{"id":"9743a623-1bed-4748-a964-5f9f9bf529d1","arxiv_id":"2412.00933","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"A morphology-plus-color pipeline decomposes and removes Galactic cirrus from deep wide-field images, improving low surface brightness galaxy detection and yielding optical DGL measurements.","lead":"This paper presents a method to separate faint Milky Way dust wisps, called cirrus, from true fuzzy galaxies in deep wide-field images. The tool first flags wispy structures by shape, then uses color information to subtract them, and the authors show it helps recover dwarf galaxies and measure the brightness of the dust-scattered light.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Single g-r color model leaves residuals in high-column-density regions, weakening the claim that cirrus is cleanly removed for LSBG searches and DGL radiometry.","rationale":"The reader's weakest_assumption identifies the fixed g-r color assumption as the central risk; the paper itself provides evidence of its failure in Sec 5.3. My stress-test confirms that this is the most load-bearing concern because it directly undermines the quantitative claims of cirrus removal and DGL precision. The suggested concrete test is a simple, data-driven check that would either validate the single-color model across the field or expose its spatial breakdown. I do not think the concern warrants moving beyond CONDITIONAL: the paper has independent support from And XXII recovery and injection-recovery tests, and the residual pattern is described as 'very faint'. However, the absence of released code/data and the in-sample nature of the flatness metrics strengthen the case for conditional acceptance rather than full acceptance. Therefore the reader's verdict remains appropriate.","tokens_in":41415,"tokens_out":6458,"duration_ms":64005,"concrete_test":"Using the source-subtracted, RHT-masked, and infilled g and r images of Field A from Sec 5.2, bin pixels by Planck radiance R (or by spatial location within the field) and compute the median g-r color in each bin. If the color varies by more than the statistical uncertainty (e.g., >0.05 mag) across R bins, especially above R ~ 2.5e-7 W m^-2 sr^-1 where residual cirrus is observed, the single-color model is falsified. Equivalently, after applying the paper's decomposition, compute the residual g+r image and measure the correlation between residual surface brightness and the Planck radiance map; a significant nonzero correlation at high R demonstrates that the decomposition leaves cirrus residuals that could compromise LSBG searches and bias DGL radiometry.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The decomposition assumes a single g-r color for all cirrus in each field, fitted via Eq. 14-17 to the same binned g and r images that are later cleaned. Any spatial variation in cirrus color (with dust column, radiation field, or grain properties) makes the linear model over- or under-subtract cirrus. The paper's own Sec 5.3 reports a 'very faint large-scale diffuse light pattern' in the residual image that spatially matches the high-intensity cirrus regions, attributed to changes in dust properties or optical depth effects. This is direct evidence that the single-color assumption is violated where cirrus is brightest. Because the RHT-extracted cirrus map is scaled by this single color, these residuals are systematic and spatially correlated with cirrus. They can bias the flatness metrics (measured in-sample) and, more importantly, create false diffuse features or mask real LSBGs located behind high-column cirrus. The injection-recovery F1=0.75 already includes failures attributed to blending with cirrus, consistent with this limitation. The claim of 'high radiometric precision' for the DGL also depends on the linear scaling; a residual correlated with the Planck radiance tracer directly biases the derived slope b_lambda and the zero-point, so the quoted DGL intensities in Sec 7.2 are not robust to color variation.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper presents a method for decomposing Galactic cirrus in deep wide-field images by combining Rolling Hough Transform (RHT) morphological filtering with a single g-r color model calibrated against Planck thermal dust radiance. Using ~10 deg^2 of Dragonfly g/r data in two fields, the authors subtract foreground and background sources, extract cirrus-like structures, fit a linear color relation (Eqs. 14-17), subtract the predicted cirrus from each band, and evaluate the resulting sky flatness with skewness, Gini, and Δ-variance metrics. They demonstrate the method's application by recovering injected mock ultra-diffuse galaxies (F1=0.75) and the known dwarf satellite And XXII, and they derive optical diffuse Galactic light (DGL) intensities that are consistent with dust scattering models.","tokens_in":41655,"tokens_out":7486,"duration_ms":71087,"significance":"If the method holds, it offers a practical, physically motivated way to separate cirrus from low surface brightness galaxies and to measure DGL in deep wide-field surveys, with timely relevance for Rubin, Euclid, and Roman. The paper is commendable for anchoring the color model to external Planck correlations, for testing recovery with realistic stellar population models, and for candidly stating several limitations. These strengths give the central claim credibility. However, the in-sample fitting of the color model, the acknowledged residual in high-column regions, and an apparent inconsistency in the injection-recovery reporting mean that the performance claims need stronger out-of-sample validation before the paper can be accepted as a reliable methods reference.","major_comments":[{"comment":"The parameters A and B of the color model are fitted by maximum likelihood to the same binned g and r images that are subsequently cleaned with that model in §5.3. The flatness metrics in §5.4 (skewness, Gini, Δ-variance) are computed on this same residual image, so the measured improvement is partly a mathematical consequence of projecting out the best-fit linear relation, not an independent measurement that cirrus has been removed. I recommend a cross-validation step: for example, fit the color model on one subregion or one field and apply it to the other, or report residuals against the Planck radiance in pixels excluded from the fit.","section":"§5.2, Eqs. (14)–(17); §5.3; §5.4"},{"comment":"The paper itself reports a 'very faint large-scale diffuse light pattern' in the residual that spatially matches the high-intensity cirrus regions and attributes it to changes in dust properties or optical depth effects. This is direct evidence that the single-color assumption is violated where cirrus is brightest. Because the RHT-extracted cirrus map is scaled by that single color, the residual is systematic and spatially correlated with the cirrus; it can both create false diffuse features and mask real LSBGs behind high-column cirrus, and it propagates into the DGL slope b_lambda in §7.2. Please quantify this residual as a function of Planck radiance or g-r color, and show its effect on recovery completeness and on the derived DGL intensities.","section":"§5.3; §6.1.2; §7.2"},{"comment":"The reported precision (0.77) and recall (0.73) appear inconsistent with the footnote statement that roughly 20 objects were detected and only one overlapped the injections within 3 arcseconds. If that statement refers to the same detection run, then TP=1 would give a precision of about 0.05, not 0.77. Please state the total number of injected galaxies, the matching radius, and clarify whether the footnote concerns a separate diagnostic; as written, the injection-recovery result is not reproducible.","section":"§6.1.2 and footnote 14"},{"comment":"The photometric zero-point is set by assuming that no dust-scattered light exists where the Planck radiance is zero, with the contribution from EBL and the diffuse ionized medium acknowledged in footnote 9 but not quantified. Because Eq. (23) is fitted to the residual images, any unmodeled EBL-like constant propagates directly into b_lambda and the quoted DGL intensities. Please provide a quantitative estimate of EBL and other diffuse background components at these Galactic latitudes and include them in the systematic error budget for the DGL measurements.","section":"§5.1.2, Eq. (12); §7.2, Eq. (23)"}],"minor_comments":[{"comment":"There is a spurious space in the header title ('F uzzy Galaxies') and a typo in §6.1.3 ('galaixes' should be 'galaxies').","section":"Title and §6.1.3"},{"comment":"The legend shows two identical 'This work (Dragonfly; l~133 b~ 35)' labels; please distinguish the g-band and r-band points and specify whether the magnitudes are in the AB or Vega system.","section":"Figure 13"},{"comment":"For Field B, the g-to-r and r-to-g slopes differ substantially (1.56 vs 2.06), while the bisector gives g-r = 0.63. A sentence explaining this asymmetry and why the bisector is the adopted value would help the reader interpret the color model.","section":"Table 3"},{"comment":"The comparison between maskfill and Gaussian process regression is only qualitative. A quantitative metric, such as the RMS difference from the ground-truth image, would make the claimed similarity concrete.","section":"Appendix B"}],"recommendation":"major_revision","confidential_remarks":"To the editor: this is a well-written and potentially useful methods paper, but the central performance claims need to be underpinned by out-of-sample validation and a clarified injection-recovery statistic. I do not think rejection is warranted; the issues appear addressable within the scope of a revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know this: the genuinely new thing here is pairing Rolling Hough Transform morphology selection with a single g-r color constraint anchored to Planck thermal dust models. That combination works well enough on two Dragonfly fields to flatten the sky noticeably, recover injected mock UDGs at F1=0.75, pull out And XXII from cirrus, and produce optical DGL intensities that sit sensibly on the literature SED. Credit is due: the flatness metrics are appropriate, the injection-recovery is a real test, and the paper does not oversell the residuals at high column density. The authors explicitly note a faint large-scale residual pattern that tracks the bright cirrus regions and attribute it to changing dust properties or optical depth effects.\n\nThe soft spots are real but proportionate. The color model is fitted to the same binned g and r images that get cleaned (Eq. 14-17), so the flatness metrics are partly in-sample; the residual sky will always look flatter when the model is tuned to it. The single-color assumption is demonstrably violated where cirrus is brightest, which weakens the LSBG completeness estimate and the radiometric precision claim for DGL. The F1=0.75 already includes failures from blending with cirrus, consistent with that. On reproducibility, no code or data are released; for a method with several tunable parameters (Dw, DM, thresholds, mixture-model priors), that is a real gap. And the DGL zero-point rests on the assumption that optical scattered light is zero where Planck radiance is zero, which ignores EBL and diffuse ionized gas, so the absolute DGL numbers are less robust than the relative removal demonstration.\n\nStill, the central argument holds up in the regime it claims. The paper is honest about its limitations and the method clearly advances the practical toolkit for removing cirrus from deep wide-field images. It is worth reading for anyone planning LSBG searches with Rubin, Euclid, or Roman, and for ISM folks interested in optical DGL. The main requested revisions should be code/data release and a more direct test of spatial color variation. This deserves peer review, not a desk rejection.","headline":"A solid methods paper that combines RHT morphology filtering with Planck-calibrated optical colors; the in-sample color fit is the main soft spot, but the demonstration and honest treatment of limitations justify sending it to referees.","tokens_in":42293,"tokens_out":1642,"would_cite":true,"duration_ms":17417,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper shows that a morphological filter isolating filamentary emission plus a single $g-r$ color constraint decomposes Galactic cirrus from deep wide-field images, flattening the sky and letting faint low surface brightness galaxies…","keywords":["Galactic cirrus","diffuse Galactic light","low surface brightness galaxies","image decomposition","Rolling Hough Transform","Planck thermal dust model","sky background subtraction","interstellar dust"],"falsifier":"Compute the best-fit $g-r$ color of the cirrus in sliding windows that span the full range of Planck radiance in one field; if the color shifts by more than the bootstrap uncertainty, the fixed-color model must leave residuals that spatially track the dust map, and those residuals should correlate with the recovery failures of injected galaxies.","tokens_in":41174,"feed_emoji":"🌌","tokens_out":9753,"duration_ms":82987,"temperature":0.7,"pith_summary":"The paper proposes that deep wide-field images can be decomposed into Galactic cirrus and everything else using two pieces of information: the filamentary or patchy morphology of the emission and a single optical color. Using about 10 square degrees of imaging from an array optimized for low surface brightness work, the authors show that after this decomposition the sky background is markedly flatter, simulated ultra-diffuse galaxies are recovered with an F-score of 0.75, and the known M33 satellite And XXII becomes visible in integrated light. The same decomposition yields optical diffuse Galactic light intensities that match dust scattering models and show no clear extended red emission in the r band. A working version of this recipe matters because cirrus is one of the major foregrounds for upcoming deep surveys such as Rubin, Euclid, and Roman, and because the decomposed cirrus is itself a clean signal for studying dust properties.","feed_headline":"One color plus filament filter strips cirrus from deep images","feed_subtitle":"A single g-r color, calibrated with Planck dust maps, flattens the sky and uncovers faint galaxies.","key_machinery":"The load-bearing object is the Rolling Hough Transform (RHT), a line-detection variant of the Hough transform that rolls a circular window over the image and records, at each pixel, the maximum response over orientation; filamentary cirrus produces a strong peak in the response while a round galaxy produces a flat response. The ratio of the median-filtered image to the RHT response separates blobby structures, which are masked, dilated, and infilled. The second piece is a linear color model: a mixture-model fit of g versus r and r versus g yields a field-averaged $g-r$ color, with zero-points pinned by the correlation of the optical surface brightness with Planck dust radiance. Together these convert a single-band morphological mask into a two-band cirrus model that can be subtracted.","core_discovery":"The central claim is that cirrus in a given field can be characterized photometrically by extracting its filamentary and patchy component with the Rolling Hough Transform, masking and infilling the blobby residuals that are candidate galaxies, and constraining the cirrus component with a single linear color model $g-r \\approx 0.7$, calibrated by correlating the optical images with the Planck thermal dust radiance. Once the color model predicts the cirrus contribution in each band, subtracting it leaves a residual image whose sky is close to flat: the Gini coefficient of the pixel intensity distribution drops from 0.28 to 0.04, and the $\\Delta$-variance power of large-scale structure falls by a factor of about 200 at scales of 5 arcminutes. The method recovers the known dwarf And XXII, whose mean color is bluer than the cirrus, and recovers injected mock galaxies with an F-score of 0.75. The paper therefore argues that morphological plus color separation is sufficient to clean cirrus with only two filters.","pith_inferences":["A direct test the paper does not run is to vary the cirrus color spatially within a field; if the local $g-r$ drifts with Planck radiance, the residual pattern the authors note in bright cirrus regions would grow, and the F-score for galaxy recovery would drop.","The zero-point assumption that no optical scattered light exists where the Planck tracer is zero makes the absolute DGL scale vulnerable to extragalactic background light; cross-calibrating with an independent measurement of the cosmic optical background would put the photometric zero-point on firmer ground.","The fixed-color assumption likely fails near optically thick cirrus, so a promising extension is to fit a two-component or column-density-dependent color model and to propagate the resulting uncertainty into source photometry.","The infilling step currently treats masked blobs without a physical model; replacing it with a local interpolation that carries uncertainties would turn cirrus removal into a per-pixel uncertainty budget for faint-source measurements."],"forward_implications":["Sky-background flattening of this kind should directly improve the completeness of low surface brightness galaxy searches in cirrus-rich fields, since the same mask-and-color step can be inserted before source detection.","The measured optical DGL intensities, normalized by 100 micron emission, sit on the expected dust-scattering model curves, so the technique can be used to constrain grain properties over much larger sky areas.","Surveys with more than two filters can generalize the single-color model to a full spectral energy distribution, which should separate cirrus from galaxies with colors close to the cirrus color.","Integrated-light detection of satellites such as And XXII becomes possible even when cirrus is present, complementing star-count searches.","The demonstrated recovery of injected mock galaxies indicates that such a decomposition can be used to quantify survey completeness in cirrus-affected fields."],"supporting_citations":[{"why":"Supplies the Rolling Hough Transform algorithm used to extract filamentary and patchy cirrus-like morphology.","marker":"Clark et al. (2014)"},{"why":"Provides the all-sky thermal dust model and radiance maps that set the color zero-points and calibrate the optical-to-FIR correlation.","marker":"Planck Collaboration XI (2014)"},{"why":"Establishes the sky-subtraction and combination scheme that preserves cirrus signal before decomposition.","marker":"Liu et al. (2023)"},{"why":"Demonstrates that cirrus can be separated from galaxies by optical colors, the physical basis for the color constraint.","marker":"Román et al. (2020)"},{"why":"Provides the dust scattering models and expected DGL spectral energy distribution against which the measured optical DGL is compared.","marker":"Brandt & Draine (2012)"},{"why":"Supplies the outlier-robust mixture model used to fit the linear $g-r$ color relation while pruning non-cirrus pixels.","marker":"Hogg et al. (2010)"},{"why":"Provides the MRF source modeling method that subtracts stars and galaxies before cirrus extraction.","marker":"van Dokkum et al. (2020)"},{"why":"Provides ArtPop, used to generate realistic mock ultra-diffuse galaxies for the injection-recovery test.","marker":"Greco & Danieli (2022)"}],"fun_headline_variants":["Two-filter trick strips cirrus, sharpens faint galaxies","Filament shape plus one color separates dust from galaxies","Sky flattens 200x after cirrus removed by color and shape","Planck dust maps calibrate single color to clean deep images","Morphology and g-r color uncover galaxies hidden in cirrus"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The decomposition assumes that all cirrus in a field has one fixed $g-r$ color, estimated from the same images that are later cleaned; if that color varies with dust column density, dust properties, or the illuminating radiation field, the subtraction will leave residual cirrus or erase parts of real galaxies.","fun_headline_variants_meta":{"raw":{"variants":["Two-filter trick strips cirrus, sharpens faint galaxies","Filament shape plus one color separates dust from galaxies","Sky flattens 200x after cirrus removed by color and shape","Planck dust maps calibrate single color to clean deep images","Morphology and g-r color uncover galaxies hidden in cirrus"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000246,"raw_usage":{"total_tokens":1555,"prompt_tokens":976,"completion_tokens":579,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":592,"completion_tokens_details":{"reasoning_tokens":492}},"tokens_in":592,"tokens_out":579,"duration_ms":6201,"temperature":1.0,"reasoning_tokens":492,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T04:51:44.684396+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the best-fit $g-r$ color of the cirrus in sliding windows that span the full range of Planck radiance in one field; if the color shifts by more than the bootstrap uncertainty, the fixed-color model must leave residuals that spatially track the dust map, and those residuals should correlate with the recovery failures of injected galaxies.","supporting_citations":[],"review_version":1}