It’s been incredibly hot here in Cambridge. It inspired me to sit down and think a little bit more about how we can prepare to live through hotter and hotter summers. Inspired by Professor @sunt05 post, I decided to also run my own warm up simulation through Codex.
The question motivating my work is:
Which neighbourhoods in the focus city become most heat-risk exposed under a +3 C hotter-future stress test, and where do social vulnerabilities make that heat most dangerous?
To start, I used the exact same prompt he used:
I’m new to urban climate modelling. I’d like to look at a typical dense residential neighbourhood in London: roughly 40% buildings around 12 m tall, 35% paved streets and car parks, and the rest split between street trees and small gardens. Using SUEWS, could you set up this neighbourhood and run a short simulation for a hot summer week, then tell me what the surface energy balance looks like in plain terms?
In Professor Sun’s run, there was a readiness check that prompted him to trust the agent. In his words, “tooling itself refused to treat the sample as my neighbourhood.” There must be a misunderstanding on my end. I assume that the agent immediately ran the simulation based on the bundled KCL London sample, without checking with me. It might have something to do with my prompting, or it might even be because I’ve been “conversing” with the agent before giving it this prompt. Might I have confused its context window? Whatever the case, in my run, it still pulled the same time horizon as Professor Sun’s. It ran for 14-21 August 2012 with 168 hourly records. Here is the figure it produced (after making Python crash twice!!!):
Figure 1. Surface Energy Balance for One Week. The top panel shows hourly heat flows through the hot week. The lower-left panel simplifies the sunny-hour energy split. The lower-right panel shows modelled air and surface temperatures.
Here are the results:
During sunny hours, the average available energy was about 285 W/m2. It was split roughly like this:
- 66% went into heating the air directly, the sensible heat flux QH.
- 24% went into warming roads, roofs, walls, and other urban materials, the storage heat flux QS.
- 10% went into evaporation and plant water use, the latent heat flux QE.
That means this dense residential setup behaves like a heat-retaining urban surface. The paved and built parts dominate, so much more energy heats the air and urban fabric than is used for evaporation.
In other words, the residential area retains heat. Interesting enough! If I understand correctly, this would form the evidence base of the hackathon deliverable: SUEWS config, output, figure, and a report file. In recognition of this, I asked the agent to generate a deliverable page, answering the hackathon prompt, as well as my motivating question. If you are interested, you may read my practice run report here. Do let me know your thoughts! I would deeply appreciate your feedback. (I really need to know where I went wrong!)
Key Lesson Learned: I will need to keep track of how I prompt the AI agent, as evidenced by the agent assuming I wanted to run sample case for my trial run. Also, not being clear with requests (e.g., in generating the figure) led to time (and resources) wasted waiting for the agent to revise something as small as a label. (P.S. I still haven’t fixed the label! Please disregard it for now. Thanks!)