Vertical responses of urban heat islands to heatwaves and their local climate zones dependence: a case study of Guangzhou, China

Luo, Fu; Yang, Yuanjian; Sun, Ting; Zheng, Zuofang; Bi, Xueyan; Ren, Guoyu; Lolli, Simone

doi:10.5194/egusphere-2026-3392

Preprints

https://doi.org/10.5194/egusphere-2026-3392

Preprints

24 Aug 2026

| 24 Aug 2026

Vertical responses of urban heat islands to heatwaves and their local climate zones dependence: a case study of Guangzhou, China

Fu Luo, Yuanjian Yang, Ting Sun, Zuofang Zheng, Xueyan Bi, Guoyu Ren, and Simone Lolli

Abstract. The interaction between urban heat island (UHI) and heatwaves (HWs) shapes urban extreme heat, but whether HW-induced UHI amplification reorganises canopy-to-boundary-layer thermal structure, and whether local climate zones (LCZs) modulate this response, remains unclear. Here, we investigated two July 2020 HW events in the Guangzhou metropolitan area in southern China, using 1 km LCZ-coupled Weather Research and Forecasting (WRF) simulations. We analysed the vertical thermal response of the urban area and individual LCZs using canopy and boundary-layer UHI intensities (CUHII, BUHII), layer-mean urban–rural air-temperature differences (Layer ∆T ), and surface-energy-balance and turbulence indicators. HWs amplified both CUHII and BUHII in a vertically non-uniform manner: CUHII increased more than BUHII, and additional warming concentrated in the canopy and lower boundary layer. Warming propagated upward as the daytime boundary layer developed but contracted toward the surface at night and during transition periods, with upper-level responses weakening or reversing. LCZ heterogeneity was largely confined to the lowest ∼250 m: LCZ 1 showed the strongest additional warming, LCZ 2 and LCZ 4 reached comparable levels through different energy–turbulence pathways, and LCZ 6 was weakest. Stronger sensible heat flux, higher Bowen ratios, sustained nocturnal heat sources, and reduced low-level stability jointly drove this diurnal reorganisation. During HW1, residual-layer warm anomalies in LCZ 1 ( 0.41 °C at 500–600 m) persisted through the night and provided an additional heat reservoir for next-day mixed-layer development, whereas LCZ 6 showed much weaker memory effects. HW2 reproduced the inter-LCZ ranking with weaker absolute amplitude.

Received: 11 Jun 2026 – Discussion started: 24 Aug 2026

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Preprint (PDF, 17179 KB)

Supplement (9475 KB)

Download & links

Preprint (17179 KB)
Metadata XML
Supplement (9475 KB)
BibTeX
EndNote

Fu Luo, Yuanjian Yang, Ting Sun, Zuofang Zheng, Xueyan Bi, Guoyu Ren, and Simone Lolli

Status: final response (author comments only)

RC1:
'Comment on egusphere-2026-3392', Anonymous Referee #3, 01 Sep 2026
This manuscript investigates the vertical structure of heatwave–UHI interactions and their LCZ dependence over Guangzhou using WRF with multi-layer urban canopy schemes. The topic is relevant and is addressed with an appropriate modelling framework and a thoughtfully designed set of diagnostics, and several results are credible. However, the paper's central conclusions on the vertical structure and its LCZ dependence are not yet adequately supported, owing to limited observational constraints, questionable comparability of the analysis periods, insufficient treatment of uncertainty, and interpretations that go beyond what the analysis design can attribute. I recommend major revision. Please see detailed comments below.
Introduction:
The land-sea breeze needs to be emphasised, as the Greater Bay Area is largely influenced by it.

The coupling between the nocturnal stable boundary layer and the residual layer is a good point; I suggest the authors also provide a brief literature review (mechanism/numerical simulation) before introducing it in question 2.

Method:
The authors used the Qingyuan sounding data to validate the model’s ability to reproduce vertical fields; however, the measurements only cover the day-night transition periods (0800 and 2000). The model’s ability to simulate the daytime convective boundary layer and the nocturnal stable boundary layer has not yet been validated. The authors should justify the selection or use other possible sources of data (e.g. Canton Tower observations, Lidar).

Model configuration: the lowest vertical level and resolution are missing. I am also concerned about the urban parameters used; does LCZ-UCP use the default table or localised parameters? Does the BEP-BEM use Guangzhou’s local properties or default values?

Case selection: The background synoptic type/circulation and cloud coverage of the NHW period are not presented. How does the author ensure that the HW-NHW difference is solely due to HW?

The definition of BUHII is quite dynamic, and the min() cap could exclude the layer between the rural and urban PBLH, where the nocturnal boundary-layer UHI (and the SBL–residual-layer coupling of question 2) is strongest; the authors need to justify this.

Line 195: z₀ = 30 m is considered the height of the above-canopy layer; however, in Guangzhou, LCZ1/4 buildings mostly exceed 30 m. Please justify; maybe use a normalised height?

Conflict in definition of disruption: ‘A layer with ∆∆T < 0 was regarded as showing a disruption in the upward propagation of HW-induced additional warming.’ That is, the situation where ∆T is weakening, but ∆∆T > 0, is considered continuity. However, the introduction section defines the disruption as: ‘marked weakening, stagnation, or even reversal’.

Result:
The simulated urban PBLH bias was 343 m (42%) higher than PBLH product.Although the authors stated that Hc is set by the rural PBLH at 98.7% of times, and at all nighttime steps, confirms that the BUHII integral almost never samples the layer between the rural and urban PBL tops. Reinforcing my concern in comment 4 above.

The validation shows that the rural 2m air temperature is underestimated during HW. This will raise all daytime ∆T values. Therefore, the ‘clear daytime amplification’ (positive ∆∆T during 1100-1500 in Fig. 6d is suspicious. Further quantification throughout the paper is needed.

Fig. 7b: The result shows that the differences in BUHII between LCZs are much smaller than in CUHII. The authors acknowledged that ‘LCZ-induced heterogeneity of UHII was concentrated in the canopy and the lower part of the boundary layer and was appreciably damped once the boundary-layer-mean was taken’. However, this makes the following discussion of residual-layer thermal memory’s LCZ differences (Fig. 9) logically flawed.
Citation: https://doi.org/10.5194/egusphere-2026-3392-RC1
RC2:
'Comment on egusphere-2026-3392', Anonymous Referee #1, 06 Oct 2026
General comments
The authors run WRF/BEP-BEM over the Guangzhou–Foshan core for July 2020, and analyse two heatwaves (HW1 12–16 July, HW2 23–29 July) against a non-heatwave period to explore the vertical extent and spatiotemporal distribution of heatwave-UHI synergy. The topic is interesting and important for the science community, and the manuscript is well written. However, there are still some problems that need to address before it can be further considered for publication. A major revision is therefore recommended.
Major comments
The manuscript tested various configurations regarding PBL schemes, BEP/BEM, and LULC, but leaves out the YSU PBL scheme that can also be coupled with BEP/BEM. Since the major concern of this manuscript is the UHI in boundary layer, it is important to test all the possible options available to determine the optimal schemes. For the same reason, the manuscript should evaluate the performance for wind, which is important for boundary layer development. In addition, some studies in this region using WRF/MUUCM/LCZ have reported that using the building drag coefficients implemented in post-V4.3 WRF model usually underestimate 10-m wind speed, while using constant drag coefficient of 0.4 can mitigate this underestimation (Xin et al, 2023, 2024). So, it is interesting if the authors can report the building drag coefficients used in this study, and if possible, if the wind speed does improve when using constant drag coefficient.

The inter-LCZ order of ΔT is confounded with the radial position within the city. As stated in Section 2.4, the analysed LCZs are spatially nested rather than interspersed: LCZ 1 is in the city centre, LCZ 2 is distributed mainly around LCZ 1, LCZ 4 surrounds LCZ 1 and LCZ 2, and LCZ 6 is “distributed around the periphery of the city centre.” Every metric in the manuscript is a difference against a rural reference outside the built-up area. The background urban–rural temperature field decays outward from the core, so a cell's ΔT depends on how far it sits along that gradient, independently of its morphology. The reported ordering (LCZ 1 > LCZ 4 ≈ LCZ 2 > LCZ 8 > LCZ 6) is exactly the ordering that radial position alone would produce, and no analysis in the manuscript separates the two. The deviation-from-CUR-mean presentation in Fig. 8f–j helps but does not remove the gradient, which persists inside the CUR. The statements such as "LCZ heterogeneity … LCZ 1 showed the strongest additional warming” should be phrased as an association with LCZ class, not as a demonstrated morphological control.

Further to the above comment, since the manuscript examines the LCZ heterogeneity’s contribution to the vertical distribution of UHI, it is important to know if the model simulations incorporated the morphological characteristics of Guangzhou-Foshan. The model parameters associated with each LCZ should be reported. Are the default LCZ parameters or the locally-derived parameters used?

The study rests on three days per period (6–8 July NHW, 13–15 July HW1). LCZ 1 has 8 grid cells, LCZ 2 has 26, LCZ 4 has 82, and the entire rural reference consists of 11 grid cells. Even worse, the connected-component screening deliberately retains contiguous “core patches”, so LCZ 1’s 8 cells are effectively a single patch: the effective number of independent spatial samples is close to one, and the grid cells are strongly correlated. Yet inter-LCZ differences of 0.05–0.15 °C (LCZ 4 at 0.47 °C versus LCZ 2 at 0.43 °C; LCZ 8 at 0.46 °C versus LCZ 6 at 0.38 °C) are presented as meaningful, and the entire ranking narrative is built on them. No confidence interval, significance test, or measure of spread is reported anywhere. Please report the number of grid cells and, more importantly, an effective sample size for each LCZ and for the rural reference; show the day-to-day and cell-to-cell spread behind each number in Table 3; and state plainly which inter-LCZ differences are actually statistically significant.

Hc = min(PBLHu, PBLHr), and the manuscript reports that the rural PBLH sets the limit at 98.7 % of times. Two consequences follow that the authors did not address. First, at night Hc is shallow, and Fig. 3d shows rural nocturnal/early-morning PBLH is itself underestimated, so BUHII at night is a mean over roughly 30–150 m and is not an independent diagnostic of CUHII; the claim that “BUHII was also enhanced at night” is close to restating the near-surface result over a slightly deeper layer. Second, during the day the layer between the rural and urban PBL tops is excluded from the integral, and that is precisely the layer where Fig. 5f shows urban ΔT extending to ~1500 m. The daytime damping of the BUHII response therefore partly results from this definition. The day–night contrast in (CUHII − BUHII), which is a major result of this manuscript, is largely geometric. If possible, please show BUHII computed with fixed upper limits (e.g. 300 m, 500 m, 1000 m) and with Hc = PBLHu as sensitivity tests, and demonstrate that the conclusion is robust.

z0 = 30 m is introduced as “the height of the above-canopy layer”, but LCZ 1 (compact high-rise) and LCZ 4 (open high-rise) are high-rise classes whose buildings routinely exceed 30 m (especially in Guangzhou's core), while LCZ 6 and LCZ 8 are low-rise (3–10 m). The same fixed 30 m lower limit therefore sits inside the canopy for some classes and well above it for others, which makes the CUHII/BUHII separation and the inter-LCZ BUHII comparison height-inconsistent. Furthermore, the lowest model level is at ~26 m (Section 2.6) and only 40 levels span the column to 50 hPa, so the 30–100 m and 100–250 m layers contain only very few model levels. So please either adopt a class-dependent lower limit or state the inconsistency explicitly. Also, interpolating TKE from 40 to 71 levels (Section 2.7) adds no information and should not be described as “providing a more refined representation of the turbulent vertical structure.”

The residual-layer memory claim exceeds what is shown in the manuscript, and is not defensible at LCZ scale. (i) The analysis is a three-day composite, which removes precisely the day-to-day sequencing needed to demonstrate memory; a memory claim requires night-by-night and following-morning pairing, not a composite. (ii) Nothing establishes that the 500–600 m anomaly originated in the previous day's mixed layer; the observed persistence is equally consistent with a quasi-stationary urban–rural difference in the residual layer that simply never mixes out. (iii) Most seriously, attributing a 500–600 m signal to individual LCZ patches is not defensible: LCZ 1 is 8 grid cells, roughly 8 km², and at 500–600 m the flow is far above the blending height (the manuscript’s own result is that LCZ heterogeneity is confined below ~250 m, which is consistent with a blending height of that order). Attributing the 0.41 °C anomaly to LCZ 1’s morphology contradicts the paper's own finding.

Minor comments
Line 177: ‘three consecutive regional hot days constituted one HW event,’ how about more than three consecutive regional hot days? Will they be split into different HW events?

Section 3.3 states “LCZ 2 was slightly stronger than LCZ 4 at night,” but Table 3 gives 30-100 m nocturnal ΔΔT of 0.49 °C for both (equal), and nocturnal ΔCUHII of 0.83 °C for LCZ 2 versus 0.84 °C for LCZ 4 (LCZ 4 marginally stronger). Two sentences earlier the same section says the nighttime values are “nearly equal (or a slightly stronger LCZ 2 in HW2)”. Please resolve, and check whether the intended statement is about HW2.

ΔΔTKE is defined in Section 3.4.1 as the CUR-rural difference of ΔTKE (Fig. 11c) and in Section 3.4.2 as the LCZ-CUR deviation of ΔTKE (Fig. 13). These are different quantities with the same symbol; please distinguish them.

(2) is described as a “thickness-weighted mean”; it is a simple vertical average over Hc - z0. Please correct the wording.

Line 165: “LCZ 8 contained the largest number of grid cells and was closer to the CUR-mean state; it was therefore used as a background reference,” but deviations are then computed against the CUR mean rather than against LCZ 8.

References
Xin, R, Li, XX, Shi, Y, et al (2023). Study of urban thermal environment and local circulations of Guangdong-Hong Kong-Macao Greater Bay Area using WRF and local climate zones. Journal of Geophysical Research: Atmospheres, 128, e2022JD038210. DOI: 10.1029/2022JD038210.
Xin, R, Li, XX, Du, Y, et al (2024). Simulation of a cold spell in Guangdong-Hong Kong-Macao Greater Bay Area with WRF: Sensitivity to PBL schemes. Atmospheric Research, 310, 107640. DOI: 10.1016/j.atmosres.2024.107640.
Citation: https://doi.org/10.5194/egusphere-2026-3392-RC2

Fu Luo, Yuanjian Yang, Ting Sun, Zuofang Zheng, Xueyan Bi, Guoyu Ren, and Simone Lolli

Supplement

https://doi.org/10.5194/egusphere-2026-3392-supplement

Data sets

The 30 m annual land cover datasets and its dynamics in China from 1985 to 2025 Yang and Huang https://zenodo.org/records/18180184

Global map of Local Climate Zones Demuzere et al. https://zenodo.org/records/6364594

A Harmonized Global Continental High-resolution Planetary Boundary Layer Height Dataset Covering 2017-2021 Guo et al. https://zenodo.org/records/6498004

NCEP GDAS/FNL 0.25 Degree Global Tropospheric Analyses and Forecast Grids National Centers for Environmental Prediction, National Weather Service, NOAA, U.S. Department of Commerce https://gdex.ucar.edu/datasets/d083003/

Integrated Global Radiosonde Archive (IGRA), Version 2 – Qingyuan station, July 2020 subset Durre et al. https://doi.org/10.7289/V5X63K0Q

Model code and software

Weather Research and Forecasting (WRF) model Skamarock et al. https://github.com/wrf-model/WRF/tree/release-v4.4

W2W: A Python package that injects WUDAPT’s LocalClimate Zone information in WRF Demuzere et al. https://github.com/matthiasdemuzere/w2w

Fu Luo, Yuanjian Yang, Ting Sun, Zuofang Zheng, Xueyan Bi, Guoyu Ren, and Simone Lolli

Viewed

Total article views: 376 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
219	90	67	376	71	74	65

HTML: 219
PDF: 90
XML: 67
Total: 376
Supplement: 71
BibTeX: 74
EndNote: 65

Views and downloads (calculated since 24 Aug 2026)

Month	HTML	PDF	XML	Total
Aug 2026	98	27	18	143
Sep 2026	95	53	44	192
Oct 2026	26	10	5	41

Cumulative views and downloads (calculated since 24 Aug 2026)

Month	HTML	PDF	XML	Total
Aug 2026	98	27	18	143
Sep 2026	95	53	44	192
Oct 2026	26	10	5	41

Viewed (geographical distribution)

Total article views: 352 (including HTML, PDF, and XML) Thereof 352 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 07 Oct 2026

Short summary

Cities are hotter than surrounding countryside, and heatwaves intensify this contrast. How combined warming reshapes air above different urban areas remains unclear. Using a high-resolution model, we examined two 2020 heatwaves over Guangzhou, China, finding extra warming strongest above densely built neighbourhoods and weakest over open low-rise areas. Dense areas retain heat overnight, adding to next-day warming and showing how urban form shapes extreme heat from surface to lower atmosphere.


Total:	0
HTML:	0
PDF:	0
XML:	0