Urban climate changes during the COVID-19 pandemic: integration of urban-building-energy model with social big data

The COVID-19 pandemic substantially altered human behaviour, especially in urban areas where stay-at-home advisories and teleworking reduced daytime populations in business districts. In Tokyo, a state of emergency in April–May 2020 reduced the daytime office-district population by more than 50%. While global studies documented improved air quality and reduced greenhouse gas emissions during lockdowns, the magnitude of local urban climate changes—particularly near-surface air temperature (T), anthropogenic heat emissions (Qf), and electricity consumption (EC)—remained uncertain. Observational detection of a clear pandemic signal in air temperature is challenging due to weather variability, motivating numerical modeling approaches. Prior urban climate models often prescribe anthropogenic heat rather than explicitly linking it to human behaviour. This study addresses these gaps by integrating real-time population and traffic data with a coupled urban canopy–building energy model (UCM-BEM) to quantify how pandemic-driven behavioural changes affected T, Qf, and EC in Tokyo, and to develop simple relationships enabling extrapolation to other Japanese cities. The broader aim is to evaluate whether behaviour change can serve as an urban climate adaptation and decarbonization strategy.

Previous research during COVID-19 documented improved air quality and reduced CO2 emissions globally and in cities, but global mean temperature impacts were projected to be small, suggesting a need for regional/local assessments. Urban temperature responses to lockdowns have been studied using land-surface temperature (LST), but near-surface air temperature (T) responses directly tied to human behaviour and energy use have been less quantified. Urban parameterizations and coupled building–urban models have demonstrated sensitivity of urban T to anthropogenic heat (Qf), including AC waste heat, and there is high confidence in their ability to simulate urban radiative exchanges. Earlier work using a UCM-BEM with real-time population/traffic indicated a modest T reduction (~0.13 °C) in Osaka but was limited by idealized spatial sampling and not aligned to the exact advisory period. Other observational/statistical studies reported pandemic-related urban T reductions in Tokyo (~−0.40 ± 0.21 °C) and Chinese cities (~−0.42 ± 0.26 °C). Related literature on urban planning scenarios (e.g., dispersed city) and weekday–weekend contrasts shows comparable magnitudes of ΔT (≈ −0.1 to −0.6 °C), highlighting the influence of human presence and activity on urban climate.

Data: The study used Mobile Spatial Statistics (MSS) from NTT Docomo (78 million devices; 1 h, 500 m resolution) to quantify population changes between 18 April–14 May 2019 and the same dates in 2020. Traffic counts at 5-min resolution from 8060 sites in the Tokyo Metropolitan Area (TMA) were used to quantify traffic changes. Additional meteorological observations were from JMA AMeDAS. Model: Simulations employed WRF ARW v3.7.1 coupled online with a comprehensive urban canopy–building energy model (WRF-CM-BEM). Three nested domains (25 km, 5 km, 1 km) covered Japan and resolved TMA; 34 vertical levels up to 50 hPa. Physics included RRTM-LW, Goddard SW radiation, Thompson microphysics, Mellor–Yamada–Janjic PBL, and Noah LSM. Urban representation used CM-BEM in TMA and a single-layer UCM elsewhere as needed. Building footprints and GIS-based land use/land cover (GIAJ, ESRI Geo Suite, Tokyo Metropolitan Government) informed urban geometry (mean building width, spacing, height distributions) and urban category classification: C (commercial/office), Rm (fireproof apartments), Rd (wooden detached dwellings). Building energy parameters (occupancy diurnal profiles, AC systems) were set from literature; appliance baseload EC was assigned per grid using observed EC data. Traffic-related anthropogenic heat (Qf,TRA) was computed from traffic census and fuel economy data and converted via combustion heat. Experimental design: Two cases were simulated for April–May 2020 (No-COVID and COVID). The No-COVID case ran 12 April 09:00 JST to 17 May 09:00 JST (first 5 days spin-up). The COVID case modified human behaviour and traffic parameters using observed change ratios during the advisory period: for working hours (08:00–22:00 JST), the number of occupants, appliance EC, and AC operation schedules were scaled by the population change ratio APCOVID (XCOVID = XNO-COVID × APCOVID). Traffic heat Qf,TRA was scaled hourly by the traffic ratio ATRA (Qf,TRA COVID = Qf,TRA NO-COVID × ATRA). A summertime simulation (27 July–1 September 2020) repeated the design using the same APCOVID and ATRA to assess seasonality under higher AC use. Model verification is described in supplementary materials. Diagnostics and extrapolation: Differences between COVID and No-COVID cases yielded ΔEC, ΔQf, and ΔT. Relationships between population change ratio (ΔP) and these variables were quantified by linear regressions for each urban category, yielding simple predictive equations: ΔECijt = aΔPijt + b; ΔQf ijt = cΔPijt + d; ΔT = eΔP + f, with slopes/intercepts provided by category (Supplementary Table 1). These equations, combined with land-use maps and observed population changes, were used to estimate impacts across major Japanese cities.

- Spring (Apr–May 2020, daytime 09–17 JST): - Central Tokyo office districts experienced large reductions in activity; at Tokyo Station the population was 39.4% of pre-COVID, EC was 30.2% of pre-COVID, Qf was 32.8% of pre-COVID, and ΔT = −0.21 °C. Abstract-level synthesis indicates EC ≈ 30% and Qf ≈ 33% of pre-COVID levels under strongly reduced population, yielding ≈ −0.2 °C cooling (~20% of historical GHG-induced warming ~1 °C in Tokyo). - Spatially, daytime EC decreased in office areas and increased in residential areas due to population redistribution. EC changes were largely from baseload appliances given limited heating/cooling demand in spring. - Daily total CO2 emissions in Tokyo’s 23 wards decreased by 8.2% (8.06 t-CO2 km−2 day−1) during the advisory period (emissions factor 0.000445 t-CO2 kWh−1). - Qf decreased in central Tokyo; total reduction ≈ −1.3 GW in office grids, comprised of 88.2% from AC-related waste heat and 11.8% from traffic. Qf decreased around Rm areas due to traffic reductions. - ΔEC and ΔQf exhibited clear linear correlations with population change; ΔT decreased with population reduction in C areas, with weak/no clear relationship in residential areas. - Nighttime impacts on EC, Qf, and T were smaller. - Summer (Jul–Sep 2020): - In office areas, EC decreased with reduced occupancy; in residential areas, EC increased markedly due to higher daytime occupancy and intensive AC use. Category-level EC changes correlated with occupancy changes. - Daily total CO2 emissions in the 23 wards decreased by 7.9% (11.07 t-CO2 km−2 day−1). - Qf reduction in office grids was much larger than in spring (≈ −3.1 GW), while Qf increased in Rm areas due to AC use despite reduced traffic. ΔQf–population correlations strengthened relative to spring. - Office-area ΔT reached −0.30 °C, exceeding spring impacts and representing ~30% of past GHG-induced warming in Tokyo; temperatures in residential areas increased relative to No-COVID due to increased AC waste heat. - Nationwide extrapolation using linear equations reproduced spatial ΔEC patterns in TMA and indicated EC, Qf, and T reductions in major city centers across Japan. Example city-level results (daytime, spring): Tokyo ΔT −0.21 °C; Osaka −0.10 °C; Nagoya −0.10 °C; Sapporo and Sendai −0.07 °C; Kyoto −0.01 °C; Kobe −0.07 °C; Fukuoka −0.08 °C. Corresponding ΔEC (W floor m−2) and ΔQf (W m−2) were negative in city centers. - Consistency with prior studies: Magnitudes of ΔT align with previous modeling (Osaka ≈ −0.1 °C) and independent observational/statistical estimates in Tokyo (~−0.40 ± 0.21 °C) and Chinese cities (~−0.42 ± 0.26 °C). The magnitude is comparable to dispersed-city scenarios and weekday–weekend differences reported in literature.

By explicitly linking real-time human behaviour (population and traffic) to building energy use and anthropogenic heat in a coupled urban climate model, the study isolates the “pandemic signal” in urban air temperature. Reductions in daytime occupancy in office districts lowered EC and Qf, leading to measurable daytime cooling (−0.2 to −0.3 °C) in city centers, while increases in residential occupancy shifted energy use and anthropogenic heat to residential areas, sometimes offsetting cooling locally. The clear linear relationships between population change and ΔEC/ΔQf/ΔT support the causal chain from behaviour to energy use to heat release to air temperature. Agreement with independent observational/statistical studies and with known weekday–weekend and dispersed-city signals strengthens confidence in the estimated ΔT. From a policy perspective, widespread remote work can emulate aspects of dispersed urban form at far lower cost, delivering localized cooling and emissions reductions; however, benefits may be redistributed, potentially increasing residential heat release, especially in summer with high AC usage. The results thus demonstrate the dual decarbonization and adaptation potential of behaviour changes while highlighting spatial trade-offs that urban planners need to manage.

This study introduces and applies an integrated framework combining social big data (real-time population and traffic) with a UCM-BEM to quantify how COVID-19-induced behaviour changes affected urban EC, Qf, and near-surface air temperature. In Tokyo, office-district EC and Qf dropped to roughly one-third of pre-COVID levels, producing daytime cooling of about 0.2–0.3 °C, and daily CO2 emissions decreased by ~8% in spring (~8.2%) and ~8% in summer (~7.9%). Simple linear equations relating population change to ΔEC, ΔQf, and ΔT were derived and successfully reproduced patterns across Japanese cities, enabling broader application where climates, building stocks, lifestyles, and AC usage are similar. Behavioural changes such as remote work can function as cost-effective urban climate adaptation and decarbonization strategies, with magnitudes comparable to established heat island countermeasures. Future research should quantify combined effects of behaviour change with “hard-type” measures (e.g., cool roofs, greening), assess seasonal and long-term dynamics under climate change and evolving AC adoption, and evaluate equity and spatial trade-offs across urban neighborhoods.

- Observational detection of pandemic-induced temperature signals is inherently difficult due to meteorological variability; while modeling helps isolate the signal, results depend on model physics, parameterizations, and input datasets. - The approach assumes linear relationships between population change and ΔEC/ΔQf/ΔT by urban category; these may vary with season, meteorological conditions, building technologies, and behavioural patterns, especially under extreme heat or evolving AC efficiency. - Generalizability is strongest for regions with climates, building materials, urban forms, lifestyles, and AC usage similar to Japan; applicability elsewhere requires benchmarking and recalibration. - Residential increases in AC use can offset or reverse cooling locally, especially in summer; the study period focuses on spring and one summer season in 2020 and may not capture interannual variability. - MSS population estimates and traffic data, while extensive, are indirect proxies and subject to sampling and extrapolation uncertainties; baseload EC and building operation schedules are parameterized from literature and prior datasets. - Nighttime and non-working-hour behaviours were modified less, so impacts there are smaller and more uncertain; other sources of anthropogenic heat (industrial processes) are not the main focus.