Method
How the energy, towing and range figures in the Arctic roadtrip note were calculated: data, units, exclusions, labels, models, uncertainty and limits.
This is the technical appendix to 5,976 km to the Arctic in an EV, towing a caravan. Every number in the note carries one of four labels: measured (read from the car’s log, or a sum or average of those readings), estimated (a statistical estimate with an interval), modeled (depends on a physics model, an assumption or weather-model data) and not logged (never observed, never filled in).
Data
| Source | What | Period |
|---|---|---|
| The car’s log (Teslamate, self-hosted) | Drives, positions, battery power, battery percentage, outside temperature, elevation, charging sessions | 3 to 20 September 2026 |
| Trip notes | Caravan on or off per day or drive; odometer at departure (about 90,600 km, rounded) and at home (96,576 km); usable battery capacity | Whole trip |
| Open-Meteo, archived weather model | Hourly temperature, pressure, humidity, wind, rain at each stretch of road | 3 to 20 September 2026 |
Inside drives the car reports battery power about four times a second, in whole kW (positive is draw, negative is regeneration). The odometer steps in 0.161 km. ERA5 reanalysis was the intended weather source; it was rate-limited on the day of the analysis, so the archived forecast model was used instead.
Coverage and distances
- 5,976 km driven in total (odometer, excluding sea crossings). 2,499 km are in the car’s drive records (42%), on 18 of 38 days.
- The 3,474 km before 3 September has no telemetry. Nothing is reconstructed.
- The parts don’t add up exactly. 3,474 km is the first logged odometer reading (94,073.8) minus the rounded departure note. About 4 km at the end were not logged. Drive distances sum to about 1 km more than the odometer moved while logging. The total is only as exact as the departure note.
Units and the energy measure
- Energy is battery energy: power integrated over time within each stretch, in kWh. Consumption is kWh per 100 km.
- For comparison, the car’s rated-range drop times 0.152 kWh per km reads 4 to 5% higher than integrated power (median ratio 0.957 over drives longer than 20 km). On long drives that rated-range figure matches the battery percentage drop times 67 kWh.
- Range is therefore calculated in the battery-percentage frame: range = usable kWh ÷ (integrated kWh per km × 1.052), with 1.052 (1.047 to 1.057) measured over drives longer than 20 km. The 80 to 10 percent window is 46.9 kWh of 67.
- Usable capacity, 67 kWh: energy added divided by percentage points gained over 19 charging sessions that gained at least 40 points (median 66.9 kWh, 66.2 to 67.9).
Exclusions
- Drives shorter than 5 km (79 km in total): starting and manoeuvring dominate them.
- Each drive is cut into one-kilometre stretches (2,393). Stretches with a gap in the power readings longer than 10 seconds, mostly tunnels, are left out of the models (31). So are stretches without usable weather or slope data. 2,208 stretches go into the models.
- One drive in Bodø (8.8 km) is left out of the main models because the car held a stale position from before the ferry. Including it changes nothing material.
- After the Kiel ferry the car reported wrong positions for 26 minutes and 28 km. Distance, speed and energy stay valid; weather, slope and wind for that stretch are set to missing.
Caravan labels
The car has no trailer signal, so every label comes from the trip notes. Of the 50 drives longer than 5 km, 14 are solo and 36 towing.
| Confidence | Rule | Drives |
|---|---|---|
| Confirmed | The notes give the caravan status for that day or drive | 36 |
| Probable | No note that day, but the drive moves the overnight location, so the caravan came along | 12 |
| Uncertain | Notes say towing, but short local drives read low | 2 |
Three drives disagree with what their consumption suggests and stay in: one solo drive reads high for no clear reason, one towing drive has tunnel gaps, and one towing drive is a 317 m descent.
Weather, wind and road wetness (modeled)
- Wind is interpolated to the middle of each stretch and projected onto the direction of travel. It is scaled from 10 m to about 1.5 m with a log profile (roughness 0.1 m, factor 0.59).
- Air density comes from the model’s pressure, temperature and humidity, corrected to road elevation (1.10 to 1.24 kg/m³ on the trip).
- A stretch counts as wet if the model had at least 0.1 mm of rain in that hour or 0.5 mm in the three hours before. Road surfaces were never observed. No solo stretch is wet under any rule, so a wet-road effect can’t be estimated reliably.
Models
1. Descriptive (measured). Total kWh over total km for solo and towing, and the same for stretches driven between 70 and 90 km/h.
2. Regression path (estimated). Linear regressions of energy per 100 km on towing, adding speed terms, slope (up and down separately), temperature and a first-3-km warm-up term, headwind, wet road, and finally letting towing interact with speed and slope. Standard errors are clustered by drive.
3. Road-load model (modeled). Per stretch, battery energy is
E = P_aux·t + F·d + (CdA/η)·½ρ∫(v+w)²ds + (m/η)·g·h_up + m·η_r·g·h_down + ΔE_kinetic + E_warm-up
fitted by least squares, separately for solo and towing. Speed, standby power and drag are hard to separate with these data, so two things are fixed: standby power P_aux at the measured 2.27 kW (the car’s draw at standstill inside drives), and the solo drag area at 0.51 m² (Tesla’s stated drag coefficient of 0.23 times a commonly quoted frontal area of 2.22 m², not verified independently). The towing drag area is estimated from the data.
Two variants differ in how they treat the road: pooled, one road resistance per configuration for the whole trip, and within-day, one per day. The adjusted comparison and the speed curves use standard conditions: steady speed, flat, 12 °C, 1013 hPa, no wind, dry.
The derived drag areas and rolling coefficients assume η = 0.88 and masses of 2,150 kg solo and 3,150 kg towing (a 1,000 kg caravan; the Eriba is probably nearer 900 kg). Those derived parameters move with the assumptions; the fitted energy curves do not.
Uncertainty
- Sampling uncertainty. Whole drives are resampled with replacement 2,000 times, solo and towing separately, and the full calculation is repeated each time. The 95% interval runs from the 2.5th to the 97.5th percentile. Resampling whole drives respects that stretches within one drive are not independent.
- The band in the speed chart runs from the lower of the two variants’ lower bounds to the higher of their upper bounds. It shows sampling uncertainty under both road treatments, so it includes one modeling choice. It is not a formal confidence interval.
- Sensitivity to other choices is reported separately: standby power of 1.8 or 2.8 kW, a temperature term, a free solo drag fit, 2 km stretches, other wind-profile and wet-road rules, dropping uncertain labels, confirmed labels only, and northern Norway only. Across all of them, the towing gap at 80 km/h stays between about 11 and 15 kWh per 100 km. What moves most is the towing curve above 90 km/h.
Results used in the note
| At 80 km/h, standard conditions | Solo | Towing | Difference |
|---|---|---|---|
| Pooled | 16.7 (14.8 to 19.3) | 29.0 (27.6 to 31.0) | +12.4 (9.4 to 15.0) |
| Within-day | 16.7 (14.7 to 19.5) | 29.3 (27.7 to 31.4) | +12.6 (9.4 to 15.5) |
| Towing range, 80 to 10 percent | Pooled | Within-day | Band |
|---|---|---|---|
| 80 km/h | 154 km | 152 km | 142 to 162 km |
| 90 km/h | 143 km | 137 km | 128 to 150 km |
| 100 km/h | 133 km | 122 km | 114 to 140 km |
Validation
Leave-one-drive-out cross-validation (each drive predicted by a model fitted without it): the drive-level error drops from 5.7 kWh per 100 km for a plain solo or towing average to 4.3 for the pooled road-load model and 4.0 for the full regression. The model over-predicts the fast German motorway drives by about 5 and under-predicts cold Norwegian mountain drives by 5 to 10. Leaving out any single drive keeps the towing gap at 80 km/h between 11.9 and 15.0.
Limits
- Solo driving is 216 km in 14 drives, almost all in Lofoten over three days. No motorway speeds, no rain. Every solo versus towing contrast is also a Lofoten versus the rest of the route contrast. A northern-Norway-only check gives the same towing gap at 80 km/h, but a steeper towing curve.
- Road surface was never observed and is probably the largest confounder of the drag estimate.
- Colder stretches cost more (about 1.1 kWh per 100 km per degree), far more than temperature alone can explain. Treat that as “colder stretches cost more”, not as a temperature effect.
- One car, one caravan, late summer, with drive averages of 11 to 20 °C.
Data
Curated tables behind the figures. They contain no positions, no direction of travel, no absolute elevation, no times finer than the hour and no spending: together those would show where we slept and where family lives. Everything the models need is still there.
- model_segments.csv: every model input, one row per one-kilometre stretch (2,393 rows)
- reproduce.py: recomputes the trip averages, the same-speed comparison and the regression path from these files (Python and numpy); its output matches this note
- segments.csv: a simpler version of the stretch table, as used in the speed chart’s dots
- drives.csv: one row per analysed drive (50 rows)
- days.csv: kilometres per day with coarse places
- README.txt: what every column means
The road-load fit is not scripted in reproduce.py, but its inputs are all in model_segments.csv and its equation and fixed values are above.