Written by an AI agent working in this repository.
The aerolog-scaling programme asks what a scaled model of a cargo drone tells you about the full-size aircraft. This step uses the public flights of Rodrigues et al. (a 3.7 kg DJI Matrice 100, 5 Hz battery logs, CC BY 4.0) to ask what fixed energy a route flight pays before it cruises.
I planned it for two altitudes. The data has four (25, 50, 75, 100 m), and the table stores altitude as text, which silently zeroed my first fit. I discarded that run and redid it as a pooled fit, labelled post hoc in the log.
Result (132 flights, R2 0.93): each extra 25 m of altitude costs 9.6 kJ (95% interval 8.9 to 10.3), about 10 times the potential energy gained, so climb and descent time at near-hover power dominates. Each payload step costs 8.0 kJ (7.0 to 8.9).
What it cannot do: give a range. Cruise windows are under 150 m, so the cruise slope is not identified (-13 to 103 J/m) and the range interval runs from about 1.4 km to over 40 km. The honest lower end, 1 to 1.4 km one way on one pack, matches the previous step. A real range needs long cruise legs, so the next step is a physical lift-plus-cruise model.
Code, data link and log: research/aerolog-scaling/experiments/altitude_energy/ and LOG.md.