Feedback on the use of Urock on a the Linux scientific computing platform

[lgzdbf]

Hello everyone,

@j3r3m1, as requested, here is my feedback regarding the use of Urock run on a the Linux scientific computing platform. Don’t hesitate to ask if you need further details or if you feel I have forgotten something important.

1. Characteristics of the computing machine and method

The algorithm was run on a computing server with the following specifications:

• CPU: 2 CPUs, 18 cores each (36 cores total)
• RAM: 512 GB
• GPU: None

Urock was executed for each tile (see point 2) and each hour (see point 3) as parallel jobs.

2. Size of the study area

The study area includes the municipalities of Grenoble and Échirolles (Isère, France), covering approximately 26 km².
To optimize computational resources, the area was divided into several sub-domains: a grid of 500 tiles with a resolution of 250 m² (expanded to 350 m² after applying a buffer to reduce edge effects during the final merging process).

3. Number of simulated hours and wind data sources

A total of 432 hours were simulated. Wind speed and direction data were derived from the U and V wind components at 10 m, provided by ERA-5 reanalysis datasets [https://cds.climate.copernicus.eu/datasets/reanalysis-era5-single-levels?tab=download]
The vertical wind profile used follows the power-law by Pardyjak and Brown (2003).

4. Computation time

I don’t have precise statistics on computation times, only rough estimates:

• For tiles with low building density or simple geometries, Urock typically ran in less than 10 minutes.
• For more complex tiles, runtimes ranged between 10 and 30 minutes.
• About 0.04% of the jobs took approximately 3 hours to complete.

5. Example result and discussion

Here is an example of the output for July 23, 2022, at 2:00 PM.

A main limitation of this work is the presence of edge effects between adjacent tiles (or, in other words, a lack of continuity) resulting from the sequentialized simulation, despite the application of a buffer beforehand to reduce these effects. This can be observed in Figure 1 below, particularly in the two zoomed-in areas. For greater clarity, a second figure (Figure 2) is provided, specifically illustrating these edge effects.

Figure 1.

Figure 2.