Measurement of the cosmic optical background using the long range reconnaissance imager on New Horizons

Abstract

The cosmic optical background is an important observable that constrains energy production in stars and more exotic physical processes in the universe, and provides a crucial cosmological benchmark against which to judge theories of structure formation. Measurement of the absolute brightness of this background is complicated by local foregrounds like the Earth's atmosphere and sunlight reflected from local interplanetary dust, and large discrepancies in the inferred brightness of the optical background have resulted. Observations from probes far from the Earth are not affected by these bright foregrounds. Here we analyse the data from the Long Range Reconnaissance Imager (LORRI) instrument on NASA's New Horizons mission acquired during cruise phase outside the orbit of Jupiter, and find a statistical upper limit on the optical background's brightness similar to the integrated light from galaxies. We conclude that a carefully performed survey with LORRI could yield uncertainties comparable to those from galaxy counting measurements.

ARTICLE

Received 29 Sep 2016 | Accepted 20 Feb 2017 | Published 11 Apr 2017

DOI: 10.1038/ncomms15003 OPEN

Measurement of the cosmic optical background using the long range reconnaissance imager on New Horizons

Michael Zemcov1,2, Poppy Immel1, Chi Nguyen1, Asantha Cooray3, Carey M. Lisse4 & Andrew R. Poppe5

The cosmic optical background is an important observable that constrains energy production in stars and more exotic physical processes in the universe, and provides a crucial cosmological benchmark against which to judge theories of structure formation. Measurement of the absolute brightness of this background is complicated by local foregrounds like the Earths atmosphere and sunlight reected from local interplanetary dust, and large discrepancies in the inferred brightness of the optical background have resulted. Observations from probes far from the Earth are not affected by these bright foregrounds. Here we analyse the data from the Long Range Reconnaissance Imager (LORRI) instrument on NASAs New Horizons mission acquired during cruise phase outside the orbit of Jupiter, and nd a statistical upper limit on the optical backgrounds brightness similar to the integrated light from galaxies. We conclude that a carefully performed survey with LORRI could yield uncertainties comparable to those from galaxy counting measurements.

1 Center for Detectors, School of Physics and Astronomy, Rochester Institute of Technology, 1 Lomb Memorial Drive, Rochester, New York 14623, USA.

2 Astrophysics and Space Sciences Section, Jet Propulsion Laboratory (JPL), 4800 Oak Grove Drive, Pasadena, California 91109, USA. 3 Department of Physics & Astronomy, University of California, Irvine, California 92697, USA. 4 Planetary Exploration Group, Space Department, Johns Hopkins University Applied Physics Laboratory, 11100 Johns Hopkins Road, Laurel, Maryland 20723, USA. 5 Space Science Laboratory, University of California at Berkeley, Berkeley, California 94720, USA. Correspondence and requests for materials should be addressed to M.Z. (email: mailto:[email protected]

Web End [email protected] ).

NATURE COMMUNICATIONS | 8:15003 | DOI: 10.1038/ncomms15003 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 1

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15003

The cosmic optical background (COB) is the summed emission from all sources outside of our Milky Way galaxy emitted at wavelengths roughly corresponding to those

visible with the human eye. It is a powerful diagnostic of the emission from known astrophysical processes in galaxies including stellar nucleosynthesis, mass accretion onto black holes and the gravitational collapse of stars13. A comparison of the COB intensity to the surface brightness arising from known galaxy populations can reveal the presence of diffuse backgrounds produced by more exotic phenomena such as the decay of particle species outside the standard model or light from objects outside of galaxies46.

Direct photometric measurement of the COB has proven to be challenging. The earths atmosphere is several orders of magnitude brighter than the COB, and accounting for the various relevant emission, absorption, and scattering effects is a daunting task. Sunlight scattered from interplanetary dust (IPD) particles in the Solar system, known as Zodiacal light when viewed from the earth, also produces a large foreground to direct measurement of the COB from vantage points in the inner Solar system. Though progress has been made in carefully accounting for the atmosphere and Zodiacal light in the optical7,8 and into the near-IR914, as it is typically 4100 times brighter than the COB small errors in this accountancy propagate to large errors on the COB15,16. It is thus desirable to measure the COB from vantage points where the earths atmosphere and the light from IPD are not appreciable components of the diffuse sky brightness, such as the outer parts of our Solar system17. Though many planetary probes have had optical-wavelength cameras, they are rarely designed with the demands of extragalactic astronomical observations in mind.

Two exceptions to this are the early NASA probes Pioneer 10 and 11, which were instrumented with imaging photopolari-meters (IPPs) that returned measurements of the sky brightness ranging from 1 to 5.3 a.u. (ref. 18). These data have been used to measure both the decrease in the IPD light with heliocentric distance19, diffuse light from the Galaxy20,21 and the brightness of the COB itself22,23 using the two IPP bands spanning 390500 and 600720 nm. The Pioneer measurements remain the most stringent constraints of the COB23, and have uncertainties dominated by errors associated with subtracting galactic components including the integrated light from stars (ISL) and diffuse galactic light (DGL).

NASAs New Horizons spacecraft24 recently performed the rst detailed reconnaissance of the PlutoCharon system. It includes as part of its instrument package the Long Range Reconnaissance Imager2527 (LORRI), an optical camera with sensitivity over a broad 440870 nm half-sensitivity passband. Importantly, rather than a scanning photometer like the IPP, LORRI is a Newtonian telescope with characteristics including excellent pointing stability, a 20.8 cm diameter Ritchey-Chrtien telescope, an 0.3 0.3 instantaneous eld of view, 100 100

pixels, and (crucially) real-time dark current monitoring. The achieved point source sensitivity of LORRI is V 17 in a 10 s

exposure in 4 4 pixel on-chip rebinning mode, making it a

sensitive astronomical instrument. As a result of this sensitivity and angular resolution, much of the starlight that challenged the earlier Pioneer measurements can be resolved out in LORRI images, providing a relatively clean measurement of diffuse astrophysical emission.

In this paper, we use archival data from the New Horizons checkout and cruise phases to measure the COB from several vantage points in the Solar system. We correct for dark current in the detectors, mask bright stars from the images, assess the amplitude of residual starlight, sunlight from interplanetary dust and diffuse galactic light, and correct for galactic extinction to

measure lICOBl 4.77.3(stat.) 10:3 11:6sys: nW m 2 sr 1, giving

a 2s statistical upper limit of lICOBlo19:3 nW m 2 sr 1, which excludes some of the early results in the literature. This measurement is based on a very limited data set with characteristics that complicate astrophysical examination. We conclude that a carefully designed survey of the COB from LORRI beyond the orbit of Pluto has the potential of denitively measuring its surface brightness away from the complicating effects of the earths local interplanetary dust cloud.

ResultsData set. In this study we concentrate on the 4 4 pixel rebin

ned LORRI exposures, for which on-chip summing has been used to improve the surface brightness sensitivity over the native-resolution data. This rebinning mode is particularly advantageous in the small signal regime where the read noise penalty is large. The rebinned images have spatial resolution of 400.3 over a full 256 256 (super-)pixel frame. We term magnitudes in this

LORRI band RL, as it is close to (though much broader than) the Johnson-Cousins R-band at 640 nm (ref. 28); in fact the LORRI bandpass covers essentially all of astronomical V, R and I bands.

New Horizons was launched 19 January 2006. The path of New Horizons through the Solar system is summarized in Fig. 1. Approximately 90 days after launch, some 359 dark data sets were acquired by LORRI while approximately 1.9 a.u. from the sun. During this time, the LORRI dust cover was in place over the telescope aperture, providing close to optically dark conditions. The cover was ejected on 2006 August 29, and shortly thereafter a series of images of the open star cluster Messier 7 were acquired. From these data, the New Horizons team determined a preliminary photometric calibration, a full-width at half maximum (FWHM) pointing jitter of 0.45 pixels, and geometric distortion o0.2 pixels across the eld of view27. Following this, a series of short exposure test images was acquired, and at the beginning of 2007 the rst science image was taken of Callirrhoe, a small irregular moon of Jupiter, with a full 10 s exposure time. A series of images of Jupiter and its immediate environment were then acquired during an encounter, which we do not consider here. Closest approach to Jupiter occurred on 28 February 2007. Following the Jupiter encounter, New Horizons entered cruise phase. LORRI data were acquired on an approximately annual basis, and consisted of 10 s observations of distant Solar system objects. The current public archival data records end 20 July 2014, approximately a year before the Pluto encounter.

We cut the data using requirements on integration time, solar elongation and thermal dust emission, following which we are left with elds 1 4 whose characteristics are summarized in Table 1.

COB measurement. The brightness in an arbitrary image of the astronomical sky acquired above the Earths atmosphere lImeasl can be expressed as:

lImeasllIIPDl lI l lIRSl lIDGLl ElICOBl lIinstl; 1

where lIIPDl is the brightness associated with interplanetary dust, lI l is the brightness associated with resolved stars, lIRSl is the brightness associated with residual starlight from stars too faint to be detected individually or the faint wings of masked sources, lIDGLl is the brightness of the DGL, lICOBl is the brightness of the

COB, E is a factor accounting for absorption in galactic dust, and Iinstl is brightness associated with the instrument including all potential contributions to the measured zero-point offset. The major difculty with COB measurements is that, with the exception of lI l, each of these sources can have an isotropic component, which is problematic since the COB itself is isotropic.

As a result, care must be taken to understand and correct for the

2 NATURE COMMUNICATIONS | 8:15003 | DOI: 10.1038/ncomms15003 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15003 ARTICLE

a

Table 1 | Data sets used in this analysis.

Field number

a (J2000) hh:mm:ss

d (J2000) hh:mm:ss

20

c () b () eb () AV

(mag)

1 13:04:02 23:57:02 345.4147 85.7384 28.2096 0.06

2 10:47:36 26:46:56 271.4532 28.4141 31.5843 0.22

3 23:04:27 7:07:00 66.2722 57.6861 1.0847 0.16

4 00:07:14 1:15:00 98.8079 62.0328 1.8651 0.10

Uranus

d y(a.u., Solar ecliptic J2000.0)

10

0

Saturn

Launch

Jupiter

Cover ejected

Jupiter encounter

Jupiter encounter

Field 2

Field 1

102

101

100

10

20

1.0

0.5

0.0

0.5

1.0

0 5 10 15 20 25

WFPC2

Field 3

Field 4

CIBER

DIRBE

COB (nW m2 sr1 )

HESS -rays

Pioneer

20

10

0

10

20

HDF/SDF

dx (a.u., Solar ecliptic J2000.0)

[afii9838]I [afii9838]

b

STIS

0.2 0.3 0.4 0.5 0.7

[afii9838] (m)

1.0 1.5 2.0

d z(a.u., Solar ecliptic J2000.0)

Figure 2 | Measurements of the COB surface brightness. The lICOBl determined in this study are shown as both an upper limit (red) and a mean (red star). We also show previous results in the literature, including direct contraints on the COB (lled symbols) and the IGL (open symbols). The plotted LORRI errors are purely statistical and are calculated from the observed variance in the mean of individual 10 s exposures; see Fig. 3 for an assessment of the systematic uncertainties in the measurement. We include the measurements from HST-WFPC2 (ref. 7; green squares), combinations of DIRBE and 2MASS1013 (diamonds; the wavelengths of these measurements have been shifted for clarity), a measurement using the dark cloud method8 (grey circles), and previous Pioneer 10/11 measurements22,23 (blue upper limit leader and circles). The gold region

indicates the H.E.S.S. constraints on the extragalactic background light29.

We include the background inferred from CIBER5 (pentagons). The IGL points are compiled from HST-STIS in the ultraviolet (UV)62 (open square), and the Hubble Deep Field63 (downward open triangles) the Subaru Deep Field64,65 (upward open triangles and sideways pointing triangles) in the optical/near-IR. Where plotted, horizontal bars indicate the effective wavelength band of the measurement. Our new LORRI value from just 260 s of integration time is consistent with the previous Pioneer values.

Launch

Field 1

Field 3 Field 4

Field 2

Cover ejected

Jupiter

Saturn

Uranus

dr (a.u., Solar ecliptic J2000.0)

Figure 1 | The trajectory of New Horizons through the solar system. Data collection periods of relevance to this study are indicated. Both the x y

and r z planes are shown (a,b, respectively), with the axes in solar ecliptic units and dr

d2x d2y

q . New Horizons was launched from Earth at 1 a.u., and the data with the LORRI dust cover in place were acquired at 1.9 a.u., just beyond Mars orbit at 1.5 a.u. (inner blue dotted lines). The dust cover was ejected near 3.6 a.u., and the data were acquired before and during an encounter with Jupiter. The data considered here were taken between 2007 and 2010 while New Horizons was in cruise phase. The red vectors indicate the relative positions of elds 1 4 compared to the sun and plane of the

ecliptic.

brightness of each component, particularly those that appear constant over angular scales similar to the eld of view of the instrument.

We isolate lICOBl using three basic steps: mask stars near or brighter than the detection threshold to remove the effect of lI l;

subtract the diffuse components either originating in the instrument or from local astrophysical emission to isolate the

diffuse residual component lIresidlElICOBl; and correct the mean

residual intensity for the effects of galactic extinction to yield lICOBl. Averaging over all the elds using inverse noise variance weighting, we determine that lICOBl4:7 7:3 nW m 2 sr 1,

where the uncertainty is purely statistical and is assessed from the scatter in the individual exposures. This gives a 2s upper limit on the COB brightness of lICOBl2so19:3 nW m 2 sr 1. Our

measurement and comparisons with previous measurements in the literature are shown in Fig. 2.

This measurement is also subject to various systematic uncertainties associated with the calibration and foreground removal. We carefully assess these errors by probing the allowed variation in each of the models and measurements to derive an overall calibration and systematic uncertainty budget,

NATURE COMMUNICATIONS | 8:15003 | DOI: 10.1038/ncomms15003 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 3

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15003

COB (nW m2 sr1 )

15

10

5

0

[afii9829][afii9838]I [afii9838]

5 Inst.

Calibration

Astrophysical

Dark current

Photometric calibration

beamuncertainty

USNO-B1 uncertainty

Star sample variance

IPD light

DGL uncertainty

Figure 3 | Summary of the various systematic errors in our determination of kICOBk. The various sources of uncertainty are labelled, with the coloured bars showing their variation from the mean value we measure (red solid lines; see also Supplementary Table 3). Most of the errors are smaller than the statistical uncertainty of the measurement (dashed red lines), except for the uncertainty in the DGL model, which is large compared to the other errors. We do not show the errors associated with the optical ghosts and extinction correction as these are substantially less than a signicant gure. The dominant uncertainties in this measurement are in fact not statistical, and to a great extent depend on the elds chosen and ancillary data available, so further observations in a dedicated survey program hold great promise.

summarized in Fig. 3. As the astrophysical errors identied in this analysis are uncorrelated, combining them in quadrature is an appropriate estimate of the total error present in the measurement. This is not the case for the calibration errors, which we add linearly and then sum in quadrature with the astrophysical foregrounds to give a conservative total systematic error estimate of { 10.3, 11.6} nW m 2 sr 1.

DiscussionThese data show the power of LORRI for precise, low-foreground measurements of the COB. The measurement presented here is not consistent with the earlier HST-WFPC2 constraints7, but is consistent with both the Pioneer23 and dark cloud8 measurements, as well as the g-ray inference29. This measurement constrains the possibility of a COB signicantly in excess of the expectation from IGL.

Though the bandwidth required to telemeter the data from the outer Solar system constrains the number of observations possible, with a carefully designed survey we should be able to produce a denitive measurement of the diffuse light in the local universe, and a tight constraint on the light from galaxies in the optical wavebands. LORRIs ability to resolve much of the starlight has signicantly reduced the potential for foreground contamination compared to measurements from the Pioneer IPP, and the LORRI eld is small enough that bespoke ground-based assessments of the faint starlight in each eld are conceivable. As a result, a future LORRI survey would benet from careful design and pre facto observations of the survey elds. Given the total integration time used in this measurement was only 260 s, a total integration time of B4.5 h would allow us to achieve B1 nW m 2 sr 1 statistical uncertainties. Because LORRI can allow 30 s integrations, this hypothetical measurement would require B500 integrations, which is not prohibitive in terms of data storage nor telemetry requirements.

It would be particularly useful to observe high galactic latitude elds at a variety of ecliptic latitudes and solar elongations to search for IPD light. Though likely to be too faint to detect, models suggest there may be an increase in the IPD population towards the Kuiper Belt from collisional material30. This increase may be observable in the IPD light intensity with a carefully designed, deep survey. At the very least, observations of the inner Solar system from New Horizons perspective may provide useful new information about the global structure of the IPD cloud. In addition, currently unpublished instrument calibration information such as susceptibility to off-axis light and detailed pointing stability assessments could improve the accuracy of this kind of measurement.

A primary lesson learned from this analysis is that, following the accurate removal of ISL, the DGL estimate becomes the largest source of uncertainty. Because of the uncertainty in the measured 100 mm background level, the DGL-I100mm scaling, and the galactic latitude dependence of the scattering, this component varies by approximately the expected brightness of the IGL in each of our elds. For example, Field 1 is very close to the galactic pole where the DGL should be faint, but even here the models and observations suggest the DGL brightness could vary from 4 to 11 nW m 2 sr 1 at one s.d. Future observations with LORRI should concentrate on the lowest I100 mm elds available on the sky to minimize the uncertainty. If many statistically independent elds are sampled, the DGL-I100 mm linear regression technique we briey explore should permit measurement of the optical-thermal infrared correlation precise enough to allow subnW m 2 sr 1 determination of the COB. Improved DGL characterization using other techniques are also of continuing importance.

This measurement of the COB brightness, while not currently as precise as those from Pioneer23, is important as it suffers from completely different instrumental and foreground uncertainties as the existing measurement. It is also the only measurement sensitive to the I-band 700900 nm wavelength range. Though some challenges remain, further data from LORRI could provide a denitive measurement of the extragalactic background light at optical wavelengths, and may be instrumental in completing our understanding of the history of stars and galaxies in the universe.

Methods

Observational data. The basic New Horizons ight timeline is given in Supplementary Table 1, including a summary of the data taken during the checkout and cruise phases. The LORRI bandpass has an effective wavelength of l 655 nm

for a at-spectrum source, with half power response from 440870 nm. Supplementary Table 1 indicates the dates of the observations, the notional targets, the number of integrations available, the exposure time per integration, as well as astrophysical information like the heliocentric distance R , solar elongation y ,

and the 100 mm specic intensity I100mm. In this work, we refer to each data set with

a eld number, running over D14, R110 and 14, using the numbering scheme presented below.

We performed cuts on the full data set to account for various factors affecting the data quality. First, we restricted our attention to the data with integration times tint41 s, eliminating sets R1 4. Second, because the large-angle response of the

LORRI telescope shows response from diffuse scattering of sunlight illuminating the light bafe31, we remove data for which yeo90. Finally, many of these LORRI data sets are taken at low galactic latitudes where the DGL is bright. We exclude data sets for which I100mm410 MJy sr 1, which removes elds R5 and R710 from analysis. Though excluding much of the most useful data is not ideal, these elds are measured very close to the Galactic plane where the contamination from the local environment precludes careful measurement of the faint signals from either the COB or interplanetary dust in reection.

Data reduction. The archival LORRI data are available in a format in which they have already been processed through several instrument calibration steps including bias correction, smear correction and relative pixel response correction27,32. In brief, the raw data consists of voltages measured at the end of the exposure reported in data numbers (DN). In the rst step of the processing, the median of the dark reference pixels is used to subtract the global reference voltage, and a

4 NATURE COMMUNICATIONS | 8:15003 | DOI: 10.1038/ncomms15003 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15003 ARTICLE

reference super-bias frame measured during ground testing is used to correct for bias variations over the array. Next, image smearing and at eld corrections measured during ground calibration are used to account for image smearing and relative pixel response. The les that are input to our processing have units data number per integration time and, though they have been partially processed, must be: astrometrically registered; masked for known detector defects, transients, and bright astronomical objects; corrected for instrumental effects; and calibrated to photometric units. Example images of the four science elds in this study are shown in Supplementary Fig. 1, calibrated to lIl.

Astrometric registration. Astrometric registration is required to allow masking of bright stars, which is necessary to account for lI l. Functionally, this means we want to ensure that each pixel in an image is accurately associated with a pair of right ascension and declination coordinates, (a, d). We determine the orientation and the scale of each image using the publicly available astrometric calibration software package http://astrometry.net

Web End =http://astrometry.net 33. The algorithm uses a four-step procedure in which: bright sources in each image are detected; the detected sources are divided into subsets whose relative positions are recorded and matched against a pre-built index; the solution is veried using predictive star position checks; and, the nal alignment information is returned to the user in a Flexible Image Transport System (ts) header for each le. For the LORRI data, this algorithm successfully solved the astrometric registration for each eld independently. Further, constant parameters of the instrument like the pixel size and image distortions are found to be consistent between observations over the entire data set, with very small uncertainties. The registration information returned by the software package is used to calculate (a, d) for every pixel in each image.

Masking. To reliably measure the diffuse sky brightness, it is necessary to exclude residual instrumental signal and detectable point sources from the images that contribute to lIinstl and lI l, respectively. We implement image masks to remove brightness associated with: stars near or brighter than the detection limit; pixels that may suffer from electronic or optical pathologies; and cosmic rays and other transient events.

To mask stars, we use the USNO-B1 catalog34, which provides photometric uxes in approximately Johnson-Cousins B, R and I bands over the entire sky. Though in some regions this catalog reaches completeness of V 21, it is

nonuniform and source uxes are calibrated to only 0.25 mag accuracy. We synthesize the RL ux by tting a linear model to the USNO-B1 measurements and compute the LORRI band-weighted integral of the ux for each source. To compromise between maximal removal of stars and minimal removal of galaxies that contribute to the COB, we mask the elds at a ux limit of RL 17.75, which is

B1 mag below the 1s point source sensitivity in the images. This threshold has the added benet of moving the mask ux threshold away from the USNO-B1 catalog completeness limit where the survey uniformity is problematic. Given LORRIs small eld of view, we calculate that the error introduced in the nal COB estimates due to accidentally masking galaxies at the bright end of the number counts35 is0.006 nW m 2 sr 1. A search of the available optical data in these elds is consistent with the number counts, and we nd that no exceptionally bright galaxies fall into these elds.

To build an appropriate source mask, we also require accurate knowledge of the instrument point spread function (PSF). We use a stacking method36 to sum the emission from all RLo16 sources in the image to form an estimate of the LORRI

PSF. Briey, for each source brighter than the magnitude limit, we interpolate the image onto a ten times ner grid centered on the cataloged source position, and sum all such postage-stamp images. In each postage-stamp image, pixels far from any star images or masked pixels are used to calculate the zero point of the image. No lower magnitude limit is required as none of the images contain stars bright enough to induce non-linear response in the detector. The stacked PSF is shown in Supplementary Fig. 2. We compute the uncertainty in the PSF by performing two stacks, one stacking on a random half of the sources in the catalog and the other on the other half. We nd a FHWM of 1.530.0500, consistent with both laboratory26 and in-ight37 measurements of the PSF of 1.500.

We mask stars in each LORRI image by using the band-weighted magnitude estimate m to calculate a radius around each source to exclude from analysis. This radius r(m) is computed from:

rm2:5

mlim

m

2; 2

where the free parameters are determined empirically from the data. Here,mlim 17.7 at RL-band and r(m) has units of pixels. To assess the efcacy of this

mask as a function of magnitude, we simulate noiseless images of the stars in each eld per magnitude bin, apply the mask, and then calculate the residual surface brightness. We nd the largest contribution to the residual brightness is from stars near the limiting magnitude, which contribute at most 0.15 pW m 2 sr 1 per source. Brighter sources have larger masks and contribute less total surface brightness as there are fewer of them. We calculate the total ux left in the images from residual unmasked star ux as part of the ISL assessment.

The data used in our COB study have a variety of Solar system objects as their primary targets. Though Haumea and Makemake should be faint in the optical (R415), even at a distance of 23.2 a.u. Neptune appears bright in the LORRI

images (RB7). To account for this, we uniformly mask the central 20.3 20.3 from

each of the science images. In the case of Neptune, this corresponds to 35R , which is signicantly beyond the outermost known ring at 2.6R and the brightest moons (including Triton at 14.3R ). The Neptunian system does extend further, with a small moon orbiting B2,000R (at this distance 2.2)

from the planet, and we cannot exclude the possibility that a dust halo far from the planet is reecting sunlight and increasing the surface brightness in the LORRI image. We do not observe residual structure in these images, so any such contamination from a circum-Neptunian dust cloud is relatively faint. In principle, Haumea and Makemake may have their own dust clouds, and we cannot exclude the possibility from these data. As a result of these considerations, formally our measurements must be considered to be upper limits to the surface brightness of the COB. Future observations away from known Solar system objects would be benecial in this regard.

For the Neptune eld, the image of the planet is bright enough to induce charge transfer artifacts in the detector, so we mask three pixel wide stripes in both the vertical and horizontal directions of each image.

LORRI is known to have optical ghosting from reections in the eld attening lens group for sources that fall between just off the eld to up to 0.37 from the eld27. This ghost is visible in the eld 1 images, but not in the other elds. The central source mask removes a large fraction of this emission, but we also manually mask the ghosts in the eld 1 images.

To reduce contamination from known defects, we mask both the outermost ve pixels in the image, as well as pixels that are consistently non-responsive or saturated in the images. Finally, we apply a clip mask which excludes pixels 43s from the mean value of each image. This excludes pixels with transient contaminants like electrical or digitization error and cosmic rays. On average, we exclude only B100 pixels in this s-clipping step, corresponding to a 0:15% loss. In

Supplementary Fig. 3 we show the same example images shown in Supplementary Fig. 1, but with the full image mask imposed.

Dark current and reference pixel behaviour. Since it cannot be separated from astronomically sourced photocurrent, an important potential contaminant in this measurement is the dark current of the detector, which contributes as an approximately isotropic component of lIinstl. The operating temperature of the

LORRI charge coupled device (CCD) is B200 K, so based on the performance of similar devices we might expect the dark current to be negligible. However, the COB measurement is more robust if the dark current contribution can be completely characterized.

An important feature of the LORRI CCD is the presence of rows of 4 1,024

(or 1 256 in rebinned mode) dark pixels. These are optically active pixels that are

shielded from incident light by means of a metal lip, but are otherwise identical to the optically active pixels. These pixels are used in the LORRI data reduction pipeline to remove the combination of dark current and voltage bias in the images. In our study, these have the added benet of giving a xed reference of the detector array performance.

To characterize the long-term performance and stability of the detector, we compare the measurements performed with the dust cover in place against those taken with the dust cover off. The photocurrent in the reference pixels should be solely a function of the temperature of the detector38, which is shown as a function of time in Supplementary Fig. 4. During post-launch operations before the cover was ejected, the detector system was passively cooled to its nal operating temperature of B193 K. As a result, the cover-on data were acquired at a signicantly higher temperature than the optical data.

The detector manufacturer has empirically determined that, in equal-integration time exposures, the dark current at a temperature T can be estimated from:

iT122 i0 T3 exp 6400=T 3 with i0B104 e s 1 per pixel. From this, we would predict i(220 K)B2.8 e s 1 per pixel and i(193 K)B0.035 e s 1 per pixel. In Supplementary Fig. 4 we show the mean of the 256 reference pixels in each of the four cover-on and four cover-off data sets. These show a decrease with temperature from a value of B545 DN to B542 DN. Assuming a model in which the bias voltage is a steady-state value where the dark current is negligible, we also plot both equation (3) and a free-amplitude t of the same equation in the gure. Given the instrumental gain of 22 e per DN, the free-amplitude t gives a mean dark current of 7 e s 1 per pixel at 220 K, which is a factor of 2.5 larger than the expectation but within the manufacturers expected device to device variance. Assuming this factor, the expected dark current at 193 K is 0.09 e s 1 per pixel.

The measured reference bias offset is subtracted from the images as part of the data processing pipeline. The value subtracted is the median of the 256 reference pixels32, which is a reasonable estimate. However, on closer investigation we nd that the median of the reference pixels can be biased due to the presence of large outlying pixels from, for example, cosmic ray hits. As a result, we measure a statistically signicant correlation between the reference row median and the mean of the processed images. To correct for this bias, we instead use the s-clipped mean of the reference row for reference subtraction. The mean is estimated by rejecting reference pixels with values 43s after two iterations of rejection. Because in these data the median reference row value has already been removed, we correct the

NATURE COMMUNICATIONS | 8:15003 | DOI: 10.1038/ncomms15003 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 5

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15003

mean value of the images by rst adding back the reference row median and then subtracting the s-clipped mean value. The correction is small, typically o0.1 DN per s. This procedure effectively removes the correlation between the subtracted reference value and the mean value of the processed image.

Photometric calibration. We calibrate the images from DN per s to Jy per pixel using aperture photometry. For each eld, we identify two stars with ux o1,000

DN per s to avoid saturation effects, and greater than four pixels away from other sources or array artifacts. The pixel values are summed across a six pixel-wide aperture, and the background in a three pixel-wide ring three pixels away from the inner aperture excluding masked pixels is calculated and subtracted. The background-corrected aperture sum is then divided by the exposure time, giving the source ux S in DN per s. Synthetic photometry is used to determine the magnitudes in the LORRI band using the USNO-B1 catalog. We calculate that the reference magnitude at RL,0 given by the equation:

RL 2:5logS RL;0 4

is 18.520.08. The calibration factors are summarized in Supplementary Fig. 5.

The conversion from ux to surface brightness relies on both the frequency of the measurement n0 and the measured solid angle of the PSF Ob through lIl nFn/Ob, where Ob is the instruments 2D image-space impulse response

function integrated over both dimensions. We estimate the effective frequency using the measured LORRI passband and assuming a at input spectrum, which yields n0 458 THz. Ob is calculated by summing the PSF shown in Supplementary

Fig. 2 over the full 10.8 10.8 image. We nd Ob2:640:18 0:16pixel2, in good

agreement with the FWHM 1.5pix Gaussian model prediction of Ob2:54pixel2.

We estimate the nal surface brightness calibration factor to be 1189 mJy/(DN s 1), which corresponds to 50.93.7 nW m 2 sr 1 per DN per s. Following multiplication by this factor, the images are calibrated in surface brightness units and have associated masks that can be used to exclude pixels containing Ro17.7 point sources. The unmasked pixels in these images can be used to estimate the diffuse sky brightness. The raw diffuse sky brightness measurements corresponding to lIdiffusellIIPDl lIRSl lIDGLl ElICOBl are shown in Supplementary Fig. 6. To

isolate the residual component of the observed emission associated with the COB, it is necessary to account for more local sources of emission, including residual interplanetary dust, residual starlight and diffuse galactic light. The contribution from each component is summarized in Supplementary Table 2.

Interplanetary dust. The population of B11,000 mm dust particles in the Solar system reects light from the sun and sources a diffuse sky brightness. Early in situ measurements of the dust distribution in the inner solar system from Helios, Galileo, Ulysses and Pioneers 8/9 show a sharp drop-off in the IPD density beyond 1 a.u., and connement of the dust particles within 30 of the ecliptic plane39. It is difcult to formulate a mechanism that produces a long-lived population of dust out of the ecliptic plane40, so the bulk of the IPD material is thought to reside at low inclination angles with respect to this plane and models of the dust distribution support this30. Interestingly, Ulysses measurements far above the ecliptic found a continuum level of particle events associated with the planar inow of interstellar dust from the local hot bubble39. These dust particles are very small, with characteristic radii 110 nm, so do not effectively reect sunlight at optical wavelengths. As a result, there is no expectation that IPD light is sourced far from the plane of the ecliptic.

In the outer Solar system, few in situ dust measurements exist. Pioneers 10 and 11 carried detectors that measured the ux of 510 mm particles41. Pioneer 10 reported data to 18 a.u. (ref. 42). Pioneer 11 made continuous measurements to

B9 a.u. and crossed the 3.75 a.u. region three times (once outbound and twice while transiting from Jupiter to Saturn), nding consistent results each time43. More recently, the Student Dust Counter (SDC) on New Horizons has measured the ux of 0.55 mm dust grains from 5 to 30 a.u. (refs 30,44,45). These measurements suggest an order of magnitude drop of the dust ux from 1 to 5 a.u., followed by a attening of the particle ux to at least 20 a.u. Recently, a model has been generated that is consistent with all of the in situ measurements30; the predictions of this model scaled to a quantity that should follow the IPD light intensity are shown in Supplementary Fig. 7. From these in situ measurements, we would infer that the IPD population in the region over which the observations are performed is small and decreasing with distance.

In addition to in situ measurements, Pioneer 10 and 11 observed the optical-wavelength intensity of the background through the Solar system with a two-band imaging photopolarimeter46. The Pioneer 10 measurements from 2.4 to 3.2 a.u. exhibit a factor of 425 decrease in the surface brightness of two survey regions19, both measured at y 4102. The measurements beyond 3.25 a.u. are individually

consistent with zero; averaging over these four measurements gives a 2s upper limit on the IPD brightness of o4.9 nW m 2 sr 1, using the known conversion47 between S10(V) and nW m 2 sr 1. The Pioneer 10 measurements are shown in

Supplementary Fig. 7, and the upper limit on lIIPDl is listed in Supplementary

Table 2. No published analyses of IPD light in the Pioneer 11 data are available.

In the full set of 255 images, the LORRI data show no surface brightness change with heliocentric distance consistent with a variable contribution from the IPD, nor with viewing angle through the plane of the ecliptic.

Residual starlight. There are two contributions to lIRSl in these data that from the unmasked wings of the PSF, and that from sources below the masking threshold.

To calculate the residual starlight from the unmasked wings, we use the USNO-B1 catalog and measured LORRI PSF to simulate each ight image. These images are then masked with the ight mask, and the mean of the unmasked pixels is computed. We estimate that the residual starlight is negligible compared to the mean sky brightness in these images (see Supplementary Table 2).

The residual starlight from stars with RL417.7 is challenging to calculate from real catalogs as they do not approach the required depth of RB25. As a result, we use the TRILEGAL model to estimate the faint star ux, which models star elds as a function of position on the sky, photometric system, assumed stellar IMF, binary fraction, the suns position and various parameters describing the Milky Ways thin disc, thick disc, halo and bulge48. The model returns catalogs of stars consistent with the observed number counts and known populations of stars. The number counts are returned to high precision, and TRILEGAL performs particularly well away from the galactic plane where the galaxy model is relatively simple. For each elds position, we generate ten TRILEGAL simulations of a 0.3 0.3 eld

corresponding to the LORRI image, complete to R 32. For each simulation, we

compute the mean surface brightness of the corresponding image. This results in the lIRSl from faint sources listed in Supplementary Table 2. Even at the relatively faint masking threshold we apply, the residual ux from faint stars contributes a surface brightness comparable to the expected COB.

Diffuse galactic light. At optical wavelengths, dust in the galaxy reects the local interstellar radiation eld, and may also luminesce49. Similarly to the ecliptic dependence of light from the IPD, the DGL is brightest in the galactic plane and relatively faint at high galactic latitudes. Early Pioneer 10 measurements50 found a factor 410 reduction between the DGL measured on the galactic plane and at the poles, and suggest a brightness of B150 and B10 nW m 2 sr 1, respectively. The implication is that nowhere on the sky can we ignore the contribution from the

DGL. Since, on small scales, the spatial variation of the DGL5 is fractions of a nW m 2 sr 1 in these LORRI data the primary effect is that of an overall surface brightness in the images. Due to LORRIs broad optical passband and the limited number of observed elds, with the LORRI data alone the DGL would be impossible to disentangle from the COB.

The dust grains responsible for the DGL are also heated by the interstellar radiation eld (ISRF) and emit this energy thermally in the far-IR. As a result, the DGL is highly correlated with 100 mm emission in the optically thin limit where the optical photon scattering is simple49. Here we take advantage of this correlation and the excellent 100 mm all-sky surface brightness maps available51,52 to estimate the contribution of the DGL to the optical surface brightness in each of the four elds. We have restricted our attention to high galactic latitude elds with I100mmo10 MJy sr 1 in order to avoid optically thick dust as part of the data cut process. This allows us to take advantage of the linear relationship between thermal emission intensity and optical surface brightness.

We estimate the absolute surface brightness of the DGL in each eld via the following relation:

lIDGLll; ; bn In100 mm

h i cl db; 5 where nhIn(100 mm)i is the 100 mm surface brightness averaged over the eld, cl is

the conversion from thermal emission intensity to optical surface brightness formulated below, and d(b) is a function that accounts for the change in cl due to scattering effects as a function of galactic latitude.

To estimate nhIn(100 mm)i we compute the Improved Reprocessing of the IRAS

Survey (IRIS) 100 mm image52 for each pixel in the LORRI images. We then subtract 0.8 MJy sr 1 to account for the CIB brightness at 100 mm (refs 53,54), yielding the brightness of the dust in each eld.

There are a variety of measurements of the scaling between the surface brightness in the optical/near-IR and at 100 mm, cllIlopt=nIn100 mm (which

is sometimes quoted as blInopt=In100 mm). We estimate bl by tting the

mean ZDA04 model55 to a compilation of measurements of bl (ref. 49), which yields a best-tting cl and its uncertainty through multiplication by a factor of 10 6 100 mm=0:655 mm. We then compute the wavelength-averaged value cl by

integrating cl weighted by the LORRI bandpass. This gives c655nm0:49 0:13.

Finally, we estimate d(b) using the relation:

dbd0 1 1:1g

sin b

j j

p

6 where d0 is a normalizing factor and the HenyeyGreenstein parameter g is the asymmetry factor of the scattering phase function56. To estimate g, we compute the bandpass-weighted mean of an observation-constrained model for the high-latitude diffuse dust component of the DGL57, and take the allowed variation in the mean from measurement and modelling errors as its uncertainty, yieldingg 0.610.10. To determine d0, we normalize d(b) at b 25 to be consistent with

previous measurements58. As a check, we investigate the effect of using a larger estimate for g consistent with scattering from dense molecular clouds below.

From this set of information, lIDGLl can be computed. The value of lIDGLl in each eld is listed in Supplementary Table 2. Due to the relatively high galactic latitude and small size of the LORRI elds, and the large effective smoothing in the IRIS maps, we nd it is unnecessary to account for the spatial variation in DGL in these elds.

6 NATURE COMMUNICATIONS | 8:15003 | DOI: 10.1038/ncomms15003 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15003 ARTICLE

As a check of the relatively uncertain direct DGL subtraction, we also compute a linear t of the diffuse sky brightness minus the residual starlight against the100 mm surface brightness in each of the four elds via:

lIdiffusel lIRSl c0l n In100 mm

h i lIresidl 7

where c0l and lIresidl are the parameters of the t. We nd a best-tting

c0l0:40 0:27 and COB results consistent with the values determined via the

direct subtraction method, but with much larger uncertainties since, in this tting method, errors in the other DGL model parameters are folded into the measurement. When systematic errors are included, the estimates from both methods are similar in both absolute value and uncertainty, as expected.

Extinction correction. After accounting for the astrophysical foregrounds in the LORRI images, we retrieve the residual sky brightness measurements shown in Supplementary Fig. 8 and listed as lIresidl in Supplementary Table 2. The per-eld measurements are computed as the weighted mean of the individual exposures, where the weights are the inverse error on the mean in each 10 s exposure calculated from the variance of unmasked pixels. We take the uncertainty on the mean to be the the s.d. of the individual exposures in each eld.

To propagate these eld averages to a measurement of the COB, it is necessary to correct for the effect of galactic extinction. One of the eld averages is negative, for which an extinction correction is unphysical, so we choose to compute the mean residual brightness and then apply an equivalently generated extinction correction to that quantity. We rst compute the uncertainty-weighted mean of the elds, and nd lIresidl4:5 6:9 nW m 2 sr 1.

Next, we estimate the extinction correction by computing the mean of the two AR predicted by two models51,59 in each eld. We then compute the mean of the four eld extinction measurements weighted by the same uncertainty weights used in the mean intensity computation. This yields an extinction correction ofAR 0.11 mag. However, galactic extinction comprises two components, namely

scattering and absorption. For point source observations, both of these remove light out of the line of sight, but since the COB is isotropic, the scattered light is replaced by light from other lines of sight in a conservative fashion. The proportion of these effects is roughly 40% absorption and 60% scattering, so our actual extinction correction is AR 0.05 mag. Applying this to lIresidl, we nd

lICOBl4:7 7:3 nW m 2 sr 1, where the errors are purely statistical. Extinction

corrections do not apply to the systematic errors as they are all due to local mechanisms.

Systematic and foreground error estimation. The errors in this measurement can broadly be categorized as: statistical; systematics in the instrument, where we include the calibration uncertainty in this category; and systematics in the astrophysical foreground accountancy. The uncertainties in the measurement are summarized in Supplementary Table 3 and Supplementary Fig. 3, and their derivation is described in detail in this section. Mean values are always computed using the same eld-to-eld statistical weights as used in the average COB brightness calculation.

The statistical uncertainties are computed from the variance of a number of independent measurements, and as a result fold in a variety of sources of noise (detector, read out, photon, etc.) in an indistinguishable way. On the basis of a cursory inspection of the data and the known photocurrent, the noise is dominated by bit noise from the analog to digital converter, which suggests that in future measurements increased integration times would be benecial.

As there are several steps in the data reduction, there are a corresponding number of potential errors in the instrumental corrections and data analysis we apply. First, the reference frame subtraction would have an error associated with it. However, because we later subtract the reference pixel values, we are removing a large part of the frame to frame variation that would lead to some offset error in this step. As a result, we fold all of the uncertainty associated with reference value/ dark current into that steps error, as described below. Second, the application of a inter-pixel gain correction may have some intrinsic error, but because we perform photometric calibration after the at eld correction and we are not interested in the spatial structure of the images, errors in the applied pixel-to-pixel response should have a negligible effect on the nal result. The only situation in which such an error could have a measurable impact is if the regions immediately surrounding the calibration stars had some local inter-pixel response different from the bulk of the detector array, and different than that measured during laboratory testing. To guard against this, we used calibration stars in random positions on the detector array and in multiple elds, and have shown the calibration to be consistent through the observation period. The photometric calibration uncertainty captures the remaining uncertainty from this effect.

As they are electrically identical to the photo-responsive pixels, and share the same read out chain, we have no expectation of or evidence for the reference pixel subtraction leading to any misestimation of the overall array offset. Extra variance in the nal image brightness from the intrinsic measurement error of the 256 reference pixels that varies frame to frame is naturally accounted in the statistical uncertainty estimate. However, as the reference pixels are shaded from incident photons by a metal slat, it is possible that there is a o20% reduction in the dark current due to electromagnetic coupling to the shade (A. Reinheimer, private communication). To estimate the uncertainty from this effect, we compute 20% of

the expected dark current in the pixels, which would cause a spurious image offset of 0.9 nW m 2 sr 1. Because this would be in the direction of an under-subtraction, this error is in the positive-going direction and should be applied uniformly through the observations.

Optical ghosts were identied to be present in the central ro50 pixel region of the LORRI images during laboratory testing27. The position and brightness of these ghosts depend on bright stars slightly off the imaged eld, and they may be fainter than can be easily detected in the images. Though we masked these ghosts from our science images, ghosts fainter than the surface brightness limit to which we masked may be present. As a check of our masking procedure, we mask the full ro50 region from the science images and recompute the resulting COB brightness through the entire analysis pipeline. When this augmented mask is applied to the images, we nd a modest change to lIinstl of 0.1 nW m 2 sr 1, that is, in the

sense of an increased lIresidl. Unmasked optical ghosts would have the opposite sign, so we conclude there is no evidence for excess surface brightness from optical ghosts in these data.

On the basis of the dispersion of the aperture photometry measurements, we estimate the raw photometric calibration uncertainty of this measurement to be 7.3%. This compares well with the 0.25 mag catalog s.d. quoted as the per source photometric accuracy of the USNO-B1 catalog, which for 8 objects would give a photometric accuracy of 7.3%. USNO-B1 was ultimately calibrated from the Tycho-2 catalog, which itself has been calibrated to an accuracy of 2% (ref. 60). We therefore estimate the absolute photometric uncertainty of this study to be 8% of lIdiffusel, which corresponds to 3.8 nW m 2 sr 1 on lICOBl.

The uncertainty in the solid angle of the beam also plays a role in the ultimate calibration uncertainty of this study. On the basis of half-half PSF stacking jack knife tests, we estimate the solid angle of the LORRI beam is known to 4%. This propagates to a 4% uncertainty on lIdiffusel, which corresponds to1.9 nW m 2 sr 1 on lICOBl. This uncertainty also includes the error inlI l lIRSl due to our imperfect knowledge of Obeam on the conversion from ux to

surface brightness.

The astrophysical foregrounds present in this study all have errors in their estimation. We have argued for a low level of IPD light in the outer Solar system, but explicitly quote the full 1s uncertainty on the R 43.3 a.u. Pioneer 10

measurements as an upper limit on the total IPD light contribution.

To account for the effect of the USNO-B1 photometric calibration uncertainty, we compute the estimate for lIdiffusel after randomizing the reported magnitude of each source from the catalog by 0.25 mag using a random Gaussian deviation per source. This has the effect of modifying the mask radius to be either inappropriately small or large, depending on the sign of the randomization. Over many masked sources, this should adequately probe the photometricerror from the catalog calibration. On the basis of this calculation, we estimate the error to be 0.1 nW m 2 sr 1 on lIRSl, resulting in an error on the COB of

0.1 nW m 2 sr 1.

To calculate the variation in lIRSl due to sample variance of the faint stars in our elds, we calculate the variance of ten realizations of the TRILEGAL model for each eld, complete to R 32. As these are relatively high galactic latitude elds, we nd

the s.d. in lIRSl is 0.6 nW m 2 sr 1 over the set, corresponding to

0.7 nW m 2 sr 1 on lICOBl.

We compute the error in our estimate for the DGL brightness by propagating the error on the parameters in the various input functions. Specically, we use nIn100 mm0:80 0:25 (refs 53,54), g 0.610.10 (ref. 61) and cl0:49 0:13

(ref. 49) as being consistent with the existing measurements. Propagating these errors through the DGL model gives s lIDGLl

8:7; 8:2

f g nW m 2 sr 1,

resulting in an uncertainty of { 9.1, 8.6} nW m 2 sr 1 on lICOBl. As a check of

the effect of the relatively uncertain value of g, we use a compilation of results based on measurements of dense molecular clouds61 that have a different mean scattering asymmetry factor g 0.750.1 and nd lICOBl3:8 nW m 2 sr 1, well within the

quoted systematic uncertainty.

Finally, the galactic extinction correction is based on models that themselves have uncertainties. To bracket these, we take the allowed difference between the two extinction models51,59 as our best estimate for sA . Over the four science elds, we compute the uncertainty-weighted allowable variation in the extinction correction to be a factor of 0.01 in surface brightness, which corresponds to a negligible error in lICOBl.

Data availability. The data that support the ndings of this study are available from the NASA Planetary Data System at http://pds-smallbodies.astro.umd.edu/data_sb/missions/newhorizons/index.shtml

Web End =http://pds-smallbodies.astro.umd.edu/ http://pds-smallbodies.astro.umd.edu/data_sb/missions/newhorizons/index.shtml

Web End =data_sb/missions/newhorizons/index.shtml .

References

1. Tyson, J. A. In Extragalactic Background Radiation Meeting. (ed. Calzetti, D. et al.) 103133 (1995).

2. Hauser, M. G. & Dwek, E. The cosmic infrared background: measurements and implications. Ann. Rev. Astron. Astrophys. 39, 249307 (2001).

3. Cooray, A. Extragalactic background light measurements and applications.R. Soc. Open Sci. 3, 150555 (2016).4. Cooray, A. et al. Near-infrared background anisotropies from diffuse intrahalo light of galaxies. Nature 490, 514516 (2012).

NATURE COMMUNICATIONS | 8:15003 | DOI: 10.1038/ncomms15003 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 7

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15003

5. Zemcov, M. et al. On the origin of near-infrared extragalactic background light anisotropy. Science 346, 732735 (2014).

6. Gong, Y. et al. Axion Decay and Anisotropy of Near-IR Extragalactic Background Light. Astrophys. J. 825, 104 (2016).

7. Bernstein, R. A. The Optical Extragalactic Background Light: revisions and Further Comments. Astrophys. J. 666, 663673 (2007).

8. Mattila, K., Lehtinen, K., Vaisanen, P., von Appen-Schnur, G. & Leinert, C. In The Spectral Energy Distribution of GalaxiesSED 2011 (eds Tuffs, R. J. & Popescu, C. C.) 429436 (2012).

9. Gorjian, V., Wright, E. L. & Chary, R. R. Tentative Detection of the Cosmic Infrared Background at 2.2 and 3.5 Microns Using Ground-based and Space-based Observations. Astrophys. J. 536, 550560 (2000).

10. Wright, E. L. DIRBE minus 2MASS: conrming the cosmic infrared background at 2.2 microns. Astrophys. J. 553, 538544 (2001).

11. Cambrsy, L., Reach, W. T., Beichman, C. A. & Jarrett, T. H. The cosmic infrared background at 1.25 and 2.2 microns using DIRBE and 2MASS:a contribution not due to galaxies? Astrophys. J. 555, 563571 (2001).12. Wright, E. L. COBE observations of the cosmic infrared background. New Astron. Rev. 48, 465468 (2004).

13. Levenson, L. R., Wright, E. L. & Johnson, B. D. DIRBE minus 2MASS: conrming the CIRB in 40 new regions at 2.2 and 3.5 mm. Astrophys. J. 666, 3444 (2007).

14. Matsumoto, T. et al. Infrared telescope in space observations of the near-infrared extragalactic background light. Astrophys. J. 626, 3143 (2005).

15. Mattila, K. Has the optical extragalactic background light been detected? Astrophys. J. 591, 119124 (2003).

16. Mattila, K. The 1-mm discontinuity in the extragalactic background light spectrum: an artefact of foreground subtraction. Mon. Not. R. Astron. Soc. 372, 12531258 (2006).

17. Cooray, A. et al. A new era in extragalactic background light measurements: the cosmic history of accretion. Nucleosynth. Reioniz. Preprint at http://arxiv.org/abs/0902.2372

Web End =http:// http://arxiv.org/abs/0902.2372

Web End =arxiv.org/abs/0902.2372 (2009).

18. Weinberg, J. L., Hanner, M. S., Beeson, D. E., DeShields, II L. M. & Green, B. A. Background starlight observed from Pioneer 10. J. Geophys. Res. 79, 3665 (1974).

19. Hanner, M. S., Weinberg, J. L., DeShields, II L. M., Green, B. A. & Toller, G. N. Zodiacal light and the asteroid belt: the view from Pioneer 10. J. Geophys. Res. 79, 3671 (1974).

20. Toller, G., Tanabe, H. & Weinberg, J. L. Background starlight at the north and south celestial, ecliptic, and galactic poles. Astron. Astrophys. 188, 2434 (1987).

21. Gordon, K. D., Witt, A. N. & Friedmann, B. C. Detection of extended red emission in the diffuse interstellar medium. Astrophys. J. 498, 522540 (1998).

22. Toller, G. N. The extragalactic background light at 4400 A. Astrophys. J. Lett. 266, L79L82 (1983).

23. Matsuoka, Y., Ienaka, N., Kawara, K. & Oyabu, S. Cosmic Optical Background: The View from Pioneer 10/11. Astrophys. J. 736, 119 (2011).

24. Weaver, H. A., Gibson, W. C., Tapley, M. B., Young, L. A. & Stern, S. A. Overview of the new horizons science payload. Space. Sci. Rev. 140, 7591 (2008).

25. Conard, S. J. et al. in Astrobiology and Planetary Missions (eds Hoover, R. B. et al.) 5906, 407420 (2005).

26. Morgan, F. et al. in Astrobiology and Planetary Missions (eds Hoover, R. B. et al.) 5906, 421432 (2005).

27. Cheng, A. F. et al. Long-Range Reconnaissance Imager on New Horizons. Space. Sci. Rev. 140, 189215 (2008).

28. Bessell, M. S., Castelli, F. & Plez, B. Model atmospheres broad-band colors, bolometric corrections and temperature calibrations for O - M stars. Astron. Astrophys. 333, 231250 (1998).

29. H. E. S. S. Collaboration et al. Measurement of the extragalactic background light imprint on the spectra of the brightest blazars observed with H.E.S.S. Astron. Astrophys. 550, A4 (2013).

30. Poppe, A. R. An improved model for interplanetary dust uxes in the outer Solar System. Icarus 264, 369386 (2016).

31. Cheng, A. F., Conard, S. J., Weaver, H. A., Morgan, F. & Noble, M. In Space Telescopes and Instrumentation 2010: Optical, Infrared, and Millimeter Wave (eds Oschmann, Jr J. M. et al.) 7731, 110 (2010).

32. Space Science and Engineering Division. New Horizons SOC to Instrument Pipeline ICD. Document No. 05310-SOCINST-01 (Southwest Research Institute, 2007).

33. Lang, D., Hogg, D. W., Mierle, K., Blanton, M. & Roweis, S. Astrometry.net: blind astrometric calibration of arbitrary astronomical images. Astrnom. J. 139, 17821800 (2010).

34. Monet, D. G. et al. The USNO-B Catalog. Astrnom. J. 125, 984993 (2003).35. Driver, S. P. et al. Measurements of extragalactic background light from the far UV to the Far IR from deep ground- and space-based galaxy counts. Astrophys. J. 827, 108 (2016).

36. Bock, J. et al. The cosmic infrared background experiment (CIBER): the wide-eld imagers. Astrophys. J. Suppl. 207, 32 (2013).

37. Owen, Jr W. M., Dumont, P. J. & Jackman, C. D. in Astrobiology and Planetary Missions: 23rd International Symposium for Space Flight Dynamics, 421432 (Pasadena, CA, USA, 2012).

38. Janesick, J. R. Scientic Charge-coupled Devices (SPIE Press, 2001).39. Grn, E., Gustafson, B. A. S., Dermott, S. & Fechtig, H. Interplanetary Dust (Springer, 2001).

40. Nesvorn, D. et al. Dynamical model for the zodiacal cloud and sporadic meteors. Astrophys. J. 743, 129 (2011).

41. Humes, D. H. Results of Pioneer 10 and 11 meteoroid experiments -Interplanetary and near-Saturn. J. Geophys. Res. 85, 58415852 (1980).

42. Landgraf, M., Liou, J.-C., Zook, H. A. & Grn, E. Origins of Solar System Dust beyond Jupiter. Astrnom. J. 123, 28572861 (2002).

43. Dikarev, V. & Grn, E. New information recovered from the Pioneer 11 meteoroid experiment data. Astron. Astrophys. 383, 302308 (2002).

44. Poppe, A., James, D., Jacobsmeyer, B. & Hornyi, M. First results from the Venetia Burney Student Dust Counter on the New Horizons mission. Geophysical Research Letters 37, L11101 (2010).

45. Szalay, J. R., Piquette, M. & Hornyi, M. The Student Dust Counter: status report at 23 AU. Earth Planet Space 65, 11451149 (2013).

46. Pellicori, S. F., Russell, E. E. & Watts, L. A. Pioneer imaging photopolarimeter optical system. Appl. Opt. 12, 12461258 (1973).

47. Leinert, C. et al. The 1997 reference of diffuse night sky brightness. Astron. Astrophys. Supplement 127, 199 1998.

48. Girardi, L., Groenewegen, M. A. T., Hatziminaoglou, E. & da Costa, L. Star counts in the Galaxy. Simulating from very deep to very shallow photometric surveys with the TRILEGAL code. Astron. Astrophys. 436, 895915 (2005).

49. Ienaka, N. et al. Diffuse Galactic Light in the Field of the Translucent High Galactic Latitude Cloud MBM32. Astrophys. J. 767, 80 (2013).

50. Toller, G. N. A study of galactic light, extragalactic light, and galactic structure using Pioneer 10 observations of background starlight. (Ph.D. thesis, State Univ. New York, 1981).

51. Schlegel, D. J., Finkbeiner, D. P. & Davis, M. Maps of dust infrared emission for use in estimation of reddening and cosmic microwave background radiation foregrounds. Astrophys. J. 500, 525553 (1998).

52. Miville-Deschnes, M.-A. & Lagache, G. IRIS: A New Generation of IRAS Maps. Astrophys. J. Suppl. 157, 302323 (2005).

53. Puget, J.-L. et al. Tentative detection of a cosmic far-infrared background with COBE. Astron. Astrophys. 308, L5 (1996).

54. Fixsen, D. J., Dwek, E., Mather, J. C., Bennett, C. L. & Shafer, R. A. The spectrum of the extragalactic far-infrared background from the COBE FIRAS observations. Astrophys. J. 508, 123128 (1998).

55. Zubko, V., Dwek, E. & Arendt, R. G. Interstellar dust models consistent with extinction, emission, and abundance constraints. Astrophys. J. Suppl. 152, 211249 (2004).

56. Jura, M. The ultraviolet phase function of interstellar grains. Astrophys. J. 231, 732735 (1979).

57. Draine, B. T. Scattering by interstellar dust grains. I. Optical and ultraviolet. Astrophys. J. 598, 10171025 (2003).

58. Lillie, C. F. & Witt, A. N. Ultraviolet photometry from the orbiting astronomical observatory. XXV - Diffuse galactic light in the 15004200 A region and the scattering properties of interstellar dust grains. Astrophys. J. 208, 6474 (1976).

59. Schlay, E. F. & Finkbeiner, D. P. Measuring reddening with sloan digital sky survey stellar spectra and recalibrating SFD. Astrophys. J. 737, 103 (2011).60. Maz-Apellniz, J. A cross-calibration between tycho-2 photometry and hubble space telescope spectrophotometry. Publ. Astron. Soc. Pac. 117, 615619 (2005).

61. Sano, K. et al. First detection of galactic latitude dedpendence of near-infrared diffuse galactic light from DIRBE renalysis. Astrophys. J. Lett. 821, L11 (2016).

62. Gardner, J. P., Brown, T. M. & Ferguson, H. C. Ultraviolet galaxy counts from space telescope imaging spectrograph observations of the hubble deep elds. Astrophys. J. Lett. 542, L79L82 (2000).

63. Madau, P. & Pozzetti, L. Deep galaxy counts, extragalactic background light and the stellar baryon budget. Mon. Not. R. Astron. Soc. 312, L9L15 (2000).

64. Totani, T., Yoshii, Y., Iwamuro, F., Maihara, T. & Motohara, K. Diffuse extragalactic background light versus deep galaxy counts in the subaru deep eld: missing light in the universe? Astrophys. J. Lett. 550, L137L141 (2001).

65. Keenan, R. C., Barger, A. J., Cowie, L. L. & Wang, W.-H. The resolved near-infrared extragalactic background. Astrophys. J. 723, 4046 (2010).

Acknowledgements

We thank both the New Horizons science team and the LORRI instrument team for their decades of dedicated effort designing, building and ying such a complex mission. Many thanks to H. Weaver and M. Richmond for useful discussions during the course of this work, and to the referees whose depth of knowledge and insights signicantly improved the work. The New Horizons launch, Jupiter y-by, and cruise phase data sets were obtained from the Planetary Data System (PDS). A.R.P. gratefully acknowledges NASA Planetary Atmospheres grant #NNX13AG55G.

8 NATURE COMMUNICATIONS | 8:15003 | DOI: 10.1038/ncomms15003 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15003 ARTICLE

Author contributions

M.Z. developed the analysis and systematic error assessment pipeline, performed the foreground analysis and wrote the rst draft of the paper. P.I. generated the data cuts and determined the photometric calibration of the instrument. C.N. collected the data from the archive and performed a variety of data quality checks. A.C., C.M.L. and A.R.P. worked on various aspects of foreground analysis and provided input on the low level analysis and workings of the instrument. All coauthors provided feedback and comments on the paper.

Additional information

Supplementary Information accompanies this paper at http://www.nature.com/naturecommunications

Web End =http://www.nature.com/ http://www.nature.com/naturecommunications

Web End =naturecommunications

Competing interests: The authors declare no competing nancial interests.

Reprints and permission information is available online at http://npg.nature.com/reprintsandpermissions/

Web End =http://npg.nature.com/ http://npg.nature.com/reprintsandpermissions/

Web End =reprintsandpermissions/

How to cite this article: Zemcov, M. et al. Measurement of the cosmic optical background using the long range reconnaissance imager on New Horizons. Nat. Commun. 8, 15003 doi: 10.1038/ncomms15003 (2017).

Publishers note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional afliations.

This work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the articles Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/

Web End =http://creativecommons.org/licenses/by/4.0/

r The Author(s) 2017

NATURE COMMUNICATIONS | 8:15003 | DOI: 10.1038/ncomms15003 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 9