GUST1.0: A GPU-accelerated 3D Urban Surface Temperature Model

Mei, Shuo-Jun; Chen, Guanwen; Hang, Jian; Sun, Ting

doi:https://doi.org/10.5194/egusphere-2025-1485

Preprints

https://doi.org/10.5194/egusphere-2025-1485

Preprints

19 May 2025

| 19 May 2025

GUST1.0: A GPU-accelerated 3D Urban Surface Temperature Model

Shuo-Jun Mei, Guanwen Chen, Jian Hang, and Ting Sun

Abstract. The escalating urban heat, driven by climate change and urbanization, poses significant threats to residents’ health and urban climate resilience. The coupled radiative-convective-conductive heat transfer across complex urban geometries makes it challenging to identify the primary causes of urban heat and develop mitigation strategies. To address this challenge, we develop a GPU-accelerated Urban Surface Temperature model (GUST) through CUDA architecture. To simulate the complex radiative exchanges and coupled heat transfer processes, we adopt Monte Carlo method, leveraging GPUs to overcome its computational intensity while retaining its high accuracy. Radiative exchanges are resolved using a reverse ray tracing algorithm, while the conduction-radiation-convection mechanism is addressed through a random walking algorithm. The validation is carried out using the Scaled Outdoor Measurement of Urban Climate and Health (SOMUCH) experiment, which features a wide range of urban densities and offers high spatial and temporal resolution. This model exhibits notable accuracy in simulating urban surface temperatures and their temporal variations across different building densities. Analysis of the surface energy balance reveals that longwave radiative exchanges between urban surfaces significantly influence model accuracy, whereas convective heat transfer has a lesser impact. To demonstrate the applicability of GUST, it is employed to model transient surface temperature distributions at complex geometries on a neighborhood scale. Leveraging the high computational efficiency of GPU, the simulation traces 10⁵ rays across 2.3×10⁴ surface elements in each time step, ensuring both accuracy and high-resolution results for urban surface temperature modeling.

Received: 29 Mar 2025 – Discussion started: 19 May 2025

Competing interests: Dr Ting Sun is a member of the editorial board of Geoscientific Model Development.

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

Shuo-Jun Mei, Guanwen Chen, Jian Hang, and Ting Sun

Status: final response (author comments only)

RC1:
'Comment on egusphere-2025-1485', Sasu Karttunen, 25 Jun 2025
General statement
The manuscript presents GUST1.0, a model for simulating urban surface temperatures using reverse Monte Carlo ray tracing (rMCRT). The model is written in CUDA Python, targeting GPU-accelerated computing environments. It models radiative, conductive, and convective heat transfer processes of complex 3D urban environments. A validation against a scale-model outdoor urban experiment is presented.
With many of the new HPC platforms relying on GPUs for much of their computational power, there is a growing need for GPU-accelerated models and tools in the Earth sciences. Solving radiative transfer within urban canopies is for its difficulty to parallelize efficiently, and further advances in models in this area are needed to fully utilize the new HPC platforms in urban climate research.
However, I have some major concerns that need to be addressed before the paper can be reconsidered for publication in GMD.
Major comments
GUST1.0 as a standalone model has very limited capabilities compared to many other urban surface models. The main novelty is within its radiative transfer modelling, whereas the other parts of the model seem extremely simplistic or even incomplete compared to many other urban surface models (e.g. building-resolving models listed in Table 1 or urban surface models in general as in e.g. Urban-PLUMBER intercomparison by Lipson et al., 2024). The authors do not discuss the scope of applicability and the model’s limitations to a necessary extent, nor they adequately justify the publication of a stand-alone model rather than integrating the rMCRT method in a pre-existing model. In my opinion, the current version of the model has too limited real-life applicability for the paper to be considered a substantial contribution.

The model is not sufficiently described in the paper to allow for, in theory, complete reimplementation of the model by others as required by the GMD policy. The model structure and numerical methods used to solve the model equations are not sufficiently documented. With respect to the detail required from the model description, I point out the following excerpt from the journal policy:

“The main paper should describe both the underlying scientific basis and purpose of the model and overview the numerical solutions employed. The scientific goal is reproducibility: ideally, the description should be sufficiently detailed to in principle allow for the re-implementation of the model by others, so all technical details which could substantially affect the numerical output should be described. Any non-peer-reviewed literature on which the publication rests should be either made available on a persistent public archive, with a unique identifier, or uploaded as supplementary information.”

The model for convective heat flux uses a bulk transfer equation, which ignores natural convection. The natural convection becomes especially important in low-wind conditions and whenever the temperature difference between the wall and the atmosphere grows large (see e.g. Fan et al., 2021).

The assumption that the indoor air temperature is equal to the outdoor air temperature is a strong assumption, often invalid. It would require total and efficient ventilation of the indoor air in all conditions, which is not realistic during the heating season or if cooling is applied. This also contradicts the authors' own measurements showing indoor temperatures reaching 40°C (L465).

The model setup for the evaluation is not sufficiently documented. Here again, scientific reproducibility should be the goal. I would also recommend presenting some sensitivity analysis with respect to model inputs, or to quantify the uncertainties in another way.

Although the authors present evaluation against real-life scale-model measurements, they do not present information on how the model code has been verified. An excerpt from the journal policy:

“... authors are expected to distinguish between verification (checking that the chosen equations are solved correctly) and evaluation (assessing whether the model is a good representation of the real system). Sufficient verification and evaluation must be included to show that the model is fit for purpose and works as expected.”

Figure 9: The systematic underestimation of west wall temperatures suggests issues with either the radiation model or convective transport. The authors attribute this to sensitivity but don't fully investigate the cause.

Although the authors have made the model code public, I cannot find any kind of user manual for the model. Inclusion of a user manual is required for a model description paper in GMD (see manuscript type policy). Without a user manual it is very difficult for the end users to actually use the model for their applications. A minimal user manual would describe the model installation (or required runtime environment), inputs, outputs, and all other necessary details regarding the model’s usage.

The validation is limited to a single day and a specific experimental setup with uniform materials. Multi-day validation and diverse material properties could strengthen confidence in the results.

Minor comments
I suggest moving Figure 5 and associated analysis into the model evaluation section. After all, analysing the sensitivity of model accuracy with respect to its inputs is part of evaluation. At the same time, I wish the authors would extend the sensitivity analysis to more input parameters.

Table 1: The authors present an overview of building-resolved models for urban surface temperature. The comparison, however, is rather shallow and does not compare the model features and limitations in depth. I suggest extending the comparison to properly contextualize the new model development.

The authors should clarify the code licensing situation. According to the manuscript, a special collaboration agreement is required to use the code. However, on Zenodo, the license is set to Creative Commons Attribution 4.0 International, which as a public license does not accommodate for such requirements. It is also worth noting that Creative Commons does not recommend their licenses to be used on software (see https://creativecommons.org/faq/#can-i-apply-a-creative-commons-license-to-software). I encourage the authors to investigate other licensing options that are suitable for open publication of the source code. GMD’s Code & Data Policy includes some useful information as well.

L452-453: “Our previous study has demonstrated that the Monte Carlo ray tracing method has good accuracy in predicting solar radiation.”

The authors should clarify this statement. Either a reference is needed or the statement needs to be justified in the paper.

References
Yifan Fan, Yongling Zhao, Juan F. Torres, Feng Xu, Chengwang Lei, Yuguo Li, Jan Carmeliet; Natural convection over vertical and horizontal heated flat surfaces: A review of recent progress focusing on underpinnings and implications for heat transfer and environmental applications. Physics of Fluids 1 October 2021; 33 (10): 101301. https://doi.org/10.1063/5.0065125
Lipson, M.J., Grimmond, S., Best, M., Abramowitz, G., Coutts, A., Tapper, N., et al. (2024) Evaluation of 30 urban land surface models in the Urban-PLUMBER project: Phase 1 results. Quarterly Journal of the Royal Meteorological Society, 150(758), 126–169. https://doi.org/10.1002/qj.4589
Citation: https://doi.org/10.5194/egusphere-2025-1485-RC1
- AC1: 'Comment on egusphere-2025-1485', Shuo-Jun Mei, 18 Sep 2025
  
  We thank both reviewers for their constructive comments and valuable suggestions. In the attached document, we provide a detailed, point-by-point response to each comment. The original reviewer comments are presented in black text, our responses are shown in blue text, and the corresponding revisions in the manuscript are highlighted in red font within a grey-shaded box.
  
  Citation: https://doi.org/10.5194/egusphere-2025-1485-AC1
RC2:
'Comment on egusphere-2025-1485', Anonymous Referee #2, 05 Aug 2025
This manuscript presents a GPU-accelerated Urban Surface Temperature model that employs the Monte Carlo method to address complex radiative exchanges and heat transfer processes. The model holds significant potential for various urban applications. It describes the main components of the model, including conduction, solar radiation, longwave radiation, outdoor convection, and an indoor sub-model. The model is validated using field measurements. However, several issues should be addressed in the revised manuscript:
Lack of Justification for Monte Carlo Method: The manuscript does not explain or justify the use of the computationally intensive Monte Carlo method. Urban surfaces are typically characterized by simple geometries, where analytical methods might suffice. A rationale for choosing Monte Carlo over simpler approaches is needed.

Limited Scale of Real-World Application: The application to a real urban configuration (p. 27) includes only 40 buildings. This raises concerns about adequacy, as cities typically comprise thousands of buildings. If computational time or hardware limitations restrict the model to such a small scale, its practical applicability may be limited.

Insufficient Detail in Model Description: Certain aspects of the model, such as the solar radiation sub-model (p. 10), are poorly described. For instance, the manuscript mentions two GPU parallel computing approaches (Fig. 4) but does not clarify what "elements" are, how they are constructed, or how "points" are selected within the domain. Additionally, there is no explanation of how shaded areas are handled, how solar irradiation is calculated over time, or how various urban objects (e.g., buildings, roads, trees, grass, water) are represented in the model.

Hardware and Computational Time Details: If the GPU-accelerated Monte Carlo method is suitable for urban surface temperature modeling, the manuscript should provide details on the hardware used for simulations and the computational times for real-world scenarios, starting with the 40-building example.

Clarification of Albedo Statement: The statement on lines 43–44, "the complex three-dimensional geometry of urban environments leads to multiple reflections, which reduce urban albedo," requires clarification. Albedo is a material property, and it is unclear how reflections reduce it. The authors should explain whether this refers to effective albedo or another phenomenon.

Figure Improvements:

Fig. 1: Revise to correct typos and include Monte Carlo references in all relevant sub-model descriptions for consistency.

Fig. 2: Correct the typo "Calculatin" to "Calculation."

Fig. 3: Standardize terminology, using either "start point" or "target point" consistently throughout the manuscript.

Terminology Consistency: Clarify the use of "direct," "directional," and "direction" in reference to solar radiation to avoid confusion.

Fig. 5: Specify whether "run time" refers to computational time or the number of model runs.

Fig. 8: Rewrite the figure caption for clarity, as its current wording is difficult to understand.
Citation: https://doi.org/10.5194/egusphere-2025-1485-RC2
- AC1: 'Comment on egusphere-2025-1485', Shuo-Jun Mei, 18 Sep 2025
  
  We thank both reviewers for their constructive comments and valuable suggestions. In the attached document, we provide a detailed, point-by-point response to each comment. The original reviewer comments are presented in black text, our responses are shown in blue text, and the corresponding revisions in the manuscript are highlighted in red font within a grey-shaded box.
  
  Citation: https://doi.org/10.5194/egusphere-2025-1485-AC1
AC1: 'Comment on egusphere-2025-1485', Shuo-Jun Mei, 18 Sep 2025

We thank both reviewers for their constructive comments and valuable suggestions. In the attached document, we provide a detailed, point-by-point response to each comment. The original reviewer comments are presented in black text, our responses are shown in blue text, and the corresponding revisions in the manuscript are highlighted in red font within a grey-shaded box.

Citation: https://doi.org/10.5194/egusphere-2025-1485-AC1
AC2: 'Comment on egusphere-2025-1485', Shuo-Jun Mei, 18 Sep 2025

The manuscript has been revised in accordance with the reviewers’ suggestions, and the attached PDF shows all changes tracked. The source code and data have also been updated on Zenodo (https://doi.org/10.5281/zenodo.17138571)

Citation: https://doi.org/10.5194/egusphere-2025-1485-AC2

Shuo-Jun Mei, Guanwen Chen, Jian Hang, and Ting Sun

Viewed

Total article views: 1,664 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
1,547	101	16	1,664	22	42

HTML: 1,547
PDF: 101
XML: 16
Total: 1,664
BibTeX: 22
EndNote: 42

Views and downloads (calculated since 19 May 2025)

Month	HTML	PDF	XML	Total
May 2025	76	17	5	98
Jun 2025	68	18	5	91
Jul 2025	50	18	2	70
Aug 2025	377	32	2	411
Sep 2025	976	16	2	994

Cumulative views and downloads (calculated since 19 May 2025)

Month	HTML	PDF	XML	Total
May 2025	76	17	5	98
Jun 2025	68	18	5	91
Jul 2025	50	18	2	70
Aug 2025	377	32	2	411
Sep 2025	976	16	2	994

Viewed (geographical distribution)

Total article views: 1,630 (including HTML, PDF, and XML) Thereof 1,630 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 27 Sep 2025

Short summary

Cities face growing heat challenges due to dense buildings, but predicting surface temperatures is complex because sunlight, airflow, and heat radiation interact. By simulating how sunlight bounces between structures and how heat transfers through materials, we accurately predicted temperatures on roofs, roads, and walls. The model successfully handled intricate city layouts thanks to GPU speed. By revealing which heat matters most, we aim to guide smarter city designs for a warming climate.


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