Model-Integration of Anthropogenic Heat for Improving Air Quality Forecasts over the Beijing Megacity

In air quality forecasting systems, failure to consider the considerably large anthropogenic heat emissions generated daily in the Beijing megacity by intensive human activities is one of the major causes of model failure. In this paper, we employ the nested air quality prediction model system coupled with the weather research and forecasting model and an urban canopy model to integrate anthropogenic heat emissions over Beijing into the modeling system and exhaustively evaluate their potential effects on air quality forecast by analyzing the wind field, boundary layer structure (height and atmospheric circulation), and surface and vertical distribution of pollutants. Consequently, the effects of anthropogenic heat on the boundary layer structure, greatly pronounced in urban areas, exhibited substantial variability at different levels depending on the time. The effects were evident during both daytime and night, but played a more prominent singular role in the night in the absence of solar short-wave radiation. Basically, anthropogenic heat acts not only by directly inducing the ascent of a warm air mass from the low parts of the atmosphere over urban areas to the top of the boundary layer, but also by indirectly driving wind convergence and inducing the descent of a cooled air mass from a high altitude to the boundary layer through a complex atmospheric circulation process. Incorporating anthropogenic heat emissions into the modeling system was effective in improving predictions by reducing the normalized mean bias by 20%–30% (for wind speed) and root mean square error by 361–558 m (for boundary layer height) and by 10–23 μg m (for surface PM10), with a significant reduction in the underestimation of ozone concentration by approximately 20 ppb at urban sites. This paper is expected to provide new insights into the improvement of model accuracy for air quality forecasts over megacities.


INTRODUCTION
Numerical modeling is a crucial approach to investigate atmospheric pollution problems, and it has been widely used to explore the spatial-temporal patterns, formation mechanisms, and control strategies of air pollutants (Streets et al., 2007;Wu et al., 2011).However, in recent years, models have been found to frequently fail in accurately forecasting air pollution episodes, possibly not only because of emission uncertainties but also because of failure to consider some important physicochemical processes such as anthropogenic heat effects in the modeling design (Li et al., 2011b;Zhang et al., 2014).The rapid economic development in China is associated with the country's increasing urban population and potential anthropogenic heat emission, which has become a critical factor affecting the atmospheric circulation and thus cannot be neglected (Ichinose et al., 1999;Hamilton et al., 2009;Quah and Roth, 2012).As one of the worldscale megacities with a population of more than 20 million, Beijing is drastically affected by anthropogenic heat emission with increasing effects on the local atmospheric circulation (Yang et al., 2014).This city is surrounded by mountains on three sides and is frequently controlled by stable synoptic conditions with temperature inversion in the boundary layer (Zhang et al., 2013b); such conditions tend to enhance the anthropogenic heat effects (Miao et al., 2009).
Basically, anthropogenic heat can increase turbulent fluxes in sensible and latent heat, resulting in the atmosphere reserving more energy (Oke, 1988), with a significant influence on the dynamics and thermodynamics of the urban boundary layer structure (Ichinose et al., 1999;Block et al., 2004;Fan and Sailor, 2005;Chen et al., 2012;Bohnenstengel et al., 2014) and variation in surface meteorological conditions (Khan and Simpson, 2001;Zhu et al., 2011;Menberg et al., 2013).Anthropogenic heat might also strengthen the vertical movement of urban surface air flow and change the urban heat island circulation, leading to enhanced turbulence and instability of the boundary layer (Bohnenstengel et al., 2014).Furthermore, anthropogenic heat in cities may induce significant and extensive warming and cause an increase in the urban air temperature by several degrees (Fan and Sailor, 2005;Ferguson and Woodbury, 2007;Zhu et al., 2011).Because urban air quality and local meteorological conditions are inextricably linked, the aforementioned findings are likely to have important implications for pollution forecast and control measures.However, anthropogenic heat emission in cities is a factor that is often neglected in atmospheric models (Mills and Arnfield, 1993;Hafner and Kidder, 1999).Regarding the thermal impact induced by intensive human activities in Beijing and surrounding areas, neglecting the huge anthropogenic heat emissions in models appears to be one of the major causes of large bias in air quality forecasting systems over the city.
To address this problem, a few recent studies attempted to analyze the effects of anthropogenic heat on surface air pollutants (Ryu et al., 2013a;Yu et al., 2014).However, further substantial investigation is necessary to elucidate the positive effects of anthropogenic heat on air pollutant vertical variations for the improvement of model predictions.
Accordingly the present study used the nested air quality prediction model system (NAQPMS) coupled with the weather research and forecasting (WRF) model and an urban canopy model (UCM) to integrate anthropogenic heat emissions over the Beijing megacity into the modeling system and exhaustively evaluate their potential effects on air quality forecast through a thorough analysis of wind speed, boundary layer structure (height and atmospheric circulation), pollutant spatial distribution, and vertical patterns.The availability of an estimate of anthropogenic heat emissions and boundary layer observations from aerosol lidar and tower air pollutant observations for Beijing provided a particular opportunity for this study.The present study is thus expected to stimulate the application of an integrated anthropogenic heat multimodal system for Beijing and provide a new insight into the improvement of model accuracy for air quality forecast over the megacity.In addition, this study provided a prototype for analyzing model uncertainties.The remainder of the paper is organized as follows.Section 2 describes the methodology, Section 3 presents the results and discussion, and Section 4 provides the conclusions and environmental implications.

Model Description
The NAQPMS is a fully modularized, three-dimensional chemical transport model for evaluating the effects of anthropogenic heat on the vertical distribution of pollutants (Wang et al., 2001).This chemical transport model reproduces the physical and chemical evolutions of reactive pollutants by solving the mass balance equation in a terrain following specific coordinates.The model incorporates advection and diffusion processes, gas/aqueous chemistry, and dry/wet deposition processes.More details on this model can be found in previous studies (Wang et al., 2002;Wu et al., 2010;Li et al., 2011a).The NAQPMS was coupled with the WRF model (V3.3), of which the parameterization schemes used to describe various physical processes are listed in Table 1.To appropriately present the urban island effect of anthropogenic heat, the advanced Noah (Ek et al., 2003) and land surface modules (LSM) (Chen and Dudhia, 2001) were integrated in the WRF model to generate a Noah-LSM module that can simulate the sensible and latent heat and temperatures of land surfaces in addition to the lower part of the boundary layer.The WRF model was also coupled with the UCM (Kusaka and Kimura, 2004) to adequately account for heat, momentum, and vapor exchanges.Parameters such as building shadows, the urban canopy, diurnal variation in solar altitude, long and short-wave radiation effects, urban canopy wind profiles, anthropogenic heat, and heat exchanger equations for multiple layers such as roofs, walls, and road surfaces were considered in the WRF model.
The modeling system was configured on four nested domains, as illustrated in Fig. 1.The first domain covered East Asia (centered at 110°E, 35°N) with a model resolution of 81 km and an 83 × 65 grid; the second domain covered North China with a resolution of 27 km and a 61 × 58 grid; the third domain covered the Beijing-Tianjin-Hebei region with a model resolution of 9 km and a 79 × 70 grid; and the fourth domain covered Beijing with a resolution of 3 km and a 73 × 64 grid.Our analysis focused on the third domain in comparison with observations.The modeling system was vertically divided into 28 layers, with 10 layers being within a depth of 1.5 km.NCEP-FNL data were employed as meteorological initial and boundary layer conditions (collected on 1° × 1° grids every 6 h).The emission inventory  used in the NAQPMS model was derived from the 2006 bottom-up regional emission inventory in Asia data with a resolution of 0.5° (Ohara et al., 2007) and an updated anthropogenic emission inventory in China (Wang et al., 2011).The simulation period was extended from 00:00 July 22 to 00:00 July 30, 2008 (Beijing Local Time), with the first 16 h as warm-up time.

Anthropogenic Heat Emission
Anthropogenic heat emissions data for Beijing used in the present study refer to the estimation provided by Tong et al. (2004).Anthropogenic heating can be divided into three components representing the major sources of waste heat in the urban environment, mainly from building energy consumption, vehicular tail-gas waste heat and industrial manufacturing energy.Anthropogenic heat emissions from building energy consumption were provided from the survey data in 2010 from China Academy of Building Research and the China heating association (80 W m -2 maximum generated in morning and evening).Vehicular exhaust anthropogenic heat emissions were estimated through vehicle fuel consumption in Beijing area, with a maximum peak of 104.3 W m -2 .There was no released survey for the industrial manufacturing energy waste.However, Tong et al. (2004) assumed the industrial manufacturing energy in Beijing to be half of that in Tokyo with a maximum peak of 59 W m -2 .In total, the maximum anthropogenic heat generated in the city was 170 W m -2 per day (base year 2000).The methodology employed for the emission inventory was standard and consistent with those previously used in Tokyo (Kimura and Takahashi., 1991) and Taiwan (Lin et al., 2008).
Anthropogenic heat emission data were integrated at a spatial resolution of 3 km with the innermost domain in this study.Two peaks (at 9:00 and 17:00) were observed in the diurnal cycle of anthropogenic heat emission over Beijing (Tong et al., 2004), similar to those found in megacities in the United States, Japan, and Korea (Ichinose et al., 1999;Sailor and Lu, 2004;Ryu et al., 2013a).The spatial distribution and hourly variation of anthropogenic heat are illustrated in Fig. 2.

Experimental Settings
Anthropogenic heat emission was integrated into the urban canyon through the canopy model coupled with the WRF model.In the WRF model, anthropogenic heat release was considered in the innermost domain.To thoroughly investigate the effects of anthropogenic heat emissions on the meteorological boundary layer structure and pollutant vertical distribution, two experiments were designed, namely Experiments A and B (Table 2).The same modeling system and land use data (moderate resolution imaging spectroradiometer, MODIS) were employed in Experiments A and B. However, anthropogenic emissions were integrated into only Experiment B. Comparing these two experiments enabled highlighting the effects of anthropogenic heat emissions.Statistical parameters such as mean bias (MB), normalized mean bias (NMB), normalized mean error (NME), and root mean square error (RMSE) were employed

Synoptic Pattern Analysis
Over the study period (July 22-30, 2008), Beijing, situated on the southeast border of the low-pressure system in Northern China, was continually affected by systematic southerly winds on the surface and was constantly under the influence of a warm tongue with weak advection at 850 hpa.No significant rainfall or cloud cover was observed; surface winds were weak, and the synoptic system remained stable.Such a synoptic condition, considered a crucial characteristic of the Beijing synoptic system, may significantly contribute to the formation of urban heat islands in summer (Wang et al., 2006).

Potential Positive Effects of Anthropogenic Heat Emission Integration Effect on Surface Wind and Temperature Patterns
Wind is the meteorological factor with the greatest impact on air pollutant transport and diffusion.As mentioned, successively running Experiments A and B enabled comparing modeling results and detecting the effects of anthropogenic heat on wind forecast through the analysis of associated key statistical parameters.Accordingly, three typical observation sites (Aoti, Chaoyang, and Guanxiangtai) were considered.The Aoti and Chaoyang stations are located in northern urban areas, whereas the Guanxiangtai site is situated in a southern rural area.A detailed statistical description of the effects of anthropogenic heat on the surface wind pattern at each station is presented in Table 3.The results revealed a 1 m s -1 increase in the simulated average wind speed at the two urban stations (Aoti and Chaoyang stations), when anthropogenic heat in Experiment B was compared with that in Experiment A. Conversely, a minor change was observed between Experiments A and B at the Guanxiangtai station (0.2 m s -1 increase in simulated wind speed), probably because of the limited effect of anthropogenic heat in rural areas.These findings are comparable to reported increments of 0.5-0.7 m s -1 over Shanghai (Xie et al., 2016).In addition, the increase in surface wind speed observed at the urban sites over Beijing was slightly higher at night than that in daytime, a finding that is similar to those reported in Shanghai (Xie et al., 2016).Such variations in the predicted wind speed could be attributed to the strengthened urban breeze circulation caused by enhanced anthropogenic heat fluxes (Chen et al., 2008;Ryu et al., 2013b;Yu et al., 2014).
Moreover, Experiment B yielded a significant reduction in the simulation bias.As illustrated in Fig. 3 and Table 3, a comparison between the observed and simulated wind speeds for Experiments A and B at the three stations revealed an evident improvement in wind speed prediction when anthropogenic heat was integrated in the modeling system.The NMB of the simulated wind speed in Experiment B was in general nearly 10%, whereas that in Experiment A reached  40%-50%.The notable persistent bias in the prediction in Experiment B might be partly due to the complexity of the geographical features of Beijing (surrounded by the Taihang Mountains), which restricts the skills of the modeling system in accurately reproducing the actual geopotential height (Jiménez et al., 2010;Zhang et al., 2013a).Furthermore, during the study period (July 22-30, 2008), Beijing was affected by onshore breeze from the Bohai Sea, mountain and valley breeze circulation caused by the Yan and Taihang Mountains, and regional heat island circulation (Beijing-Tianjin-Hebei), which enhanced the uncertainties in simulating the city's low-level atmospheric circulation (Liu et al., 2011).Besides Surface wind pattern, the surface and vertical temperature also presents improved simulation results induced by anthropogenic heat emission.Validation of predicted surface temperature for the two simulated experiments against observations, and statistical evaluation are shown in Fig. 4 and Table 4, respectively.In addition, the model validation with observed vertical temperature profile at the two typical moments is presented in Fig. 5.

Effects on the Boundary Layer Structure
As mentioned, the effects of anthropogenic heat on the boundary layer structure were assessed by analyzing the boundary layer height and atmospheric circulation.
In air quality modeling, the boundary layer height is one of the key factors affecting the vertical distribution and  spatial-temporal variations of pollutants.Thus, analyzing the effect of anthropogenic heat emissions on the boundary layer height is crucial in air quality forecast.In this study, observation data of the boundary layer height were provided by the aerosol lidar set at the Institute of Atmospheric Physics (IAP), Chinese Academy Sciences, located in a typical urban area between the North 3rd and 4th Ring Roads.The boundary layer height was retrieved using the cubic root gradient method proposed by Yang et al. (2017).The diurnal cycle of the boundary layer height on three typical polluted days (July 24, 27, and 28, 2008) is presented in Fig. 6, revealing a valley in the boundary layer height at 07:00-08:00, with a drastic increase by 200-300 m, and a peak value at 16:00-17:00.Such an increase appears to be higher than that reported for Shanghai (160-m increase in boundary layer height in July; Xie et al., 2016), plausibly because of the larger amount of anthropogenic heat emission in Beijing.Basically, the variation in the height was caused by the enhancement of vertical air movement in the boundary layer induced by the warming up of surface air temperature (Fan and Sailor, 2005;Ferguson and Woodbury, 2007;Bohnenstengel et al., 2014).The effects of anthropogenic heat are actually more pronounced during the daytime due to the high intensity of human activities.However, in the absence of the influence of solar radiation during nighttime, anthropogenic heat emission plays a potentially singular role by triggering thermodynamic motion within the boundary layer to maintain the increase in the height (Fan and Sailor, 2005).Further comparing the observed results with the boundary layer height simulated in Experiments A and B (Fig. 6) indicated substantially improved consistency and agreement of the results of Experiment B with the observed diurnal cycle pattern and valley and peak values.Such an improvement in the predicted boundary layer height was also statistically illustrated by the significant reduction in the simulation bias, as indicated by an analysis of the RMSE (significant reduction from 361 to 558 m) and correlation coefficient (slight increase from 0.83 to 0.85) as illustrated in Table 5.Clearly, similar to the aforementioned effects on wind speed, integrating anthropogenic heat in the modeling system could induce an increase in the boundary layer height and result in a more accurate prediction.
Anthropogenic heat emission could lead to further changes in atmospheric circulation in the boundary layer through an increase in the sensible heat and surface temperature (Lin et al., 2008;Feng et al., 2012).To elucidate the effects of anthropogenic heat emissions on the variations in the boundary layer structure, a cross-sectional temperature difference caused by anthropogenic heat in downtown Beijing (116°23′17″E, 39°54′27″N) was captured, as illustrated in Fig. 7; the temperature difference between "with" and "without" anthropogenic heat is illustrated by the shaded color and the boundary layer height by the white line.Warming caused by anthropogenic heat emissions was observed to mostly occur in the medium-and lowlevel parts of the boundary layer, particularly evident at night and in the early morning, with a maximum temperature increase of 3°C.This increase illustrates the prominence of the singular effect of anthropogenic heat during these two periods (night and early morning), because of the absence of solar short-wave radiation as mentioned (only anthropogenic heat may significantly induce the progressive warming of the boundary layer under weak solar short-wave radiation conditions).The present finding is comparable to the temperature increase values reported worldwide for megacities (Fan and Sailor, 2005;Ferguson and Woodbury, 2007;Chen et al., 2008;Menberg et al., 2013).However, notably, anthropogenic heat appeared to indirectly induce a reduction in the temperature at the top of the boundary layer in the afternoon, with a maximum reduction of 2°C (Fig. 7).Such a situation may be attributed to two complex processes.First, warm air ascending from the land surface may have been adiabatically cooled at the top of the boundary layer because of the possible reduction in atmospheric pressure in the neighboring areas.On the other hand, air mass converging at the top of the boundary layer (Fig. 8(b)) may have driven high-level cold air toward the boundary layer around noon (when the height was at its maximum and the boundary layer was critically exposed to descending cold air from a high altitude).Such a decreasing tendency of the temperature at the top of the boundary layer, as highlighted by the modeling system, appears to be consistent with a phenomenon previously reported by Shahgedanova et al. (1997) during an observation campaign in Moscow.Indirectly induced air mass cooling and descent at the top of the boundary layer could be considered as additional key effects of anthropogenic heat on the boundary  0.85 M 0 denotes the mean observed value; M m denotes the mean simulated value.layer structure and might significantly affect the transport, diffusion, and pollutant exchanges inside and outside the boundary layer.
A schematic is presented in Fig. 9 to recapitulate the effects of anthropogenic heat on the atmospheric circulation in the boundary layer.In brief, anthropogenic heat emissions increase the surface temperature in the urban area and lead to ascending air flow on the surface layer.Such warm air enhances the increase in temperature in low layers of the atmosphere and rises up to the top of the boundary layer, where it diverges toward neighboring areas and subsequently descends with a significant reduction in temperature.In addition, cold air on the surface of suburban areas moves toward urban areas to compensate the urban air mass loss.A converging cooled air mass can be observed on the top of boundary layer (Fig. 8(b)).

Effect on Surface and Vertical Distribution of Pollutants
As mentioned, by affecting the wind field and boundary layer structure (height and atmospheric circulation), anthropogenic heat could certainly induce changes in the distribution of air pollutants.For convenience and availability of pollutant observation data, this study considered two typical observation sites (Beida and Changping) for the analysis of pollutant surface distribution.The Beida site is located in an urban area, whereas Changping is a rural site situated in the northern area of Beijing.We analyzed PM 10 , which is considered a typical pollutant over Beijing.A detailed description of the modeled anthropogenic heat effects on PM 10 concentration variations (based on Experiments A and B) is presented in Fig. 10.The modeling system exhibited an obvious overestimation of PM 10 concentration (without integration of anthropogenic heat emissions) at the Beida station (urban area) compared with the trend at the Changping site (rural area site).Integrating anthropogenic heat emissions in the model certainly induced an elevation of boundary layer heights, increased the surface pollutant distribution space, and significantly reduced the discrepancy  between the observed and modeled surface PM 10 concentrations.The effects of anthropogenic heat on the PM 10 distribution were more conducive in the urban area probably because the high intensity of human activities in urban Beijing is associated with a stronger impact on the boundary layer height than that in the rural area.Accordingly, the RMSE was significantly reduced by 10-23 µg m -3 , and the correlation coefficient increased from 0.45-0.60 to 0.70-0.76over the Beida and Changping sites, respectively (Table 6).Clearly, integrating anthropogenic heat in the modeling system substantially improved the prediction of the surface distribution of air pollutants, supporting the finding reported by Ryu et al. (2013b) for the Seoul metropolitan area.
Regarding the analysis of the vertical distribution of pollutants, the present study did not obtain valid PM 10 observations of vertical distribution because of the limited observation campaign.However, the vertical distributions of various pollutants share some similarity; thus, the effects of anthropogenic heat on the ozone vertical distribution were investigated instead of those on PM 10 .Valid observations of ozone at the IAP station were used for comparison with the model results.and an increase in the correlation coefficient from 0.63 to 0.65.Experiment A yielded periods of O 3 concentration at an altitude of 280 m as well as at the ground level, whereas Experiment B reproduced a more realistic and accurate prediction compared with observations (Fig. 11(b)).This improvement in ozone prediction was also supported by a reduction in RMSE from 34 to 30 ppb and an increase in the correlation coefficient from 0.57 to 0.60 (Table 7).In brief, integrating anthropogenic heat emissions in the model enabled a significant correction of the prediction bias for both surface and vertical distributions of pollutants.

CONCLUSIONS AND ENVIRONMENTAL IMPLICATIONS
The considerably large anthropogenic heat emissions generated daily from intensive human activities in Beijing (more than 20 million habitants) are often neglected in the design of air quality forecasting systems, which critically enhances model uncertainties and prediction bias over the megacity.Therefore, the present study proposes an anthropogenic heat-integrated multimodal system for evaluating the effects of anthropogenic heat emissions on  -The anthropogenic heat effects were more pronounced and particularly obvious in urban areas than in rural areas, because urban areas have much more intensive human activities than rural areas.The effects were evident during both daytime and nighttime; however, a more prominently singular role was observed at night in the absence of solar shot-wave radiation.These effects on the boundary layer structure exhibited significant variability at different levels.
The results reveal a warming effect in the low and middle parts of the boundary layer, which was most notable at night, with a maximum temperature increase of 3°C; by contrast, a cooling effect was observed at the top of the boundary layer, with a maximum temperature decrease of 2°C.-Anthropogenic heat was determined to directly impact atmospheric circulation over the urban area by increasing the surface temperature and inducing the ascent of a warm air mass from the low parts of the atmosphere to the top of the boundary layer, thus significantly changing the tropospheric structure; however, it was also observed to indirectly drive wind convergence and the descent of a cooled air mass from a high altitude into the boundary layer through a complex atmospheric circulation process.This complexity significantly affected all the key parameters with evident correction of model predictions.-Integrating anthropogenic heat emissions in the modeling system was effective in improving predictions by reducing the NMB by 20%-30% (for wind speed) and RMSE by 361-558 m (for boundary layer height) and by 10-23 µg m -3 (for surface PM 10 ), in addition to significantly reducing the underestimation of ozone column by approximately 20 ppb at urban sites.
In terms of environmental implication, such an innovation is expected to technically contribute to the improvement in the accuracy of regional air quality modeling systems for pollution forecasts and control measures, particularly over numerous megacities in China.However, because of the limitations of observations, this study focused only on the effects of anthropogenic heat during a specific period.The seasonal variability of such effects on the vertical structures of various pollutants should be explored in future investigations.

Fig. 2 .
Fig. 2. (a) The Spatial distribution of anthropogenic heat at 7:00 am in Beijing area, (b) the diurnal cycle of the anthropogenic heat induced by building energy consumption, vehicle exhaust, industrial manufacturing energy waste and total anthropogenic heat in Beijing.
Fig. 11 presents a comparison between modeled (Experiments A and B) and observed ozone concentrations at the ground level and at an altitude of 280 m over the IAP site.As illustrated in Fig. 11(a), multiple periods of zero O 3 concentration at the ground level (marked by gray boxes) were simulated in Experiment A, whereas no zero concentration periods were identified in Experiment B, which exhibited good consistency with the observed ozone levels.Neglecting anthropogenic heat could therefore lead to an overestimation of nighttime titration effects (i.e., falling O 3 concentrations).Statistically, Experiment B was associated with a reduction in RMSE from 40 to 34 ppb

Fig. 11 .
Fig. 11.Comparison between model simulated and observed ozone concentrations for the IAP site on July 23-29, 2008 at the (a) surface and (b) at 280 m.

Table 1 .
Physical Processes and Parameterization Schemes.

Table 2 .
Experiments settings for the two simulations.
SimulationMeteorological Model Land use data Air pollution Model Consideration of anthropogenic heat

Table 3 .
Statistical parameters of wind speed during 23-29 th July 2008.
m denotes the mean simulated value.

Table 6 .
Statistical parameters of PM 10 during 23-29 th July 2008.the mean observed value; M m denotes the mean simulated value.