the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 09 Sep 2026
URAY v1.0: a 3D Monte Carlo model for urban shortwave radiation with explicit buildings and tree canopies
Abstract. Urban canopy models with trees often represent radiation exchange with idealized street canyon geometry. These models clarify tree shading, canopy interception, and multiple reflection, but they cannot directly resolve the 3D arrangement of buildings and tree crowns across real city districts. URAY addresses this limitation as a city-scale 3D Monte Carlo model for urban shortwave radiative transfer that resolves both buildings and tree crowns. The model extends a building-resolving shortwave solver by adding ellipsoidal tree crowns as transmissive participating media, Woodcock tracking for stochastic canopy interactions, bounding volume hierarchy (BVH) acceleration for complex urban meshes, and GPU parallel photon tracing. The coupled framework resolves explicit building surfaces and tree crowns in one 3D radiative transfer domain and diagnoses both scene albedo and absorbed shortwave radiation for ground, walls, roofs, and tree canopies. Two validation experiments test the canopy transmissivity formulation and the coupled building-tree albedo response in a 3D urban block. URAY reproduces the observed radiative behavior in both cases. We then apply URAY to Guangzhou, a subtropical city with high building density and dense urban tree cover, using resolved 3D building and canopy geometry. This city-scale simulation demonstrates that GPU-parallel URAY can efficiently resolve component-level shortwave absorption across an entire city.
Competing interests: At least one of the (co-)authors 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.- Preprint
(14722 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 04 Nov 2026)
- CC1: 'Comment on egusphere-2026-5226', Sivarajah Mylevaganam, 17 Sep 2026 reply
-
CC2: 'Comment on egusphere-2026-5226', Sivarajah Mylevaganam, 17 Sep 2026
reply
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2026/egusphere-2026-5226/egusphere-2026-5226-CC2-supplement.pdf
-
RC1: 'Comment on egusphere-2026-5226', Anonymous Referee #1, 09 Oct 2026
reply
URAY traces direct and diffuse photons through triangulated buildings and ellipsoidal tree crowns treated as turbid media, runs on a GPU, and is applied to 6761 one-kilometre tiles of Guangzhou after two validation cases, an isolated tree and a scaled outdoor block. While a building-and-tree Monte Carlo model that runs over a whole city is useful, and the open archive of code and validation files is welcome, the evidence that the canopy and component-absorption parts work is thin, the city-scale results say less than the Conclusions claim, and what is new relative to existing 3D models is not pinned down. I would be glad to see the paper in Geoscientific Model Development after a major revision.
Major comments
- Regarding Section 3.1, the IsoTree test is one tree and one PAR sensor, placed 1.8 m from the trunk, so the comparison checks transmission along roughly one geometry at each sun position. Bailey et al. (2014) report a measured LAD of 0.8 ± 0.7, while SOMUCH uses 3.5 and Guangzhou 3.0 m2 m-3, and the text does not say which value drove Fig. 6 or how its uncertainty was treated. ωLS = 0.30 is set by the authors and not measured. Please state the values used, show how MAE, RMSE and r change across the LAD range and ωLS, and add a plain Beer–Lambert transmission as a reference, so that the reader sees what the stochastic crown treatment adds.
- The SOMUCH comparison (Section 3.2, Table 1) reports MAE of 0.0172 (trees) and 0.0209 (empty street) for dry, high-sun records, with a tree-case bias of +0.0121. The quantity the experiment is meant to test is the tree minus no-tree albedo difference, and neither the observed nor the simulated difference is reported in Table 1 or the text, so one cannot judge whether an error of about 0.02 is small. Ground, wall and roof albedo are all set to 0.24, and it is not said whether this was measured or adjusted, nor whether 3.5, 0.30 or 0.80 were tuned on these same four weeks. Also, the model albedo is the escape fraction of the whole periodic tile (Eq. 5) while a CNR4 at 2.0 m sees a limited footprint. How were the two made equivalent? The Boland split of K↓ adds an error of its own that is not propogated.
- Both tests check a reflected or transmitted flux. Neither checks the thing the paper is built around, namely the split of absorbed shortwave between ground, walls, roofs and canopy (abstract). The authors cite models that could serve as references for it: Overby et al. (2016), Villefranque et al. (2022), Landier et al. (2018) and Wang et al. (2024) with DART, and Krayenhoff et al. (2014) for trees. A comparison of component absorption in one shared scene, an idealised canyon with trees and the SOMUCH layout for instance, would do more for confidence than any further tile map. If the SOMUCH platform had wall or ground pyranometers, they should be used.
- The Guangzhou run covers four winter days with one albedo (0.28) for ground, walls and roofs, one LAD (3.0) for all crowns, and G(θ) = 1. The spatial pattern in Figs. 13 and 14 therefore comes entirely from the counts and positions of surfaces and crowns, which should be said when the maps are described as showing “strong spatial heterogeneity”. The sensitivity tests in Section 3.3 vary LAD and χ on one tile, and albedo moves from 0.2064 to between 0.2237 (0.25 LAD0) and 0.2053 (2.0 LAD0), a spread of about 0.018, about the size of the SOMUCH MAE. Yet the Discussion lists the optical parameters first among its limitations: ωLD = 0.80 is an unweighted average of ρ + τ over 400–2500 nm, and ωLS is prescribed. Please add ωLD, ωLS and surface albedo to the sensitivity analysis, and use more than one tile.
- The correlation of r = 0.942 between albedo and ground absorption (Section 4) is between two outputs of the same photon histories: every photon that does not escape is absorbed by ground, wall, roof or crown, so the quantities are tied by closure. The morphology split into median classes of surface count (945) and tree count (1071) then shows that this correlation is higher in tiles with more trees (0.733 to 0.914), which is a statement about scatter, not about what trees do to ground absorption. The Conclusions say the results “demonstrate the importance of explicit three-dimensional geometry in accurately modeling urban shortwave radiation”, but no run uses a non-3D model on the same tiles, and the Discussion admits there is no independent city-wide validation. Paired runs per tile with and without trees, and against a canyon-based or flat-albedo treatment, would give the effect directly and the model can do it cheaply.
- Footprints are replaced by convex hulls, crowns are ellipsoids fitted to segments of a global canopy-height product (Tolan et al., 2024), and each tile is closed with periodic lateral boundaries. The Discussion mentions the crown fitting, but nothing in the paper measures the effect of any of these simplifications. A convex hull fills courtyards and the inside of L-shaped blocks, which changes canyon width and wall area exactly where multiple reflection matters. Please test a subset of tiles with the original footprints, with a 3 × 3 neighbourhood instead of periodic wrapping, and with crowns perturbed within the error of the canopy-height data.
- The speed-up of 198 is a sequential CPU against a GPU, both with BVH off, for one time step in one tile. The city-scale runs presumably use BVH, so the relevant comparison is CPU and GPU with BVH, and a multi-threaded CPU rather than a sequential one. Further, with σmax = σt in homogeneous ellipsoids there are no null collisions, so Woodcock tracking reduces to ordinary free-path sampling here. It would earn its place only with crowns of varying LAD, and as it stands the abstract promises more than the implementation uses. A short table of what URAY does that Overby et al. (2016) and Villefranque et al. (2022) do not would also help with the novelty question.
Minor comments
- Line 75: “The manuscript are organized”; lines 137–138 (“If the ray leaves the top of the domain, reaches the maximum reflection count …”) are garbled; please also say how photons stopped at the maximum reflection count are counted in Eq. 5; line 299 “set to equal to 1”.
- Section 3.1 gives the measured LAD as 0.8 ± 0.7 m-1, while LAD is in m2 m-3 elsewhere. Please check the units.
- Eq. 8 averages albedo and absorption over the 44 sunlit time steps with equal weight, so low-sun hours count as much as noon. An irradiance-weighted mean is the usual choice, and it could change the maps and the regression.
- Section 3.3 says LAD factors are applied to “each tile’s own baseline LAD”, but the Guangzhou run uses a uniform 3.0. Also, the text says G(θ) = 1 is assumed in the simulation, yet varies χ, which acts only through G(θ) (Eq. 11). Which G(θ) is used in the baseline and in the χ test?
- Line 362, “calculated from the unrounded runtimes before BVH acceleration was introduced”, is hard to follow. Please say plainly which runs are BVH-off.
- The Discussion mentions mapped materials and moisture-dependent properties as future work but not how URAY output would reach a climate or weather model. Wang et al. (2025), “Ultrafine-resolution urban climate modeling: Resolving processes across scales”, Journal of Advances in Modeling Earth Systems, discusses modelling of this kind across scales and could help the authors place the next step.
Citation: https://doi.org/10.5194/egusphere-2026-5226-RC1 -
RC2: 'Comment on egusphere-2026-5226', Anonymous Referee #2, 15 Oct 2026
reply
General comments
The authors present a mature, comprehensive, and polished model development and validation study that introduces the URAY model for urban shortwave radiation characterization that resolves complex three-dimensional geometry, including infrastructure and tree canopies, albeit with parameterized trees canopies.
URAY is implemented in Python using a CUDA-based NUMBA backend. By using Python, the authors ensure URAY is an accessible model that users and developers from a wide variety of backgrounds can utilize. In contrast, reliance on CUDA somewhat limits the portability of the code. It would be interesting to see future development explore hardware-agnostic GPGPU solutions, but this comment is more looking towards future versions of URAY than the present study.
The validation simulations are executed on an NVIDIA RTX 5090 GPU, which the authors do note is a consumer-grade GPU. If the authors have access to HPC resources, it would be a valuable portability, scalability, and performance test to re-run the city-scale validation test on datacenter-grade GPU hardware. Nevertheless, the model’s reasonable runtimes on a consumer-grade workstation demonstrate its accessibility, particularly for under-resourced users.
I recommend minor revisions.
Specific comments
Lines 141-142: What are the implications of periodic lateral boundaries?
Equation 4: Please define all terms.
Equation 9: Please define all terms.
Equation 11: Please define all terms.
Lines 223-230: When a major point of this study has to do with trees, please be clear when discussing trees/leaves in the context of BVH.
Section 3: For model validation experiments, tables summarizing model parameters may be useful in place of in-text descriptions.
Line 346: “... confirming …” --> “... suggesting that a small number of extreme ratio residuals dominate …”
Figure 8: For such a long and fine time series, a model vs. observations plot could provide a quick at-a-glance visual summary of model performance. This could be added, for example, as an additional panel in this figure, rather than replacing it.
Figure 9: Are CUDA kernels launched with fixed thread blocks? Do the thread blocks scale dynamically with the number of photons? How much time is spent in compute versus overhead? Please profile the kernels and discuss bottlenecks.
Line 389: Previously, the quantity “nphotons” was labeled as N. Please maintain notation consistency, or explain how these quantities are distinct.
Citation: https://doi.org/10.5194/egusphere-2026-5226-RC2
Data sets
Published August 29, 2026 | Version v1 Software Open URAY v1.0: a 3D Monte Carlo model for urban shortwave radiation Mei and Sun https://doi.org/10.5281/zenodo.22153464
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 256 | 89 | 42 | 387 | 54 | 42 |
- HTML: 256
- PDF: 89
- XML: 42
- Total: 387
- BibTeX: 54
- EndNote: 42
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Decades of theoretical advancements, paired with rapid technological progress, have driven researchers to develop mathematical frameworks capable of addressing complex problems across numerous disciplines. Because these problem domains are inherently dynamic rather than static, there is a continuous need to evaluate existing methodologies and pioneer new models with advanced functionalities.
In this context, the authors claim that current urban canopy models fail to adequately resolve the three-dimensional configuration of buildings and vegetation within urban environments. To overcome this limitation, the authors introduce a novel model to the scientific community named URAY (Version 1.0). Based on the current version of the manuscript, the following evaluation points are brought forward.
1. The manuscript states that building footprints and heights are transferred from a GIS environment into a 3D model where buildings are represented as explicit surface geometry (Lines 88–90). However, the exact workflow and technical methodology for this process are missing. The manuscript merely states that it was done, rather than explaining how it was executed. The authors should explicitly clarify the underlying mechanisms: did they utilize existing extensions within proprietary software (e.g., Esri products), or did they develop custom, open-source code from scratch?
2. The analytical necessity of translating GIS data into explicit 3D surface geometry remains unclear. For example, consider a simple building with a 10×10 footprint and uniform wall heights of 10. Extruding this shape yields four vertical surfaces of 10×10. Because the model ultimately represents these buildings as flat surface geometries [line 90-91], this translation appears to loop back to the initial dimensional inputs. The authors need to clearly justify the necessity of this 3D conversion and explain what unique analytical value the explicit surface representation adds.
3. In Line 156, the authors state that URAY records whether each photon is absorbed by the ground, a wall, or a roof for each radiation component and time step, but the exact methodology supporting this statement is entirely missing from the current manuscript. Because the radiation tracking process is treated as a black box, it remains unclear how the model actually derives and tracks these specific absorption values without explicit detail on the underlying algorithms or tracking mechanisms.
4. In its current form, the manuscript functions more like a conceptual framework than an operational model. Too many technical details are omitted, rendering the model a "black box" and leaving readers with insufficient information to fully understand, evaluate, or replicate the work.
5. Section 2.0 is dedicated to model development, where readers expect a precise, step-by-step workflow and implementation methodology. However, several statements appear out of place. For instance, the discussion of the Guangzhou application in Line 86 belongs in a model application or case study section rather than under model development. The authors should restructure this section to separate core model logic from its empirical application.
6. Figure 1 illustrates the workflow of the proposed URAY model. The left side of the diagram positions "model development" and "model validation" as simultaneous inputs feeding into the model. Conceptually, this sequence is flawed. Model validation is a post-development evaluation process used to verify that a completed model behaves as expected. Validation cannot act as an input to the model it is meant to test. The authors must revise the flowchart sequences to accurately reflect standard modeling logic.
7. The mathematical formulations for Ag, Aw, and Ar in Equation (4) are incomplete. The authors have omitted the definitions for the core variables I0 and Id. All variables introduced in equations must be explicitly defined in the surrounding text.
8. The definitions of Fjd and Qd represent distinct variables in the manuscript, where Fjd denotes the number of diffuse photons absorbed by a specific surface class and Qd represents the total diffuse photon counts terminating in absorption by urban surfaces or tree crowns. Clarification is required to determine whether summing all individual Fjd values yields Qd, and how Equation (5) relates mathematically to Equation (3).
9. Equation (7) appears to be highly repetitive of Equation (4) [Line 73]. It seems that Equation (7) is simply a generic or compact version of Equation (4), rendering its inclusion redundant. The authors should either clearly justify the theoretical necessity of formulating Equation (7) or remove it entirely to streamline the mathematical presentation.
10. The handling editor should closely evaluate the co-author contribution statements outlined in Lines 552–555. Given the numerous clarity issues, missing details, and structural oversights in the current text, the claimed contributions regarding "reviewing and editing" from the secondary authors are questionable. If this draft is indeed the final product of a rigorous review by multiple co-authors, it raises concerns regarding the overall quality control of the team's collaborative process.