Traffic restrictions during the 2008 Olympic Games reduced urban heat intensity and extent in Beijing

The study investigates how reductions in urban traffic volume affect the urban thermal environment in Beijing, focusing on the 2008 Olympic Games traffic restriction (odd–even license plate rule) that halved daily vehicle volume. Urbanization intensifies the Urban Heat Island (UHI) effect via low-albedo, high-heat-capacity materials and anthropogenic heat from stationary (buildings, industry) and mobile (vehicles) sources. Prior work emphasized land cover change (albedo, vegetation) and suggested anthropogenic heat is smaller or negligible in many urban contexts. However, the distinct contribution of mobile anthropogenic heat from automobiles has been difficult to quantify because it is confounded with stationary sources. Beijing, with rapid growth, 3.3 million vehicles in 2008, and strong warming trends, implemented strict traffic controls from July 20 to September 20, 2008. This natural experiment enables a quantitative assessment of traffic volume impacts on daytime urban heat extent and intensity using satellite-derived land surface temperatures (LST) and statistical models that separate meteorological forcing (net radiation) from policy-induced traffic variation.

Past studies identified UHI drivers including reduced albedo and vegetation, urban morphology, and anthropogenic heat. Anthropogenic heat can raise temperatures 2–3 °C day and night, but many works deem it secondary to surface properties. Transport-related effects on UHI were previously assessed qualitatively through relationships with road networks and traffic density; quantitative estimates of mobile anthropogenic heat impacts have been lacking. Transportation is a major greenhouse gas source (~28.9% in the U.S.), but separating mobile from stationary heat contributions is challenging. This study addresses that gap by exploiting a large, time-bounded traffic reduction to quantify vehicle heat effects relative to net radiation.

Design: A quasi-experimental design compared a case period with traffic restriction (July 20–September 20, 2008) to control periods without restriction (July 1–September 30, 2007; July 1–19 and September 21–30, 2008). Time windows were restricted to adjacent summers to minimize land cover and stationary anthropogenic changes. Data: Daily MODIS LST v6 from Terra (10:30 a.m.) and Aqua (1:30 p.m.) at 1 km resolution were used to derive daytime LST; nighttime LST from Terra (10:30 p.m.) and Aqua (1:30 a.m.) were also processed. Images with >10% cloud cover or high wind (>5 m/s) were excluded. Total 77 images were used: morning (Terra) 36 cloud-free (16 case, 20 control); afternoon (Aqua) 41 cloud-free (13 case, 28 control). Ancillary data: Two cloud-free Landsat-7 ETM+ scenes (Sept 9, 2007 and Sept 11, 2008) were used to estimate broadband albedo via narrowband-to-broadband conversion. Albedo slightly decreased from 0.145 (2007) to 0.136 (2008). Meteorological observations from CLDAS station in Beijing provided incoming shortwave radiation, net longwave radiation, wind speed, and were used to compute daily net radiation. Zones: The Beijing metropolitan area was partitioned into six concentric zones by ring roads: Zone 1 (inside 2nd ring) through Zone 5 (between 5th and 6th ring); Zone 6 is a 20 km outward buffer beyond the 6th ring. Traffic restrictions applied inside the 5th ring, with spillover reductions anticipated in Zone 6 due to flow changes. Metrics: Urban heat spatial extent (SE) was defined as the area exceeding a temperature threshold. The primary threshold was 35 °C, selected as the 81st percentile of the maximum temperature distribution; sensitivity analyses used 34 °C and 36 °C. Urban heat intensity (HI) was defined as the mean LST per zone per day. Predictors: Net radiation Rn was computed as Rn = (1 − α)I + L, incorporating daily incoming shortwave (I), net longwave (L), and albedo (α). Traffic restriction was represented by a dummy variable RSRt (1 during restriction days, 0 otherwise), capturing approximately a 50% reduction in daily traffic volume. Models: Two multiple linear regression families were fitted separately for morning and afternoon: - SEt = a0 + a1 Rn + a2 RSRt + et, for each threshold (34, 35, 36 °C), estimating traffic impacts on high-temperature spatial extent. - HIzt = b0 + b1 Rn + b2 RSRt + ezt, for each zone z = 1..6, estimating traffic impacts on mean LST. Nighttime analogs were also modeled for SE and HI using nighttime overpasses. Assumptions and controls: Stationary anthropogenic heat and vegetation cover were assumed stable between summers 2007–2008, supported by similar electricity production/consumption and minimal observed land cover change. Latent heat flux was treated as effectively constant over the short study window. Wind screening reduced convective variability. Sensitivity analyses tested threshold dependence of SE results.

- Traffic restriction effects on daytime urban heat extent: - Morning (Terra): For thresholds 34/35/36 °C, SE decreased by 945.7/748.8/551.8 km²; models explained 57%/55%/54% of variance, with both Rn and RSR significant (p ≤ 0.05 for RSR). - Afternoon (Aqua): For thresholds 34/35/36 °C, SE decreased by 863.7/905.7/908.2 km²; R² ≈ 0.58–0.60 with significant predictors. Average shrinkage due to restriction ≈ 820 km². - Traffic restriction effects on daytime urban heat intensity (mean LST): - Morning reductions (Zones 1–5): 1.69, 1.65, 1.63, 1.59, 1.56 °C; models explained >50% of variance in Zones 1–5; Zone 6 less responsive in the morning (more vegetation/agriculture). - Afternoon reductions (Zones 1–6): 2.44, 2.37, 2.27, 1.83, 1.71, 1.50 °C; models explained 60–63% of variance; effects stronger in the afternoon and decline with distance from city center. - Nighttime: For spatial extent at 10:30 p.m. and 1:30 a.m., R² < 0.20; net radiation significant, traffic restriction not significant (p ≫ 0.05). For nighttime HI, R² < 0.45 across zones; R significant, RSR not significant. Indicates negligible nighttime traffic effects, consistent with unchanged nighttime traffic volumes. - Additional quantitative interpretations: - With an assumed linear relationship, the observed ~1.95 million vehicle reduction implies an average heat intensity effect of ~1–1.64 °C per 1 million vehicles; total potential traffic-related contribution for 3.3 million vehicles estimated at ~3.3–5.4 °C. - For SE at 35 °C threshold, a 1 million vehicle decrease corresponds to an average shrinkage of ~394 km² (morning) and ~477 km² (afternoon). The observed SE reductions represent ~12.4–15.0% of the metropolitan study area. - Model performance: Daytime SE models R² ≈ 0.54–0.60; daytime HI models R² ≈ 0.50–0.63 (morning) and 0.60–0.63 (afternoon). Nighttime models exhibited lower explanatory power and nonsignificant traffic terms.

The study directly addresses the research question by isolating the effect of a large, policy-driven reduction in mobile anthropogenic heat on urban thermal metrics, independent of meteorological variability captured via net radiation. The significant daytime reductions in both spatial extent and intensity of urban heat demonstrate that traffic volume has a substantial and previously underappreciated role in shaping urban thermal environments. The stronger afternoon effects and the radial gradient (larger in central zones, smaller outward) reflect spatial heterogeneity in urban functions, density, and impervious cover. The lack of nighttime effects is consistent with minimal changes in nighttime traffic and highlights the diurnal dependence of traffic-related heating. Compared with prior literature that downplayed anthropogenic heat relative to albedo/vegetation, these results suggest mobile sources alone can exceed earlier combined anthropogenic estimates, implying that mitigation strategies should incorporate traffic and vehicle energy use management alongside surface property modifications. The timing mismatch between MODIS overpasses and rush hours suggests that peak-hour heat effects could be even larger than estimated, reinforcing the significance of traffic control in heat mitigation policies.

This work quantifies, using satellite observations and statistical modeling, that halving traffic volume during Beijing’s 2008 Olympic traffic restrictions substantially reduced daytime urban heat intensity (by ~1.5–2.4 °C) and high-temperature spatial extent (~820 km² on average). The findings reveal mobile anthropogenic heat as a major, spatially and temporally variable contributor to urban heating, larger than previously assumed. Policy implications include expanding urban heat mitigation portfolios to explicitly manage traffic volumes, vehicle efficiency, and fuel types, in addition to high-albedo and greening interventions. Future research should: (1) refine quantitative attribution among anthropogenic sources (mobile vs. stationary), (2) capture peak traffic periods with higher-temporal-resolution observations, (3) incorporate additional variables explaining the remaining variance (e.g., humidity, synoptic conditions, aerosol loading), and (4) test generalizability across cities, seasons, and policy contexts.

- Unexplained variance: Daytime models account for ~50–60% of variance, leaving 40–50% potentially due to other factors (e.g., humidity, aerosols, synoptic variability) and measurement errors. - Nighttime insensitivity: No significant traffic effect at night, partly because nighttime traffic volumes were unchanged; models had low R². - Assumptions: Stationary anthropogenic heat and latent heat flux treated as constant over the study window; linearity assumed for extrapolations (e.g., per-million-vehicle effects). - Temporal sampling: MODIS overpasses miss rush-hour peaks (10:30 a.m. and 1:30 p.m.), possibly underestimating peak traffic heat effects. - Potential confounders: Regional factory shutdowns during the Olympics could have indirect thermal effects via reduced aerosols/greenhouse gases; land cover/albedo changes were minimal but not zero. - Spatial resolution: 1 km MODIS LST may smooth fine-scale urban heterogeneity.