subsection_galaxy_clustering_requirement_simulation.tex

\subsection{Requirements}
\label{sec:grs_requirements}
\Auth{Yun, Lado}

The most important task of the SIT in guiding
development of the WFIRST HLS spectroscopy is to set and validate the
requirements of the instrument, the data reduction software, and the survey.
The GRS Lead Co-I Wang will work closely with PI Dor\'e and Co-Is
Hirata and Teplitz on setting requirements for the GRS. To make fully
informed decisions, the team requires high fidelity simulations of
both
instrument performance and the observable sky that the instrument will
measure.  The team must also ensure that the analysis of these data by
the reduction pipeline will be of sufficient quality to enable
measurement with the high precision needed for cosmology.
These simulation and pipeline activities will require the team to
coordinate with the WSC.  Several members of our team (Wang, Teplitz,
Capak, Helou) are located at the Infrared Processing and Analysis
Center (IPAC), and work closely with the WSC.
%\Oli{Mention George too if he stays in}%The WSC task plans are still under discussion, but will
%most likely include development of a pixel-level simulator for imaging
%and spectroscopy.  They will (certainly) include the construction of a
%production-quality analysis pipeline to remove detector effects and
%(probably) extract and measure emission-line spectra.
To the extent practical, we will draw on tools created by the WSC and
design our own software and simulations to be useful to them.
We note that our work primarily demands the ability to quickly and flexibly
simulate different configurations and analyze the results with different
algorithms, while the WSC has the task of developing tools for the
community and production
ready pipelines that integrate with the full WFIRST data system.

\subs{Deriving requirements.} As with WL (\S\ref{sec:wl_requirements}), we will focus first on GRS
requirements that may drive hardware choices, i.e., those that may be
demanding in terms of grism design, detector properties, stability and repeatability of pointing, or dedicated calibration hardware.
We will include a prioritized list of effects to
incorporate in grism simulations.  Over time, we will use our
increasingly realistic network of simulations to evaluate the impact
of requirements and possible trades from the pixel level through
to cosmological inferences. The starting points for this
process are the WFIRST ETC and survey planning software
and the \CoLi\ forecasting tool.

The maximum achievable statistical power of the GRS is determined
mainly by the telescope aperture, throughput, detector area and pixel
size, and allotted observing time.  However, the statistical power and
uniformity of the GRS are further affected by numerous aspects of
instrument performance and survey design, e.g., spectral resolution,
detector read noise and persistence, dither and roll angle pattern,
image quality, complexity of non-1$^{\rm st}$ order features,
spatially varying thermal background due to the warm telescope,
scattered light from bright stars, repeatability of the grism
positioning, and accuracy of calibration of the wavelength-dependent
PSF and distortion map.

Non-uniformity of the survey, which is inevitable to some degree, can
be corrected in clustering measurements by weighting galaxies to
account for incompleteness.  However, large corrections typically come
at a cost in statistical power, and imperfect knowledge of the
non-uniformity leads to systematic errors in the inferred clustering.
The other important source of observational systematics is
contamination of the redshift catalog by artifacts or objects with
incorrectly determined redshifts, and loss of objects from the catalog
because of catastrophic redshift errors or uncertainties in the flux
calibration.  We will define requirements such that (a) the
statistical power of the GRS is close to the maximum allowed by the
telescope aperture and detector area and (b) the impact of uncorrected
observational systematics is small compared to the statistical
errors. The expected precision of the galaxy power spectrum provides a
useful guide to the statistical power of the GRS, but the full
question of cosmological constraining power depends on the
astrophysical modeling techniques used to interpret the measured
clustering, as discussed in \S\ref{sec:cmethods} below.  By the end of our
investigation we will have a complete set of tools to evaluate the
impact of hardware or strategy trades, changes in requirements, or
changes in astrophysical inputs on the expected cosmological return
from the GRS.

\subs{Simulations.} Our development of requirements and a prototype spectroscopic
pipeline will rely critically on realistic simulations of the
pixel-level grism images.  We will work closely with the WSC on
developing these simulations for a variety of cases, ranging from
simple widely separated sources to realistically clustered galaxy
populations drawn from the cosmological simulations described in
\S\ref{sec:cmethods}.  Our team has extensive experience in producing
such simulations for the Hubble Space Telescope (HST) and Euclid.  Co-I Tepliz is one of the leaders
of the HST Wide-Field Camera 3 (WFC3) IR Spectroscopic Parallel survey (WISPs), for which
pixel simulations are vital in assessing completeness and other
parameters \cite{Colbert13}; Co-I Wang (with Teplitz and Capak) is
developing simulation techniques for the Euclid grism survey.
Co-I Teplitz will lead our grism simulations for WFIRST.
A critical astrophysical input for these simulations is the
redshift-dependent luminosity function of H$\alpha$ and [OIII]
emitters, which is currently uncertain at levels that have an
important impact on WFIRST strategy and performance forecasts.  Co-Is
Teplitz and Wang are part of an HST archival study to reprocess
existing data from multiple HST projects to mitigate systematic
uncertainties of the H$\alpha$ luminosity function (LF) measurement.  Through WISPs, Teplitz
is also working to obtain significantly more HST data to improve the
H$\alpha$ LF measurement.  In addition, realistic galaxy templates are
vital to the forecasting of grism measurements, and we are working to
improve both line diagnostics and prediction of line vs. continuum
properties.

\subs{Prototype pipeline.} We will build a prototype spectroscopic pipeline for the analysis of
slitless spectroscopic data to produce a redshift catalog.  This is a
complex, multi-step process.  Co-I Teplitz has extensive experience
with HST slitless spectroscopy using the WFC3, HST Near Infrared Camera and
Multi-Object Spectrometer (NICMOS), and HST Space Telescope Imaging
Spectrograph (STIS) instruments \cite{Atek10,Shim09,Teplitz03}; he
will lead our work on prototype pipelines for the GRS. We will take the basic steps
implemented for WFC3 processing as the starting point for a prototype
WFIRST pipeline. This pipeline must clean the grism images of contamination
from detector artifacts and cosmic rays, register and combine images
from separate dithers and roll angles, match objects in the dispersed
and direct imaging exposures, extract wavelength- and flux-calibrated
2D spectra, infer redshifts based on detected emission lines, and
measure emission-line fluxes and other spectroscopic characteristics.
The resulting catalogs are the input for the clustering analyses
discussed further in \S\ref{sec:cmethods}.

\subs{Analysis challenges.} Slitless spectroscopic analysis presents several important challenges.
First, the pipeline must mitigate the confusion caused by overlapping
spectra.  The standard solution (used by the HST data pipeline) is to
subtract a model of neighboring objects from each source as it is
extracted.  We will investigate the use of HST-like algorithms for
WFIRST, as well as more sophisticated solutions that could produce
better results, such as fitting the full pixel set for regions of the
frame.  Model dependent solutions, with iterative fitting, are
potentially promising, but biases would have to be carefully
understood.  While the HLS obtains exposures at multiple roll angles,
these may be greatly separated in time.  This could introduce new
problems for variable sources, or in fields with foreground moving
objects.  We will also develop methods to automate quality assessment
and flagging of extracted spectra, as the sheer volume of WFIRST GRS
will make human review of spectra (standard practice in current grism
surveys) impossible.

A second major challenge is the need to mitigate catastrophic redshift
errors.  Such failures arise from the misidentification of redshifts
(e.g., confusing [OIII] for H$\alpha$\ in low signal to noise (S/N) spectra) or
false-positive line detections caused by noise peaks or unflagged
cosmic rays.  Redshift fidelity can be greatly improved by using the
photometric redshift estimates derived from the multi-band photometry as a prior
in the redshift determination.  Co-I Capak is spearheading
multidimensional analysis of galaxy color information for WFIRST
photometric redshifts, and that work will be folded into the
spectroscopic pipeline.
% (\Oli{PETER SHOULD DOUBLE CHECK THE WORDING HERE}).

\subs{Completeness maps.} Nearly as important as the redshift catalogs themselves is the production of
completeness maps that characterize the spatially varying depth of the survey,
the level of contamination, and regions that should be masked because the
catalog is unreliable.  These completeness maps are used to weight galaxies in
clustering analysis and/or to create random catalogs such that the local number
density of points is proportional to the likelihood of successfully measuring a
redshift of a galaxy if it were at that point.  Co-I Samushia is
leading a similar effort in DESI and has previously worked on quantifying and removing
systematic effects associated with inaccuracies in random catalogues
\cite{Samushia2012}.
Co-I Ho has also led the effort in creating the SDSS-BOSS LSS catalog and randoms \cite{Reid2015} and led the effort in removing observational effects in BOSS LSS catalog \cite{Ross2011}.
Co-I Samushia will lead our work to develop tools for creating these completeness maps
by a full forward-modeling method, where artificial sources are assigned random
angular positions and redshifts, added to grism images, and pushed through the
data pipeline.  Compared to existing large redshift surveys (from ground-based
fiber spectroscopy), the WFIRST completeness map will have much more complex
small scale structure because of the varying numbers of exposures at individual
points on the sky and sensitivity variations across the focal plane.  Because of
source confusion and sky background effects, the completeness and contamination
will be a function of the local galaxy surface density.  We will develop
strategies and tools for recording these large and complex completeness maps in
formats that can be efficiently used to create random catalogs and weight
galaxies for clustering analyses.  By the end of the investigation period, we
will be able to create full pixel-level simulations from an input cosmological
simulation (see \S\ref{sec:cmethods}), run them through our proto-type pipeline to create a
redshift catalog and completeness map, and analyze the resulting artificial data
set with our clustering analysis tools to compare to the idealized case that has
the complete galaxy catalog of the cosmological simulation.

\subs{Calibration strategies.}
We will define the absolute and relative calibration requirements for the GRS, such as the angular scale and temporal stability.
We will develop methods for calibrating the relative and absolute flux measurements along with wavelength calibration and redshift accuracy and completeness.
For flux and wavelength calibration we will set requirements on the ground testing, in flight calibration sources, and calibration observations based on experience
with other missions including Spitzer, HST, and Euclid.  Furthermore, we will investigate self calibration strategies based on optimizing dither patterns and exposure
times for the science observations and the use of touch-stone fields that both calibrate and provide long-term trending of the data.  Both the primary and
self calibration procedures will be tested with simulations specified by this SIT and conducted by the WSC.  Finally, we will use the large spectroscopic surveys
necessary for the weak lensing photo-$z$ calibration to verify the calibration by directly testing the redshift accuracy and completeness estimates from the simulations.
Co-Is Capak and
Padmanabhan, both with extensive experience from similar work for
Euclid and BOSS, will lead our calibration work.

\subsection{Simulations}
\label{sec:grs_simulations}
\Auth{Shirley, Elena, Andrew, Alina, Alex, Yun}


%===
% \subsection{Requirements, Simulations, and Proto-type Pipelines}


% The most important task of the SIT in guiding
% development of the WFIRST HLS spectroscopy is to set and validate the
% requirements of the instrument, the data reduction software, and the survey. To make fully informed
% decisions, the team requires high fidelity simulations of both
% instrument performance and the observable sky that the instrument will
% measure.  The team must also assure that the analysis of these data by
% the reduction pipeline will be of sufficient quality to enable
% measurement with the high precision needed for cosmology.
% These simulation and pipeline activities will require the team to work
% closely with the WFIRST Science Centers (WSC).  Several members of our team
% (Wang, Teplitz, Capak) are located at IPAC, and they work closely with
% the WSC. %The WSC task plans are still under discussion, but will
% %most likely include developmet of a pixel-level simulator for imaging
% %and spectroscopy.  They will (certainly) include the construction of a
% %production-quality analysis pipeline to remove detector effects and
% %(probably) extract and measure emission-line spectra.
% To the extent practical, we will draw on tools created by the WSC and
% design our own software and simulations to be useful to the WSC.
% We note that our work primarily demands the ability to quickly and flexibly
% simulate different configurations and analyze the results with different
% algorithms, while the WSC has the task of developing production
% ready pipelines that integrate with the full WFIRST data system.

% As with WL (\S\ref{sec:wl_requirements}), we will focus first on GRS requirements that may drive hardware choices, i.e., those that may
% be demanding in terms of grism design, detector properties, stability
% and repeatability of pointing, or dedicated calibration hardware.
% Our initial assay will include a prioritized list of effects to incorporate in grism simulations.  Over time, we will use our
% increasingly realistic network of simulations to evaluate the
% impact of all requirements and possible trades from the pixel level through to cosmological inferences.  The starting points for this
% investigation are the WFIRST ETC and survey planning software (both written by Co-I Hirata) and the CosmoLike forecasting tool
% described in \S\ref{sec:wl_requirements}.

% The maximum achievable statistical power of the GRS is determined
% mainly by the telescope aperture, throughput, detector area and pixel size,
% and allotted observing time.  However, the statistical power and uniformity of the GRS
% are further affected by numerous aspects of instrument performance and survey design:
% spectral resolution, detector read noise and persistence, dither and roll angle pattern,
% blurring due to diffraction by the glass elements of the grism, spatially varying thermal
% background due to the warm telescope, scattering due to bright
% stars, repeatability of the grism positioning, and accuracy of
% calibration of the wavelength-dependent PSF and distortion map.

% Non-uniformity of the survey, which is inevitable to some degree,
% can be corrected in clustering measurements by weighting galaxies
% to account for incompleteness.  However, large corrections typically come
% at a cost in statistical power, and imperfect knowledge of the
% non-uniformity leads to systematic errors in the inferred clustering.
% The other important source of observational systematics is
% contamination of the redshift catalog by artifacts or objects with incorrectly
% determined redshifts, and loss of objects from the catalog
% because of catastrophic redshift errors.
% We will define requirements such that (a) the statistical power of
% the GRS is close to the maximum allowed by the telescope aperture
% and detector area and (b) the impact of uncorrected observational
% systematics is small compared to the statistical errors. The expected precision of
% the galaxy power spectrum provides a useful guide to the statistical power of the GRS, but the full
% question of cosmological constraining power depends on the
% astrophysical modeling techniques used to interpret the measured
% clustering, as discussed in \S 5.2 below.
% By the end of our investigation we will have a complete set of
% tools to evaluate the impact of hardware or strategy trades,
% changes in requirements, or changes in astrophysical inputs
% on the expected cosmological return from the GRS.

% Our development of requirements and a proto-type spectroscopic pipeline
% will rely critically on realistic simulations of the pixel-level
% grism images.  We will work closely with the WSC on developing
% these simulations for a variety of cases, ranging from simple
% widely separated sources to realistically clustered galaxy populations
% drawn from the cosmological simulations described in \S 5.2.
% {\bf Can we say more here?  Do we have any experience producing
% grism simulations for HST analysis?}
% A critical astrophysical input for these simulations
% is the redshift-dependent luminosity function of H$\alpha$ and [OIII]
% emitters, which is currently uncertain at levels that have
% an important impact on WFIRST strategy and performance forecasts.
% Co-I's Teplitz and Wang are part of an HST archival study to reprocess existing
% data from multiple HST projects to mitigate systematics uncertainties of the
% H$\alpha$ LF measurement.  Teplitz is also a key member of the WISP team,
% which aims to obtain significantly more HST data to improve the H$\alpha$ LF measurement.

% We will build a prototype spectroscopic pipeline for the
% analysis of slitless spectroscopic data to produce a redshift catalog.
% This is a complex, multi-step process.
% Co-I Teplitz has extensive experience with the HST grism data
% pipeline (e.g., \cite{REFS}), and we will take the algorithms
% implemented in this pipeline as the starting point for a
% proto-type WFIRST pipeline. This pipeline must clean the grism images for contamination from
% detector artifacts and cosmic rays, register and combine images from
% separate dithers and roll angles, extract wavelength- and flux-calibrated
% 2D spectra, match these spectra
% to individual objects from the photometric imaging data,
% infer redshifts based on detected emission lines, and measure
% emission-line fluxes and other spectroscopic characteristics.
% The resulting catalogs are the input for the clustering analyses
% discussed further in \S 5.2.

% The biggest challenge for slitless spectroscopic analysis is
% eliminating confusion from overlapping spectra.
% The standard solution (used by the HST data pipeline) is to subtract a
% model of neighboring objects from each source as it is extracted.
% We will investigate the use of HST-like algorithms for WFIRST,
% as well as more sophisticated solutions that could produce better results,
% such as fitting the full pixel set for regions of the frame.
% Model dependent solutions, with iterative fitting, are potentially
% promising, but biases would have to be carefully understood.
% While the HLS obtains exposures at multiple roll angles, these may
% be greatly separated in time.  This could introduce new problems for variable sources, or in fields with foreground moving objects.
% We will also develop methods to automate quality assessment
% and flagging of extracted spectra, as the sheer volume of WFIRST GRS will make human review of spectra
% (standard practice in current grism surveys) impossible.

% Another challenge is the misidentification of noise peaks as emission lines, which leads to
% catastrophic redshift errors. This can be mitigated by using the photometric redshift derived from
% the multi-band photometry as a prior in the redshift determination.

% Nearly as important as the redshift catalogs themselves is the
% production of completeness maps that characterize the spatially
% varying depth of the survey, the level of contamination, and
% regions that should be masked because the catalog is unreliable.
% These completeness maps are used to weight galaxies in clustering
% analysis and/or to create random catalogs such that the local number
% density of points is proportional to the likelihood of successfully measuring a redshift of a galaxy if it were at
% that point.  We will develop tools for creating these completeness
% maps by a full forward-modeling method, where artificial sources
% are assigned random angular positions and redshifts, added to
% grism images, and pushed through the data pipeline.
% Compared to existing large redshift surveys (from ground-based
% fiber spectroscopy), the WFIRST completeness map will have much
% more complex small scale structure because of the varying numbers
% of exposures at individual points on the sky and sensitivity
% variations across the focal plane.
% Because of source confusion and sky background effects,
% the completeness and contamination will be a function of
% the local galaxy surface density.
% We will develop strategies and tools for recording these large and
% complex completeness maps in formats that can be efficiently
% used to create random catalogs and weight galaxies for clustering
% analyses.  By the end of the investigation period, we will be able
% to create full pixel-level simulations from an input cosmological
% simulation (see \S 5.2), run them through
% our proto-type pipeline to create a redshift catalog and completeness
% map, and analyze the resulting artificial data set with our
% clustering analysis tools to compare to the idealized case
% that has the complete galaxy catalog of the cosmological simulation.