---
title: ExoREM Atmospheric Model
url: https://www.emergentmind.com/topics/exorem
type: topic
---

# ExoREM Atmospheric Model

ExoREM, often written Exo-REM, is a one-dimensional atmospheric forward model and family of pre-computed grids for self-luminous giant exoplanets and brown dwarfs. It was introduced as the “Exoplanet Radiative-convective Equilibrium Model” to interpret the photometry and spectroscopy of directly imaged planets, with inputs given by surface gravity, effective temperature, and elemental composition, and outputs that include an equilibrium temperature profile, gas mixing-ratio profiles, and synthetic spectra [1504.04876]. In later applications, ExoREM is described as a radiative–convective equilibrium model developed specifically for young giant companions and L/T-transition objects, with cloud treatments, variable metallicity and C/O, and, in more recent grids, explicit cloud sedimentation control and a two-column approximation for heterogeneous cloud cover [2507.08961; 2509.22803; 2605.29070].

## 1. Origins and model scope

In its 2015 formulation, Exo-REM was designed for directly imaged, self-luminous giant exoplanets, especially young, low-gravity objects observed with instruments such as SPHERE and GPI. The model was intentionally constructed with a “minimal number of assumptions and parameters,” and the initial implementation neglected stellar irradiation, assumed thermochemical equilibrium for the major species, and added absorption by iron and silicate cloud particles above the expected condensation levels with a fixed scale height and a given optical depth at a reference wavelength; scattering was not included at that stage [1504.04876].

Subsequent work broadened both the scientific domain and the internal physics. Papers using the Charnay et al. public grids describe ExoREM as a 1D radiative-convective equilibrium model for brown dwarfs and young giant exoplanets, with a cloud prescription tuned to reproduce dusty L/T-transition objects and, in some versions, disequilibrium chemistry induced by vertical mixing with “simple microphysics” [2310.00148; 2509.22803]. The model family has therefore evolved from a simplified direct-imaging forward model into a broader atmospheric framework used across warm dusty planets, cool T dwarfs, and JWST-era mid-infrared studies.

The 2026 “Exo-REM k26” release extends this trajectory explicitly. It introduces a cloudy grid with a free sedimentation parameter $f_{\rm sed}$, updated opacities and abundances, strict convergence criteria that remove unstable solutions, and a two-column framework intended to emulate patchy clouds in highly variable substellar atmospheres [2605.29070]. This suggests that ExoREM has become not merely a static grid library but a modeling platform whose later generations are shaped by the failure modes exposed by JWST-quality data.

## 2. Physical formulation

ExoREM is fundamentally a 1D, plane-parallel, hydrostatic radiative–convective equilibrium model. In the original description, the atmosphere is discretized on 64 levels equally spaced in $\ln p$ between 50 bar and 0.01 mbar, with net flux constrained to remain constant with depth and equal to the intrinsic thermal flux of a self-luminous planet [1504.04876]. The two central structural relations are hydrostatic balance,
$$
\frac{dP}{dz} = -\rho g,
$$
and constant total flux,
$$
\pi F(p) = \sigma T_\mathrm{eff}^4,
$$
where $\sigma$ is the Stefan–Boltzmann constant [1504.04876].

The radiative transfer is solved without scattering in the original implementation, using correlated-$k$ opacities over the interval 20–16000 cm$^{-1}$ in 20 cm$^{-1}$ bins, with 16 quadrature points per spectral interval [1504.04876]. Later descriptions preserve the same conceptual basis: ExoREM iterates the pressure–temperature profile until radiative and convective energy transport balance through the column, assuming hydrostatic equilibrium and net-flux conservation, and generally neglecting stellar heating for widely separated, self-luminous companions [2507.08961; 2509.22803].

The model therefore belongs to the class of self-consistent forward atmosphere solvers rather than parametric retrieval frameworks. For a chosen set of bulk parameters—typically $T_{\rm eff}$, $\log g$, metallicity, and C/O—it computes a self-consistent atmospheric structure and emergent spectrum. Derived quantities such as radius, luminosity, and mass are usually obtained outside the atmospheric solver by scaling the model flux to the measured absolute flux and then applying relations such as
$$
L = 4\pi R^2 \sigma T_{\rm eff}^4.
$$
This workflow is explicit in several ExoREM-based analyses, including HIP 65426 b and HR 2562 B [2310.00148; 2409.04524].

## 3. Chemistry, opacities, and cloud treatments

Chemistry and clouds are the defining axes along which ExoREM has evolved. In the initial model, gas abundances were computed in thermochemical equilibrium, with opacity from H$_2$–He collision-induced absorption and molecular or atomic lines from eight main compounds, including CH$_4$ with an ExoMol line list, while cloud opacity was supplied by iron and silicate condensates [1504.04876]. The 2015 paper emphasized simplicity: no scattering, no explicit vertical mixing, and a highly parameterized cloud prescription.

Later public grids used in applications to dusty young planets are described as self-consistent atmospheres with clouds and chemistry coupled to the temperature–pressure structure. In HIP 65426 b, ExoREM is treated as computing radiative–convective equilibrium and cloud formation self-consistently for given $T_{\rm eff}$, $\log g$, metallicity, and C/O, with metallicity and C/O treated as explicit grid parameters and a cloud model tuned for dusty L/T-transition atmospheres [2310.00148]. In AB Pic b, ExoREM is described as including iron and silicate clouds and non-equilibrium chemistry for CO, CH$_4$, CO$_2$, and H$_2$ due to vertical mixing, with equilibrium abundances for the remaining species [2211.01474]. In Ross 458c, the adopted public ExoREM grid is summarized as combining clouds, vertical mixing, variable metallicity, and variable C/O in radiative–convective equilibrium, making it the most physically elaborate forward grid among those compared in that study [2509.22803].

The cloud treatment is especially important. Early ExoREM parameterized cloud absorption with a reference optical depth and mean particle size, extending each cloud between its condensation pressure and $p_{\rm c}/100$ [1504.04876]. Later work instead highlights a self-consistent treatment tuned to L/T-transition objects, but also notes that the cloud microphysics remains simplified and can miss important physics, especially in the 3–5 $\mu$m and 8–10 $\mu$m regimes [2310.00148; 2507.08961]. In TWA 27b, ExoREM’s clouds are strong enough to redden the near-infrared SED and suppress the CH$_4$ absorption feature at 3.3 $\mu$m, yet the same grids do not reproduce the observed 8–10 $\mu$m silicate absorption band [2507.08961].

Exo-REM k26 formalizes cloud sedimentation through the Ackerman–Marley-style parameter
$$
f_{\rm sed} = \frac{v_{\rm sed} H}{K_{zz}},
$$
and embeds this within ExoREM’s radiative–convective framework [2605.29070]. Low $f_{\rm sed}$ yields vertically extended, optically thick clouds with small particles; high $f_{\rm sed}$ yields thin, rapidly sedimenting clouds. The k26 release also adds a “simple microphysics” alternative, updates molecular and atomic opacities, corrects an erroneous CH$_3$D abundance that had biased spectra of low-$T_{\rm eff}$ objects, and uses a total extinction coefficient
$$
\kappa_{\rm ext}(\lambda,z) = \kappa_{\rm CIA} + \kappa_{\rm Rayleigh} + \kappa_{\rm lines} + \kappa_{\rm clouds}
$$
in its modern radiative transfer [2605.29070].

## 4. Fitting workflows and software ecosystems

ExoREM is not itself an inference engine; in practice it is embedded in external fitting pipelines. The resulting workflows vary substantially across studies, but most use ExoREM as a discrete or interpolated grid of self-consistent spectra.

For HIP 65426 b, ExoREM spectra were fit with the Python package **species** and **pyMultinest**, using 1000 live points and a Gaussian likelihood over all photometric and spectroscopic data. The fit jointly sampled $T_{\rm eff}$, $\log g$, $[\mathrm{Fe/H}]$, C/O, radius, parallax, two Gaussian-process hyperparameters for correlated SPHERE IFS noise, and a nuisance radial velocity for the SINFONI spectrum. The SPHERE covariance was modeled with a squared-exponential kernel,
$$
k(\lambda,\lambda') = A_{\rm SPHERE}^2 \exp\left(-\frac{(\lambda-\lambda')^2}{2l_{\rm SPHERE}^2}\right),
$$
and the paper stressed that interpolation errors in the non-linear ExoREM grid were not propagated into the formal parameter uncertainties [2310.00148].

For TWA 27b, ExoREM was again interfaced through **species**, but the nested sampling used **PyMultinest** with 2000 live points and fit the full 1–15 $\mu$m SED, explicitly weighting sparse MIRI photometric points to contribute comparably to the likelihood relative to dense NIRSpec spectra [2507.08961]. In AB Pic b, ExoREM was incorporated into **ForMoSA**, a nested-sampling forward-modeling framework using 500 living points, with parameters including $T_{\rm eff}$, $\log g$, $[\mathrm{M/H}]$, C/O, radius, radial velocity, $v\sin i$, and extinction $A_V$ [2211.01474]. For Ross 458c, the authors first completed the ExoREM grid by interpolating missing nodes, then performed a least-squares scan, continuous optimization with SciPy, and an **emcee** run with explicit error inflation,
$$
s^2(\lambda) = \sigma^2(\lambda) + (x_{tol}F_\lambda(\lambda))^2 + 10^b,
$$
inside the log-likelihood [2509.22803]. For HR 2562 B, ExoREM was used in a simpler $\chi^2$-map workflow over a coarse grid recently incorporated into **species**, without MCMC and with uncertainties dominated by grid spacing $\Delta T=50$ K and $\Delta\log g=0.5$ [2409.04524].

These heterogeneous inference strategies are themselves instructive. They show that ExoREM is most often used as a physically constrained forward prior on atmospheric structure, while the statistical machinery—nested sampling, MCMC, Gaussian processes, or coarse $\chi^2$ scans—resides outside the atmospheric code. A plausible implication is that reported ExoREM parameter uncertainties are often as sensitive to interpolation strategy, grid completeness, and nuisance-noise modeling as to the underlying atmosphere physics.

## 5. Representative applications

The literature supplied here uses ExoREM across a wide span of objects: young L-type directly imaged planets, L/T-transition companions, cool T dwarfs, and highly variable dusty planetary-mass companions observed with JWST. The table summarizes representative uses.

| Object | Role of ExoREM | Salient result |
|---|---|---|
| $\beta$ Pictoris b | Forward model for GRAVITY K band and GPI YJH | With a mass prior and variable C/O, ExoREM gave $T=1590\pm20$ K, $\log(g/g_0)=4.0$, $[\mathrm{Fe}/\mathrm{H}]=+0.5$, and $\mathrm{C/O}=0.43\pm0.05$ [1912.04651] |
| AB Pic b | ForMoSA forward modeling of broad SED and high-resolution bands | ExoREM gave $T_\mathrm{eff}=1700\pm50$ K, $\log g=4.5\pm0.3$ dex, $[\mathrm{M/H}]=0.36\pm0.20$, and $\mathrm{C/O}=0.58\pm0.08$ [2211.01474] |
| HIP 65426 b | Multi-instrument SED fit with species/pyMultinest | Best-fit ExoREM model had $T_{\rm eff}=1337\pm9$ K, $\log g=3.52^{+0.03}_{-0.02}$, $R=1.51^{+0.03}_{-0.02}R_{\rm J}$, $[\mathrm{Fe/H}]=0.15^{+0.08}_{-0.10}$, and ${\rm C/O}=0.595^{+0.008}_{-0.009}$ [2310.00148] |
| TWA 27b (2M1207 b) | Full 1–15 $\mu$m JWST/NIRSpec+MIRI fit | ExoREM+ext gave $T_\mathrm{eff}=1400$ K, $\log g=4.0$, $R_p=1.2R_J$, $\mathrm{C/O}=0.55$, $[\mathrm{Fe/H}]=-0.2$, and $A_V=3.7$ mag [2507.08961] |
| Ross 458c | Preferred forward grid in a grid-vs-retrieval comparison | ExoREM favored $T_{\mathrm{eff}}\approx650$ K, $\log g\approx4.4$, $R\approx1.3R_J$, $[\mathrm{M/H}]\approx+0.25$, and C/O $\approx0.64$ [2509.22803] |
| HR 2562 B | Cloudy comparison model against ATMO | Full-SED ExoREM fit gave $T_\mathrm{eff}=1252^{+25}_{-25}$ K, $\log g=4.80^{+0.25}_{-0.25}$, $R=1.04^{+0.10}_{-0.10}R_{\rm Jup}$, and $M=28^{+11}_{-11}M_{\rm Jup}$ [2409.04524] |
| VHS 1256 b | Exo-REM k26 two-column fit for patchy clouds | A $\sim$60–40% split of thick and thin clouds, with $f_{\rm sed,1}\simeq0.7$ and $f_{\rm sed,2}\simeq2.4$, reproduced the strong 10 $\mu$m silicate absorption [2605.29070] |

Across these applications, ExoREM repeatedly serves as a bridge between atmospheric spectroscopy and formation or evolution arguments. For $\beta$ Pic b, agreement between ExoREM and a petitRADTRANS retrieval on $\mathrm{C/O}=0.43$ supported the interpretation that the planet formed by core accretion with strong planetesimal enrichment [1912.04651]. For AB Pic b, ExoREM’s K-band fit to the CO bandheads yielded the first C/O estimate for the companion and was incorporated into a discussion of core accretion, gravitational instability, and possible outward scattering [2211.01474]. For HIP 65426 b, slightly super-solar metallicity and C/O $\approx0.6$, combined with GRAVITY astrometry disfavouring high eccentricity, were used to argue for a tentative picture of core accretion beyond the CO snowline and scattering before disk dispersal, rather than scattering after disk dispersal [2310.00148].

In the JWST era, ExoREM has also become a diagnostic of model inadequacy. In TWA 27b, it matched the global SED better than ATMO, suppressed the unobserved 3.3 $\mu$m methane feature, and provided the best formal fit among the tested self-consistent grids, yet still failed at the 8–10 $\mu$m silicate band and required an external extinction component plus a separate blackbody to describe the 15 $\mu$m excess interpreted as circumplanetary-disk emission [2507.08961]. In VHS 1256 b, by contrast, Exo-REM k26 was explicitly designed to address precisely that class of failure by allowing very low $f_{\rm sed}$ values and heterogeneous cloud cover, enabling reproduction of the strong JWST silicate feature [2605.29070].

## 6. Performance, limitations, and current directions

A consistent theme in the supplied literature is that ExoREM is physically informative but often systematics-limited. HIP 65426 b is the clearest statement of this point: the abstract explicitly warns that the metallicity and C/O values are “subject to interpolation errors as well as potentially missing model physics,” and the paper notes that the very tight formal posteriors do not include interpolation uncertainty in the non-linear spectral grid [2310.00148]. The same study emphasizes missing cloud microphysics, equilibrium-chemistry assumptions, and possible opacity incompleteness, with underprediction of flux beyond 3 $\mu$m treated as evidence for such shortcomings.

The mid-infrared has been a particularly stringent stress test. In TWA 27b, ExoREM’s silicate clouds were sufficient to redden the continuum and mute methane, but “none of the models reproduce the 8–10 µm silicate cloud feature,” and the authors left that problem to future retrieval work [2507.08961]. In HR 2562 B, ExoREM’s cloud-rich framework yielded masses and gravities larger than the astrometric constraints supported, leading the authors to conclude that a cloud-dominated ExoREM atmosphere is not the most representative description of that object and that HR 2562 B is more likely near cloud-free in the observed layers [2409.04524]. In Ross 458c, ExoREM was the preferred forward grid, but it still required substantial error inflation and gave a metallicity enhancement relative to the host that the authors ultimately judged less reliable than the near-stellar metallicity obtained with the POSEIDON retrieval [2509.22803].

These issues have motivated two distinct responses. One response is methodological: several papers explicitly call for atmospheric retrievals, either to validate ExoREM grid-based metallicity and C/O inferences or to add flexible cloud and chemistry parameterizations not available in the pre-computed grids [2310.00148; 2507.08961; 2509.22803]. The other response is architectural: Exo-REM k26 modifies the model itself by adding a sedimentation parameter, updated alkali and methane-isotopologue opacities, strict convergence criteria that reject unstable solutions, and a two-column cloud-heterogeneity prescription [2605.29070]. The GJ 504 b revision in that paper, driven by a corrected CH$_3$D abundance and updated opacity treatment, shows that apparently small changes in line lists or isotopologue bookkeeping can materially alter inferred $T_{\rm eff}$ and $\log g$ for cool objects [2605.29070].

ExoREM therefore occupies a specific position in current substellar-atmosphere research. It is neither a purely phenomenological spectral interpolator nor a free retrieval framework. Its distinctive contribution is to provide self-consistent radiative–convective structures with cloud physics embedded directly in the forward model, which can then be confronted with photometry and spectroscopy from the near-infrared through the mid-infrared. The literature here suggests that this strategy remains powerful for temperature, radius, and broad compositional diagnostics, but that precise metallicities, C/O ratios, and cloud mineralogy increasingly require either upgraded ExoREM generations or cross-checks against retrieval-based approaches [1912.04651; 2509.22803; 2605.29070].

Source: https://www.emergentmind.com/topics/exorem