Cosmic Explorer Location Search (CELS)

• CELS is a multi-criteria framework integrating scientific performance, cost, and socio-cultural factors for Cosmic Explorer site evaluation.
• It employs Python-based GIS and cost surrogate modeling to analyze terrain, land cover, and interferometer geometry for optimal sensitivity.
• The framework supports a two-phase search—from national-scale screening to community-engaged site visits—for sustainable, high-performance observatory development.

Cosmic Explorer Location Search (CELS) is the location-evaluation framework developed for Cosmic Explorer (CE), the proposed U.S. next-generation ground-based gravitational-wave observatory. In the CE literature, the term denotes both a Python-based national-scale screening code for candidate interferometer geometries and the broader multidisciplinary siting program that integrates scientific performance, construction cost, access, and social criteria. Its immediate purpose is to identify locations in the conterminous United States that can host CE’s reference observatory concepts—one 40 km L-shaped detector and one 20 km L-shaped detector—while preserving the low-frequency sensitivity, network geometry, and long-term operational stability required of a third-generation observatory (Datrier et al., 5 Sep 2025, Daniel et al., 2024).

1. Origins in the Cosmic Explorer observatory concept

CELS emerged from the fact that Cosmic Explorer is not a straightforward scale-up of the existing 4 km LIGO facilities. CE’s reference design calls for two widely separated L-shaped detectors in the United States, with arm lengths of 40 km and 20 km, using technology extrapolated from Advanced LIGO-class interferometers. The scientific motivation is the usual arm-length scaling of strain sensitivity: increasing the baseline from 4 km to 20–40 km increases reach by roughly an order of magnitude in distance and much more in accessible source volume, but it also creates a siting problem of fundamentally different scale (Datrier et al., 5 Sep 2025).

The underlying observatory requirements make site selection a first-order systems problem. CE is conceived as a surface installation on “flat and seismically quiet land in the U.S.” with two 40 km arms in the original white-paper framing, later generalized into the 40 km plus 20 km two-site reference concept. Over a 40 km straight arm, Earth curvature produces a sagitta of about 30 m relative to the geoid, so arm construction is coupled directly to terrain, excavation, embankment design, and long-term geotechnical stability. At the same time, CE’s targeted sensitivity down to approximately 5 Hz makes seismic motion, Newtonian noise, infrasound, and anthropogenic disturbance central site variables rather than secondary engineering nuisances (Reitze et al., 2019, Evans et al., 2021).

This context explains why CE siting was formalized as a staged program rather than an isolated civil-engineering exercise. The CE site-evaluation effort uses remote suitability analysis, site visits and relationship building, and then on-site physical and socio-cultural assessments. Within that architecture, the National Suitability Analysis (NSA) emphasizes scientific suitability together with quality-of-life and socio-cultural factors, while CELS supplies the geometry- and cost-aware screening of candidate interferometer layouts (Datrier et al., 5 Sep 2025).

2. Scope, criteria, and staged search process

As a programmatic framework, CELS evaluates candidate locations through three interconnected classes of criteria: science performance requirements, cost boundaries and access, and social factors. The stated objective is not only to identify technically buildable sites, but to develop, through engagement and co-design with local and Indigenous communities, a short list of mutually acceptable and scientifically excellent candidate regions for the 40 km and 20 km observatories (Daniel et al., 2024).

The physical footprint requirements are unusually explicit. A candidate site must support a straight L-shaped geometry with nominal arm lengths of approximately 40 km or 20 km; a cleared corridor of about 75 m width is assumed along each arm for the beam tubes, access roads, mid-stations, vacuum pump stations, and environmental or engineering features; and the campus area near the vertex is about $300\times 300~\mathrm{m}$, likely offset by about $1~\mathrm{km}$ from the interferometer vertex to lessen human noise coupling. CELS also treats modest deviations from the nominal geometry as admissible design degrees of freedom: opening angles outside exact orthogonality and small arm-length reductions can enlarge the candidate set, but are scientifically disfavored once they drive more than roughly a 10% reduction in representative sensitivity metrics (Daniel et al., 2024).

The search itself is organized in two phases. Phase 1 is a national-scale suitability analysis based on public datasets over the continental United States. It overlays digital elevation models, land cover, seismic datasets, seismic hazard maps, climate-risk indices, geological information, transportation and utility networks, land ownership, and demographic or cultural data to identify regions of interest. Phase 2 consists of introductory visits, relationship building, and—only with community permission—local physical assessments such as seismic-array deployment, geotechnical investigation, environmental scoping, and refined costing. The pace of progression is stated to depend not only on technical milestones but also on the pace at which trust and mutual interest develop (Daniel et al., 2024).

A central consequence of this framing is that CELS is not a single scalar optimization over topography. The literature explicitly treats site evaluation as a coupled problem in which construction geometry, environmental noise, land access, workforce viability, Indigenous rights, environmental review, and long-term community relationships can all function as constraints or ranking criteria. This suggests a multi-criteria decision structure even where a final explicit global figure of merit is not written down (Daniel et al., 2024).

3. The CELS code: GIS representation, search strategy, and cost surrogates

In its narrower technical sense, CELS is the Python package used for national-scale screening of CE layouts. The code represents candidate corner-station locations over the conterminous United States, places an L-shaped interferometer of specified arm length and opening angle at each grid point, samples GIS layers along the arms and site footprint, and computes a scalar score combining land-cover cost, earthworks cost, and science-related geometric penalties. Its current published focus is a 40 km detector with nominally $90^\circ$ opening angle, evaluated across many arm orientations and reduced at each map pixel to the minimum cost over rotations (Datrier et al., 5 Sep 2025).

The data inputs in the current implementation are deliberately limited and uniform at national scale. Elevation is taken from the USGS National Map 3D Elevation Program (3DEP) at 1 arc-second resolution, corresponding to roughly 30 m horizontal resolution. Land cover is taken from the National Land Cover Database (NLCD) 2021 at 30 m resolution. These layers are used to compute the cut-and-fill required to realize a level optical path, assess the bowl geometry induced by Earth curvature, and assign land-cover-dependent costs with prohibitively high penalties for open water and densely developed areas, and more favorable values for rural, agricultural, or undeveloped land. The papers emphasize that full lithology and soil-mechanics layers are not yet included, although work is underway to include soil and rock type (Datrier et al., 5 Sep 2025).

The cost model is explicitly surrogate rather than detailed. For land cover, the description is

$C_{\rm land} = \int_A c_{\rm LC}(x,y)\, dA,$

implemented in practice as a raster sum over the relevant footprint. For elevation, the paper writes

$$C_{\rm elev} = \frac{V_{\rm cut} + V_{\rm fill} + \left|V_{\rm cut} - V_{\rm fill}\right|}{10~\text{USD}},$$

where $V_{\rm cut}+V_{\rm fill}$ represents material moved on site and $|V_{\rm cut}-V_{\rm fill}|$ represents surplus or deficit that must be hauled away or imported. The code also imposes a cap for very deep cuts, beyond which tunneling is assumed cheaper than trenching. The authors state that these are relative screening scores rather than final engineering estimates (Datrier et al., 5 Sep 2025).

CELS also incorporates geometric science penalties. For a 40 km arm on a perfectly spherical Earth, the minimum mirror tilt required even at a zero-elevation site is

$$\theta_0 = \frac{L_{\rm arm}}{2R_\oplus} \approx 3~\text{mrad},$$

and the tilt score is defined as

$$C_{\rm tilt} = 10\left[\left(\frac{\theta_x}{\theta_0}\right)^2 + \left(\frac{\theta_y}{\theta_0}\right)^2\right].$$

This term is not a monetary cost; it is a science penalty associated with vertical-to-horizontal coupling induced by tilt. The current code combines land-cover cost, elevation cost, and tilt score into a scalar configuration score that is minimized over orientations for each corner-station pixel (Datrier et al., 5 Sep 2025).

A related cost analysis developed for the broader siting program supplies an explicit trench-volume scaling that clarifies why geometry relaxation is valuable. For a 40 km arm cut as a straight trench on a constant-elevation surface, the maximum mid-arm depth is approximately

$$d_{\max}\approx \frac{L^2}{8R}\approx 31~\mathrm{m},$$

and an approximate excavation volume is

$1~\mathrm{km}$0

with $1~\mathrm{km}$1. This steep scaling underlies the CELS interest in slightly shorter arms, modestly non-right opening angles, and naturally bowl-shaped terrain that can reduce earthworks by orders of magnitude (Daniel et al., 2024).

4. Science-driven geometry and network constraints

CELS is constrained not only by local buildability but also by the role of CE in a global third-generation detector network. The CE Science Traceability Matrix motivates the integration of astrophysical and network-level requirements into site evaluation. In the published CELS implementation, the science factors already included or under active integration are tilt, deviations of the opening angle from $1~\mathrm{km}$2, and arm-length reductions from 40 km. The sensitivity proxy for a geometry relative to a 40 km, $1~\mathrm{km}$3 reference detector is written as

$1~\mathrm{km}$4

reflecting the proportionality of interferometer response to $1~\mathrm{km}$5. The current search considers opening angles in the range $1~\mathrm{km}$6 and arm-length reductions up to about $1~\mathrm{km}$7, corresponding to an approximately 10% strain-amplitude penalty when combined with angle deviations (Datrier et al., 5 Sep 2025).

More broadly, the siting literature treats CE as part of a multi-observatory network whose science depends on baselines, orientations, and detector count. The Horizon Study is explicit that the reference concept is two widely separated U.S. facilities, one 40 km and one 20 km, and that the observatories should neither be co-located nor parallel if accurate sky localization, polarization disentangling, and transient-noise rejection are to be achieved. It further contrasts CE-alone configurations, CE with second-generation detectors, CE with Einstein Telescope (ET), and CE plus ET plus a southern-hemisphere partner, framing siting as a coupled network-design problem rather than a search for a locally optimal single site (Evans et al., 2021).

The network trade studies sharpen this point quantitatively. One analysis concludes that two Cosmic Explorer observatories are indispensable for precise localization of binary neutron star events and for transforming multimessenger astronomy, while the addition of ET is critical for accomplishing all identified science metrics. In that study, the 4020A network—CE-A at 40 km, CE-B at 20 km, and one A$1~\mathrm{km}$8 detector—yields far more well-localized binary neutron star events than one-CE networks, and the 4020ET network improves those numbers again by roughly another factor of several for the sharpest localization metrics (Gupta et al., 2023).

A separate Bayesian parameter-estimation study isolates what a single 40 km CE can do without network triangulation. When Earth rotation and finite arm-length response are fully modeled, a single CE can localize some high-SNR binary neutron star mergers to 90% sky areas as small as approximately $1~\mathrm{km}$9–$90^\circ$0, but typical events span a few hundred to a few thousand square degrees, and the worst cases reach roughly $90^\circ$1. This result is significant for CELS because it shows that CE geometry influences localization even at a single site, but it also reinforces the network literature’s central claim that precise localization depends primarily on having multiple third-generation baselines (Baral et al., 2023).

The practical implication is that CELS must select not only individually favorable land parcels but also a pair of observatories with continental-scale separation, non-degenerate relative orientations, and favorable complementarity with ET, LIGO-India, and any future southern-hemisphere detector. The siting criteria therefore include both local metrics—topography, tilt, seismicity, anthropogenic noise—and network metrics such as latitude, longitude, and arm orientation relative to the existing and expected global array (Daniel et al., 2024).

5. Environmental, access, and socio-cultural criteria

The published siting criteria treat the environmental and human landscape as co-equal with construction geometry. For local physical performance, CELS considers seismic hazard from the USGS 2023 National Seismic Hazard Model, ambient seismic spectra from approximately $90^\circ$2 to $90^\circ$3, local geology and subsurface structure, hydrology, wind climatology, climate-vulnerability indices, and future land-use change. The CE sensitivity model assumes Rayleigh-wave-dominated ground acceleration of about $90^\circ$4 and ambient infrasound of about $90^\circ$5, while the target environmental background is described as being at or near the Peterson Low Noise Model for seismic noise. These assumptions link siting directly to the feasibility of low-frequency isolation, Newtonian-noise subtraction, uptime, and long-term resilience under climate change (Daniel et al., 2024, Evans et al., 2021).

Access and infrastructure are treated as both cost drivers and operational constraints. CE requires year-round road access, the ability to ship large vacuum and observatory components, access to regional airports, grid infrastructure capable of supplying roughly $90^\circ$6 average observatory power excluding large computing, and high-speed data networking with fiber redundancy. Yet the site must remain far enough from major transport corridors and dense development to preserve a quiet environmental-noise budget. The literature repeatedly frames this as a remoteness-versus-accessibility trade: very remote sites may be quiet but logistically fragile, whereas locations that are too close to population centers become scientifically compromised (Daniel et al., 2024, Evans et al., 2023).

The distinctive feature of CELS, however, is the explicit elevation of social and Indigenous criteria into the core search logic. The criteria paper states that CE aims to approach the lifecycle of a large scientific facility in a way that prioritizes mutually beneficial relationships with local and Indigenous communities. Site evaluation therefore includes demographic, ethnographic, economic, quality-of-life, land-tenure, and cultural data, together with formal attention to Indigenous sovereignty, historical and current ties to land, treaty and jurisdictional issues, and culturally significant places. The Indigenous and Place-based Partnerships & Responsible Siting (IPP-RS) team is described as a central component of the search, guided by UNDRIP, ADRIP, and Free, Prior, and Informed Consent principles, and committed to the possibility of walking away from otherwise favorable sites if Indigenous communities deem the project incompatible with their values or rights (Daniel et al., 2024).

This social framing is not peripheral to the location search. The CE policy and planning documents state that site identification and evaluation must consider environmental, cultural, socio-economic, and political impacts alongside science and cost, and that CE cannot be built and operated without ongoing local consent. A common misconception is therefore that CELS is simply a topographic optimizer to be followed later by community engagement. The published framework rejects that sequencing: relationship building begins during the design phase, and locations that fail the relational threshold need not advance to expensive on-site surveys (Evans et al., 2023, Daniel et al., 2024).

6. National-scale results, limitations, and planned evolution

The first published national-scale application of the CELS code presents maps for a 40 km CE with $90^\circ$7 opening angle in which each pixel represents the logarithm of the minimum cost score over all arm rotations at a candidate corner-station location. Yellow regions are relatively favorable and blue regions highly unfavorable. The paper does not provide a textual ranking of U.S. regions, but it states that the broader CE site-evaluation team has identified an initial list of 26 suitable locations for a 40 km CE, shown as pink buffer outlines in combination with results from the National Suitability Analysis. These locations are explicitly characterized as draft and subject to change as further searches and site visits proceed (Datrier et al., 5 Sep 2025).

The published authors also emphasize that CELS remains a screening tool rather than a final design or site-selection instrument. Its current layers are limited chiefly to elevation and land cover; soil and rock type are not yet included in the implemented code; roads, utilities, and explicit infrastructure layers are not yet part of the main scoring function; and only a subset of science drivers—tilt and simple geometric response penalties—have been encoded as formal configuration scores. Likewise, the use of approximately 30 m raster data is adequate for national comparison but not for site-scale design, where local ravines, drainage, property boundaries, and detailed geotechnical conditions become decisive (Datrier et al., 5 Sep 2025).

Planned extensions follow directly from these limitations. The literature identifies richer cost modeling, explicit inclusion of soil and rock types, automated variation of arm length and opening angle through the $90^\circ$8 metric, tighter integration of STM-derived science factors such as seismic and anthropogenic noise maps and network performance with ET, deeper coupling to NSA outputs, and a robust national-scale search for the 20 km CE20 configuration. In programmatic terms, the intended trajectory is a progression from GIS-based national screening to a smaller set of well-characterized candidates in the late 2020s, with NSF-facing reports that integrate quantitative physical ranking, cost and access analysis, and documentation of community partnerships and consent status (Datrier et al., 5 Sep 2025, Daniel et al., 2024).

CELS is thus best understood as the formalization of CE siting into a reproducible, multi-layered research program. At one level it is a geometry- and cost-aware GIS engine that scans the United States for plausible interferometer layouts. At another, it is the institutional framework by which CE attempts to reconcile low-frequency gravitational-wave performance, large-scale civil construction, network optimization, environmental stewardship, and responsible relations with host communities. The published work suggests that the mature version of CELS will not merely identify where CE can be built, but where it can be built in a way that remains scientifically effective, socially legitimate, and operationally resilient over the observatory’s full lifetime (Datrier et al., 5 Sep 2025, Daniel et al., 2024).