feat: add OKF knowledge bundle for hackathon challenges

Capture all six Future City Hackathon 2026 challenges as OKF v0.2 concepts, with dataset and organizer concepts, indexes, and a log.

Configure release-please with the simple strategy so the first release PR bumps the repository from 0.0.0 to 0.1.0.
This commit is contained in:
Noel Sabia committed 2026-11-13 18:23:00 +01:00
commit 7b9893a89e
45 files changed
+45531

No files matched your search

@@ -0,0 +1,89 @@
---
type: Hackathon Challenge
title: "Measure where it matters: sensor placement for accurate heat maps"
description: Design an algorithm that places stationary heat sensors and plans mobile measurement routes so a coarse satellite map becomes a meaningful fine-grained heat map.
resource: ../../AI%20TRAQC/AI-TRAQC_Challenge_Measure-where-it-matters_Click_EN_v2.pptx
tags: [hackathon, future-city-hackathon-2026, ai-traqc, heat-mapping, sensor-placement, remote-sensing, optimization]
status: stable
generated: { by: opencode/deepseek-v4.1-flash, at: 2026-11-13T17:13:56Z }
sources:
- id: ai-traqc-deck
resource: ../../AI%20TRAQC/AI-TRAQC_Challenge_Measure-where-it-matters_Click_EN_v2.pptx
title: "Measure where it matters — AI TRAQC challenge deck (including speaker notes)"
author: team:ai-traqc
---
# At a glance
| Field | Value |
|---|---|
| Organizer | [AI TRAQC](/organizers/ai-traqc.md) |
| Core question | Where, when, and at how many locations must we measure for a city heat map to be meaningful? |
| Output | Sensor positions, mobile-unit routes with timing, and the number of measurement locations |
| Input | Open satellite data and open city data (NASA/USGS, ESA/Copernicus) |
| Bonus | Show that the approach also works for humidity or noise |
# The problem
Cities do not heat evenly. A river, a park, dense residential streets, and an industrial area can all behave differently on the same hot day. The available satellite data forces a trade-off between resolution and repetition:
- **Sentinel-3** provides a land surface temperature map every day, but at roughly 1 km per pixel. Streets and squares disappear inside a few pixels.[^ai-traqc-deck]
- **Landsat 8 and 9** provide finer images at roughly 100 m, but only every eight days, only around midday, and only when the sky is clear.[^ai-traqc-deck]
- Measuring everywhere with ground sensors and mobile units is too expensive because sensors and robot time are scarce.[^ai-traqc-deck]
The challenge is to decide where limited measurements buy the most information.
# Your task
Design an algorithm that starts **from scratch, without existing sensors**, and outputs:[^ai-traqc-deck]
1. **Where** stationary sensors are placed.
2. **Which routes** mobile units such as the robot dog take, and **when**.
3. **How many locations** need to be measured at all.
Example constraints are the number of sensors (for example `5+5` or `10+2`), the mission time, and priority areas such as day-care centres or care homes. The deck calls these examples, not fixed rules.
## Suggested workflow
1. **Choose placements from coarse data only.** Derive the placement from the Sentinel-3 baseline (1 km) and open city data. Do not peek at the fine reference while choosing locations.
2. **Optionally segment the city into climate zones.** Areas with similar buildings and surfaces behave similarly. The global Local Climate Zone map and OpenStreetMap are suggested open inputs.[^ai-traqc-deck]
3. **Simulate measurements.** Sample values at the chosen locations from the high-resolution Landsat reference (100 m, same day). The simulated readings correspond to a downward-facing infrared sensor and represent land surface temperature, not air temperature.[^ai-traqc-deck]
4. **Reconstruct a finer map.** Combine the simulated readings with the coarse baseline to produce a map on a finer grid, for example 100 m.
5. **Test against the reference.** Compare the reconstructed map with the full Landsat image. A curve of map error versus number of measurement points is one way to justify how many locations are needed.[^ai-traqc-deck]
## Timing
Landsat and Sentinel-3 are snapshots around midday, so the "when" needs a separate time axis. The speaker notes suggest ECOSTRESS scenes at different times of day or a modelled diurnal cycle. ECOSTRESS only covers up to about 52 degrees north; Sentinel-3 and Landsat cover every city.[^ai-traqc-deck]
# What you get
- Open satellite data from NASA/USGS and ESA/Copernicus, including Sentinel-3, Landsat 8/9, and optionally ECOSTRESS.
- An example city layout with a river and park, residential areas, a dense city centre, and an industrial area.
- Suggested open data: the global Local Climate Zone map (about 100 m) and OpenStreetMap for the path network.
- Suggested starting points from the deck: optimal sensor placement and informative path planning.
No challenge dataset is stored in this repository. The deck describes the data sources and the simulation workflow rather than shipping a file.
# Expected result
The deck is explicit that the result is **positions and routes, not just a map**: the algorithm itself should transfer to any city. The corrected heat map is the proof of how good the placement is, not the only deliverable.[^ai-traqc-deck]
# What makes a strong solution
- Placement uses only the coarse baseline and open data, then is evaluated honestly against the fine reference.
- The number of measurement locations is justified quantitatively, for example with an error-versus-count curve.
- The approach is transferable beyond the example city.
- Assumptions, limitations, and uncertainty are visible.
- Bonus: the same approach works for humidity or noise.
# Open questions
- How should priority areas be weighted against overall map error?
- How much does the time of day change the optimal placement?
- How should the algorithm handle clouds or missing satellite scenes?
# Sources
The challenge deck, including its detailed speaker notes, is the primary source for this concept.[^ai-traqc-deck]
[^ai-traqc-deck]: AI TRAQC, "Measure where it matters — sensor placement for accurate heat maps", challenge deck and speaker notes, 2026.