{"id":"87893cfa-8cf5-43d6-ac54-c4a1abeaedcb","arxiv_id":"2608.04136","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Large-eddy simulations of a realistic Barcelona district for 16 wind directions show pedestrian flow is strongly direction-dependent while double-averaged profiles stay quasi-isotropic, with two inflection points in the canopy.","lead":"This study simulates wind flow over a real Barcelona district from 16 directions and shows that local wind patterns change a lot with wind direction while neighborhood-averaged profiles stay nearly the same. The result matters for city wind comfort, air quality, and for deciding whether simple wind-direction-independent models are good enough at the neighborhood scale.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The claimed lower inflection point at z ≈ 0.08–0.10H_avg lies only 1–2 grid cells above the ground at ~1 m resolution, is identified visually without derivative or convergence quantification, and is not tested by the single-direction mesh study; it may be a near-wall numerical artifact.","rationale":"This paper's central claim is twofold: (i) double-averaged profiles collapse across 16 wind directions, and (ii) the mean velocity profile has two consistent inflection points, the lower one at 0.08–0.10H_avg marking a transition from recirculating to connected transport. Claim (i) is plausible and visible in Fig. 10; it is a broad, vertically integrated feature, and the u*-normalization may account for some of the collapse, so I do not see a fatal issue there. Claim (ii) is the more novel and more quantitative sub-claim, and it is the one most exposed to numerical error. The lower inflection point sits at z ≈ 1.5–1.9 m, while the stated mesh spacing at pedestrian height is ~1 m (Section 3.1.1): the feature is determined by at most two or three resolved heights where the ground wall-model (Section 3.1.2) also imposes a strong near-wall gradient. The paper identifies the inflection visually from Fig. 10, with no derivative-based method, no confidence interval, and no grid-convergence test for this specific feature. Appendix A's mesh assessment uses only Φ = 180°, only p = 2 vs p = 4, and only pointwise profile comparisons; it never quantifies the curvature of the double-averaged profile below z ≈ 0.2H_avg. Thus the load-bearing evidence for the second inflection point is missing exactly where numerical uncertainty is largest. A concrete test—computing the second derivative across all directions and a locally refined run at one or two directions—would settle this. If the inflection persists and is grid-converged, the claim stands; if not, the paper's novelty reduces to the direction-collapse, which is still interesting but less specific. The reader's weakest assumption (rotating geometry vs wind, mesh differences across orientations) is related but not identical; I find the near-wall resolution issue more direct and more damaging. The verdict remains CONDITIONAL: the paper should be asked to provide derivative-based inflection detection and a targeted resolution check before the quantitative headline is accepted.","tokens_in":20228,"tokens_out":6606,"duration_ms":62928,"concrete_test":"Compute the second derivative d²⟨ū⟩/dz² of the double-averaged streamwise velocity from the raw LES output on the actual vertical grid for all 16 directions, using a consistent finite-difference stencil, and report the zero-crossing heights. Additionally, rerun Φ = 180° on a mesh with vertical refinement to Δz ≈ 0.3 m below z = 2H_avg (and also a p = 2 vs p = 4 comparison at one oblique direction, e.g., Φ = 247.5°), then check whether a lower inflection point at 0.08–0.10H_avg persists and whether its height varies by more than ±0.02H_avg across directions. If the inflection is absent, or shifts to a height outside the reported band, the central claim is not supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing concern is the reliability of the lower inflection point at z ≈ 0.08–0.10H_avg (Section 4.4, Fig. 10). With H_avg = 19.27 m, this feature lies at z ≈ 1.5–1.9 m, i.e., only one or two cells above the ground given the stated ~1 m pedestrian-level resolution (Section 3.1.1). The inflection point is identified visually from the double-averaged profile; no derivative computation, uncertainty interval, or resolution-convergence test is reported for it. The only mesh sensitivity study (Appendix A) is conducted at a single inflow direction (Φ = 180°) and compares p = 2 vs p = 4 profiles at eight pointwise stations; it does not examine the double-averaged profile's curvature below z ≈ 0.2H_avg, nor does it verify that the lower inflection point persists or is grid-converged. The near-ground flow is further affected by the equilibrium rough-wall model applied at the ground (Section 3.1.2), which can impose a log-law-like gradient in the first cells. Consequently, the second inflection point—a headline novelty—could be a numerical artifact of the wall model and/or insufficient vertical resolution rather than a physical transition from recirculating to connected flow. The across-direction collapse is less threatened, as it is visible in Fig. 10 over a broad height range and is normalized by u*; but the lower inflection point is precisely where resolution is coarsest and wall-model influence is strongest.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents building-resolving large-eddy simulations of neutral atmospheric boundary-layer flow over the Pedralbes district in Barcelona, using the SOD2D spectral-element solver with the Vreman subgrid model and a precursor inflow. Sixteen wind directions are simulated by rotating the city geometry relative to a fixed inlet, with meshes of roughly 5.0e8 points and pedestrian-level resolution near 1 m. The authors report strong directional sensitivity at pedestrian level, with preferential ventilation corridors forming for wind directions aligned with the street network, while double-averaged vertical profiles of mean velocity and streamwise turbulence intensity collapse across all wind directions. They further identify two inflection points in the mean velocity profile: one slightly below H_avg and a second at z approximately 0.08-0.10 H_avg, interpreted as a transition between near-ground recirculating flow and more connected intra-canopy transport. Subdomain analyses show that the lower inflection point is morphology- and direction-dependent.","tokens_in":20514,"tokens_out":4427,"duration_ms":50379,"significance":"If the central claims hold, the paper provides a valuable and unusual dataset: sixteen systematically varied inflow directions over a realistic urban district, with long sampling times (160-170 eddy turnover times), high-order spatial discretization, and direct statistics that do not rely on fitted parameters for the collapse or inflection-point observations. The proposed scale separation—strong local directionality coexisting with quasi-isotropic double-averaged profiles—would be practically relevant for reduced-order urban parameterizations and for interpreting local versus neighborhood-scale ventilation metrics. The paper also makes a concrete, falsifiable prediction about the existence and location of a lower inflection point, which is a useful starting point for future canopy-flow studies. The strengths include the precursor-inflow methodology, the explicitly stated mesh-sensitivity study for one direction, and the careful comparison of sampling durations with earlier urban LES work.","major_comments":[{"comment":"The lower inflection point at z = 0.08-0.10 H_avg, i.e., approximately 1.5-1.9 m for H_avg = 19.27 m, is asserted from visual inspection of the double-averaged profiles, without any derivative computation, curvature quantification, confidence interval, or grid-convergence test at that height. This height range is precisely where the vertical resolution is coarsest relative to the canopy and where the equilibrium rough-wall log-law model is applied at the ground. The only mesh assessment (Appendix A) compares p = 2 and p = 4 pointwise profiles at eight stations for a single inflow direction and does not report the double-averaged profile below z/H_avg around 0.2. As a result, the headline novelty—the second inflection point—could be a numerical artifact of the wall model or of insufficient near-wall resolution. I ask the authors to quantify the curvature of the double-averaged profile in this layer, to report uncertainty or spread across the 16 directions, and to demonstrate that the inflection persists under mesh refinement (including the lowest 0.2 H_avg) for at least a second wind direction.","section":"Section 4.4, Fig. 10; Sections 3.1.1 and 3.1.2"},{"comment":"The across-direction collapse is a central claim, but the comparison is made across 16 different meshes (4.8e8 to 5.2e8 points) that are produced by rotating the city geometry rather than by rotating the inflow. The mesh sensitivity study is performed only for Phi = 180 degrees, so there is no direct evidence that the observed directional invariance of the double-averaged profiles is independent of resolution differences between orientations. I request a per-orientation description of the refinement zones (e.g., element sizes in the pedestrian layer and in the vertical column) and, ideally, a second mesh-sensitivity check for a contrasting direction, or an explicit demonstration that the refinement pattern is statistically equivalent across rotations. Without this, the apparent collapse could partly reflect numerical smoothing in less resolved orientations rather than a physical property of the urban morphology.","section":"Sections 2 and 3.1.1; Appendix A"},{"comment":"The statement that the double-averaged profiles 'collapse remarkably across all wind directions' is not quantified. The visual spread in Fig. 10 appears modest, but the paper makes a strong practical claim—that direction-agnostic parameterizations may be sufficient at neighborhood scale—without reporting a quantitative measure of the collapse, such as the maximum or RMS normalized deviation of the 16 profiles from their ensemble mean, or the sampling uncertainty of each profile given the finite averaging time. I ask the authors to add such a metric and to state how the collapse compares with the estimated statistical error bars. This would also place the lower inflection point on a firmer footing, since a spread of a few percent in the near-wall region could be comparable to the shear variation that defines the inflection.","section":"Section 4.4, Fig. 10"}],"minor_comments":[{"comment":"The domain distances are quoted inconsistently: upstream distance is given as 70 H_avg in one place and 65 H_avg later, downstream as 140 H_avg versus 124 H_avg, and lateral distance as 97 H_avg versus 91 H_avg. Please reconcile these numbers.","section":"Section 2"},{"comment":"The text states 'with H_avg/H_max = 0.42', but the reported values H_avg = 19.27 m and H_max = 67.29 m give H_avg/H_max = 0.286. If H_std/H_avg = 0.42 was intended, the notation should be corrected.","section":"Section 4.4"},{"comment":"The appendix is called 'Annex A' in Section 3.1.1 and 'Appendix A' elsewhere; please use one consistent label.","section":"Appendix A"},{"comment":"The caption says 'The black dashed line denotes the average building height, H_avg, and the red dashed line the maximum building height, H_max,' but it would be clearer to also state in the caption that the lower inflection point appears near z/H_avg = 0.08-0.10, since the reader cannot infer the inflection location from the printed figure alone.","section":"Fig. 10 caption"}],"recommendation":"major_revision","confidential_remarks":"The paper is a strong candidate for the journal if the authors can convert the visually asserted lower inflection point into a quantitatively supported feature and can address the mesh-rotation equivalence concern. The current evidence for the headline lower inflection point is not yet convincing, but the issue is fixable within the scope of the manuscript. I would also encourage the editor to ask for the quantitative collapse metric, since the 'quasi-isotropic' claim is likely to be cited by parameterization studies."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is worth reading and worth refereeing. The effort behind the dataset is real: sixteen directions over the full 360°, a realistic Barcelona district, about 5e8 spectral-element points, long sampling (160–170 ETT), and careful use of a precursor inflow. The main message comes through clearly: pedestrian-level flow is strongly direction-dependent, while double-averaged profiles of mean velocity and turbulence intensity collapse across all directions. That distinction, and the suggestion that direction-agnostic neighbourhood-scale parameterizations may be feasible, is useful for the urban CFD community and is supported by the data shown.\n\nWhat is genuinely new is the systematic full-circle sweep on a realistic geometry, and the explicit decoupling of local directional sensitivity from quasi-isotropic double-averaged behaviour. The zone-level analysis, showing that local morphology controls the directional response, is also a nice addition and helps explain why the neighbourhood average masks so much.\n\nThe soft spots are real but not fatal. The lower inflection point at z ≈ 0.08–0.10H_avg is the most important one. It lies at 1.5–1.9 m above ground, and with near-wall mesh spacing around 1 m (even with fourth-order elements) this is close to the resolution limit. It is identified visually from averaged profiles; no derivative, no uncertainty interval, and no grid-convergence test specific to that height are reported. The Appendix A mesh study covers only Φ = 180° and does not examine profile curvature below about 0.2H_avg, so it does not settle whether the second inflection point is physical or a numerical artifact of the wall model and near-wall resolution. I would not treat this as a refutation of the main collapse claim, which is visible over a broad height range, but the inflection point is a headline novelty and needs quantitative support.\n\nThe rotating-geometry approach is standard, but the mesh differs slightly between orientations (4.8–5.2e8 points). That the collapse persists across all sixteen is encouraging; still, a second mesh-sensitivity direction would strengthen the claim against numerical bias. There is also a clear internal inconsistency in the reported domain dimensions (337H_avg vs. later numbers like 70/140 and 65/124 and lateral distances 97 vs. 91), which needs fixing. And data/code are only available on request, which limits reproducibility.\n\nOverall: a solid, informative study with one over-interpreted feature. The neighbourhood-scale collapse is likely robust and supports the parameterization message. The lower inflection point should be either defended with quantitative analysis or softened. I would send this to peer review, expecting major revision on the inflection-point evidence.","headline":"A valuable, carefully executed LES campaign whose central collapse claim is credible, but the headline lower inflection point is under-supported and needs quantitative backup before it becomes a definitive finding.","tokens_in":21093,"tokens_out":2546,"would_cite":true,"duration_ms":28929,"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":"A 16-direction LES of a Barcelona district shows pedestrian-level flow is strongly wind-direction dependent, yet double-averaged velocity and turbulence profiles collapse across all directions, with two persistent inflection points.","keywords":["large-eddy simulation","urban canopy layer","wind direction","pedestrian-level wind","double-averaging","roughness sublayer","ventilation corridors","atmospheric boundary layer"],"falsifier":"Re-run the same district for one wind direction using two different geometry rotations that should be physically identical (e.g., the 180-degree case and the same geometry rotated by 22.5 degrees with inflow adjusted accordingly) and compare the double-averaged profiles; if the two representations differ by more than the inter-direction scatter seen in the present profiles, the collapse is partly a numerical artifact.","tokens_in":1618,"feed_emoji":"🌬️","tokens_out":2017,"duration_ms":55689,"temperature":0.7,"pith_summary":"This paper uses 16 large-eddy simulations of the atmospheric boundary layer over a real Barcelona neighborhood, one per 22.5-degree increment around the compass, to separate local from neighborhood-scale effects of wind direction. It seeks to establish that although pedestrian-level airflow reorganizes dramatically as the wind rotates, forming or destroying accelerated street-corridor flows, the double-averaged vertical profiles of mean velocity and turbulence intensity are nearly direction-independent. It further claims that these averaged profiles consistently show two inflection points: one slightly below the average building height, marking a shear-driven mixing layer, and a second near pedestrian level at about 0.08 to 0.10 average building heights, marking the transition from recirculating near-ground flow to connected intra-canopy transport. This matters because it suggests that district-scale aerodynamic parameterizations may safely ignore wind direction, while local ventilation, comfort, and dispersion assessments still require direction-specific simulations.","feed_headline":"Wind direction scrambles street air but not the district profile","feed_subtitle":"Pedestrian-level wind reorganizes with every rotation, while double-averaged velocity and turbulence profiles barely move.","key_machinery":"The machinery is high-resolution large-eddy simulation combined with double-averaging. Sixteen cases are produced by rotating the city geometry while keeping the inlet and precursor turbulent boundary-layer inflow fixed, so all orientations share the same boundary-condition implementation; the meshes use fourth-order spectral elements with roughly 5.0e8 degrees of freedom and sub-meter pedestrian-level resolution. Double-averaging (Reynolds time averaging followed by spatial averaging) is the tool that separates local directional heterogeneity from the neighborhood-scale response, and the effective frontal area density λ_f,eff ≡ λ_f H_avg/H_std is used to connect local morphology to pedestrian-level wind metrics.","core_discovery":"The central discovery is the coexistence of strong local directionality with quasi-isotropic neighborhood-scale statistics. At pedestrian level, wind direction activates or suppresses preferential channeling corridors, switching between continuous accelerated bands aligned with major streets and fragmented cellular patterns under oblique inflow. Yet when the time-averaged fields are spatially averaged across the whole district, the double-averaged vertical profiles of mean streamwise velocity and streamwise velocity variance collapse for all sixteen wind directions. The mean velocity profile exhibits two distinct inflection points: one slightly below the average building height, consistent with a mixing-layer regime above the canopy, and a second near z ≈ 0.08–0.10 H_avg that separates a sheltered, recirculating near-ground layer from more connected transport within the canopy. The lower inflection point is morphology-dependent and direction-dependent when the analysis is restricted to subdomains, while the turbulence intensity peak is systematically located between the average and maximum building heights, implicating the tallest structures as dominant sources of turbulent kinetic energy.","pith_inferences":["If the profile collapse is generic, it should persist for other heterogeneous districts; a testable extension is to apply the same sixteen-direction protocol to a regular street grid, where stronger directional anisotropy in the double-averaged profiles would be expected.","The direction-dependence of the lower inflection point could be developed into a quantitative ventilation-connectivity metric, e.g., by correlating its height with pollutant residence time or scalar exchange rates in each subdomain.","Because only the city is rotated and the precursor inflow stays fixed, the equivalence of geometry rotation and wind rotation is assumed; an explicit test would be to run one orientation twice with two different but statistically equivalent precursor inflows.","The near-isotropy appears to emerge from building-height disorder and orientation scatter, suggesting that the averaging area needed to reach quasi-isotropy may itself be a morphological property worth quantifying."],"forward_implications":["District-scale aerodynamic parameterizations of roughness and drag can plausibly be direction-agnostic for this class of heterogeneous urban morphology.","Local studies of pedestrian comfort, pollution dispersion, or ventilation must retain direction-specific simulations because corridor activation depends on the alignment between wind and street network.","The lower inflection point at about 0.08–0.10 H_avg can serve as a diagnostic of the transition between sheltered recirculation and connected transport within a real canopy.","The effective frontal density λ_f,eff orders pedestrian-level wind exposure better than plan area density alone in the analyzed zones.","Turbulence intensity peaks between H_avg and H_max across all directions, pointing to tall buildings as the main source of turbulent kinetic energy above the canopy."],"supporting_citations":[{"why":"Supplies the validated LES methodology, mesh design, and precursor-inflow setup that the sixteen direction simulations reuse.","marker":"Teng et al. (2025)"},{"why":"Provides realistic-urban LES reference data for double-averaged profiles and for inflection points near the average building height.","marker":"Giometto et al. (2016)"},{"why":"Provides idealized cube-array double-averaged profiles against which the present neighborhood profiles are compared.","marker":"Coceal et al. (2006)"},{"why":"Offers oblique-incidence street-system data used to interpret Zone 4 profile behavior at 180 degrees inflow.","marker":"Castro et al. (2017)"},{"why":"Establishes the double-averaging decomposition that separates temporal fluctuations from spatial deviations.","marker":"Raupach and Shaw (1982)"},{"why":"Fixes the vertical range used to fit displacement height and roughness length for each inflow direction.","marker":"Kanda et al. (2013)"},{"why":"Supplies the effective frontal area density scaling that orders pedestrian-level wind metrics better than plan area density.","marker":"Duan et al. (2023)"},{"why":"Documents wind-dependent pedestrian-level ventilation corridors in a real city, the phenomenon the paper quantifies locally.","marker":"Wang et al. (2020)"},{"why":"Provides the δ/H > 50 criterion justifying the domain's vertical extent for a fully developed rough-wall flow.","marker":"Jimenez (2004)"}],"fun_headline_variants":["Local wind shifts, but district turbulence profiles hold","Streets vary with wind, but district averages don't budge","Pedestrian wind flips per direction; district profiles unchanged","Urban wind: local chaos, neighborhood calm","Sixteen wind directions, one uniform district profile"],"cache_read_input_tokens":23168,"weakest_assumption_plain":"The collapse of the double-averaged profiles depends on rotating the city geometry instead of the wind direction being dynamically equivalent and numerically unbiased across all sixteen orientations, while the mesh sensitivity check is performed for only one direction (Φ = 180 degrees).","fun_headline_variants_meta":{"raw":{"variants":["Local wind shifts, but district turbulence profiles hold","Streets vary with wind, but district averages don't budge","Pedestrian wind flips per direction; district profiles unchanged","Urban wind: local chaos, neighborhood calm","Sixteen wind directions, one uniform district profile"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000538,"raw_usage":{"total_tokens":2536,"prompt_tokens":853,"completion_tokens":1683,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":469,"completion_tokens_details":{"reasoning_tokens":1607}},"tokens_in":469,"tokens_out":1683,"duration_ms":13926,"temperature":1.0,"reasoning_tokens":1607,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-08T00:26:26.521930+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-run the same district for one wind direction using two different geometry rotations that should be physically identical (e.g., the 180-degree case and the same geometry rotated by 22.5 degrees with inflow adjusted accordingly) and compare the double-averaged profiles; if the two representations differ by more than the inter-direction scatter seen in the present profiles, the collapse is partly a numerical artifact.","supporting_citations":[{"cited_title":", Dur\\'o Diaz , J.M","cited_arxiv_id":null,"evidence_quote":"Supplies the validated LES methodology, mesh design, and precursor-inflow setup that the sixteen direction simulations reuse."},{"cited_title":", Thomas , T.G","cited_arxiv_id":null,"evidence_quote":"Provides idealized cube-array double-averaged profiles against which the present neighborhood profiles are compared."},{"cited_title":", Xie , Z.T","cited_arxiv_id":null,"evidence_quote":"Offers oblique-incidence street-system data used to interpret Zone 4 profile behavior at 180 degrees inflow."},{"cited_title":", Shaw , R.H","cited_arxiv_id":null,"evidence_quote":"Establishes the double-averaging decomposition that separates temporal fluctuations from spatial deviations."},{"cited_title":", Inagaki , A","cited_arxiv_id":null,"evidence_quote":"Fixes the vertical range used to fit displacement height and roughness length for each inflow direction."}],"review_version":1}