Large-scale evidence of behavioral responses and adaptation to wildfire smoke

Using 371 million bikeshare trips across 12 US cities from 2010 to 2024, we find that wildfire smoke reduces urban cycling substantially and nonlinearly, and that populations with more prior smoke exposures avoid cycling more, not less - a pattern consistent with adaptation rather than habituation. Leveraging day-to-day variation in smoke in a distributed lag nonlinear model with high-dimensional fixed effects to absorb spatiotemporal confounding, we estimate that increasing from 0 ug/m3 to 25 ug/m3 of wildfire-attributable PM2.5 reduces daily rides by 8.2% (95% CI: 7.2-9.3%) and increasing to 50 ug/m3 reduces rides by 20.8% (19.0-22.7%). Wildfire smoke also shortens trip duration and distance among those who ride. Weekend trips are five times as responsive as weekday trips, consistent with recreational cycling being more deferrable than commuting. At the 90th versus 0th percentile of prior-year smoke exposure, the acute response at 25 ug/m3 is 7.9 percentage points larger; this difference is predominantly driven by within-city, between-year changes rather than permanent geographic differences between historically smoky and cleaner-air regions, consistent with behavioral learning rather than geographic selection. Three smoke events - the 2023 Canadian wildfire transport to the Northeast, the 2020 western fires, and the 2018 Camp Fire - produced peak daily ridership declines of 25-82%. As wildfire smoke intensifies, our results suggest populations will increasingly avoid outdoor exercise and, as cycling often offsets driving, may increase emissions of greenhouse gases and other hazardous pollutants. Consequences for physical health and the climate are not captured in existing estimates of wildfire smoke-related harms.