DRAFT VERSION FEBRUARY 11, 2021Typeset using LATEX twocolumn style in AASTeX63

Evaluation of New Submillimeter VLBI Sites for the Event Horizon Telescope

ALEXANDER W. RAYMOND,1 DANIEL PALUMBO,1, 2 SCOTT N. PAINE,1 LINDY BLACKBURN,1, 2 RODRIGO CÓRDOVA ROSADO,1, 3

SHEPERD S. DOELEMAN,1, 2 JOSEPH R. FARAH,1, 4 MICHAEL D. JOHNSON,1, 2 FREEK ROELOFS,5 REMO P.J. TILANUS,6 ANDJONATHAN WEINTROUB1, 2

1Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA2Black Hole Initiative, Harvard University, 20 Garden Street, Cambridge, MA 02138, USA

3Department of Astrophysical Sciences, Princeton University, 4 Ivy Lane, Princeton, NJ 08544, USA4University of Massachusetts Boston, 100 William T, Morrissey Blvd, Boston, MA 02125, USA

5Department of Astrophysics, Institute for Mathematics, Astrophysics and Particle Physics (IMAPP), Radboud University, P.O. Box 9010, 6500 GL Nijmegen,The Netherlands

6Department of Astronomy and Steward Observatory, University of Arizona, 933 N. Cherry Avenue, Tucson, AZ 85721, USA

ABSTRACT

The Event Horizon Telescope (EHT) is a very long baseline interferometer built to image supermassive blackholes on event-horizon scales. In this paper, we investigate candidate sites for an expanded EHT array withimproved imaging capabilities. We use historical meteorology and radiative transfer analysis to evaluate siteperformance. Most of the existing sites in the EHT array have median zenith opacity less than 0.2 at 230 GHzduring the March/April observing season. Seven of the existing EHT sites have 345 GHz opacity less than 0.5during observing months. Out of more than forty candidate new locations analyzed, approximately half have230 GHz opacity comparable to the existing EHT sites, and at least seventeen of the candidate sites would becomparably good for 345 GHz observing. A group of new sites with favorable transmittance and geographicplacement leads to greatly enhanced imaging and science on horizon scales.

1. INTRODUCTION

The Event Horizon Telescope (EHT) is a verylong-baseline interferometry (VLBI) array operating at230 GHz (Doeleman et al. 2009; Event Horizon TelescopeCollaboration et al. 2019a). In 2017, eight telescopes atsix sites participated in the EHT observing campaign thatyielded the first horizon-scale images of the supermassiveblack hole in the M87 galaxy (or M87∗; Event HorizonTelescope Collaboration et al. (2019b)1). The planned 2021EHT array includes 10 telescopes at nine sites plus the theAtacama Large Millimeter/submillimeter Array (ALMA):the Atacama Pathfinder Experiment (APEX), the GreenlandTelescope (GLT), the Instituto de Radioastronomía Mil-imétrica 30 m telescope (IRAM-30 m), the James ClerkMaxwell Telescope (JCMT), the 12 m radio telescope at KittPeak (KP) operated by the University of Arizona, the LargeMillimeter Telescope Alfonso Serrano (LMT), the NorthernExtended Millimeter Array (NOEMA), the SubmillimeterArray (SMA), the Submillimeter Telescope (SMT), and theSouth Pole Telescope (SPT). The coordinates for these sta-tions are listed in Table 1. Generally, these stations were

Corresponding author: Alexander W. Raymond1 This black hole is also named Powehi by A Hui He Inoa (Kimura et al.

2019)

developed as standalone submillimeter facilities and performnon-VLBI observations throughout most of the year.

There is an active effort (Blackburn et al. 2019) to expandthe array to new sites and higher frequency (e.g., 345 GHz),which will improve imaging and modeling of M87∗ and theother primary science target, Sgr A∗, the black hole candi-date at the center of the Milky Way (Doeleman et al. 2008).These improvements will lead to a new instrument, the next-generation EHT (ngEHT), that is designed to deliver trans-formative science capabilities. In this paper, we identify a setof candidate ngEHT sites, characterize their meteorologicalsuitability for 230 and 345 GHz observations, and evaluatethe additional VLBI baseline coverage they provide beyondthe existing EHT array. New analysis for the existing EHTsites is also presented.

There are many factors to consider in the selection of newsites for ngEHT stations. The elevation of the primary sci-ence targets above the horizon, mutual visibility of thosesources with the existing sites in the array, and the incre-mental Fourier coverage contributed by a new site are all keyconsiderations. New Fourier coverage improves the VLBI in-strument in several ways. Dense sampling improves imagingfidelity, and baseline redundancy in dense arrays is a pow-erful tool for calibration (Pearson & Readhead 1984). Asufficiently dense array may even allow imaging to rely oncalibration-independent closure quantities (three- and four-

arX

iv:2

102.

0548

2v1

[as

tro-

ph.I

M]

10

Feb

2021

2

station products) instead of visibilities (two-station products;Chael et al. (2018b); Blackburn et al. (2020)). Finally, an ar-ray producing dense sampling is robust against losing one ortwo stations because of poor weather on a particular day.

The meteorological conditions are a key consideration inthe evaluation of a new site. Water is the atmosphericconstituent that primarily governs atmospheric transmissionand brightness temperature at millimeter and submillime-ter wavelengths. We characterize precipitable water vapor(PWV, expressed in millimeter units) and liquid water path(LWP, expressed in microns) in our analysis. Although sub-millimeter site evaluations usually emphasize the importanceof PWV statistics, clouds consisting of liquid water dropletscan also contribute significantly to opacity (Matsushita &Matsuo 2003). Analysis of EHT data indicates that the imag-ing capability of the ngEHT will depend primarily on in-creasing the number of VLBI baselines, and only secondar-ily on sensitivity, so lower altitude sites with variable atmo-spheric conditions will be considered viable. Our interest insites that achieve good Fourier coverage motivates our anal-ysis of meteorological conditions at suboptimal places, anddistinguishes the present study from previous submillimetersite surveys, e.g., Tremblin et al. (2012).

Historically, EHT observations have been scheduled forMarch or April. Our analysis specifically addresses observ-ing conditions during those months, which are the time ofyear when the sources are above the horizon at nighttime (thetime of day when observing conditions are generally best formost sites), and when PWV paths are favorable across theNorthern and Southern hemispheres.

There are relatively few undeveloped locations for newtelescopes in dry places that provide new VLBI baselinesof hundreds to thousands of kilometers, which is the stationspacing required to fill gaps in the existing EHT coverage.To illustrate this point, we divide the globe into 1◦ latitudeby 1◦ longitude squares, and within each square, the PWVand LWP values are calculated for the location of the highestpeak based on the Modern-Era Retrospective analysis for Re-search and Applications, version 2 (MERRA-2; Gelaro et al.(2017)). In Fig. 1, peaks are highlighted between 60◦ lati-tudes that have median PWV and LWP values less than 5 mmand 5 µm, respectively. The locations of the dry peaks agreewith previous PWV maps, e.g., Suen et al. (2015), whichhave shown that the driest places tend to be nonequatorialhigh planes. Maunakea, Parque Nacional Pico de Orizaba,the Atacama Desert, and other places hosting existing sub-millimeter telescopes emerge on the plot. The Himalayas,Alps, Andes, and southwest United States each have manylow-PWV sites. The latter three regions have sustained cov-isibility of the primary science targets with ALMA, which isthe most sensitive submillimeter station. The large signal-to-noise ratio (S/N) achieved on ALMA baselines enables the

Table 1. Locations of Existing Sites (2021) in the Event Horizon Tele-scope Array (Event Horizon Telescope Collaboration et al. 2019a).

Site Location Lat. (◦) Lon. (◦) Alt. (m)

(Region, Country)

ALMA Antofagasta, CL -23.03 -67.75 5070APEX Antofagasta, CL -23.01 -67.76 5100GLTa Avannaata, GL 76.54 -68.69 90

IRAM-30m Granada, ES 37.07 -3.39 2920JCMT Hawaii, US 19.82 -155.48 4120

KP Arizona, US 31.96 -111.61 1900LMT Puebla, MX 18.98 -97.31 4600

NOEMA Pr.-Alpes-Côted’Azur, FR 44.63 5.91 2620

SMA Hawaii, US 19.82 -155.48 4110SMT Arizona, US 32.70 -109.89 3160

SPTb South Pole, -90.00 45.00 2820Antarctica

asite cannot observe Sgr A∗

bsite cannot observe M87∗

atmospheric phase correction that improves and calibratesdetections on non-ALMA baselines (Event Horizon Tele-scope Collaboration et al. 2019c). Selecting sites that havecovisibility with the existing EHT stations also capitalizeson the N (N − 1)/2 growth in the number of baselines, whereN is the number of stations that can simultaneously observea given source.

Using the PWV map as a guide together with knowledgeabout existing infrastructure and previous astronomical siteevaluations, e.g., Suen et al. (2014), we identify 45 sites tostudy in detail. Those sites are plotted in Fig. 2 and theirlongitude, latitude, and altitude coordinates are listed in Ta-ble 2. These sites cover a broad sampling of longitudes andlatitudes, regions, and climates. Our main objective in thispaper is to catalog potential sites based primarily on weatherand Fourier coverage. The analysis we present is an initialstep toward array expansion. Many of the places we analyzealready host an observatory or existing roads and infrastruc-ture; however, some of the sites we present are remote andwould require responsible assessments of the environmental

3

Figure 1. Worldwide March PWV for peaks above 2000 m in each 1◦ latitude/longitude square between 60◦ S and 60◦ N latitude. MERRA-2datasets for years 2009 through 2012 were averaged. A cutoff of 5 mm is applied.

and cultural impact of new construction to determine if de-velopment is viable 2.

We perform a radiative transfer analysis of the sites usinghistorical meteorological statistics. From that analysis, wepresent transmittances at 230 GHz and higher frequenciesup to 500 GHz. Observations at 345 GHz are of particularinterest as a way to improve angular resolution and reducescattering by the interstellar medium on the line of sight toSgr A∗. In general, 345 GHz observations demand sites withbetter weather than sites that are suitable for 230 GHz obser-vations. Our calculations show which sites, new and existing,are feasible for high-frequency observing.

We evaluate the Fourier coverage contributed by the can-didate stations in a few ways. First, we present the geometricFourier coverage contributed by each station while payingparticular attention to the baselines with ALMA. Second, weuse the meteorological statistics for each site to calculate theprobabilistic Fourier coverage from a set of Monte Carlo ob-serving trials. Finally, we perform test imaging reconstruc-tions of M87∗ and Sgr A∗ simulations to illustrate how newstations would enhance the imaging performance of the EHT.We find that a collection of sites chosen for their favorableprobabilistic Fourier coverage and good conditions is able tosignificantly improve the fidelity of the reconstructions.

2. METHODS

We use atmospheric state data from MERRA-2 as inputsto our calculations. Each site is analyzed using a 10 year his-torical dataset ending on 2019 January 1. The datasets con-sist of 3 hr averages, gridded to 0.5◦ latitude and 0.625◦ lon-

2 The FUJI, KEN, and KILI sites we analyze in this study are UNESCOWorld Heritage Sites. In each case the site has either been previously con-sidered or used for astronomical purposes (Sekimoto et al. 2000; Grahamet al. 2016). At the KILI site there are local efforts to explore the locationfor possible astronomical research (private communication, Dr. Noorali T.Jiwaji). We present these sites to assess their meteorological viability butacknowledge that considerations beyond just the scientific suitability willdetermine the future possibilities for ngEHT telescopes at these locations.

gitude and 42 vertical pressure levels. Each time, position,and altitude datum contains the state of the atmosphere in-cluding water quantities (vapor, liquid, and ice) and temper-ature. The gridded data are interpolated to the precise lo-cation of a candidate site. We calculate water statistics aswell as submillimeter radiative properties with the aid of theam atmospheric modeling and radiative transfer code (Paine2019). For each site, we compute 10 year monthly statisticsof column-integrated water vapor and liquid clouds, and weuse those to calculate the spectral opacity τ (ν) and bright-ness temperature Tb as a function of frequency ν followingthe methodology described in Paine (2018, 2019). The liquidand ice opacity calculations utilize the Rayleigh approxima-tion.

Submillimeter wavelength observations are affected bymeteorology in several ways, and we consider various factorsin our site evaluation. First, PWV statistics are frequentlyused as proxies for submillimeter observing conditions, andwe present monthly PWV statistics for both existing EHTand candidate ngEHT sites. The LWP statistic is another im-portant metric we consider. The LWP contribution to zenithopacity will be approximately 2.5×10−3 and 3.5×10−3 µm–1

at 230 and 345 GHz, respectively, with weak temperature de-pendence (Paine 2019). In other words, 100 µm of liquidwater path contributes 0.25 to the opacity at 230 GHz. Fig-ure 3 shows the 230 GHz effective system temperature (de-fined later in this section) at a typical lower bound on tele-scope elevation angle during VLBI observations; at 100 µmLWP column, liquid clouds become the dominant source ofsystem noise. Consequently, we include the effects of all wa-ter phases in our radiative transfer analysis.

An additional metric that is important in interferometry isthe coherence time of the atmosphere. Turbulence cells inthe moist atmosphere cause phase fluctuations in the inter-ferometric response of a VLBI baseline resulting in signalloss for extended coherent integrations. Here, we define thecoherence time to be the integration interval beyond which

4

NZ

Dome C

KVNYS FUJINOB

NOR

APEXALMA

GLT

IRAM-30m

JCMTLMT

NOEMA

SMA

SPT

BOL

ERB

GAMPAR

HAY

KENKILI

LAS

SPX

YAN

BGK

CNI

Dome ADome F

SUF

LLA

HAN

GLT-S

BAN

BRZ

SGO

BGA

CAT

CAS

FAIR

YBG

GARS

BMACKPSMT

HOP

LOS

BAROVRO

BLDR

BAJA

SOC

PIKE

ORG

SAN

JELM

Figure 2. Map of planned 2021 EHT sites (blue, bold) listed in Table 1 and 45 potential new sites (red) for the ngEHT listed in Table 2. Themap is centered on the 2021 EHT sites.

the VLBI signal loss rises above 10%. During the 2017EHT observations, the coherence time was typically in therange of 10–20 seconds on ALMA baselines (Event Hori-zon Telescope Collaboration et al. 2019a). Strong S/N detec-tions to ALMA are used to correct the phase data and enablelonger coherent integration time. The spatiotemporal gridfor MERRA-2 is not suitably fine for calculating coherencetimes, so we do not consider that effect in this analysis; how-ever, atmospheres with small PWV and LWP characteristicswill generally have small wet path delays and consequentlybetter coherence times (Thompson et al. 2017).

We calculate the incremental Fourier coverage, or Fourierfilling, of the M87∗ and Sgr A∗ sources for new sites com-pared to a fiducial array, i.e., the 2021 EHT array. As theEarth rotates during an observation, each baseline samplesa set of Fourier components that traces tracks in the u-v(east–west and north–south) Fourier plane. The incremen-tal Fourier coverage is the fractional increase in the u-v areasampled by an additional station compared to the fiducial ar-ray. The width of each each track is given by 0.71 dividedby the field of view (FoV) adopted for imaging, which corre-sponds to the half-width of the visibility response to a filleddisk occupying the FoV on the sky (Palumbo et al. 2019).If a new site primarily samples Fourier space that is alreadycovered by existing baselines, then its incremental coverageis small. If a new site samples spatial frequencies that arenot covered by the existing array, then its incremental cov-erage will be large. There are plans to position an antennaat the Greenland Summit, which has excellent atmosphericconditions (Matsushita et al. 2017). We assume that the GLT

200 225 250 275 300frequency (GHz)

0

500

1000

1500T* sy

s (K)

added 100 m of LWPadded 10 m of LWPreceiver+PWV thermal noise

325 350 375

clear10 m

100 m

clear

10 m

100 m

Figure 3. Effective system temperature along a line of sight at 20◦

elevation for the ALMA/APEX site. The indicated liquid water pathis added to the meteorological conditions of the April clear-sky days(i.e., to the days with less than 10 µm LWP) at the 450 mbar pres-sure level. The liquid component can double the effective systemtemperatures at 230 or 345 GHz.

site remains in the fiducial array and evaluate the incrementaladdition of a GLT-S station.

To incorporate the effects of weather in the Fourier fill-ing metric, we perform 45 repeated Monte Carlo trialsimulations of the weather-dependent observations usingeht-imaging (Chael et al. 2016, 2018b) based on the sta-tistical distribution of opacity. In each trial, we flag low S/Nportions of the Fourier coverage (S/N < 3, in accordance withthe long baseline scans reported in Event Horizon TelescopeCollaboration et al. (2019c)), and report the average cover-

5

Table 2. Next-generation Event Horizon Telescope Candidate Sites.

Site Location Lat. (◦) Lon. (◦) Alt. (m)

(Region, Country)

BAJA Baja California, MX 31 -115 2500BANa Alberta, CA 53 -118 2000BAR California, US 38 -118 3500BGA Sofia, BG 42 24 2500

BGKa Westfjords, IS 66 -23 500BLDR Colorado, US 40 -105 2500BMAC Eastern Cape, ZA -31 28 2500BOL La Paz, BO -16 -68 5000BRZ Espírito Santo, BR -20 -42 2500CAS Tierra del Fuego, AR -55 -68 500CAT Río Negro, AR -41 -71 2000CNI La Palma, ES-CN 29 -18 2000

Dome Ab Antarctica -80 77 4000

Dome Cb Antarctica -75 123 3000

Dome Fb Antarctica -78 39 3500ERB Kurdistan, IQ 37 44 2000

FAIRa Alaska, US 65 -145 1000FUJI Fujinomiya &

Yamanashi, JP 35 139 3500GAM Khomas, NA -23 16 2000GARS Antarctica -63 -58 0

GLT-Sa Northeastern, GL 73 -38 3000HAN Jammu & Kashmir, IN 33 79 4000HAY Massachusetts, US 43 -71 0HOP Arizona, US 32 -111 2000JELM Wyoming, US 41 -106 2500KEN Meru, KE -0 37 4000KILI Kilimanjaro, TZ -3 37 4500

KVNYS Seoul, KR 38 127 0LAS Coquimbo, CL -29 -71 2000LLA Salta, AR -24 -66 4500LOS New Mexico, US 36 -106 2000NOB Nagano, JP 36 138 1000NOR Gifu, JP 36 138 2500NZ Canterbury, NZ -44 171 2000

ORG Oregon, US 42 -118 2000OVRO California, US 37 -118 1000PAR Antofagasta, CL -25 -70 2500PIKE Colorado, US 39 -105 4000SAN California, US 34 -117 2500SGO Santiago, CL -33 -70 3500SOC New Mexico, US 34 -108 2000SPX Bern, CH 47 8 3500SUF Jizzakh, UZ 40 68 2000YAN Huanca Sancos, PE -14 -75 4500YBG Yangbajing Tibet, CN 30 91 4000

asite cannot observe Sgr A∗

bsite cannot observe M87∗

age across all trials. We make some simplifying assumptionslike neglecting losses resulting from the beam pointing offsource or mechanical defocusing. Those effects could be in-corporated in the future using synthetic data tools developedfor the EHT (Roelofs et al. 2020). The M87∗ source modelwe use in the simulations has 0.6 Jy flux density at 230 GHzon the spatial scale of the event horizon. A scattering ker-nel is applied to the Sgr A∗ source model to simulate theeffects of the interstellar medium (Johnson et al. 2018), anda short baseline 230 GHz flux density of 2.3 Jy is used forthat source.

The simulated S/N values for each synthetic observationare governed by the system-equivalent flux density for eachsite (SEFDi),

SEFDi = 2kT ∗sys,i/Aeff,i, (1)

where Aeff,i is the effective antenna area and k is the Boltz-mann constant. For the 2021 array as well as the new sitesHAY, KVNYS, LAS, NOB, SUF, and GAM, we base thecollecting area on the antenna diameter of the existing orplanned telescopes, which are 37 (Rogers et al. 1993; Usoffet al. 2014), 21 (Lee et al. 2011), 15 (Booth et al. 1989),45 (Ishiguro et al. 1986), 70, and 15 m, respectively. Forall other sites, we assume a 10 m diameter antenna. Knownaperture efficiencies are used, and for the new 10 m sta-tions we assume a Ruze-formula (Thompson et al. 2017)aperture efficiency based on a 64 µm rms surface accu-racy, which is the quadrature sum of a 40 µm primary sur-face accuracy and an effective 50 µm focus offset. TheSEFDi values depend on the effective system temperature,T ∗

sys,i =[TR + Tb,i (θ)

]eτi(θ) (Thompson et al. 2017), and the

opacity for a given site. τi (θ) is the opacity toward the sourceas a function of elevation, θ, where a lower limit of 10◦ in el-evation is applied.

We adopt a receiver temperature, TR, of 60 and 100 K forall sites at 230 and 345 GHz, respectively, and use the in-terquartile ranges for sky-brightness temperature (Tb,i) andopacity (τi) to generate a Gaussian random variable for T ∗

sys,i.The random variables at the ALMA/APEX, JCMT/SMA,and KP/SMT EHT sites are assumed to be correlated whilethose for candidate ngEHT sites are assumed to be uncorre-lated. The S/N of the visibility measured on a single baselineis (Thompson et al. 2017)

S/N = ηSν (~u)

√2∆ντc

SEFD1SEFD2, (2)

where η is the digital and processing efficiency, Sν (~u) is thecorrelated flux density on baseline ~u, ∆ν is the bandwidth, τc

is the integration time. We specify integration times of 300 s(M87∗) and 100 s (Sgr A∗, shorter than the smallest expectedperiod of the innermost stable circular orbit). The bandwidthspecified for S/N calculations is 2 (as for the current EHT)or 8 GHz (ngEHT), and corresponds to the Nyquist band

6

of a single digitizer (Event Horizon Telescope Collabora-tion et al. 2019a). We call this the fringe-finding bandwidth.Multiple fringe-finding sidebands are aggregated for imag-ing: two of 2 GHz each in 2017, four of 2 GHz each in 2021,and two of 8 GHz each for the ngEHT. Those correspondto a total bandwidth across two polarizations of 8, 16, and64 GHz in the respective years (assuming simultaneous dual-frequency 230+345 GHz observing with the ngEHT). Thedual-frequency approach increases the Fourier coverage, andwith the appropriate model, the two frequencies can be com-bined to synthesize a single image (Sault & Wieringa 1994).We account for the slightly different center frequencies ofeach sideband when calculating the Fourier filling.

Finally, we perform imaging reconstructions of M87∗ andSgr A∗ using a possible ngEHT array chosen by pickingfrom among the best sites from the Fourier filling metric.The reconstruction is meant as an example of how new siteswill affect the imaging capabilities of the EHT instrument.The source models for those reconstructions are general-relativistic magnetohydrodynamic (GRMHD) simulations ofaccretion flows (Rowan et al. 2017; Chael et al. 2018a, 2019).We calculate the noise for each baseline in the same man-ner described above under median opacity conditions. ForM87∗, we generate synthetic data with a 40% duty cycle,and for Sgr A∗, we use a brief 100 s snapshot observationthat begins at a time of day when the Fourier filing is great-est, which can be different for different arrays. We applyan imaging process similar to the published eht-imagingscript released alongside Event Horizon Telescope Collabo-ration et al. (2019d).

3. RESULTS AND DISCUSSION

3.1. Weather and Observing Statistics

Figure 4 compares the 225 GHz opacity calculated withthe am code using MERRA-2 data against the logged tip-ping radiometer measurements made at three different EHTsites. The tipping radiometer datasets are each multiyear:1995-2004 for ALMA/APEX3, 2013-2017 for LMT (Ferr-usca & R. 2014), and 2009–2014 for SMA (Radford & Pe-terson 2016). The tipper data were minimally flagged forτ225 extremes, and more than 95% of entries were kept foreach site. The am calculation is for the same dates as thetipper entries except for APEX, which is done for our nomi-nal 10 year period ending on the first day of year 2019. Theagreement between the medians of the calculations and fieldmeasurements supports our methods. The calculation repro-duces the wide range of seasonal opacities between medianτ225 = 0.04 at APEX in July to τ225 = 0.47 at LMT. Seasonalfeatures, like the dip in July/August opacity at the LMT, are

3 http://legacy.nrao.edu/alma/site/

Jan

Dec0.0

0.1

0.2

0.3

0.4

0.5

0.6

opac

ity a

t 225

GHz

ALMA/APEX

observ.am calc.

Jan

Dec

LMT

Jan

Dec

JCMT/SMA

Figure 4. Monthly 225 GHz zenith opacity (τ225) for three existingEHT sites. The calculated values (solid) from the reanalysis and225 GHz radiometer observations (dashed) are comparable. Theshaded region bounds the interquartile range, i.e., the middle 50%between the 25th (lower) and 75th (upper) percentiles for the re-analysis.

also reproduced. Figure 4 confirms that the calculation meth-ods accurately reflect ensemble observing conditions at geo-graphically varied sites.

Figure 5 plots the PWV statistics throughout the year at ex-isting and candidate new sites. The northern sites (top row)generally exhibit minimum PWV values during Decemberand January. Southern sites in the bottom row exhibit min-imum PWV values during July and August. Historically,EHT observations are performed in March or April, whenthe northern sites still have reasonably dry atmospheres andALMA has a median PWV column of 1–2 mm. All of the ex-isting EHT sites have a median PWV column in March andApril that is less than 5 mm. Most of the sites we analyzehave comparable PWV statistics in those months.

The PWV averages we report, which are derived from theMERRA-2 data, agree with measurements available in theliterature for both the 2021 array sites (plus ALMA) and forthe candidate new sites. At the South Pole, which is presentlythe driest site in the array, radiosonde measurements col-lected over several decades show PWV ranges from about0.25 mm in July to just below 0.8 mm in January (Cham-berlin 2001). The medians we derive for the SPT site are0.24 mm in July and 0.8 mm in January. At Kitt Peak, multi-year infrared measurements done in the 1980s found medianPWVs ranging from about 4 mm in winter to 27 mm in Au-gust during monsoon season (Wallace & Livingston 1984).We find 3.8–4.2 mm in winter and about 23 mm in August.Thus, for the driest and wettest sites in the EHT array, wefind that the interpolated MERRA-2-based PWV agrees withwhat has been measured in the field.

7

10 1

100

101

102PW

V (m

m)

1. GLT

1

2. IRAM-30m2. IRAM-30m

2

3. KP34. LMT

4

5. NOEMA

5

6. JCMT/SMA

6

7. SMT

7

existing north

1. BAJA

1

2. BAN

2

3. BAR

3

4. BGA

4

5. BGK

5

new, north 11. BLDR

1

2. CNI

2

3. ERB

3

4. FAIR

4

5. FUJI

5

6. GLT-S

6

new, north 2

1. HAN

1

2. HAY

2

3. HOP

3

4. JELM

4

5. KVNYS

5

new, north 3

1. LOS

1

2. NOB

2

3. NOR

3

4. ORG

4

5. OVRO

5

6. PIKE

6

new, north 4

1. SAN

1

2. SOC

2

3. SPX

3

4. SUF

4

5. YBG

5

new, north 5

Jan

Dec10 1

100

101

102

PWV

(mm

)

1. ALMA/APEX

1

2. SPT

2

existing south

Jan

Dec

1. BMAC

1

2. BOL

2

3. BRZ

3

4. CAS

4

new, south 1Ja

n

Dec

1. CAT

1

2. Dome A

2

3. Dome C

3

new, south 2

Jan

Dec

1. Dome F

1

2. GAM

2

3. GARS

3

4. KEN

4

new, south 3

Jan

Dec

1. KILI

1

2. LAS

2

3. LLA

3

4. NZ

4

new, south 4

Jan

Dec

1. PAR

1

2. SGO

2

3. YAN

3

new, south 5

Figure 5. Median monthly precipitable water vapor during the year at existing and candidate new EHT sites. Northern sites are in the top rowand southern sites are in the bottom row. In April, existing sites are below 5 mm, so a dotted line is added at that level.

PWV field measurements have also been made at many ofthe candidate new sites, and in most of those cases, we alsofind that the MERRA-2 PWV values are reliable. At Dome Aand Dome C, the yearly median PWV values reported fromsatellite data from 2008 to 2010 are 0.21 and 0.28 mm, re-spectively (Tremblin et al. 2012). We find yearly medians of0.19 and 0.29 mm for those sites, which are in good agree-ment as expected since the satellite data used would havebeen assimilated into MERRA-2. Ground-based radiometermeasurements between 2009 and 2014 (Ricaud et al. 2015)observed mean seasonal PWV values of about 0.3 mm duringJune–August and about 0.7 mm in December–February. Wefind a similar range: 0.2 mm in June–August and 0.6 mm inDecember–February.

Precipitable water vapor measurements for Teide andRoque de los Muchachos Observatory in the Canary Islandswere made using Global Positioning System path delays andradiosondes (Castro-Almazán et al. 2016). For the years2012 and 2013, monthly median PWV values at CI ranged

from about 2 mm in February to about 9 mm in August com-pared with our 10 year medians of 2.3 mm and 7.9 mm inFebruary and August, respectively.

The European Southern Observatory performed a multi-instrument campaign involving optical, far infrared, andradiosonde measurements at Paranal and La Silla (Querelet al. 2010). The study concluded that between 2005 and2009, those sites had mean PWV columns of 2.3± 1.8 and3.4± 2.4 mm, respectively. Those values are close to themean of the monthly medians we obtain from interpolatingMERRA-2 data: 2.6 mm at PAR and 3.6 mm at LAS. Boththe Querel et al. (2010) result and our value agree with atwo-year field measurement made using a 183 GHz radiome-ter (Kerber et al. 2015), which found 3.0 mm.

While the agreement with the literature is good for mostof the sites we analyze, there are a few cases where theagreement is not as close. Far-infrared radiometer mea-surements made on intervals between 1984 and 1987 atPico Veleta (Quesada 1989) returned median PWV values

8

of 1.2 mm in January up to 4.1 mm in September. We findslightly more water for IRAM-30 m: 2.5 mm in January and6.7 mm in September. At JELM, field measurements begin-ning in the late 1970s found 1.4 mm of water vapor in winterup to 6.8 mm in August (Grasdalen et al. 1985). We findabout 3.0 mm in winter and almost 12 mm in August. Inthese cases, discrepancies could be caused by local effectsthat the MERRA-2 profiles do not resolve. Future field mea-surements will be needed to determine if such anomalies ex-ist at particular sites.

As we have already established, cloud liquid water alsoaffects submillimeter opacity. In Fig. 6, the LWP and PWVvalues at the existing EHT sites are plotted for each 3 hr bin inthe MERRA-2 dataset, and the median of each axis is markedwith a red line. The correlation between the LWP and PWVvariables ranges from weak to strong depending on the site,with the Pearson coefficient varying from about 0.3 to a littlemore than 0.7, and the maximum LWP values range from50 µm at the South Pole to approximately 200 µm at severalof the sites. Figure 7 shows the same information but for thecandidate new sites. The LWP/PWV correlation coefficientshave a similar range as the existing sites: approximately 0.3to 0.75 with the exception of CAS. At the BAN, BGA, BGK,NOR, NZ, and SPX sites, there are a significant number ofdays when the vapor content is less than 5 mm, but the liquidcontent is approaching 100 µm.

At some of the sites in Figs. 6 and 7, there is only mod-erate correlation between the LWP and PWV statistics. Thefrequency of appreciable cloud cover is therefore importantin its own right. In Table 3, we show the number of days permonth with at least 50 µm of LWP column on a bimonthlybasis, which corresponds to a 0.13 and 0.18 opacity increaseabove the clear-sky conditions at 230 and 345 GHz, respec-tively. In March, the existing EHT sites have an averageof five or fewer days each month with heavy cloud cover.Twenty-five of the candidate new sites have a comparablefrequency of clouded days and would therefore be similarlyreliable for observing.

Unlike PWV statistics, which have a seasonality that de-pends strongly on hemisphere, the seasonality of cloud coveris more regional. For example, the sites in the southwesternUnited States have increased cloudiness in late summer be-cause of the North American monsoon. Except for NOEMA,which does not exhibit much seasonal variation to begin with,none of the existing EHT sites has its greatest number ofclouded days in March. This supports the idea that Marchand April are good months for EHT observations. Of thecandidate new sites, some, like BGK and YAN, probablyhave a prohibitive number of clouded days throughout theyear. Other sites show remarkable seasonal dependence. Forexample, GAM goes from eight clouded days in January tozero days throughout June and August. If the EHT eventu-

ally moves from once-per-year, campaign-style observationsto remotely controlled observations that target the optimalconditions for each source, then it may be possible to capi-talize on such variability.

Figure 6. Precipitable water vapor and liquid water path for ex-isting EHT sites during March and April. The Pearson correlationcoefficient between PWV and LWP is listed below the site label.The vertical and horizontal lines show the median PWV and LWP,respectively.

While PWV and LWP are good indicators of observingconditions, radiative transfer at a particular site also de-pends on local vertical properties of moisture and tempera-ture (Paine 2019) and must therefore be calculated. Table 4presents the zenith 230 GHz opacity. Opacities are calculatedusing the local thermodynamic conditions above each site.Neither March nor April are extreme seasons for the 230 GHzzenith opacity. Of the candidate new sites, 21 of them havemedian zenith opacity of less than 0.2 during either of thosemonths. The sites with median opacity below 0.10 (Dome A,Dome C, Dome F, FUJI, GLT-S, LLA, NOR, and PIKE) haveinterquartile ranges that vary from less than 0.01 at Antarcticsites to 0.2 elsewhere. A second tier of sites have opacitiesbetween 0.1 and 0.15 (BAJA, BAR, HAN, PAR, SGO, SPX,and YBG).

The opacities we report agree reasonably well with pub-lished field measurements. At the Greenland Summit site, amultiyear 225 GHz radiometer campaign beginning in 2010found that the median opacity during the winter months wasabout 0.06 (Matsushita et al. 2017). We find a similar me-dian opacity: 0.05 to 0.06. The median in July is also similar:0.16–0.19 in the Matsushita et al. (2017) field measurementand 0.16 for our retrieval.

9

Figure 7. Same as Fig. 6 but for candidate new sites.

10

Table 3. Days with at Least 50 µm Liquid Water Path.

Site Jan. Mar. May Jul. Sep. Nov.

ALMA/APEX 120 01

0 000 00

0 000 00

0GLT 24

1 241 911

8 10139 1316

12 796

IRAM-30m 242 46

3 352 00

0 242 46

3JCMT/SMA 12

1 352 13

1 010 23

1 242

KP 242 12

0 010 911

8 463 12

1LMT 01

0 231 710

6 10129 1518

14 242

NOEMA 453 57

4 10129 68

5 685 57

4SMT 23

1 121 12

0 10129 68

5 121

SPT 000 00

0 000 00

0 000 00

0

BAJA 131 01

0 010 34

2 242 01

0BAN 68

5 11139 1315

11 161915 1113

9 9117

BAR 120 12

1 352 35

3 121 01

0BGA 34

2 685 1315

12 10129 79

6 352

BGK 232522 2325

21 192218 1618

15 222421 2224

21BLDR 24

1 353 79

6 796 56

4 231

BMAC 10139 810

6 241 12

0 121 46

3BOL 1719

15 131612 57

4 342 810

7 10129

BRZ 131612 1719

15 796 35

3 574 1618

15CAS 2628

26 232622 1921

17 171916 1821

17 242623

CAT 574 68

5 9128 810

7 8107 810

7CNI 12

1 231 00

0 000 01

0 242

Dome A 000 00

0 000 00

0 000 00

0Dome C 00

0 000 00

0 000 00

0 000

Dome F 000 00

0 000 00

0 000 00

0ERB 57

4 796 35

3 000 01

0 574

FAIR 353 34

2 121411 1922

18 10139 46

3FUJI 12

0 242 46

3 10129 79

6 242

GAM 8107 79

6 010 00

0 010 12

1GARS 2224

21 212420 1820

16 141713 1517

14 212320

GLT-S 000 00

0 120 12

1 131 01

0HAN 01

0 121 35

2 7106 45

3 010

HAY 9117 1013

9 131511 1113

10 8107 911

7HOP 24

2 120 01

0 121511 68

5 121

JELM 231 35

2 796 68

5 463 24

1KEN 46

3 9117 1012

9 111410 810

7 151814

KILI 574 912

8 463 00

0 120 1113

10KVNYS 45

3 574 68

5 131511 79

6 8107

LAS 000 01

0 131 12

0 010 00

0LLA 57

4 231 00

0 000 00

0 000

LOS 231 23

1 342 911

8 685 23

1NOB 57

4 9118 1012

9 192118 1618

14 10129

NOR 463 68

5 8107 1618

14 11139 46

3NZ 1214

10 9118 1113

10 10128 1114

10 121411

ORG 463 810

6 796 12

1 242 57

4OVRO 34

2 352 46

3 453 13

1 231

PAR 000 00

0 010 00

0 000 00

0PIKE 12

0 131 35

3 8107 35

3 020

SAN 231 12

0 120 35

2 131 01

0SGO 12

0 010 23

1 131 13

1 010

SOC 231 23

1 242 1113

10 8106 13

1SPX 24

2 353 1214

10 121411 79

5 342

SUF 453 68

5 9118 12

1 020 35

2YAN 1921

17 202319 911

8 463 911

8 121410

YBG 000 01

0 463 1618

14 10129 00

0

Note. Sub/superscripts are 25 / 75th percentiles, respectively.

At the HAN site, 220 GHz tipping radiometer measure-ments beginning at the end of year 1999 until the middle of2001 have been published (Ananthasubramanian et al. 2002).The monthly median opacities reported are 0.070.09

0.05 in Jan-uary and 0.370.46

0.26 in July. For 2009–2019, we calculate opac-ities of 0.070.12

0.05 and 0.530.690.37 in those months. The agreement

in wintertime is excellent. In summer, additional field mea-surement data is needed to constrain the level of overlap be-tween the experimental and model distributions.

The wintertime conditions at FUJI are very good andagree with previous 220 GHz radiometer measurements done1994–95 (Sekimoto et al. 1996). In December of 1994, anopacity of about 0.04 was measured, which matches exactlywith the 0.040.11

0.03 range from our calculation. The agreementin March is also within the statistical variation: 0.11 observedcompared with 0.060.16

0.03 from our 10 year average. Mt. Fujihas good submillimeter conditions in winter.

Finally, our opacity calculations agree with the 225 GHztipping radiometer measurements for 2001 September toNovember, reported by Marvil et al. (2006) for the BAR site.The median measured opacity was 0.11 over that time frame,which is similar to the 0.11 to 0.17 10 year median values wecalculate for the same months.

3.2. Applicability to 345 GHz and Higher Frequencies

The highest-frequency Fourier components (~u) sampledduring the 230 GHz observation in 2017 of the M87∗

source were measured on the baseline between Hawaii andSpain (Event Horizon Telescope Collaboration et al. 2019a)and corresponded to 25µas instrumental angular resolutiondefined by the fringe spacing λ/D, where λ is the observ-ing wavelength and D is the projected baseline length to thesource. There is interest in operating at 345 GHz or poten-tially higher frequencies at the stations with suitable condi-tions, which could improve the nominal resolution by 50%or more. Most of the 2021 EHT sites are capable of 345 GHzoperation during at least part of the year, and the transmit-tances (Tr = − lnτ ) of each existing site during March andApril are plotted in Fig. 8. Although the transmittance thresh-old for 345 GHz capability depends on the telescope sensi-tivity and other details like the atmospheric coherence, wecan make a reasonable estimate of the transmittance that isrequired. Consider the case of 0.1 Jy correlated flux den-sity at 45◦ elevation with two-bit digital efficiency, and a345 GHz zenith transmittance of 0.6 (opacity of 0.5) and sky-brightness temperature of 150 K. Under those conditions,two stations with 10 m diameter antennas having 64 µm ef-fective surface-accuracy apertures and 100 K receiver tem-peratures integrating for 100 s at 8 GHz fringe-finding band-width would achieve an S/N 3 detection. Stronger detectionswould be achieved on baselines to large apertures like ALMAor NOEMA.

11

Table 4. Zenith Opacity at 230 GHz.

site Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec.

ALMA/APEX 0.130.260.06 0.150.31

0.07 0.080.160.04 0.060.11

0.03 0.040.070.03 0.040.06

0.03 0.030.060.02 0.030.05

0.03 0.040.060.03 0.040.06

0.03 0.040.070.03 0.070.15

0.04GLT 0.160.26

0.13 0.150.250.12 0.150.24

0.13 0.210.310.16 0.370.58

0.29 0.650.940.53 0.841.17

0.69 0.821.270.62 0.530.82

0.39 0.370.600.24 0.240.40

0.17 0.170.300.13

IRAM-30m 0.140.240.08 0.130.24

0.07 0.150.290.08 0.200.34

0.12 0.240.370.14 0.300.43

0.20 0.260.400.16 0.330.49

0.21 0.290.450.19 0.260.42

0.15 0.190.360.11 0.150.27

0.08JCMT/SMA 0.060.12

0.04 0.080.170.04 0.080.20

0.04 0.080.160.05 0.080.16

0.05 0.090.150.06 0.110.17

0.06 0.090.210.06 0.080.17

0.05 0.110.220.06 0.090.21

0.05 0.070.170.04

KP 0.210.400.13 0.210.36

0.13 0.240.370.15 0.250.37

0.16 0.290.430.20 0.450.78

0.26 1.471.771.11 1.311.66

0.97 0.921.310.54 0.400.65

0.25 0.260.420.16 0.240.40

0.14LMT 0.090.15

0.05 0.090.160.05 0.130.21

0.06 0.190.330.12 0.250.45

0.15 0.400.630.20 0.370.56

0.22 0.390.580.26 0.490.69

0.31 0.260.460.11 0.140.27

0.07 0.090.160.05

NOEMA 0.160.290.09 0.160.30

0.09 0.190.330.12 0.270.45

0.17 0.370.620.22 0.460.71

0.31 0.480.690.31 0.450.67

0.29 0.390.600.24 0.290.51

0.16 0.250.450.14 0.170.31

0.09SMT 0.120.20

0.07 0.110.190.06 0.130.20

0.07 0.150.220.09 0.170.26

0.12 0.300.510.16 0.821.05

0.60 0.730.990.51 0.480.73

0.29 0.190.340.12 0.130.22

0.07 0.130.210.07

SPT 0.060.070.05 0.050.06

0.05 0.050.050.04 0.040.05

0.04 0.040.050.04 0.040.04

0.03 0.040.040.03 0.040.04

0.03 0.040.040.03 0.040.05

0.04 0.040.050.04 0.060.07

0.05

BAJA 0.130.250.07 0.110.20

0.06 0.120.210.07 0.130.21

0.07 0.140.210.09 0.190.39

0.12 0.670.940.40 0.560.82

0.31 0.340.620.16 0.160.27

0.09 0.130.210.07 0.130.22

0.07BAN 0.240.42

0.14 0.200.370.12 0.270.45

0.16 0.300.490.20 0.430.65

0.29 0.620.890.45 0.690.99

0.51 0.650.940.49 0.480.73

0.33 0.360.580.22 0.270.46

0.16 0.210.360.13

BAR 0.110.170.05 0.080.14

0.05 0.110.160.06 0.120.19

0.07 0.160.250.11 0.190.29

0.12 0.290.480.16 0.210.37

0.13 0.170.270.11 0.140.21

0.08 0.110.160.06 0.090.16

0.05BGA 0.150.26

0.09 0.160.270.11 0.190.33

0.11 0.250.430.14 0.370.60

0.24 0.460.730.29 0.420.67

0.27 0.390.580.26 0.360.58

0.21 0.260.450.13 0.210.33

0.12 0.150.260.08

BGK 0.580.940.34 0.540.89

0.31 0.510.870.29 0.540.94

0.30 0.621.080.37 0.761.20

0.51 0.891.390.65 0.921.47

0.63 0.891.430.54 0.731.17

0.43 0.601.020.34 0.530.92

0.29BLDR 0.210.30

0.13 0.200.330.14 0.240.36

0.16 0.310.480.24 0.460.63

0.33 0.540.760.39 0.841.11

0.62 0.730.990.54 0.530.76

0.37 0.360.510.25 0.240.34

0.15 0.200.300.13

BMAC 0.630.990.40 0.621.02

0.36 0.490.760.29 0.300.53

0.16 0.220.390.13 0.120.24

0.07 0.130.220.07 0.160.29

0.08 0.210.390.12 0.300.51

0.17 0.370.600.22 0.490.80

0.30BOL 0.390.51

0.27 0.420.540.29 0.340.49

0.22 0.240.400.14 0.150.29

0.08 0.120.240.05 0.090.20

0.05 0.110.220.05 0.170.34

0.08 0.270.430.14 0.270.43

0.16 0.370.510.25

BRZ 0.711.350.34 0.631.14

0.36 0.971.470.54 0.631.14

0.31 0.430.760.21 0.330.58

0.15 0.240.460.11 0.220.45

0.09 0.360.690.13 0.711.17

0.36 0.941.470.49 0.941.51

0.54CAS 0.801.27

0.53 0.761.240.48 0.711.17

0.46 0.601.020.37 0.490.82

0.33 0.480.780.31 0.450.73

0.27 0.460.760.29 0.480.78

0.30 0.540.890.34 0.631.05

0.42 0.731.140.48

CAT 0.310.530.20 0.340.58

0.20 0.310.540.19 0.290.54

0.15 0.310.620.14 0.300.54

0.15 0.260.480.13 0.290.54

0.13 0.240.480.13 0.270.49

0.15 0.290.510.17 0.310.58

0.20CNI 0.160.27

0.09 0.130.250.08 0.160.27

0.08 0.160.290.11 0.190.30

0.12 0.220.370.15 0.260.46

0.15 0.390.600.20 0.330.54

0.20 0.300.530.17 0.220.42

0.12 0.170.340.09

Dome A 0.030.040.03 0.030.03

0.03 0.030.030.02 0.020.03

0.02 0.020.030.02 0.020.03

0.02 0.020.030.02 0.020.02

0.02 0.020.020.02 0.020.03

0.02 0.030.030.03 0.030.04

0.03Dome C 0.050.06

0.04 0.040.050.04 0.040.04

0.03 0.030.040.03 0.030.04

0.03 0.030.030.03 0.030.03

0.03 0.030.040.03 0.030.03

0.03 0.030.040.03 0.040.04

0.03 0.050.050.04

Dome F 0.040.050.04 0.040.04

0.03 0.030.030.03 0.030.03

0.03 0.030.030.03 0.030.03

0.02 0.030.030.02 0.030.03

0.02 0.030.030.02 0.030.03

0.03 0.030.040.03 0.040.04

0.03ERB 0.240.40

0.15 0.270.430.15 0.310.51

0.20 0.420.620.27 0.540.73

0.39 0.480.650.33 0.490.71

0.33 0.460.630.31 0.430.58

0.30 0.450.670.29 0.330.54

0.19 0.260.460.15

FAIR 0.210.340.11 0.200.31

0.13 0.170.290.11 0.270.42

0.19 0.490.780.36 0.801.14

0.60 1.021.430.78 0.871.27

0.65 0.560.840.39 0.360.53

0.24 0.200.360.13 0.220.36

0.14FUJI 0.040.08

0.03 0.050.130.03 0.060.16

0.03 0.080.220.04 0.130.34

0.05 0.290.630.12 0.490.80

0.29 0.490.760.29 0.340.63

0.13 0.130.390.05 0.070.19

0.03 0.040.110.03

GAM 0.941.350.60 1.021.51

0.60 0.841.240.53 0.560.82

0.34 0.330.480.21 0.190.29

0.11 0.190.300.11 0.200.33

0.12 0.300.510.14 0.420.63

0.21 0.480.730.25 0.620.94

0.39GARS 0.711.14

0.51 0.761.200.51 0.781.24

0.48 0.601.050.34 0.581.05

0.33 0.530.940.27 0.400.82

0.21 0.400.820.21 0.480.94

0.25 0.581.050.34 0.621.05

0.42 0.671.050.48

GLT-S 0.060.090.05 0.060.08

0.05 0.050.080.04 0.060.11

0.05 0.090.150.07 0.130.20

0.09 0.160.220.12 0.150.24

0.12 0.130.200.08 0.080.14

0.06 0.060.110.05 0.050.08

0.04HAN 0.070.12

0.05 0.110.150.07 0.120.19

0.08 0.170.250.12 0.220.31

0.15 0.300.430.21 0.530.69

0.37 0.560.710.37 0.260.45

0.15 0.120.160.08 0.080.13

0.06 0.070.110.05

HAY 0.420.730.22 0.450.82

0.26 0.540.990.30 0.781.35

0.48 1.352.210.89 1.772.66

1.20 2.213.001.56 2.042.81

1.43 1.512.410.94 1.051.83

0.65 0.651.140.39 0.530.97

0.31HOP 0.170.31

0.11 0.170.300.11 0.190.30

0.12 0.210.310.14 0.240.36

0.16 0.420.730.25 1.271.61

0.99 1.141.510.87 0.801.17

0.49 0.330.560.20 0.210.34

0.13 0.190.330.11

JELM 0.190.270.11 0.170.30

0.12 0.210.310.14 0.270.42

0.20 0.390.540.29 0.450.60

0.31 0.650.890.46 0.580.80

0.42 0.430.620.29 0.300.45

0.21 0.200.300.13 0.170.27

0.11KEN 0.160.31

0.09 0.170.340.09 0.220.48

0.11 0.460.710.26 0.390.60

0.24 0.330.540.20 0.340.53

0.19 0.310.530.16 0.260.45

0.16 0.340.540.21 0.430.63

0.27 0.300.510.17

KILI 0.190.370.09 0.170.39

0.09 0.260.490.13 0.370.58

0.21 0.260.420.15 0.150.24

0.08 0.090.170.06 0.120.21

0.06 0.130.210.08 0.170.33

0.09 0.310.510.17 0.310.51

0.17KVNYS 0.240.40

0.16 0.290.490.17 0.390.63

0.24 0.651.050.42 1.021.56

0.67 1.972.531.43 3.223.91

2.41 3.003.912.21 1.562.30

1.05 0.801.240.53 0.560.99

0.31 0.290.530.17

LAS 0.290.460.17 0.270.42

0.17 0.220.340.15 0.190.29

0.13 0.200.300.13 0.160.26

0.11 0.140.220.08 0.140.22

0.09 0.140.220.09 0.150.24

0.11 0.170.250.12 0.200.29

0.13LLA 0.220.40

0.11 0.260.460.12 0.130.22

0.06 0.070.130.04 0.050.08

0.03 0.040.060.03 0.040.06

0.03 0.040.060.03 0.050.07

0.03 0.040.070.03 0.060.09

0.04 0.120.250.06

LOS 0.240.360.16 0.260.36

0.17 0.270.390.19 0.340.46

0.24 0.420.580.30 0.560.87

0.34 1.171.470.92 1.081.35

0.80 0.801.110.53 0.430.63

0.29 0.270.400.17 0.250.37

0.16NOB 0.240.40

0.16 0.250.480.16 0.310.63

0.19 0.460.890.27 0.711.24

0.42 1.352.120.89 2.042.66

1.51 2.122.811.61 1.512.30

0.89 0.731.430.43 0.430.80

0.26 0.290.490.19

NOR 0.080.170.04 0.080.21

0.04 0.090.260.05 0.140.37

0.06 0.210.540.09 0.480.99

0.22 0.841.310.51 0.871.35

0.53 0.531.050.17 0.160.53

0.06 0.120.290.05 0.090.21

0.05NZ 0.430.87

0.24 0.400.760.24 0.340.71

0.19 0.340.690.17 0.300.58

0.17 0.240.480.13 0.240.45

0.12 0.290.530.14 0.300.56

0.15 0.340.620.17 0.370.69

0.21 0.430.820.24

ORG 0.240.420.13 0.240.40

0.15 0.270.490.17 0.300.49

0.20 0.420.620.27 0.460.65

0.33 0.530.730.37 0.510.69

0.37 0.420.580.29 0.360.51

0.22 0.270.450.15 0.240.43

0.13OVRO 0.420.69

0.24 0.420.560.31 0.490.65

0.33 0.510.690.39 0.670.89

0.48 0.670.940.48 0.971.43

0.60 0.761.170.51 0.670.97

0.48 0.560.800.40 0.450.65

0.31 0.400.620.26

PAR 0.200.330.12 0.210.39

0.14 0.160.260.12 0.140.22

0.09 0.130.210.09 0.110.17

0.07 0.090.140.06 0.090.15

0.07 0.090.150.06 0.110.15

0.07 0.110.150.08 0.140.21

0.09PIKE 0.070.12

0.05 0.070.110.04 0.080.13

0.05 0.120.190.07 0.160.26

0.11 0.200.310.13 0.360.54

0.22 0.310.480.21 0.200.33

0.12 0.120.190.07 0.080.12

0.05 0.070.110.05

SAN 0.150.270.08 0.130.22

0.07 0.150.250.08 0.160.26

0.11 0.210.330.13 0.210.36

0.13 0.630.920.27 0.480.78

0.26 0.310.540.17 0.200.33

0.12 0.150.250.09 0.150.26

0.08SGO 0.160.26

0.11 0.150.250.09 0.130.20

0.08 0.120.190.07 0.130.21

0.07 0.110.190.06 0.110.17

0.06 0.110.190.06 0.110.17

0.06 0.120.190.07 0.120.19

0.08 0.130.200.08

SOC 0.240.370.15 0.250.37

0.16 0.270.390.19 0.310.45

0.22 0.390.540.26 0.560.89

0.34 1.311.661.05 1.241.56

0.94 0.941.270.62 0.450.69

0.30 0.290.430.17 0.260.42

0.16SPX 0.090.17

0.05 0.090.170.05 0.110.19

0.06 0.150.270.08 0.240.40

0.13 0.300.490.16 0.300.53

0.16 0.290.510.16 0.210.39

0.12 0.150.270.08 0.140.25

0.07 0.090.190.06

SUF 0.220.370.13 0.250.43

0.13 0.370.600.22 0.460.71

0.29 0.620.890.43 0.650.87

0.49 0.620.780.48 0.540.73

0.40 0.420.540.30 0.360.54

0.22 0.290.450.16 0.200.36

0.11YAN 0.560.67

0.42 0.600.710.48 0.600.71

0.46 0.490.630.34 0.370.53

0.24 0.290.430.19 0.240.39

0.14 0.270.420.16 0.370.51

0.24 0.430.560.29 0.450.60

0.29 0.530.650.39

YBG 0.070.090.05 0.080.12

0.06 0.120.190.08 0.170.26

0.12 0.290.430.20 0.580.80

0.36 0.821.020.71 0.780.94

0.63 0.630.780.45 0.200.33

0.14 0.090.130.07 0.070.09

0.06

Note. Sub/superscripts are 25 / 75th percentiles, respectively.

12

The existing sites in Chile, Hawaii, and the South Polehave median March 345 GHz transmittance of 0.75 orgreater. The South pole, in particular, has excellent 345 GHztransmittance even at the 25th percentile. The SPT site alsohas relatively good transmittance at 410 GHz (above 70%).Zenith transmission spectra for the new sites are plotted inFig. 9. Of these, Dome A , Dome C, and Dome F havehigh-frequency conditions that are similar to the SPT site.The FUJI, GLT-S, HAN, LLA, and PIKE sites are similarto the ALMA/APEX and JCMT/SMA sites: more opaquethan Antarctica and more variable. BAJA, BAR, NOR, SGO,SPX, and YBG have comparable 345 GHz transmittance tothe LMT site and would probably be viable during a usefulfraction of time.

Figure 8. Zenith transmittance spectra at 200−500 GHz for existingEHT sites in March (black) and April (red). The shaded regionbrackets the interquartile range, and 8 GHz-wide high-frequencybands centered at 345 (green) and 410 GHz (yellow) are markedwith vertical lines.

The high-frequency transmittances we calculate are inagreement with published values. At Dome C, the 345 GHzvalue for an atmosphere with the median PWV of 0.28 mmhave been reported to be between 0.85 and 0.90 (Tremblinet al. 2012). At Dome A, that same study found 345 GHztransmittance in median conditions to be greater than 0.9.Both of those findings agree with the spectra we reportspecifically for March. At Yangbajing Observatory, for amedian atmosphere containing 2.8 mm of water vapor, thatsame Tremblin et al. (2012) study concluded that YBG has345 GHz transmittance of 55%–60%. In March, we derive a

similar median PWV of 2.4 mm for the YBG site and corre-spondingly greater transmittance of about 0.67.

3.3. Improved Sampling of Fourier Components

The 230 GHz Fourier coverage for each candidate new siteis plotted in Fig. 10 for M87∗ and in Fig. 11 for Sgr A∗.These coverages are based on the source position and arraygeometry; we have not flagged low S/N visibilities as wedo in the subsequent Monte Carlo filling-factor calculations.The coverage is for a full 24 hr of observing including timeswhen ALMA cannot observe the source. Tracks correspond-ing to baselines with ALMA are plotted in bold. The tracksof the 2021 EHT array are underlaid in gray for reference.The 2021 array with ALMA lacks coverage between about4 and 7 Gλ in the north–south direction for both M87∗ andSgr A∗. Sgr A∗ also lacks mid-baseline coverage between 1and 3 Gλ as well as long-baseline coverage in the east–westdirection. In these plots, an elevation cutoff of 10◦ was spec-ified, which is especially important for the very short tracksat large u-v spacing, e.g., for FUJI and SPX toward M87∗.

The number of new baselines is determined by the covis-ibility of the new site with the 2021 array with ALMA. ForM87∗, all of the considered sites that can observe that sourceform at least five new baselines. For Sgr A∗, some sites formas few as two new baselines. The mean longitude of the 2021EHT sites is 76◦ west. Sites near that longitude have signifi-cant covisibility with the existing EHT sites and with ALMAand therefore tend to make long u-v tracks.

Some site locations are more desirable than others becausethey sample obvious gaps in Fourier coverage. For M87∗, theBOL, BRZ, CNI, BMAC, GAM, HAY, NZ, and YAN loca-tions fill portions of the 4-7 Gλ gap in the north–south direc-tion. The BMAC, CAS, CAT, GAM, GARS, NZ, SGO, andYBG sites extend the longest baselines. For Sgr A∗, the 2021coverage contains gaps between 4 and 7 Gλ in the north–south direction, which are filled by CAS, CAT, GARS, KEN,and KILI. Sites in central and southern Africa, e.g., KILI to-gether with GAM, could significantly improve the resolutionof the array in the east–west direction.

The geometric Fourier coverage contributed by a new siteis important, but only if the weather conditions at that newsite are adequate for making detections. The Monte Carlo in-cremental coverage, which flags intervals with low S/N val-ues based on the probabilistic weather at each site, is pre-sented in Fig. 12 for a specific set of observing parame-ters: 10 m antennas at new stations, simultaneous 230 and345 GHz frequency coverage, and 8 GHz bandwidth in eachof two sidebands. For the simulations in Figs. 12 and 13,the observing schedule is limited to times when ALMA cansee the source, which is consistent with previous EHT sched-ules (Event Horizon Telescope Collaboration et al. 2019c).The bandwidth per frequency band is a projected doubling of

13

Figure 9. Zenith transmittance spectra at 200 − 500 GHz for candidate new EHT sites for March (black) and April (red). Shaded spectra are theinterquartile range, and high-frequency 8 GHz-wide high-frequency bands centered at 345 and 410 GHz are marked with vertical lines.

14

Figure 10. Fourier coverage for M87∗ at 230 GHz compared to the 2021 EHT array (gray) for baselines containing the indicated site (red).The large markers denote baselines to ALMA. Sidebands are omitted for clarity.

15

Figure 11. Fourier coverage for Sgr A∗ at 230 GHz compared to the 2021 EHT array (gray) for baselines containing the indicated site (blue).The large markers denote baselines to ALMA. Sidebands are omitted for clarity.

16

the present-day EHT (Event Horizon Telescope Collabora-tion et al. 2019a) and helps compensate for some of the addi-tional noise at 345 GHz. For M87∗, an extended 500 µas FoVis specified, which is motivated by future efforts to connecthorizon-scale structure with the M87∗ jet (Blackburn et al.2019). For Sgr A∗, a 150 µas FoV is used. We assume anobserving time commensurate with the variability timescaleof each source. M87∗ evolves slowly, so a full observation isspecified to maximize the coverage on each baseline. Sgr A∗

evolves on a timescale of minutes (Event Horizon TelescopeCollaboration et al. 2019a), so a 100 s snapshot observationis simulated, which truncates the Fourier coverage. Our ap-proach to flagging low S/N scans accounts for the noise con-tributed by the physical temperature of the atmosphere. Thisis important at low-altitude polar sites like GLT and GARS,where the cold atmosphere compensates for marginal opac-ity.

The Monte Carlo incremental filling factor is calculated inthe following way. For the full observations, we report theFourier filling of the augmented and fiducial arrays using theentirety of the Fourier coverage tracks, Faug/Ffid − 1. For thesnapshot observations of Sgr A∗, we vary the start time forthe fiducial and augmented arrays independently to maximizethe filling in each case: max

(Faug)/max(Ffid)−1. Depending

on the coordinates of a new site, there might be no start timesthat produce greater snapshot Fourier filling than the fiducialarray, which would make the incremental filling for that sitezero. The maximum filling metric is a proxy for the bestinstantaneous coverage achievable by a new site.

The sites with the greatest incremental filling are added se-quentially to the fiducial array as indicated beginning withthe 2021 EHT sites. For M87∗, SGO emerges as the sitethat adds the greatest Fourier coverage to the 2021 array withALMA. Repeating the calculation for a 12 station array con-sisting of the 2021 EHT+SGO, CNI emerges as the next bestsite for increasing the Fourier coverage. Similarly for Sgr A∗

with a full observation, Dome A, KEN, BMAC, and Dome Cyield the most incremental coverage. For Sgr A∗ with snap-shot coverage, NZ adds the most coverage to the 2021 array,followed by Dome C, then CAT, and GARS. In general, theincremental coverage for a particular new site decreases asnew sites are added to the fiducial array because there arefewer gaps in the Fourier coverage to fill. Our approach ofadding the best locations in sequence makes the problem ofselecting a group of sites computationally manageable andleads us to an array with excellent coverage. As the ngEHTdevelops, a comprehensive search for the global optimum ar-ray could lead to a different selection of sites.

The Fourier filling factor we present here is one metricfor approaching VLBI array design. While we favor the fill-ing metric for its simplicity, other considerations and metrics

could prioritize a different set of new sites and should be vet-ted in future design efforts.

3.4. Imaging with an Expanded Array

To demonstrate the improvement of horizon-scale imagingthat would be possible with an expanded array under real-istic meteorological conditions, we compare reconstructionsmade using the 2017, 2021, and an ngEHT array. In the 2017and 2021 arrays, we combine multiple 2 GHz-wide fringe-finding bands for an aggregate bandwidth over two polar-izations of 16 GHz in 2017 and 32 GHz in 2021. For thengEHT array, we assume the same parameters as the fillingfactor calculation: dual-frequency 230 and 345 GHz observ-ing with an aggregate bandwidth of 128 GHz over two polar-izations. The bandwidth of the EHT backend has increasedat regular intervals (Doeleman et al. 2008; Fish et al. 2011;Doeleman et al. 2012; Johnson et al. 2015; Event HorizonTelescope Collaboration et al. 2019a), and it is reasonable toanticipate continued improvement on the timescale of arrayexpansion (Blackburn et al. 2019). The image reconstruc-tions are shown in Fig. 13. The thermal noise used for thesereconstructions are calculated under median conditions.

For M87∗, the FoV is extended and the temperature scaleis logarithmic to highlight the base of the low-brightness for-ward jet. For this particular source model with the 2017 and2021 arrays, the jet base is not faithfully reconstructed. Incontrast, the additional stations in the ngEHT array and theaddition of the 345 GHz band reproduces the jet out to morethan five shadow diameters, and the diffuse flux surroundingthe jet is more tightly constrained. The sites in the ngEHTpanel were primarily chosen based on the leading sites fromthe Monte Carlo incremental filling analysis in Fig. 12: threeor four sites each for M87∗ and snapshot Sgr A∗. BAJA,GAM, LLA, OVRO, and NOR were also included either be-cause they sample interesting Fourier components or becausethey have significant infrastructure already in place. ThengEHT Fourier coverage on M87∗ extends to approximately12 Gλ under nominal meteorological conditions.

The Sgr A∗ reconstruction is shown in linear scale. We se-lect a start time for the Sgr A∗ simulated observation whenthe Fourier filling is greatest. With 100 s snapshot observing,the Fourier tracks normally obtained through Earth-aperturesynthesis become points. For the 2017 and 2021 arrays, theFourier points are sparse, and the arrays and eht-imagingalgorithms do not converge on the known structure; however,the large number of baselines in the the ngEHT simulationfills a significant fraction of the Fourier plane even in snap-shot mode. The ngEHT is capable of reconstructing the ringemission of Sgr A∗ including the morphology of the disconti-nuity in the left-hand site of the image. A full multifrequencyreconstruction with scattering mitigation is an active researchtopic; however, these short-timescale reconstructions suggest

17

BA

JAB

AN

BA

RB

GA

BG

KB

LDR

BM

AC

BO

LB

RZ

CA

SC

AT

CN

ID

om

e A

Dom

e C

Dom

e F

ER

BFA

IRFU

JIG

AM

GA

RS

GLT

-SH

AN

HA

YH

OP

JELM

KEN

KIL

IK

VN

YS

LAS

LLA

LOS

NO

BN

OR

NZ

OR

GO

VR

OPA

RPIK

ESA

NSG

OSO

CSPX

SU

FYA

NYB

G

2021 EHT+SGO+CNI

+GLT-S+BOL

M87 * : 230 + 345 GHz, full observation, 500 as FoV

1

2

3

4

5

0

5

10

15

20

incr

em

enta

l filli

ng (

%)

BA

JAB

AN

BA

RB

GA

BG

KB

LDR

BM

AC

BO

LB

RZ

CA

SC

AT

CN

ID

om

e A

Dom

e C

Dom

e F

ER

BFA

IRFU

JIG

AM

GA

RS

GLT

-SH

AN

HA

YH

OP

JELM

KEN

KIL

IK

VN

YS

LAS

LLA

LOS

NO

BN

OR

NZ

OR

GO

VR

OPA

RPIK

ESA

NSG

OSO

CSPX

SU

FYA

NYB

G

2021 EHT+Dome A

+KILI+BMAC

+Dome F

Sgr A * : 230 + 345 GHz, full observation, 150 as FoV

1

2

3

4

5

0

5

10

15

20

incr

em

enta

l filli

ng (

%)

BA

JAB

AN

BA

RB

GA

BG

KB

LDR

BM

AC

BO

LB

RZ

CA

SC

AT

CN

ID

om

e A

Dom

e C

Dom

e F

ER

BFA

IRFU

JIG

AM

GA

RS

GLT

-SH

AN

HA

YH

OP

JELM

KEN

KIL

IK

VN

YS

LAS

LLA

LOS

NO

BN

OR

NZ

OR

GO

VR

OPA

RPIK

ESA

NSG

OSO

CSPX

SU

FYA

NYB

G

2021 EHT+NZ

+CAT+Dome C

+GARS

Sgr A * : 230 + 345 GHz, 100 s snapshot observing, 150 as FoV

1

2

3

4

5

0

10

20

30

40

incr

em

enta

l filli

ng (

%)

Figure 12. Incremental filling highlighting the sites with the best performance as the array grows incrementally. The simulated observationsuse April probabilistic meteorological conditions, and the filling reported is the mean of the Monte Carlo instantiations. Beginning with the2021 EHT array, the incremental filling is calculated for the addition of each new site. The fiducial array is then augmented by the new site withthe greatest incremental filling and the calculation is repeated again. For M87∗ (top panel), the 2021 coverage is improved most by SGO(1)followed by SGO+CNI(2) then by SGO+CNI+GLT-S(3), and so on. The same procedure was done for Sgr A∗ with a full observation (middle),and Sgr A∗ with a 100 s snapshot observation (bottom). The Sgr A∗ snapshot observing performance is indicative of the best instantaneouscoverage. The ngEHT observations are assumed to be simultaneous dual-frequency 230 + 345 GHz with 10 m antennas at new stations and an8 GHz fringe-finding bandwidth in each of two sidebands. The M87∗ fractions are calculated assuming a 500 µas FoV to encompass the baseof the forward jet while the Sgr A∗ fractions assume a 150 µas FoV. Snapshot fill fractions of zero have no more coverage than the fiducialarray.

that detailed movie-making of dynamical features at Sgr A∗

could be possible with the ngEHT.

4. CONCLUSIONS

We have evaluated potential new sites for the ngEHT andexamined how they improve the capability of the EHT ar-ray to study M87∗ and Sgr A∗ in increasing detail. We haveused meteorological statistics derived from the MERRA-2reanalysis to develop a list of 45 potential new sites. Our sitewater vapor reanalyses of meteorological data are in closeagreement with available in-situ measurements. We find thata number of the candidate sites have, first, good conditionsfor submillimeter observations and, second, also sample newspatial frequencies on both large and small angular scales,which significantly enhances imaging fidelity of the EHT ar-ray. The Monte Carlo incremental filling calculation sug-gests that the priority sites strongly depend on details like thesource and the FoV. Future work is planned for assessing thebest array across a range of science cases and metrics; how-ever, our example imaging reconstruction shows that underrealistic median meteorological conditions, additional sta-tions in the ngEHT will improve the ability to image low-

flux, dynamical features for M87∗ and Sgr A∗. Future workwill also include field measurements over at least a full sea-sonal cycle to validate MERRA-2 against local conditions.

ACKNOWLEDGMENTS

We thank several members of the EHT Collaboration andT. K. Sridharan for suggesting potential new sites for study,Simon Radford for useful discussions about the SMA opacitylogs, Andrew Chael for providing GRMHD source models,Antonio Fuentes whose MPI parallelization framework weadapted for the fill fraction calculations, Dom Pesce and GregLindahl for their assistance with the filling fraction code,and Sara Issaoun, Mareki Honma, Lynn Matthews, and PeterGalison for their useful feedback on the manuscript. We ac-knowledge the significance that Maunakea, where the SMAand JCMT EHT stations are located, has for the Hawaiianpeople. This work was supported in part by the NationalScience Foundation (AST-1935980, AST-1716536, OISE-1743747), the Gordon and Betty Moore Foundation (GBMF-5278), and the Black Hole Initiative at Harvard University,through a grant (60477) from the John Templeton Founda-tion.

REFERENCES

Ananthasubramanian, P. G., Yamamoto, S., & Prabhu, T. P. 2002,International Journal of Infrared and Millimeter Waves, 23, 227,doi: 10.1023/A:1015018200990

Blackburn, L., Pesce, D. W., Johnson, M. D., et al. 2020, TheAstrophysical Journal, 894, 31, doi: 10.3847/1538-4357/ab8469

18

Figure 13. Imaging reconstructions of GRMHD simulations for M87∗ (from Chael et al. (2019), plotted in logarithmic scale flux density)and Sgr A∗ (from Chael et al. (2018a), plotted in linear scale flux density). The simulation panels show the combined 230 and 345 GHz fluxdistributions where the high-frequency band has been scaled to the lower one using a spectral index obtained from the source model. We showSgr A∗ as it would appear on the sky, scattered by the interstellar medium (Johnson 2016). Using SEFDs calculated in this meteorological studyunder median conditions for April, the 2017 and 2021 arrays are unable to reconstruct much of the forward-propagating jet in the M87∗ modelor reconstruct the Sgr A∗ brightness distribution with a 100 s snapshot observation. The ngEHT reconstructions use the array shown on theglobe, which augments the 2021 EHT with the leading sites based on the filling factor calculation in Fig. 12 plus sites that have good existinginfrastructure (BAJA, CAT, CNI, Dome C, GAM, GARS, GLT-S, LLA, NOR, NZ, OVRO, and SGO). The ngEHT reconstruction assumes a10 m antenna diameter for new stations and an 8 GHz bandwidth per sideband. The 230 and 345 GHz bands are combined in the ngEHTreconstructions using the spectral indices. The ngEHT dramatically improves the Fourier coverage and reconstructions, both for M87∗ and forSgr A∗ in snapshot mode.

Blackburn, L., Doeleman, S., Dexter, J., et al. 2019, arXiv e-prints.

https://arxiv.org/abs/1909.01411

Booth, R. S., Delgado, G., Hagstrom, M., et al. 1989, A&A, 216,

315

Castro-Almazán, J. A., Muñoz-Tuñón, C., García-Lorenzo, B.,

et al. 2016, in Observatory Operations: Strategies, Processes,

and Systems VI, ed. A. B. Peck, R. L. Seaman, & C. R. Benn,

Vol. 9910, International Society for Optics and Photonics

(SPIE), 227 – 236, doi: 10.1117/12.2232646

Chael, A., Narayan, R., & Johnson, M. D. 2019, Monthly Notices

of the Royal Astronomical Society, 486, 2873,

doi: 10.1093/mnras/stz988

Chael, A., Rowan, M., Narayan, R., Johnson, M., & Sironi, L.2018a, Monthly Notices of the Royal Astronomical Society,478, 5209, doi: 10.1093/mnras/sty1261

Chael, A. A., Johnson, M. D., Bouman, K. L., et al. 2018b, TheAstrophysical Journal, 857, 23, doi: 10.3847/1538-4357/aab6a8

Chael, A. A., Johnson, M. D., Narayan, R., et al. 2016, TheAstrophysical Journal, 829, 11,doi: 10.3847/0004-637x/829/1/11

Chamberlin, R. A. 2001, Journal of Geophysical Research:Atmospheres, 106, 20101, doi: 10.1029/2001JD900208

Doeleman, S., Agol, E., Backer, D., et al. 2009.https://arxiv.org/abs/0906.3899

Doeleman, S. S., Weintroub, J., Rogers, A. E. E., et al. 2008,Nature, 455, 78, doi: 10.1038/nature07245

19

Doeleman, S. S., Fish, V. L., Schenck, D. E., et al. 2012, Science,338, 355, doi: 10.1126/science.1224768

Event Horizon Telescope Collaboration, Akiyama, K., Alberdi, A.,et al. 2019a, ApJL, 875, L2, doi: 10.3847/2041-8213/ab0c96

—. 2019b, ApJL, 875, L1, doi: 10.3847/2041-8213/ab0ec7—. 2019c, ApJL, 875, L3, doi: 10.3847/2041-8213/ab0c57—. 2019d, ApJL, 875, L4, doi: 10.3847/2041-8213/ab0e85Ferrusca, D., & R., J. C. 2014, in Ground-based and Airborne

Instrumentation for Astronomy V, ed. S. K. Ramsay, I. S.McLean, & H. Takami, Vol. 9147, International Society forOptics and Photonics (SPIE), 1044 – 1053,doi: 10.1117/12.2055005

Fish, V. L., Doeleman, S. S., Beaudoin, C., et al. 2011,Astrophysical Journal Letters, 727,doi: 10.1088/2041-8205/727/2/L36

Gelaro, R., McCarty, W., Suárez, M. J., et al. 2017, Journal ofClimate, 30, 5419, doi: 10.1175/JCLI-D-16-0758.1

Graham, E., Vaughan, R., Buckley, D. A., & Tirima, K. 2016,Theoretical and Applied Climatology, 124, 425,doi: 10.1007/s00704-014-1366-x

Grasdalen, G. L., Gehrz, R. D., Hackwell, J. A., & Freedman, R.1985, Publications of the Astronomical Society of the Pacific,97, 1013, doi: 10.1086/131657

Ishiguro, M., Tanaka, H., Morita, K. ., et al. 1986, in 1986Antennas and Propagation Society International Symposium,Vol. 24, 531–534, doi: 10.1109/APS.1986.1149653

Johnson, M. D. 2016, The Astrophysical Journal, 833, 74,doi: 10.3847/1538-4357/833/1/74

Johnson, M. D., Fish, V. L., Doeleman, S. S., et al. 2015, Science,350, 1242, doi: 10.1126/science.aac7087

Johnson, M. D., Narayan, R., Psaltis, D., et al. 2018, ApJ, 865,104, doi: 10.3847/1538-4357/aadcff

Kerber, F., Querel, R. R., & Neureiter, B. 2015, Journal of Physics:Conference Series, 595, 012017,doi: 10.1088/1742-6596/595/1/012017

Kimura, K., Aton, K., Baybayan, K., et al. 2019, in BAAS, Vol. 51,135

Lee, S.-S., Byun, D.-Y., Oh, C. S., et al. 2011, Publications of theAstronomical Society of the Pacific, 123, 1398,doi: 10.1086/663326

Marvil, J., Ansmann, M., Childers, J., et al. 2006, New Astronomy,11, 218, doi: 10.1016/j.newast.2005.06.005

Matsushita, S., & Matsuo, H. 2003, Publications of theAstronomical Society of Japan, 55, 325,doi: 10.1093/pasj/55.1.325

Matsushita, S., Asada, K., Martin-Cocher, P. L., et al. 2017,Publications of the Astronomical Society of the Pacific, 129, 1,doi: 10.1088/1538-3873/129/972/025001

Paine, S. 2018, BH PIRE Winter School 2018 tutorial: RadioClimatology. https://bhpire.arizona.edu/education/tutorials/

—. 2019, The am atmospheric model, 11.0, Zenodo,doi: 10.5281/zenodo.3406496

Palumbo, D. C. M., Doeleman, S. S., Johnson, M. D., Bouman,K. L., & Chael, A. A. 2019, The Astrophysical Journal, 881, 62,doi: 10.3847/1538-4357/ab2bed

Pearson, T. J., & Readhead, A. C. S. 1984, ARA&A, 22, 97,doi: 10.1146/annurev.aa.22.090184.000525

Querel, R. R., Kerber, F., Curto, G. L., et al. 2010, in Ground-basedand Airborne Telescopes III, ed. L. M. Stepp, R. Gilmozzi, &H. J. Hall, Vol. 7733, International Society for Optics andPhotonics (SPIE), 1501 – 1512, doi: 10.1117/12.856480

Quesada, J. A. 1989, Publications of the Astronomical Society ofthe Pacific, 101, 441, doi: 10.1086/132454

Radford, S. J. E., & Peterson, J. B. 2016, Publications of theAstronomical Society of the Pacific, 128, 075001,doi: 10.1088/1538-3873/128/965/075001

Ricaud, P., Grigioni, P., Zbinden, R., et al. 2015, Antarctic Science,27, 598, doi: 10.1017/S0954102015000334

Roelofs, F., Janssen, M., Natarajan, I., et al. 2020, Astronomy &Astrophysics, 636, A5, doi: 10.1051/0004-6361/201936622

Rogers, A., Barvainis, R., Charpentier, P., & Corey, B. 1993, IEEETransactions on Antennas and Propagation, 41, 77,doi: 10.1109/8.210118

Rowan, M. E., Sironi, L., & Narayan, R. 2017, The AstrophysicalJournal, 850, 29, doi: 10.3847/1538-4357/aa9380

Sault, R. J., & Wieringa, M. H. 1994, A&AS, 108, 585.https://ui.adsabs.harvard.edu/abs/1994A&AS..108..585S

Sekimoto, Y., Yoshida, H., Hirota, T., et al. 1996, InternationalJournal of Infrared and Millimeter Waves, 17, 1263,doi: 10.1007/BF02088912

Sekimoto, Y., Arikawa, Y., Aso, Y., et al. 2000, 4015, 185,doi: 10.1117/12.390411

Suen, J. Y., Fang, M. T., Denny, S. P., & Lubin, P. M. 2015, IEEETransactions on Terahertz Science and Technology, 5, 299,doi: 10.1109/TTHZ.2015.2399694

Suen, J. Y., Fang, M. T., & Lubin, P. M. 2014, IEEE Transactionson Terahertz Science and Technology, 4, 86,doi: 10.1109/TTHZ.2013.2294018

Thompson, A. R., Moran, J. M., & Swenson, G. W. 2017,Interferometry and Synthesis in Radio Astronomy, Astronomyand Astrophysics Library (Cham: Springer InternationalPublishing), doi: 10.1007/978-3-319-44431-4

Tremblin, P., Schneider, N., Minier, V., Durand, G. A., & Urban, J.2012, Astronomy and Astrophysics, 548, 1,doi: 10.1051/0004-6361/201220420

Usoff, J. M., Clarke, M. T., Liu, C., & Silver, M. J. 2014, LincolnLaboratory Journal, 21, 83. https://www.ll.mit.edu/sites/default/files/page/doc/2018-05/21_1_6_Usoff.pdf

20

Wallace, L., & Livingston, W. 1984, Publications of theAstronomical Society of the Pacific, 96, 182,doi: 10.1086/131320