{"id":"e24f5af0-2636-48cc-9937-34ecafe414a3","arxiv_id":"1908.06027","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Wind tunnel tracer tests over a regular building array show stable air roughly doubles in-street concentrations, convective air cuts them by up to three, and plume width changes little.","lead":"Wind tunnel experiments on a miniature city show that warm or cold air layers change where pollution goes: stable air keeps pollutants near the ground, while convective air lifts them away. The openly released dataset gives urban dispersion models a rare benchmark for non-neutral conditions, which most models currently ignore.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"K-theory confirmation and Kz parametrisations (Eqs. 5-6) rest on Gaussian-fit gradients with no uncertainties and only seven mean values; the qualitative width-vs-depth claim is likely robust, but the fitted laws are not.","rationale":"The reader's weakest assumption was the adequacy of the Gaussian representation in Eq. (3) without residuals or uncertainties. I agree that this is a genuine weakness, but I would locate the most load-bearing consequence in Section 5 and the resulting parametrisations, because those are the quantitative deliverables of the paper and are directly used to support the K-theory claim. The raw concentration profiles and the systematic reversal of in-canopy concentration levels between stable and convective cases provide independent support for the qualitative central claim, so I do not see grounds to reject or move the verdict away from CONDITIONAL. The concern is addressable by re-analysis of the already published data, and the verdict should remain CONDITIONAL pending that check, which is why I recommend UNCHANGED rather than a new verdict. I did not find grounds for a stronger objection such as internal inconsistency or a circular argument; the fitted-gradient procedure is a methodological limitation, not a soundness failure.","tokens_in":14614,"tokens_out":2495,"duration_ms":29647,"concrete_test":"Recompute Kz from the archived figshare dataset using finite-difference gradients of the raw vertical mean-concentration profiles at the three Table 2 locations, with bootstrap confidence intervals derived from the reported standard errors, and compare to the Gaussian-fit values. If Kz varies by more than a factor of two with height, or if the finite-difference-based Kz values differ systematically from Table 2, then the K-theory confirmation and Eqs. (5)-(6) should be downgraded to illustrative rather than predictive.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim includes the confirmation of K-theory, expressed quantitatively in Eqs. (5)-(6). This confirmation depends on the proportionality between measured vertical turbulent fluxes and the vertical concentration gradient. However, the gradient is not measured directly: it is obtained by differentiating a Gaussian fit to the mean concentration vertical profile (Fig. 16 and surrounding text), and the paper itself notes that within the canopy the concentration is approximately constant with height, so a Gaussian is not obviously the right shape there. No residuals, goodness-of-fit statistics, or confidence intervals are reported for these fits, and the Kz values in Table 2 are means over only three locations, with no uncertainty estimates. The parametrisations in Eqs. (5)-(6) are second-order polynomials fitted to seven mean values (two neutral, three stable, two convective), with three free parameters each, so they risk overfitting and have no independent validation. Because Kz enters directly into the advertised quantitative conclusion, the lack of uncertainty and the reliance on fitted gradients make the strength of the K-theory claim and the predictive use of Eqs. (5)-(6) the most load-bearing weak point. The qualitative conclusions about plume width versus vertical spread are supported by raw profiles independent of the Gaussian fits, so they are not the primary concern.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper reports wind-tunnel experiments on passive scalar dispersion from a ground-level point source in a regular array of rectangular buildings, under three stable, two convective, and two neutral reference boundary layers (bulk Richardson numbers from about -1.5 to 0.29). The central observational claims are that stratification affects the vertical spread of the plume much more than its lateral width, that stable stratification increases in-canopy mean concentrations by up to roughly a factor of two while convective stratification lowers them by up to roughly a factor of three, that the plume axis inside the canopy is unaffected by stable stratification but deviates under the strongest convective case, and that the vertical turbulent pollutant flux is proportional to the vertical mean concentration gradient, with fitted eddy-diffusivity parameterizations given in Eqs. (5) and (6). The paper also emphasizes the value of the dataset for validating numerical models and developing parameterizations for non-neutral urban dispersion.","tokens_in":14857,"tokens_out":4302,"duration_ms":43842,"significance":"If the quantitative conclusions are accepted, this is a valuable contribution to a sparse experimental literature: there are very few laboratory datasets on stratified flow and dispersion over urban-like arrays, and the authors provide concentration means, variances, and fluxes with reported standard errors and a public data repository. The qualitative result that vertical spread is more sensitive to stratification than lateral spread is physically plausible and supported by the raw profiles, and the contrast with Kanda and Yamao (2016) is worth documenting. The K-theory confirmation and the parameterizations in Eqs. (5)-(6), however, are not yet supported to the quantitative level claimed, because the analysis depends on Gaussian fits whose quality and uncertainty are not reported and because the polynomial fits use seven mean values with three free parameters. The paper would be a strong experimental contribution after the quantitative claims are either properly qualified or backed with fit diagnostics and uncertainty estimates.","major_comments":[{"comment":"The claim that K-theory is 'confirmed' in the stratified cases is not quantitatively supported as presented. In Fig. 16 the vertical concentration gradient is obtained by differentiating a Gaussian fit rather than from measured differences, yet no residuals, goodness-of-fit measures, confidence intervals, or propagation of fitting uncertainty are reported. Table 2 lists Kz values from three locations with no standard errors or ranges, and the values vary by roughly a factor of three across locations (e.g., from 0.02 to 0.035 in the neutral SBL case). Without such uncertainty information, the visual proportionality in Fig. 16 cannot be distinguished from an artifact of a smooth fitted profile, especially inside the canopy where Section 4.1 and Fig. 4 indicate the mean concentration is approximately constant with height.","section":"§5, Fig. 16, Table 2"},{"comment":"The polynomial parameterizations Kz(δ/L) and Kz(Riδ) are descriptive fits to seven mean values (two neutral, three stable, two convective) with three free coefficients each, and no uncertainties are reported for the coefficients or the fitted curve. This is too few points for a second-order polynomial to be presented as a predictive parameterization, and there is no independent validation or cross-validation. The paper should either present these as purely illustrative fits with explicit caveats, or add uncertainty bounds, a validation subset, or a physically motivated functional form with fewer free parameters.","section":"§5, Eqs. (5)-(6), Fig. 17"},{"comment":"All quantitative plume width and depth statistics (σh, σz), the plume-axis angles, and the vertical concentration gradient used in the K-theory test are derived from Gaussian fits, but the manuscript provides no fit diagnostics. Section 4.1 states the fit was 'remarkably satisfactory' without showing residuals or goodness-of-fit statistics. Given that Fig. 4 shows near-constant concentration with height inside the canopy, the vertical Gaussian assumption is not self-evident. The qualitative width-versus-depth conclusion is supported by the raw profiles and is likely robust, but the quantitative σ values and the fitted gradient should be reported with uncertainty estimates or replaced by direct estimates from the measured profiles.","section":"§4.1, Eq. (3) and similar vertical fits"},{"comment":"The abstract and conclusions state that in the unstable case the plume central axis 'appeared to deviate' from the neutral direction inside the canopy, but this rests on a single quantified case (Riapp = -1.5). The text notes that the weaker convective case (Riapp = -0.5) gives a plume direction inside the urban model 'close to the neutral reference case' and that figure is not shown. This is a thin empirical basis for a headline claim, and the statement should be softened to report the one-case observation or supplemented with the weaker-case data.","section":"§4.2, Fig. 9"}],"minor_comments":[{"comment":"The sentence 'while in convective conditions they were to three times smaller' should read 'they were up to three times smaller'.","section":"Abstract"},{"comment":"In the paragraph describing the model, 'Al the experiments' is a typo and should be 'All the experiments'.","section":"§2"},{"comment":"The phrase 'BL detpthδ' contains a typo; it should be 'BL depth δ'. Also, the row in Table 1 for u*/UREF appears to merge two values ('0.0590.081') without a separator.","section":"§3"},{"comment":"The σh and σz plots have no error bars or uncertainty intervals even though the manuscript reports standard errors for the underlying concentration statistics; adding them would help the reader judge whether the small differences between stratification cases are meaningful.","section":"Figs. 6, 7, 12, 13"},{"comment":"The citation 'Dezs˝o-Weidinger, G.' should be rendered with the correct diacritic, e.g., 'Dezső-Weidinger'.","section":"References"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The dataset here is the thing. Marucci and Carpentieri have measured dispersion in a regular building array under three stable and two convective boundary layers, with lateral and vertical mean concentration profiles, variances, and vertical fluxes. That is more coverage of non-neutral stratification than anything else I know in a wind tunnel, and the qualitative result—stable air roughly doubles in-canopy concentrations, convective air cuts them by up to three, and stratification hits vertical plume spread much harder than lateral spread—is supported by the raw profiles and consistent with Briggs. If you work on urban dispersion or validation for LES of stratified canopies, this is a useful paper.\n\nThe soft spots are real but mostly optional. Plume widths, depths, and axis angles all come from Gaussian fits, and the paper reports no residuals or fit uncertainties. The same is true for the vertical concentration gradient used in the K-theory test: it is obtained by differentiating a Gaussian fit to the mean profile, and the authors themselves note that concentration is nearly constant with height inside the canopy, so a Gaussian is not obviously the right shape there. The Kz values in Table 2 are means over three locations with no error bars, and the two second-order polynomial parametrisations (Eqs. 5–6) are each fitted to seven mean values with three free parameters. That is overfitting, not prediction, and I would not use those equations as they stand. I would also be cautious about the claim that the plume axis shifts in the CBL: the quantified example is a single case (Riapp = −1.5), and the weaker case gives a similar shift only above the canopy. None of this undermines the central experimental contribution, but the paper overstates how cleanly the K-theory confirmation comes out. The abstract and conclusion should carry the same caveat.\n\nThe experimental methods look careful: standard errors are quoted for first and second moments, the source is passive, and the companion papers describe the flow-field characterisation. The writing is clear. The dataset is openly available on figshare, which is a real plus.\n\nWorth a serious referee. I would ask the authors to add fit diagnostics and uncertainty intervals for the Gaussian parameters, to either drop or heavily qualify the Kz fits, and to soften the K-theory wording. Even with open data, this needs peer review rather than desk rejection. Bring it to the reading group if you want a concrete example of how experimental papers should report stratified-canopy dispersion data.","headline":"New stratified-canopy dispersion data are the genuine contribution; the Kz parametrisation is overfitted and should not be taken as predictive.","tokens_in":15388,"tokens_out":2453,"would_cite":true,"duration_ms":23770,"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":"Wind-tunnel tests show stable stratification can double pollutants inside a building canopy, convective can cut them by two-thirds, with plume width barely changing.","keywords":["stable boundary layer","convective boundary layer","wind tunnel experiment","urban building array","pollutant dispersion","plume width","eddy diffusivity","Richardson number"],"falsifier":"Take the raw concentration profiles and recompute the reported quantities without any shape assumption: estimate $\\sigma_h$ and $\\sigma_z$ from the second moments of the measured distributions by numerical integration, recover $\\partial\\bar{C}/\\partial z$ by finite differences of the measured mean concentrations, and locate plume-axis positions from the measured peak values rather than from Gaussian fits. If these direct estimates agree with the fitted values, the claims stand; if they drift apart near the source inside the canopy, the central comparisons — width nearly unaffected, vertical gradient proportional to flux — rest on the Gaussian assumption rather than on the measurements.","tokens_in":14381,"feed_emoji":"💨","tokens_out":20353,"duration_ms":159970,"temperature":0.7,"pith_summary":"This wind-tunnel study asks whether atmospheric stratification changes how a pollutant plume spreads through a regular array of rectangular buildings. The paper's answer is that the stratification effect on plume width is significantly smaller than the effect on vertical profiles: in stable incoming flows, measured concentrations inside the canopy were up to two times larger than in the neutral reference, and in convective flows they were up to three times smaller, while the lateral width of the plume barely changed. These measurements across five stratified boundary layers (Richardson number from -1.5 to 0.29) compared with two neutral references provide a rare experimental check of the common assumption that urban dispersion models can ignore stratification because building-generated turbulence dominates. The paper also confirms that the flux–gradient closure known as K-theory holds in the stratified cases and supplies fitted eddy-diffusivity coefficients for modelling it.","feed_headline":"Stable air doubles street pollution; convection cuts it to a third","feed_subtitle":"Plume width barely moves with weather layering, so models that ignore stratification miss the real urban risk.","key_machinery":"The argument runs on two devices. First, Gaussian fits to the measured concentration profiles, $C = A \\exp\\!\\left(-(y_{\\mathrm{plume}}-\\mu)^2/(2\\sigma_h^2)\\right)$ for lateral cuts and the analogous form with $\\sigma_z$ for vertical cuts: every derived quantity — the plume central-axis angle, the width $\\sigma_h$, the depth $\\sigma_z$, and the vertical gradient $\\partial\\bar{C}/\\partial z$ used in the flux test — is read from these fitted curves rather than from direct moments of the data. Second, the K-theory flux–gradient identity $K_z\\,\\partial\\bar{C}/\\partial z = -\\overline{w'c'}$, which the paper tests by comparing measured vertical turbulent pollutant fluxes with the gradient of the Gaussian-fitted mean concentration at the same measurement locations, then collapses the fitted proportionality constant $K_z$ onto the bulk stability parameters $\\delta/L$ and $\\mathrm{Ri}_\\delta$ through quadratic parametrisations.","core_discovery":"On the paper's own terms, the central claim is that “the stratification (either stable or unstable) effect on the plume width is significantly lower than the effect on the vertical profiles.” Stable stratification doubled the mean concentrations measured inside the canopy (up to two times the neutral values, increasing with Richardson number) while leaving the plume's central axis unchanged — the axis stayed channelled by the street geometry at about 14.7° — and convective stratification cut canopy concentrations by up to a factor of three while deflecting the in-canopy plume axis by about 20% more than neutral flow. Above the canopy both types of stratification increased the plume deflection angle slightly (8.6° to 10.8° for the stable case). The fitted plume depth $\\sigma_z$ was reduced by up to 30% under stable layers and markedly deepened under convective ones, whereas the fitted width $\\sigma_h$ changed only marginally, reproducing the pattern of earlier field experiments. Finally, the vertical turbulent pollutant flux scaled with the vertical mean-concentration gradient, confirming K-theory ($K_z\\,\\partial\\bar{C}/\\partial z = -\\overline{w'c'}$) under stratification, with fitted mean eddy diffusivities $K_z(\\delta/L) = 0.0202\\,(\\delta/L)^2 - 0.0425\\,(\\delta/L) + 0.0306$ and $K_z(\\mathrm{Ri}_\\delta) = -0.0064\\,\\mathrm{Ri}_\\delta^2 - 0.0839\\,\\mathrm{Ri}_\\delta + 0.0294$.","pith_inferences":["The mean $K_z$ values in the parametrisations hide a factor-of-three spread across the three measurement locations (0.009–0.14), which suggests that a single scalar eddy diffusivity may be too crude for street-scale models and that location-dependent or turbulence-based $K_z$ formulations would be needed to exploit the confirmed K-theory.","The in-canopy axis deflection under convection is quantified for only the strongest unstable case ($\\mathrm{Ri}^{\\mathrm{app}}_\\delta = -1.5$); testing whether the axis shift scales with the convective velocity ratio $w_*/U_{\\mathrm{REF}}$ across several instability levels would show whether it is a general convective-canopy feature.","Because the array is regular and aligned at 45°, the width-insensitivity result is tied to this geometry; the natural generalisation — which the authors themselves gesture at with tall buildings — is to re-test whether lateral spread stays neutral-like in staggered or heterogeneous urban layouts before treating it as a universal stratification result."],"forward_implications":["Urban dispersion models that neglect stratification will mis-predict in-canopy pollutant levels by factors of two to three in the tested range, because the vertical concentration response is large while the lateral plume width stays almost neutral-like.","Neutral-case lateral dispersion parametrisations can probably be kept in stratified urban modelling, with stratification corrections applied to vertical spread and to in-canopy concentration, since $\\sigma_h$ changed only slightly in all five stratified cases.","Models that use K-theory closures can carry the same flux–gradient proportionality into stratified conditions, with eddy diffusivity reduced under stable layers and increased under convective ones according to the fitted quadratic forms.","Scaled to full size, the tested stratification range corresponds to roughly the conditions observed about 75% of the time over London ($-1 < z'/L < 1$), so the results bear on common urban conditions rather than only extreme ones.","The reported profiles of mean concentration, variance, and total and turbulent vertical fluxes form a benchmark for validating large-eddy simulations and fast dispersion models under non-neutral stratification."],"supporting_citations":[{"why":"Earlier field experiments over urban roughness that found plume depth more sensitive to stratification than plume width; the pattern this paper reproduces.","marker":"Briggs (1973)"},{"why":"The prior wind-tunnel study of dispersion over cube arrays in stable and unstable flow that found width more affected than depth; the dataset this paper explicitly contrasts with.","marker":"Kanda and Yamao (2016)"},{"why":"Companion experiments supplying the flow, turbulence and temperature fields of the same five boundary layers, against which the dispersion results are interpreted.","marker":"Marucci and Carpentieri (2019b)"},{"why":"The methodology for generating thick stable and convective boundary layers in the wind tunnel that every stratified case depends on.","marker":"Marucci et al. (2018)"},{"why":"Neutral-case wind-tunnel confirmation of the flux–gradient proportionality at urban intersections that the stratified K-theory test extends.","marker":"Carpentieri et al. (2012)"},{"why":"The demonstration that vertical turbulent pollutant flux is proportional to the mean concentration gradient, which the K-theory identity rests on.","marker":"Dezső-Weidinger et al. (2003)"},{"why":"Large-eddy simulation of stable stratification over an aligned cube array predicting larger street concentrations and a shallower internal boundary layer, expectations the stable measurements agree with.","marker":"Tomas et al. (2016)"},{"why":"A London climatology showing weakly stratified conditions prevail about 75% of the time, justifying the tested stratification range as representative of real urban conditions.","marker":"Wood et al. (2010)"}],"fun_headline_variants":["Stable air doubles street pollution; convection cuts to a third","Stable stratification doubles urban pollution; convective cuts","Wind tunnel: stable air doubles street levels, convective cuts","Plume width unchanged, but stable air doubles street pollution"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Everything quantitative in the study — plume angles, widths, depths, and the vertical gradients used to confirm K-theory — is read off Gaussian curves fitted to the measured concentration profiles, and the paper reports no goodness-of-fit statistics, so if those profiles are not genuinely Gaussian (especially inside street canyons one building height from the source) the comparisons lose their quantitative footing.","fun_headline_variants_meta":{"raw":{"variants":["Stable air doubles street pollution; convection cuts to a third","Stable stratification doubles urban pollution; convective cuts","Wind tunnel: stable air doubles street levels, convective cuts","Plume width unchanged, but stable air doubles street pollution"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000414,"raw_usage":{"total_tokens":2200,"prompt_tokens":1065,"completion_tokens":1135,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":681,"completion_tokens_details":{"reasoning_tokens":1069}},"tokens_in":681,"tokens_out":1135,"duration_ms":9121,"temperature":1.0,"reasoning_tokens":1069,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:58:08.040205+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the raw concentration profiles and recompute the reported quantities without any shape assumption: estimate $\\sigma_h$ and $\\sigma_z$ from the second moments of the measured distributions by numerical integration, recover $\\partial\\bar{C}/\\partial z$ by finite differences of the measured mean concentrations, and locate plume-axis positions from the measured peak values rather than from Gaussian fits. If these direct estimates agree with the fitted values, the claims stand; if they drift apart near the source inside the canopy, the central comparisons — width nearly unaffected, vertical gradient proportional to flux — rest on the Gaussian assumption rather than on the measurements.","supporting_citations":[{"cited_title":", year 1973","cited_arxiv_id":null,"evidence_quote":"Earlier field experiments over urban roughness that found plume depth more sensitive to stratification than plume width; the pattern this paper reproduces."},{"cited_title":", author Yamao, Y","cited_arxiv_id":null,"evidence_quote":"The prior wind-tunnel study of dispersion over cube arrays in stable and unstable flow that found width more affected than depth; the dataset this paper explicitly contrasts with."},{"cited_title":", author Carpentieri, M","cited_arxiv_id":null,"evidence_quote":"The methodology for generating thick stable and convective boundary layers in the wind tunnel that every stratified case depends on."},{"cited_title":", author Hayden, P","cited_arxiv_id":null,"evidence_quote":"Neutral-case wind-tunnel confirmation of the flux–gradient proportionality at urban intersections that the stratified K-theory test extends."},{"cited_title":", author Stitou, A","cited_arxiv_id":null,"evidence_quote":"The demonstration that vertical turbulent pollutant flux is proportional to the mean concentration gradient, which the K-theory identity rests on."},{"cited_title":", author Pourquie, M.J.B.M","cited_arxiv_id":null,"evidence_quote":"Large-eddy simulation of stable stratification over an aligned cube array predicting larger street concentrations and a shallower internal boundary layer, expectations the stable measurements agree with."},{"cited_title":", author Lacser, A","cited_arxiv_id":null,"evidence_quote":"A London climatology showing weakly stratified conditions prevail about 75% of the time, justifying the tested stratification range as representative of real urban conditions."}],"review_version":1}