixel level (cf. Gruen et al. 2015 for our implementation in DECam). The second effect does not cause correlation of Poissonian noise in flat fields and therefore has to be measured and corrected by a different scheme. While sub-dominant for DECam, this may be an open issue for other instruments and more stringent requirements.

Interpixel capacitance and other detector effects in GalSim

In the era of large galaxy imaging surveys like LSST and WFIRST, the statistical errors on the weak lensing measurements will become small enough that we must scrutinize various systematic effects that were negligible in the past. One such class of systematics consists of various effects that the telescope detectors introduce. Detectors with CMOS architecture are becoming increasingly common in astronomy and they introduce artifacts that are different from the conventional CCD detectors. One such effect is the interpixel capacitance (IPC), a form of electrical crosstalk between neighboring pixels. The IPC is predominantly isotropic, increasing the measured size of the PSFs and galaxies with a small anisotropic component that changes the shapes as well. Additionally, IPC correlates the noise in the images. We present a brief summary of the effects of IPC on PSFs and preliminary results on the effects of IPC on shear calibration biases. The software framework we use to calibrate the effects of IPC on WFIRST PSFs and shear measurements could be applied in an LSST context to simulations that include optical detector effects. Morever, a clear understanding of these effects is essential for carrying out the WFIRST lensing analysis alone, or a joint analysis of WFIRST and LSST data.

Measurements and Simulation of the Brighter-Fatter Effect

The Brighter-Fatter Effect causes both the size and shape of star images to vary depending on the star's brightness. Since calibrating the PSF in the LSST images will be done using stars of varying brightness, it is crucial that we have an accurate model of the B-F effect in order to remove this effect at the pixel level so we can accurately determine the PSF. This poster will describe measurements of the B-F effect made on the UC Davis LSST Optical Simulator, as well as a device physics based model of the B-F effect that accurately captures the major components of the effect.

Sensor effects in astronomical data with LSST prototypes

I am going to present analysis of existing astronomical data taken with LSST sensors and will also discuss plans to continue these activities.

Predictable differences between sensor incident flux and recorded signal distributions in science data - with availability of well characterized correlation coefficient maps distilled from flat field exposures

Dynamic transfer functions that map sensor entrance window positions to pixel addresses can be decomposed into a long range, Coulombic (radial) term and a more complex (anisotropic) short range term. We identify these to be the underlying mechanism responsible for the brighter-fatter (BF) effect. We describe and demonstrate a toolkit for performing detailed, dynamic modeling of these transfer functions, which can help with interpretation of flat field pixel signal covariance evolution, improved background level subtraction and of course PSF estimation.

Modeling the Optical PSF

Measurements of weak gravitational lensing must account for the point spread function (PSF). We model the optics contribution to the PSF of the Dark Energy Camera, which is largely stable in time and able to be modeled by relatively few parameters. We present an optics PSF model based on a Zernike polynomial expansion of the wavefront in the pupil plane propagated to the CCDs which accounts for the misalignment of the optics on a per-exposure basis with only 18 parameters for the entire wavefront. Open questions still under research include how to best deconvolve the optics contribution to the PSF and how to empirically model the remaining, atmosphere-dominated contribution to the PSF.

Temporal Characterization of Zernike Decomposition of Atmospheric Turbulence

Atmospheric turbulence limits the performance of ground-based surveys such as the Dark Energy Survey and Large Synoptic Survey Telescope. Zernike decomposition of phase fluctuations due to atmospheric turbulence allows for the temporal characterization of the resulting phase fluctuations. Using pseudo-open loop phase maps, reconstructed from deformable mirror commands obtained from the Gemini Planet Imager Adaptive Optics telemetry, the temporal power spectrums for each Zernike coefficient and time averaged covariances are calculated and compared to Kolmogorov atmosphere simulations and analytic models. Longer time-series exposures allow for a characterization of how the optical aberrations due to atmospheric turbulence average over different time-scales, through study of the Zernike variances as a function of averaging time. Preliminary results show deviations from the predicted Kolmogorov model in both the Zernike temporal power spectral densities and time averaged covariances, as well as in the spatial PSD and spatial correlation function of the phase fluctuations.

Creating The Right Atmosphere: An Autoregressive Method For Generating Realistic Atmospheres in Simulations

We present an autoregressive (AR) method for the generation and time evolution of atmospheric phase screens that are tailored to match telemetry gathered by Adaptive Optics (AO) systems such as the Gemini Planet Imager (Gemini-S telescope) and ShaneAO (Shane 3-meter, Lick Observatory). Closed-loop telemetry from the wavefront sensors of these Adaptive Optics (AO) systems is analyzed using a Fourier wind identification technique to extract wind layers, and the wind vector in each layer. Wind vector information is fed into the AR atmosphere generator which can modify the power in individual Fourier modes and whose computational efficiency enables more realistic simulations of arbitrary length. Fidelity to real-world conditions is verified through spatial and temporal PSDs and the comparison of extracted parameters such as structure functions and coherence length. The utility of this method for simulating optical pipelines of extremely large telescopes is explored.
Open questions:
What is best method for extracting meaningful parameters from telemetry? Candidates include: Zernike fits, Fourier domain PSD fits, spatial autocorrelation.
Are the parameters (like r0, L0) extracted from telemetry credible and correlated with other measurements (via MASS, SCIDAR for example)?
What is the time evolution of these parameters (r0 particularly)?
How many wind layers are needed to accurately capture a snapshot of the atmosphere?

MetaCalibration: a direct empirical approach to weak lensing systematics

Estimates of systematic errors in weak lensing have often relied on fully simulated data, for which matching the real galaxy population and distribution of PSF properties can be a challenge. MetaCalibration is a method for calibrating out shear calibration biases, selection biases, and other systematics via direct re-simulation of real data. Work in progress indicates its promise for calibrating out biases due to, e.g., invalid assumptions in shear estimation methods. Further work would be needed to extend the method to include more complex sensor effects. Open questions:

(1) Is assessment of the impact of shear systematics due to sensor effects via MetaCalibration desirable?
(2) For which sensor effects would this approach be feasible?

Weak Lensing Magnification

Magnification is a complementary approach to weak lensing shear, and is appealing as a means for extracting additional lensing signal from survey data. However, studies using magnification still face hurdles if this technique is to reach its full potential. I will show some direct comparisons between masses measured with magnification and shear in CFHTLenS, and discuss some of the likely sources of bias in the magnification approach. Magnification (using lensed number densities) needs to answer open questions including how should we correct for survey systematics that affect number densities? and what is the best way to account for redshift contamination of the samples?

Mitigating Systematics in Weak Lensing Magnification and Galaxy Clustering

Variations in galaxy sample selection caused by inhomogeneous observing conditions such as survey depth, seeing, and extinction imprint themselves on the N-point correlations of the galaxy sample making interpretation of these these correlations difficult. We have developed a method of mitigating these effects using machine learning to generate corrective weights as a function of survey position. Upon application to the CFHTLenS analysis of weak lensing magnification we find that the spurious correlations that were once greater than 10% of the expected signal are significantly reduced, becoming consistent with zero in the best cases. The effectiveness of this technique can be increased by developing methods to identify which combination of survey systematics contribute most to the spurious galaxy correlation amplitudes.

Magnification in the COSMOS field

We use magnitude differences in Cosmos photometry and combine the results of B.G, V, R, I, and K bands to measure intergalactic dust and magnification using only photometric redshifts. We present several systematic and nulls test for this measurements.