Roadmap / 0.1.2.6
Complex weather
v0.1.2.6, commit f3db6ae. File docs/RESULTS_v0.1.2.6_complex_weather.md.
Written at this version and not revised since. Later versions may change these numbers; see the roadmap.
Switch: weather.model: simple | complex (with weather.enabled), so off / simple / complex can be compared directly. Optional weather.objective_weights per scenario.
Runs: 2,000 per variant, seed 20261105; all other modules on, no embargoes. Backtests unchanged (Iraq invasion and occupation pass). 14 tests pass.
What the complex model does
- Daily weather in five regions: corridor (Toronto–Montréal), prairies (Winnipeg–Calgary), oil sands (Fort McMurray), Pacific (Vancouver) and north (Yellowknife–Iqaluit). Built from ECCC 1991–2020 normals; Fort McMurray and Yellowknife are new.
- Temperatures carry day-to-day persistence (AR(1)), so cold snaps last.
- ENSO regime per run: La Niña makes Prairie and western winters ~1.5 °C colder, El Niño milder. Environment Canada: La Niña brings colder-than-normal Prairie winters; El Niño a warmer, drier West.
- Warming offset vs the normals: Canada warmed 2.0 °C over 1948–2023, the north 2.6 °C, mostly in winter (CCCR 2026).
- Events, all aggregate:
- Snow and blizzard days; cold snaps; heat waves.
- Freezing rain, tuned to Montréal's 12–17 days a year.
- Rare 1998-scale corridor ice storms.
- Spring thaw.
- Wildfire smoke, with a severity draw per season (2023: ~17M ha; 2025: ~8.9M ha).
- Oil-sands wildfire shut-ins, using the 2016 analogue: ~0.8 Mb/d lost on average in May, peak >1.1 Mb/d (EIA).
- Daylight by latitude and day of year.
- Where the weather counts:
- The campaign uses weather where the fighting is (objective weights).
- Occupation and the economy use weather where people live (population weights).
- Cold-weather capability: each side's penalties scale with (1 − proficiency).
- US: 0.25–0.55. The 11th Airborne is the Army's only Arctic-capable division.
- Canada: 0.55–0.85.
- The simple model implicitly assumed 0.5 for both.
- Hooks into other modules:
- Air denial (blizzard, freezing rain, smoke, darkness) cuts the US air advantage.
- Blizzards give insurgents cover; extreme cold lowers their tempo.
- Cold strips occupier strength.
- Heating costs.
- Ice storms hit Canadian GDP and halve power exports that month.
- Cold snaps raise Canadian blackout risk; heat waves raise US blackout risk.
- Oil-sands fires feed the oil module.
Results — full annexation
| Variant | Campaign days | US deaths (median) | US cold/non-battle casualties (mean) | US quits |
|---|---|---|---|---|
| Weather off | 28 | 1,129 | 0 | 65.2% |
| Simple, January | 38 | 1,218 | 1,457 | 63.2% |
| Complex, January | 39 | 1,231 | 2,526 | 63.4% |
| Complex, January, La Niña | 40 | 1,214 | 2,888 | 62.7% |
| Complex, January, El Niño | 39 | 1,208 | 2,311 | 64.4% |
| Complex, January, US force cold-trained (0.7) | 33 | 1,114 | 1,075 | 63.0% |
| Complex, July | 27 | 1,147 | 0 | 60.8% |
Weather denies air operations on 27% of January campaign days, against 4% in July. Alberta detachment and the Arctic seizure show the same pattern on a smaller scale. In the Arctic the cold-snap share of campaign days reaches ~45%.
Findings
- The simple model understated the cold by about 70%. Weighting the weather by where the fighting is (Kevin's point about northern objectives) and using realistic US cold proficiency raises US cold casualties in a January invasion from ~1,460 to ~2,530. They would be ~2,900 in a La Niña winter.
- Training is the US lever, not timing. Bringing the whole force to roughly 11th Airborne standard cuts cold casualties by ~60% and the campaign by six days. In practice the US can fix this only for a small part of the force.
- Weather still doesn't decide the war. The chance the US quits stays at 61–65% across every variant. Cold casualties are mostly non-fatal, and US opinion tracks cumulative deaths. Weather shapes the campaign's cost and length, not whether the US stays.
- ENSO changes the cost by about ±10–15%, not the outcome.
- Rare events turn up. An oil-sands wildfire hits ~13% of full-annexation runs and a 1998-scale ice storm ~6%. They matter for the energy modules and for Canadian economic pain, not for outcomes.
Weakest parameters
Cold proficiency for both sides, daily_sd_*, blizzard and smoke thresholds, ice_storm_prob, occupation_cold_strength_loss. Snow is a heuristic spread of monthly normals, not a calibrated daily precipitation model.
Sources
ECCC Canadian Climate Normals 1991–2020 (station weatherboxes, incl. Fort McMurray and Yellowknife); Canada's Changing Climate Report 2026, ch. 2; Environment Canada on ENSO via Western Producer (2022, 2024); CIFFC data via The Weather Network and This Magazine (2025); EIA and NEB on the 2016 Fort McMurray fire; freezing-rain climatology and the 1998 ice storm (ECCC, Canadian Encyclopedia); USU and Bloomberg (2026) on the 11th Airborne Division; Task & Purpose (2025) on Army cold-weather doctrine.