{"id":"01ac5817-787f-404e-b9fb-24b2f4be846a","arxiv_id":"2507.16687","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":11,"one_line_summary":"A 1D retrieval on 3D GCM emission spectra of WASP-76 b does not recover a homogeneous average of the atmosphere; it is most sensitive to high thermal gradient regions that need not be the brightest.","lead":"This paper runs standard 1D atmospheric retrievals on simulated high-resolution spectra built from a 3D circulation model of the hot Jupiter WASP-76 b. It finds that the 1D retrieval is biased toward regions with strong thermal gradients, which are not necessarily the regions emitting the most light.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The gradient-bias headline is not separated from the Madhusudhan–Seager P-T parameterization: without a flexible-profile control, the retrieved profile's match to maximum-gradient GCM regions could be a parameterization artifact rather than a property of HRS.","rationale":"The strongest claim in the paper is not just that 1D retrievals are imperfect on 3D data—that would be unsurprising—but that HRS is specifically sensitive to high-thermal-gradient regions rather than to the brightest emitting regions. This is a mechanistic statement about the information content of high-resolution emission spectra, and it is the basis for the practical warning that retrieved P-T and abundance profiles are biased. The paper's support for that mechanism is a comparison of one retrieved profile (from a four-parameter analytic P-T family) to a set of GCM profiles. The authors themselves flag the family's insufficient flexibility, both as a limitation and as a possible explanation for the mismatch, so the load-bearing question is exactly whether a more flexible 1D model would retrieve a different structure. This is not a disagreement with consensus; it is an internal control that the paper does not run. I agree with the reader's weakest_assumption: the Madhusudhan–Seager parameterization is the least secure link in the inference chain. The mock retrieval on 1D data, the consistency across three retrieval experiments, the released software, and the explicit caveats are real strengths and should be acknowledged; they are why the paper should remain publishable as a conditional result rather than be rejected. The concrete test proposed above is cheap relative to the claim: one additional retrieval run with the same simulated data, changing only the P-T basis. If the result is robust to that change, the paper's interpretation is much stronger. If not, the headline should be downgraded from a property of HRS to a property of this parameterization applied to this GCM. Verdict: unchanged from the reader's CONDITIONAL, with the condition made explicit.","tokens_in":28919,"tokens_out":6416,"duration_ms":67518,"concrete_test":"Run retrieval experiment 1 (H2O dissociated + rotational broadening) on the identical simulated IGRINS dataset, replacing the Madhusudhan–Seager P-T profile with free-floating P-T nodes plus a smoothness prior (e.g., 5–7 nodes, Bazinet et al. 2024), keeping forward model, likelihood, and PCA reprocessing unchanged. Compare the posterior P-T profile and the line-forming-region gradient metric directly against Fig. 7. If the profile still aligns with the maximum-gradient westward GCM longitudes and stays well below hot-spot temperatures, the gradient-sensitivity claim survives. If it shifts toward the hot-spot or toward the geometric/emission-weighted mean profile, the headline conclusion is a parameterization artifact and must be weakened.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim is that 1D HRS emission retrievals preferentially characterize spatial regions with large thermal gradients in the line-forming region rather than the bulk emitting disk. The evidence for this is the alignment of the retrieved P-T profile with GCM profiles having the largest mean temperature gradient (Fig. 7, bottom panels). That alignment is established using only the Madhusudhan & Seager (2009) parameterization, whose flexibility limits are acknowledged in §5.1 and §5.4: it 'cannot sufficiently capture the different thermal gradients in the CO and H2O line forming regions,' and the lack of flexibility is offered as an alternative explanation for the deeper-atmosphere mismatch. Within this model family, P2, alpha1, and alpha2 jointly set the inversion and deep non-inversion slopes, and the corner plots show log P2 strongly correlated with retrieved abundances; a restricted family can easily steer a best-fit profile toward a particular GCM locus even if the data are equally consistent with other structures. A free-floating P-T node parameterization could fit the same line contrasts with a hotter or more emission-weighted profile, which would invalidate the inference that the bias is an intrinsic property of HRS. The picket-fence GCM's near-isothermal hot spot is a secondary modeling dependence; the parameterization issue alone is enough to make the interpretation conditional.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper investigates whether 1D atmospheric retrievals run on high-resolution emission spectra of a 3D hot Jupiter atmosphere produce unbiased constraints. The authors use the RM-GCM simulation of WASP-76 b to generate phase-dependent spectra over orbital phases 0.54-0.64, inject these into a simplified IGRINS-like observational simulator with stellar and telluric components, apply PCA post-processing, and run 1D retrievals with the Madhusudhan & Seager (2009) P-T parameterization in three configurations that vary water dissociation and rotational broadening. The retrieved P-T and abundance profiles fall within the range of GCM conditions, but the retrieved P-T profile aligns most closely with GCM columns that have the largest mean temperature gradients in the spectral line-forming region, rather than with the hottest or most emissive regions. The authors conclude that 1D HRS retrievals are biased toward high-gradient subregions and are not a homogeneous average of the 3D atmosphere, and they discuss implications for joint low-plus-high-resolution fits and for species-dependent Doppler shifts. The paper includes a 1D mock retrieval validation in Appendix A and releases the software on GitHub/Zenodo.","tokens_in":29256,"tokens_out":5050,"duration_ms":53908,"significance":"If the gradient-sensitivity conclusion is robust, it is an important caution for the exoplanet HRS retrieval community: 1D emission retrievals could be characterizing localized high-gradient atmospheric columns rather than the disk-averaged emitting atmosphere, which would affect abundance and P-T interpretations and complicate joint retrievals with low-resolution data. The paper's methodological care is a strength: the retrieval pipeline is validated on a 1D mock dataset, the three retrieval experiments are mutually consistent, and the authors explicitly enumerate caveats in Section 5.4. The software release and the use of a realistic observational framework strengthen reproducibility. However, the central inference is currently entangled with the choice of P-T parameterization, because the parameterization is acknowledged to be too inflexible to capture the differing CO and H2O line-forming gradients; a control experiment with a more flexible profile parameterization is needed before the headline claim can be regarded as a property of HRS rather than of the model family.","major_comments":[{"comment":"The central claim that the retrieved 1D profile is most sensitive to GCM columns with the largest thermal gradients in the line-forming region is established using only the Madhusudhan & Seager (2009) P-T parameterization. Section 5.4 states that this parameterization 'cannot sufficiently capture the different thermal gradients in the CO and H2O line forming regions,' and Section 5.1 offers parameterization inflexibility as an alternative explanation for the deeper-atmosphere mismatch. Since the bottom panels of Figure 7 are the primary evidence for the gradient-alignment conclusion, the alignment may be an artifact of the restricted model family: with only alpha1, alpha2, and P2 controlling the inversion and deeper non-inversion slopes, and with log P2 strongly correlated with the retrieved abundances (Figure 11), the best-fit profile could be steered toward a particular GCM locus even if the data are equally consistent with other thermal structures. I request a control retrieval with a more flexible P-T description, such as free P-T nodes (e.g., Bazinet et al. 2024 or Smith et al. 2024b, which the authors themselves suggest), to test whether the retrieved profile still maps onto the maximum-gradient columns. This is load-bearing for the headline conclusion.","section":"Section 5.1 and Figure 7"},{"comment":"The abstract and conclusion state that 'Doppler offsets among opacity sources' impact retrieval results, but this is not directly tested in the paper. The evidence in Section 5.3 consists of the rightmost panels of Figure 8, where a single retrieved Delta(Vsys) appears to align the water lines while leaving the CO lines offset, combined with the observation that the retrieved offsets are within 1 sigma of zero. A retrieval experiment that fits species-dependent velocity offsets, or injects a known molecular offset and checks recovery, is needed before this can be reported as a demonstrated result. As written, it is a reasonable hypothesis rather than a validated finding.","section":"Section 5.3 and abstract/conclusion"},{"comment":"The near-isothermal hot spot that makes the hot-spot region a weak line emitter is tied to the picket-fence radiative scheme of this particular GCM, as the authors acknowledge. The specific manifestation of the gradient bias, in which cooler westward columns dominate the line contrast, may therefore be model-specific. The generalization that 1D HRS retrievals are intrinsically biased toward high-gradient regions would be strengthened by a second GCM or by a simpler radiative-transfer experiment that varies the hot-spot temperature gradient while holding other quantities fixed. This is not a fatal flaw, but it is part of the same inference that needs support beyond the single GCM.","section":"Section 5.4"}],"minor_comments":[{"comment":"The eccentricity row reads '01'; this appears to be a typo for '0', and the footnote marker should be typeset as a superscript.","section":"Table 1"},{"comment":"The sentence 'following equation (2) and where where Kp and Vsys are the values reported in Table 1' contains a duplicated 'where'.","section":"Section 4.1"},{"comment":"The notation '1p/X' in Equation (4) is not defined; if it is intended to denote 1/sqrt(X), please state this explicitly and use the standard radical notation for clarity.","section":"Equation (4)"},{"comment":"The text refers to 'the bottom panels of Figure 6' when describing the mean temperature-gradient comparison, but the relevant panels are in Figure 7; the cross-reference should be corrected.","section":"Section 5.1 and Figure 7"},{"comment":"The color bar is labeled 'Mean dP/dT [mbar K^-1]', which is dimensionally mbar per K, while the text describes 'mean temperature gradient' and 'dT/dP'; the label and the text should use consistent notation so that the sign and meaning of the gradient are unambiguous.","section":"Figure 7 caption"}],"recommendation":"major_revision","confidential_remarks":"The paper is well within the scope of the journal and addresses an important methodological question. The main concern is whether the gradient-bias conclusion is robust to the choice of P-T parameterization; I would ask the editor to require the flexible-profile control experiment described in major comment 1 before publication. The self-citation pattern is not excessive and mostly reflects the authors' use of their own previously published GCM tools. The manuscript is otherwise careful and well documented."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a useful, carefully run simulation study, the first controlled test I know of that runs 1D retrievals on 3D GCM high-resolution emission spectra. The mock retrieval validates the pipeline, the three retrieval variants agree with each other, and the comparison against GCM profiles is detailed. Public code and data on GitHub/Zenodo. The headline result—that the retrieved P-T profile does not match a homogeneous disk average but aligns with regions of large thermal gradient in the line-forming layers—is new and relevant to the growing use of 1D HRS retrievals.\n\nThe soft spot is real but not disqualifying. The gradient-sensitivity claim rests on a single P-T parameterization, Madhusudhan & Seager (2009), and the authors themselves say it cannot separately capture the CO and H2O line-forming gradients. A more flexible parameterization, such as free-floating P-T nodes, might land on a different GCM subregion, which would weaken the specific 'large thermal gradients' mechanism. That control is not run. So the paper establishes a robust result—1D retrievals are biased toward some subregion, not a disk average—but the causal story is conditional on the chosen profile family. The authors flag this in Sections 5.1 and 5.4, which is honest, but they don't close the loop.\n\nThe picket-fence GCM's near-isothermal hotspot is a secondary caveat, not a fatal one. The broader point that the brightest emitting region and the line-forming region do not coincide in this GCM survives that modeling choice.\n\nBottom line: this deserves a proper referee. The central phenomenon is likely real; the causal interpretation needs one more experiment, ideally with a flexible P-T parameterization or a second GCM, to solidify. I would engage with it and cite it for the bias result.","headline":"A careful controlled simulation study showing 1D HRS emission retrievals do not recover a disk average but align with high-temperature-gradient regions; the gradient interpretation is conditional on the P-T parameterization but the core bias result holds.","tokens_in":29793,"tokens_out":2505,"would_cite":true,"duration_ms":26846,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A 1D retrieval on simulated high-resolution spectra of WASP-76 b is biased toward the steepest-gradient regions of the atmosphere, not its brightest emitting regions.","keywords":["High resolution spectroscopy","Exoplanet atmospheric composition","Exoplanet atmospheric structure","Astronomical simulations","Atmospheric retrieval","Global circulation models","WASP-76 b","Retrieval bias"],"falsifier":"Repeat the retrieval suite with a flexible, non-parametric P-T profile (e.g., free-floating P-T points) on the same simulated dataset: if the retrieved profile still aligns with the maximum-gradient GCM profiles, the gradient-sensitivity claim stands; if it shifts toward the hotter, brighter regions, the bias is an artifact of the parameterization. A second check: rerun with a GCM variant in which the hottest spot also carries steep line-forming gradients, since the paper's mechanism predicts the bias (cooler-than-brightest retrieved profile and low CO abundance) should weaken or disappear.","tokens_in":28724,"feed_emoji":"🪐","tokens_out":12292,"duration_ms":109026,"temperature":0.7,"pith_summary":"Atmospheric retrievals are the standard way to turn high-resolution spectra of exoplanets into temperature and chemistry constraints, yet they use one-dimensional models on atmospheres that are genuinely three-dimensional. This paper asks whether that mismatch silently biases the answer, and shows, using simulated high-resolution emission observations built from a 3D global circulation model of the hot Jupiter WASP-76 b, that it does. The 1D retrieval returns a profile that sits inside the range of true atmospheric conditions, so it is not wrong in an obvious sense, but it is not a homogeneous average either. High-resolution spectroscopy removes the continuum, so the retrieval matches line contrast, which is set by the temperature difference between where line cores and continuum form; the retrieved atmosphere therefore tracks the regions with the steepest thermal gradients near line-forming pressures, which in this model are cooler westward longitudes rather than the brightest eastward hot spot. If correct, published 1D HRS retrievals may be describing a steep-gradient subregion of the planet rather than its dominant emitting layers.","feed_headline":"1D exoplanet retrievals chase steep gradients, not hot spots","feed_subtitle":"High-resolution spectra of WASP-76 b encode sharp thermal slopes, biasing retrieved temperatures and chemistry.","key_machinery":"The test apparatus is a simulated high-resolution time series: phase-dependent emission spectra from the RM-GCM global circulation model of WASP-76 b (picket-fence radiative transfer, drag-free winds, Doppler-on post-processing), interpolated across 107 frames at R = 45,000, combined with PHOENIX stellar and Telfit telluric models, then reduced with PCA exactly as real observations are. The load-bearing diagnostic is geometric: each local spectrum is weighted by a viewing factor $f = \\cos^2(\\mathrm{lon})\\cos(\\mathrm{lat})$, and the $\\tau = 2/3$ surfaces map where the continuum, CO, and H2O line cores form. Because the high-resolution pipeline removes the continuum, the Brogi & Line (2019) log-likelihood matches only relative line contrast, which is set by the temperature difference between line-core and continuum-forming pressures. The paper then ranks GCM profiles by the mean temperature gradient $dT/dP$ inside the line-forming region and shows the retrieved profile tracks the strongest net inversions, not the highest flux.","core_discovery":"The paper's central claim is that a 1D retrieval applied to inherently 3D high-resolution emission data does not return a homogeneous average of the atmosphere: the retrieved pressure-temperature (P-T) and chemical profiles are biased toward spatial regions with large thermal gradients at the pressures where spectral lines form, which need not coincide with the regions emitting the most radiation. In the WASP-76 b GCM studied here, the brightest region is a nearly isothermal, eastward-offset hot spot with shallow lines, while the cooler westward longitudes carry strong inversions between the line-core and continuum-forming pressures; the retrieved profile aligns with those steep-gradient profiles. The retrieved CO and H2O abundances come out slightly below the GCM values, which the authors attribute to degeneracies between the P-T parameters and abundances, the limited flexibility of the Madhusudhan & Seager (2009) parameterization, and small Doppler offsets between CO and H2O lines that make CO lines appear shallower, with rotational broadening ($v_{\\rm rot}\\sin i \\approx 6.5$ km/s) partially masking the mismatch. The three retrieval experiments (with and without water dissociation, with and without rotational broadening) are mutually consistent within 1σ, so the bias is not driven by those modeling choices.","pith_inferences":["The gradient-weighting mechanism should generalize: any retrieval that fits only relative line contrast should preferentially weight the region maximizing line-core-to-continuum temperature contrast, so the bias should be strongest for planets whose bright spots are isothermal and should weaken for planets whose hottest regions also have the steepest gradients.","Multi-species retrievals may be silently probing different spatial regions for different molecules, so apparent abundance inconsistencies between species could encode 3D structure rather than chemistry; allowing per-molecule velocity offsets is a cheap partial correction that this paper's data already hint at.","Observables that anchor absolute flux, such as independently calibrated spectra or simultaneous photometry, could break the degeneracy that pushes 1D solutions toward steep-gradient regions, and combining those with HRS in a joint fit using two P-T profiles (one for continuum, one for lines) is a concrete next step.","A validation protocol suggests itself: before trusting abundance or thermal constraints from a 1D HRS retrieval of a real planet, run the same retrieval against GCM spectra of that planet and check whether the recovered profile maps onto a high-gradient subregion rather than the emitting disk."],"forward_implications":["Published 1D HRS emission retrievals may describe a steep-gradient subregion of a hot Jupiter rather than its dominant emitting layers, so retrieved compositions and thermal structures should be read as region-weighted rather than disk-averaged.","Joint high- and low-resolution fits that use a single P-T profile can be biased whenever the steepest-gradient region and the brightest region are spatially distinct, because the two data types carry different information (line contrast versus continuum).","The P-T parameterization choice is consequential: the Madhusudhan & Seager (2009) form cannot simultaneously represent the thermal gradients at CO and H2O line-forming pressures, and more flexible profiles could shift the retrieved solution.","Molecule-dependent Doppler offsets matter: water lines dominate the retrieved velocity, misaligning CO lines and pulling the retrieved CO abundance low, while rotational broadening partly masks the effect.","Water dissociation has negligible impact on retrieved abundances in emission at the pressures probed here, in contrast to its impact in transmission spectroscopy."],"supporting_citations":[{"why":"Supplies the P-T parameterization used by every retrieval experiment; its limited flexibility to fit different CO and H2O line-forming gradients is offered as an alternative explanation for the mismatch.","marker":"Madhusudhan & Seager (2009)"},{"why":"Provides the log-likelihood (Eq. 6) and the model-injection and PCA-reprocessing scheme that the 1D retrieval framework is built around.","marker":"Brogi & Line (2019)"},{"why":"Provides the 1D forward model (radiative transfer through 107 pressure layers) from which the retrieval is adapted, and the IGRINS emission-spectroscopy precedent.","marker":"Line et al. (2021)"},{"why":"Presents the RM-GCM model of WASP-76 b whose spectra are used as the injected 3D truth, and gives the expected wind and rotation broadening of 8-10 km/s.","marker":"Beltz et al. (2022)"},{"why":"Compares the double-gray and picket-fence GCM modes and predicts species-dependent velocity offsets, framing the retrieval's velocity results.","marker":"Beltz & Rauscher (2024)"},{"why":"Develops the picket-fence radiative transfer scheme that produces the near-isothermal hot-spot P-T profiles central to the gradient-sensitivity interpretation.","marker":"Malsky et al. (2024)"},{"why":"Provides the Doppler-on ray-tracing post-processing and the tau = 2/3 method used to locate the pressures where continuum, CO, and H2O lines form.","marker":"Zhang et al. (2017)"},{"why":"Supplies the parameterized water-dissociation profile (Eqs. 4-5) used in one retrieval experiment and the transmission-spectroscopy comparison.","marker":"Gandhi et al. (2024)"},{"why":"Establishes the model reprocessing through PCA-injection that the retrieval pipeline follows to handle signal erosion.","marker":"Gibson et al. (2022)"},{"why":"Supplies the WASP-76 b system parameters (radius, log g, stellar properties) used to build the simulated high-resolution observations.","marker":"West et al. (2016)"}],"fun_headline_variants":["1D retrievals favor steep slopes, not hot spots","Retrievals bias toward steep thermal gradients","Steep gradients skew 1D exoplanet retrievals","High-res retrievals chase thermal slopes, not emission peaks"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The headline bias, that the retrieved profile maps onto the steepest-gradient regions, is read off through the Madhusudhan & Seager (2009) P-T parameterization, which the authors themselves note cannot capture the different thermal gradients in the CO and H2O line-forming regions, and through a picket-fence GCM whose hot spot is nearly isothermal; with a more flexible parameterization or a GCM with a steeper hot-spot profile, the retrieved solution might land on different regions.","fun_headline_variants_meta":{"raw":{"variants":["1D retrievals favor steep slopes, not hot spots","Retrievals bias toward steep thermal gradients","Steep gradients skew 1D exoplanet retrievals","High-res retrievals chase thermal slopes, not emission peaks"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000218,"raw_usage":{"total_tokens":1509,"prompt_tokens":1086,"completion_tokens":423,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":702,"completion_tokens_details":{"reasoning_tokens":358}},"tokens_in":702,"tokens_out":423,"duration_ms":4526,"temperature":1.0,"reasoning_tokens":358,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T15:04:12.666992+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Repeat the retrieval suite with a flexible, non-parametric P-T profile (e.g., free-floating P-T points) on the same simulated dataset: if the retrieved profile still aligns with the maximum-gradient GCM profiles, the gradient-sensitivity claim stands; if it shifts toward the hotter, brighter regions, the bias is an artifact of the parameterization. A second check: rerun with a GCM variant in which the hottest spot also carries steep line-forming gradients, since the paper's mechanism predicts the bias (cooler-than-brightest retrieved profile and low CO abundance) should weaken or disappear.","supporting_citations":[{"cited_title":"2017, , 837, L27, 10.3847/2041-8213/aa62fc","cited_arxiv_id":null,"evidence_quote":"Provides the Doppler-on ray-tracing post-processing and the tau = 2/3 method used to locate the pressures where continuum, CO, and H2O lines form."}],"review_version":1}