The coronavirus SARS-CoV-2 (COVID-19) pandemic significantly changed human behaviour worldwide. In many major cities globally, the ensuing lockdowns greatly restricted human activity. In Japan, because the number of COVID-19 cases in Tokyo tended to increase clearly, a state of emergency was declared in April 2020, which reduced the daytime population in city office districts by more than 50% due to the promotion of teleworking at home (Supplementary Fig. 1a). After the declaration, the number of cases tended to decrease, so the state of emergency was lifted on 14 May (Supplementary Fig. 1a). This change in human behaviour during the state of emergency may continue to some extent after COVID-19, with the development of a ‘new-normal’ situation. Note that although the state of emergency in Japan was a self-restraint campaign, exactly not lockdown, the magnitude of human behaviour change was like that of lockdown.

Previous studies have reported how the lockdown and stay-at-home advisories have changed environments worldwide. For example, air quality has improved1,2,3,4,5 and greenhouse gas (GHG) emissions have been reduced6,7,8,9,10 due to lockdowns and stay-at-home advisories. According to Friedlingstein et al.11, global CO2 emissions from fossil fuels decreased by 5.4% due to the COVID-19 lockdown in 2020. However, the global temperature response due to COVID-19 restrictions will probably be small, and climate scientists are encouraged to look for regional climate signatures3. Now shifting the perspective from global to local, changes in CO2 emissions have been directly observed in the urban areas where we live7,9. So, what was the impact on local temperature (urban heat islands) in cities where human activities strongly affect them? In this context, changes in the land-surface temperature (LST) due to lockdowns have also been reported12,13. In addition, the lockdown and stay-at-home advisories could have changed the urban climate in terms of urban near-surface air temperature (T), anthropogenic heat emissions (QF) and electricity consumption (EC) because the urban climate is strongly affected by human behaviour14. However, only a few studies have attempted to estimate this effect based on the local climate15,16,17. The technical difficulty of detecting changes in urban T due to the pandemic is a limitation in such studies. From an observational perspective, it is not easy to clarify a pandemic signal (temperature change due to the stay-at-home advisories) in observed temperatures because the weather changes constantly due to natural variability. It is not possible to conclude that an urban temperature decrease/increase during lockdown was not just due to the natural variation in urban temperature.

Under such a limitation, a numerical simulation using urban parameterisations is a powerful tool, which could extract the pandemic signal. For example, urban parameterisation could estimate the impact of QF on urban T18,19,20,21,22,23,24,25,26. There is robust evidence and high agreement that urban parameterisations can be used to generally simulate radiation exchanges in a realistic way27. However, there is another limitation in the use of a numerical model due to its incomplete modelling of the relationship between urban climate and human behaviour. For example, a single-layer urban canopy model (UCM)28 can handle the input of QF, which relates to human behaviour but does not directly express human behaviour (the avarage diurnal change in QF is input into the model). In this study, a coupled model, UCM with the building energy model (BEM) (hereafter, UCM–BEM)16,29,30,31 was used to express human behaviour and simulate its effects on the urban climate (see the Methods section for details of how human behaviours were input into the model).

Using the UCM–BEM, a previous study attempted to estimate the impact of stay-at-home advisories due to COVID-19 on the urban climate16. The authors developed a simulation system that used real-time population and traffic data. More specifically, they estimated population and traffic changes due to COVID-19 and input these changes as the UCM–BEM parameters related to human behaviour. This new modelling system showed that the COVID-19 stay-at-home advisories resulted in a QF of 76.3 W m–2 and a 0.13 °C air temperature reduction in urban areas in Osaka, the second-largest city in Japan. Although this work indicated the impact of COVID-19 on urban climate based on numerical simulations, there were some limitations. They did not consider the actual stay-at-home advisories period of COVID-19, and they used population change data on only one ideal office grid point in Osaka. To estimate the impact of COVID-19 on urban climate with more accuracy, it is necessary to conduct a simulation during the actual stay-at-home advisories period based on population change data for all of the grid points in a target area.

Estimations made using a numerical model are essential in areas related to global climate change, urban climate, sustainability, environment, human behaviour and urban planning. With cities predicted to be home to more than 66% of the global population by 205032, urban warming due to global warming and the urban heat island effect are critical issues26,33,34,35,36,37,38. Changes in human behaviour due to lockdowns and stay-at-home advisories could be an adaptation strategy to climate change in urban areas, but it remains unclear how effective those would be. In urban areas significantly affected by climate change27, many ‘hard-type’ climate change adaptation strategies (traditional urban heat island mitigation techniques such as greening and cool roofs) have been proposed and evaluated37,39,40. A ‘soft-type’ strategy, such as a change in human behaviour, has yet to be assessed sufficiently, although some studies have already estimated weekday–weekend temperature differences41,42,43,44. The impact of changes in human behaviour on urban climate remains unknown. The COVID-19 pandemic could provide an opportunity to understand how human behaviour affects the urban climate based on scientific principles.

We estimated the impact of changes in human behaviour related to the COVID-19 pandemic on the urban climate in the world’s biggest metropolitan area of Tokyo, which has a hot and humid climate during summer, using the UCM–BEM and real-time population and traffic data (social big data). Through this sophisticated estimation, we developed simple equations to estimate the impact on the whole of Japan and other countries with similar climates, building materials and human behaviour. The aim was to understand whether changes in human behaviour could be an adaptation strategy to climate change. To sum up, the novelties of our study are: (1) proposing a new method which is an integration of urban climate-building-energy model and social big data, (2) showing the COVID-19 impact on not only urban ‘air’ temperature (T), but also important urban climate-related variables: QF and EC (CO2 emission), (3) proposing simple equations to estimate the COVID-19 impact on a more wide area and (4) providing basic information on whether human behaviour changes could be an adaptation strategy to climate change.

Results and discussion

Changes in human behaviour due to the COVID-19 pandemic

The population densities of the Tokyo Metropolitan Area (TMA) from 18 April to 14 May 2019 and 2020 (Fig. 1a, b) and its change ratio (2020/2019) (Fig. 1c) were plotted. Human activity was significantly changed by the pandemic. The population in the centre of Tokyo and its surrounding area decreased and increased, respectively (Fig. 1c). The traffic count differed from the population data (Fig. 1d, e, f). The traffic counts decreased in almost all areas of the TMA due to the stay-at-home advisories (Fig. 1f). These changes in population and traffic counts were used as human behavioural change parameters in the numerical model (see details in the Methods section).

Fig. 1: Changes in human behaviour due to the COVID-19 pandemic.
figure 1

Horizontal variation in population densities (pop km−2) for 15:00 Japan Standard Time (JST) a 18 April to 14 May 2019, b 2020, and c the 2020/2019 change ratio. Traffic counts for d 18 April to 14 May 2019, e 2020, and f the 2020/2019 change ratio and site locations of analysed point: To (Tokyo), Yo (Yokohama), Ch (Chiba), Om (Omiya), Mi (Mito), Ta (Takasaki), and Ut (Utsunomiya) on the Kanto Plain including the Tokyo Metropolitan Area (TMA).

Urban climate change in the Tokyo Metropolitan Area

After verification of the numerical model (see Supplementary Notes 2), we estimated the impact of COVID-19 induced human behavioural changes on the urban climate (Fig. 2).

Fig. 2: Impact of COVID-19 on urban climate.
figure 2

Horizontal variation in the impact of COVID-19 on a energy consumption (EC) (ΔEC), d anthropogenic heat (QF) (ΔQF) and g near-surface air temperature (T) (ΔT) as a daytime average for the period of 18 April to 14 May 2020. Bar charts and box plots of b ΔEC, e ΔQF and h ΔT for each urban category and the total for the 23 wards of Tokyo (bold black line in a, d, and g). The relationship between population changes and c ΔEC, f ΔQF and i ΔT for each urban category.

We calculated the changes in EC (COVID minus No-COVID: ΔEC). The daytime (09:00–17:00 Japan Standard Time [JST]) EC in the centre of Tokyo (office area) decreased (Fig. 2a, b) due to the reduced population (Fig. 1c). For example, the number of people using the Tokyo railway station during the pandemic was only 39.4% of the pre-COVID level, and the EC during the pandemic was estimated to be 30.2% of the pre-COVID level (Table 1a). By contrast, the EC increased in residential areas (Fig. 2a, b) due to the increase in the daytime population (Fig. 1c). These ΔEC values were mainly due to changes in the baseload (EC for appliances that is not affected by changes in T) during the stay-at-home advisories period (April–May). This period was not the season when urban residents typically use space heating or cooling (Supplementary Fig. 1b). The EC was directly correlated with population change in terms of building use in commercial and business grids (office: hereafter, C), residential grids that predominantly consisted of fireproof apartments (Rm) and residential grids that were chiefly covered by wooden detached dwellings (Rd) (Fig. 2c). Based on the ΔEC, the changes in CO2 emissions were estimated using the CO2 emissions coefficient of 0.000445 (t-CO2 kWh−1) ( The estimation showed that daily total CO2 emissions in the 23 wards of Tokyo decreased by 8.2% (8.06 t-CO2 km−2 day−1) due to the COVID-19 pandemic.

Table 1 Summary of the relationships between the changes in population and urban climate.

The ΔEC affected the daytime changes in QFQF). As with the ΔEC, the QF decreased due to the pandemic in the centre of Tokyo (Fig. 2d), with a reduction of about −1.3 GW in the total office grids (Fig. 2e), which is contributed by 88.2% from air-conditioning (AC) use (ΔQF, AC) and by 11.8% from traffic (ΔQF, TRA). The QF in the areas surrounding Rm was also reduced by a decrease in traffic (Fig. 1f). Figure 2f shows the relationship between the population change ratio and QF, and indicates that, as with the ΔEC, the ΔQF was also correlated with population change.

The ΔQF causes changes in the urban TT). Figure 2g shows the impact of the pandemic-induced human behavioural changes on daytime urban T and indicates that the temperature in the centre of Tokyo (C areas) decreased due to the pandemic. The temperature decrease reached −0.21 °C (Fig. 2h and Table 1a, Tokyo railway station). By contrast, there were no significant temperature changes in the surrounding areas (residential areas) (Fig. 2g, h). The relationship between the population change ratio and T change is shown in Fig. 2i, which indicates that T tended to decrease with population reduction in the office area (C). There was no clear relationship in the residential areas (Rd and Rm).

As in the daytime, the EC, QF and T were also impacted by the changes in human behaviour in the nighttime, but the impact was lower because nocturnal human behaviour during that time did not change significantly.

Application to the urban climate of all Japanese cities

Based on the detailed estimation using the model and the mobile spatial statistics (MSS) population data shown in the section ‘Urban climate change in the Tokyo Metropolitan Area’, we estimated the impact of the pandemic on the urban climate of all Japanese cities. There was a clear relationship between the population change ratio (ΔP) and, ΔEC, ΔQF, and ΔT (Fig. 2c, f, i), which could be used to estimate the impact of COVID-19 on the urban climate of other cities. We proposed the following equations:

$$\Delta EC_{i,j,t} = a\Delta P_{i,j,t} + b$$
$$\Delta Q_{F\,i,j,t} = c\Delta P_{i,j,t} + d$$
$$\Delta T_{i,j,t} = e\Delta P_{i,j,t} + f$$

where i and j are the grid numbers in the zonal and meridional direction, t is hours, and a (c, e) and b (d, f) are the slope and intercept, respectively (Supplementary Table 1). These relationships were used for the simple estimation of the impact of COVID-19 on the urban climate of all Japanese cities. For example, it was possible to estimate the impact of changes in human behaviour on the EC of all Japanese cities using just a linear Eq. (1), actual population change data and land-use data (urban categories: C, Rd, and Rm) (Supplementary Fig. 3b) for all Japan cities. Similarly, the impact on QF and T could also be estimated.

Fig. 3: Change in electricity consumption (ΔEC) for six major metropolises in Japan.
figure 3

a Sapporo (Hokkaido), b Sendai (Miyagi), c Tokyo (Tokyo) and Yokohama (Kanagawa), d Nagoya (Aichi), e Osaka (Osaka), Kyoto (Kyoto) and Kobe (Hyogo), and f Fukuoka (Fukuoka).

Figure 3c shows the estimated ΔEC using Eq. (1) for TMA. The horizontal distribution of ΔEC (Fig. 2a) simulated by the sophisticated model was reproduced using equation (1). Based on the above, Fig. 3 and Table 1b indicate that EC decreased due to the pandemic in all major city centres in Japan. Similarly, it was possible to estimate the impact on CO2 emissions and QF, with similar results to those obtained for EC (Table 1b). These simple equations we developed were used to estimate the impact on cities in Japan and other countries with similar climates, building materials, lifestyles and AC usage.

Comparison with observations and previous estimations

The results of this study and previous studies are summarised in Table 2. Note that we do not compare our result with previous works focused on the COVID-19 impact on LST as it is impossible to directly compare our estimated urban ‘air’ temperature and LST. A maximum temperature decrease of about −0.2 °C was estimated for Tokyo’s office district due to the pandemic. This temperature decrease was consistent with the reported temperature decrease of about −0.1 °C in central Osaka, which was previously estimated using the same method16. More importantly, the calculated results were consistent with the temperature decrease in central Tokyo and cities in China previously estimated from observations and a statistical analysis (−0.40 ± 0.21 °C15 and −0.42 ± 0.26 °C17), which was an entirely different approach from the present study. The consistency between results suggests that ΔT could be estimated realistically.

Table 2 Summary of the results of previous studies and this study.

As described in the introduction, only a few studies have evaluated ΔT induced by the COVID-19 pandemic, and comparisons with the results of previous studies are therefore limited. Instead, our results are compared to those of studies that have assessed the effects of introducing urban planning scenarios on T. Specifically, Adachi et al.45 and Kusaka et al.46 estimated the impact of a dispersed city, i.e., an urban planning scenario in which offices and other functions are distributed around the city centre, rather than being concentrated in the city centre. Because commuting to the city centre was significantly reduced during the state of emergency period (Fig. 1), the voluntary restraint on going out due to the pandemic could cause a similar situation in terms of commuter behaviour compared with that in a dispersed city. Therefore, we compared our results of a temperature decrease of about −0.2 °C, with those of Adachi et al.45 and Kusaka et al.46. The dispersed city scenario was estimated to induce negative ΔT reaching −0.2–0.3 °C in the city centre (Table 2). Our estimate is almost the same. Thus, the stay-at-home advisories can be viewed as a kind of social experiment similar to the introduction of dispersed cities. To create a dispersed city, offices need to be moved from the city centre to the suburbs, which costs money. Promoting remote work would therefore be more cost-effective than developing a dispersed city.

Furthermore, the temperature difference between weekdays and holidays was compared to the T drop of −0.2 °C recorded in this study. Temperatures are lower in office areas in central Tokyo on weekends than on weekdays because fewer working people on weekends there41,42,43,47,48. The stay-at-home advisories could induce a situation resembling a holiday almost every day. Therefore, comparing the temperature difference between weekdays and holidays is essential. In estimates made in previous studies, the temperature on holidays was −0.1–0.6 °C lower than on weekdays (Table 2). Our values are consistent, which implies that the estimated impacts of this study are reasonable. The numerical model used in this study reproduced the difference between observed weekday and holiday temperatures well and was used for calculations in cities49, and it was therefore reasonable to make the above comparison. Finally, the Chinese New Year (CNY) influenced the impact of changes in human behaviour on T. The impact was about −0.6 °C44,50,51,52, and therefore the impact of the stay-at-home advisories in Japan tended to be smaller than that of CNY, although the order of magnitude was comparable.

In terms of CO2 emissions, we estimated that emissions during COVID were about 7% smaller than during the No-COVID period in the residential area of Tokyo, which qualitatively agrees with the changes in the observed CO2 flux in the same area reported by Sugawara et al.9.

Could changes in human behaviour be a climate change adaptation in urban areas?

The section ‘Urban climate change in the Tokyo Metropolitan Area’ shows the urban climate change due to the pandemic in the spring season when AC is rarely used. In the future, it is likely that AC usage will increase as urban temperatures increase leading to high levels of heat stress26,38,53,54,55. To assess the potential for climate change adaptation in cities, it was necessary to estimate how the urban climate changed due to the pandemic during summer when AC is often used. Although the state of emergency was lifted by 25 May, social activity was suppressed in summer due to the re-spread of infection. Therefore, we estimated the impact of COVID-19 stay-at-home advisories on the summer urban climate and compared the effect with widely adopted climate change adaptation strategies (e.g., greening and cool roofs). In the summertime simulation, we conducted ran from 09:00 JST on 27 July to 09:00 JST on 1 September 2020 and used the same as the population change ratio (ΔPCOVID) and the traffic count ratio (Δtra) (see Methods) as used in April–May simulation shown in the section ‘Results and discussion’.

Supplementary Fig. 2 shows the impact of the pandemic on urban climate in terms of ECEC), QFQF) and TT) during the summer period. The daytime EC in the centre of Tokyo decreased (Supplementary Fig. 2a) due to the population reduction (Fig. 1c) and limited AC use. By contrast, there was a large EC increase in the residential areas (Supplementary Fig. 2a, b) due to the daytime population increase (Fig. 1c) and high levels of AC use. This trend was supported by observations using a smart meter in residential houses in Japan (not shown). These ΔEC values were caused by changes in the baseload and AC use (Supplementary Fig. 2b). The EC was directly correlated with the changes in the occupancy of buildings (Supplementary Fig. 2c). The estimation showed that the daily total CO2 emissions in the 23 wards of Tokyo decreased by 7.9% (11.07 t-CO2 km−2 day−1) due to the COVID-19 pandemic.

The QF in summer decreased much more than in spring due to the pandemic in the centre of Tokyo (Supplementary Fig. 2d versus Fig. 2d), with the amount being about −3.1 GW in the office grids (Supplementary Fig. 2e). By contrast, the QF in the Rm areas increased due to AC use, although the traffic count decreased (Supplementary Fig. 2e and Fig. 1f). Supplementary Fig. 2f shows the relationship between the population change ratio and ΔQF, which indicated that the ΔQF was much more strongly correlated with population change than the temperature in spring (Supplementary Fig. 2f versus Fig. 2f).

The ΔT in the centre of Tokyo (office area) was −0.30 °C, which was a more significant impact than in spring (Supplementary Fig. 2h versus Fig. 2h) and was equivalent to 30% of the past GHG-induced warming in Tokyo: the warming in Tokyo is largely influenced by urbanisation, and the contribution of GHGs is likely to be 1 °C or so56. In contrast to office areas, the temperature increased in the surrounding areas (residential areas) (Supplementary Fig. 2g, h). This increase was also much more significant than in spring (Supplementary Fig. 2h versus Fig. 2h). The relationship between the population change ratio and temperature change is shown in Supplementary Fig. 2i, which suggests that temperature tended to decrease/increase with population changes in office area C and residential areas (Rd and Rm). respectively.

The magnitude of the temperature change (ΔT) may increase under future climate change because many urban residents will begin to use AC, and the thermal heat load will be increased by the increase in outdoor temperature26,38,57. The summertime ΔT presented above is comparable/similar to that attained by adopting the well-known urban heat island countermeasure techniques37,39 (e.g., cool roofs and greening58). An advantage of the change in human behaviour induced by the pandemic as an adaptation strategy is that there are no/small introduction and running costs. In addition, such changes in human behaviour could reduce exposure to high-heat stress in the outdoor environment. Future studies should estimate the combined effects of this change in human behaviour (a ‘soft-type’ strategy) and other well-known adaptation strategies, such as greening and cool roofs (a ‘hard-type’ strategy)37,39.


Human behaviour and traffic data (social big data)

We used MSS data regarding the stay-at-home advisories (18 April to 14 May 2020) and the same dates in 2019 (18 April to 14 May 2019) provided by NTT Docomo, Inc. The MSS data provided population statistics based on the location of 78 million users of NTT Docomo’s mobile terminals in the whole Japan. The MSS data are created from this information16. First, the number of mobile terminals in each base station area was aggregated. Second, the total number of mobile terminals was extrapolated based on the adoption rates of NTT Docomo mobile terminals. Finally, the estimated population was re-aggregated into each grid section. More details of the MSS estimation method are provided by Terada et al.59. The temporal and spatial resolutions of MSS data were 1 h and 500 m, respectively. Previously, MSS data have been used to estimate human activity in urban climates16. Other population data such as location trends were provided by KDDI, Inc. and the Japan Travel Guide of NTT Advertising, Inc.; however, their sample numbers were relatively small60. Thus, MSS provides data that can be used to examine real-time changes in specific populations.

The level of traffic congestion is related to the QF derived from automobiles (QF, TRA). To estimate the changes in traffic volumes during the stay-at-home advisories, we used traffic counts for public roads at 5-min intervals for 8060 sites in TMA in April–May 2019 and 2020. These data were provided by the Japan Road Traffic Information Center and were distributed by the Japan Traffic Management Technology Association.


The study used the Advanced Research WRF (ARW) ver. 3.7.161 and the online coupling of WRF and CM-BEM29 (hereafter, WRF–CM–BEM16,62). Supplementary Fig. 3 shows the model domain that covered the area of Japan, including TMA, which was the focus of our study. The d01, d02 and d03 domains consisted of 89, 110 and 250 grid points, respectively, in the x and y directions. We set the horizontal grid spacing to 25, 5 and 1 km in the domains d01, d02 and d03, respectively. The model top was 50 hPa, with 34 vertical sigma levels. In this simulation, the initial and boundary conditions were derived from the National Centers for Environmental Prediction–National Center for Atmospheric Research reanalysis data63.

The following schemes were used in the simulation: the updated Rapid Radiation Transfer Model longwave scheme64, the Goddard shortwave scheme65,66, the Thompson microphysics scheme67, the Mellor–Yamada–Janjic atmospheric boundary-layer scheme68,69,70, the Noah land surface model71, and the CM-BEM29,62 for the TMA and a single-layer urban canopy28,72,73 in some urban grids around the city centre.

As in Takane et al.23 and Nakajima et al.16, building footprint (polygon) data from a geographical information system in the TMA were used to identify urban canopy geometry were being used. The building use and total floor area for each building in the TMA were recorded. Land use and land cover (LULC) data and topographic datasets of the Geospatial Information Authority of Japan (GIAJ) were used in this study. In addition to the GIAJ LULC data, the ArcGIS Geo Suite provided by ESRI-Japan Co., Ltd. ( and a land-use survey performed in the 2016 fiscal year by the Bureau of Urban Development of the Tokyo Metropolitan Government were used for classification. The urban grids were classified into three categories (C, Rm and Rd) based on the dominant building type, as shown in Supplementary Fig. 3b. In the study, when we analysed each category, we only used the grids for those categories within the TMA.

The model used in this study required the identification of parameters for the detailed building energy calculations. For the geometric parameters of urban canopies, the mean building width, mean distance between buildings, and distribution of building height had to be set in every urban grid. These settings were derived from the same building footprint data and were summarised in Supplementary Table 2. For the building energy simulation, the diurnal variation in the number of occupants, and AC systems were based on the values in the literature for the three-urban categories (Supplementary Fig. 3). More details of the parameters are provided in previous studies23,26,38,62,74,75,76. Note that we set the EC for appliances (baseload EC) in every urban grid based on actual EC data presented by Nakajima et al.16,77.

As with Kikegawa et al.62 and Nakajima et al.16,77, the hourly total consumption of motor fuels in each grid, with a temporal resolution of 1 h, was estimated using traffic census data78 and fuel economy data79. The combustion heat was obtained as the QF, TRA.

We also used the Automated Meteorological Data Acquisition System (AMeDAS) data for TMA provided by the Japan Meteorological Agency (JMA) as meteorological data.

The simulation was carried out from 09:00 JST (00:00 UTC = 09:00 JST) on 12 April to 09:00 JST on 17 May 2020 (Supplementary Fig. 1b). The first 5 days were discarded and considered the model spin-up. In this study, we referred to this simulation as the No-COVID case. For comparison with the April–May simulation, we conducted a summertime simulation, which ran from 09:00 JST on 27 July to 09:00 JST on 1 September 2020 (see the section ‘Model’, Supplementary Fig. 1c).

In addition to the No-COVID case, another case (COVID) was also conducted to estimate the impact of COVID-19 stay-at-home advisories in the city. Because the population and traffic count during the COVID case differed from those of usual days (Fig. 1), we modified the human behaviour parameters (the number of occupants, EC for appliances and AC operation schedule) and QF,TRA (Fig. 4b) using the population change ratio of the stay-at-home advisories period to the usual days (ΔPCOVID) (Fig. 4a) and the traffic count ratio of the stay-at-home advisories period to the usual days (Δtra) (Fig. 4c). The ΔPCOVID and Δtra values are plotted in Fig. 1c, f, respectively. The modified human behavioural parameters (XCOVID) of each grid were calculated for the working hours (08:00–22:00 JST) on the stay-at-home advisories period as follows (Fig. 4b):

$$X_{i,j,t}^{{\rm{COVID}}} = X_{i,j,t}^{{\rm{No}} - {\rm{COVID}}} \times \Delta P_{i,j,t}^{{\rm{COVID}}}\,for\,t = 8,9,....,21,22\,{\rm{JST}}$$

where i and j are the grid numbers in the zonal and meridional directions, respectively, t is hours and XNo-COVID is the human behaviour parameter in the No-COVID case (see Fig. 4b).

Fig. 4: Schematic view of an integration of urban-building-energy-model with social big data.
figure 4

a MSS, b UCM-BEM, and c traffic count data. The red and blue lines indicate human behaviours in the COVID and No-COVID cases, respectivery.

Similarly to the aforementioned parameters, the diurnal variation in QF, TRA in the stay-at-home advisories period (QF, TRACOVID) was calculated as follows:

$$Q_{{\rm{F}},{\rm{TRA}}\,i,j,t}^{{\rm{COVID}}} = Q_{{\rm{F}},{\rm{TRA}}\,i,j,t}^{{\rm{No}} - {\rm{COVID}}} \times \Delta t_{{\rm{TRA}}i,j,t}for\,t = 0,1,....,22,23\,{\rm{JST}}$$

where QF, TRANo-COVID is QF, TRA in No-COVID.

Any other setting of COVID is the same as No-COVID except for the above human behaviour parameters.