The metropolitan region of Hangzhou (30°7′N−30°24′N, 120°2′E−120°22′E) is located at the south of the Yangtze River Delta (see Figure 1a), covering a surface area of 1,600 km² with an urbanization rate of 78.5% at the end of 2019. The landscape in this region is quite complex, which comprises the world-famous West Lake at the center of the region, the Qiantang River across the region from west to east, the urban area of downtown Hangzhou on the east and north side of West Lake and the subcenter of the city (Bingjiang and Xiaoshan) on the south side of the Qiantang River, the forest area of the west-lake scenic area on the southwest side of the West Lake, the Xixi wetland in the west of the region, and the Qiaosi farmland in the east of the region (see Figure 1b).

The default land use data (MODIS 20-category) with one urban land category (Figure 2A) used in WRF were obtained in 2004, which is outdated and cannot represent the inhomogeneity in urban areas of Hangzhou (Chen et al., 2014). In this study, the urban and rural settlement data from the Global Urban Footprint (GUF) project (https://www.dlr.de/eoc/en/desktopdefault.aspx/tabid-11725/20508_read-47944/) were used to generate urban land use data of 100-m resolution (Figure 2B). The population density data at 100-m spatial resolution (Ye et al., 2019) were utilized to specify detailed urban land use categories (low-intensity residential with population density <10 people/ha, high-intensity residential with population density between 10 and 100 people/ha, and commercial with population density larger than 100 people/ha).

The UCPs are calculated based on vector-format building data obtained from Hangzhou Surveying and Mapping Bureau, which is a vector-format data containing polygon feature classes existing in a geodatabase (Figure 1b). The high-intensity residential and commercial are mainly distributed in the downtown Hangzhou on the east side of West Lake and the Binjiang/Xiaoshan District on the south bank of the Qiantang River. The basic geographic information of buildings can be derived from a series of attributes (e.g., the floors and footprint outline of the building, the area and the perimeter of the building base, etc.) of these polygon feature classes. The UCPs are then calculated according to building information and used in WRF/urban modeling system to represent the geometrical characteristics of urban morphology.

The gridded (resolution at 100 m and 1 km) UCPs are derived by using the ArcGIS-embedded algorithms (Yu et al., 2018), and seven parameters are actually used to calculate WRF/urban morphology: (1) mean building height, (2) distribution of building heights, (3) area weighted mean building height, (4) standard deviation of building height, (5) plan area fraction, (6) building surface to plan area ratio, and (7) frontal area index. The calculation of these parameters is listed in Table 1, and the results are shown in Figure 3. Based on these parameters, the geometrical urban canyon parameters were estimated following the formulations in Macdonald et al. (1998). Previous studies suggested that the incorporation of gridded anthropogenic heat release (AHR) data can improve the modeling skill of WRF/urban system in Hangzhou (Chen et al., 2016; Yang et al., 2019). In this study, a newly developed gridded AHR data from Chen et al. (2020) were used.

The Weather Forecast and Research urban model version 4.0.2 (Chen et al., 2011; Skamarock et al., 2019) is used in this study, and the forecasts are produced on four one-way nested computational domains (Figure 1a) with the grid spacing (grid numbers) of 9 km (580 × 637), 3 km (637 × 577), 1 km (502 × 502), 200 m (211 × 211), respectively. The coarsest domain (d01) comprises a major part of Asia, while the most inner domain (d04) covers the metropolitan region of Hangzhou. The model employs 51 vertical terrain-following hybrid sigma-pressure layers from the surface up to a pressure level of 50 hPa. The RRTMG shortwave/longwave radiation scheme (Iacono et al., 2008), the Kain–Fritsch cumulus scheme (only used in d01) (Kain and Kain, 2004), the WSM6 microphysics scheme (Lim and Hong, 2010), the Bougeault and Lacarrere (1989) TKE PBL scheme, the Revised MM5 Monin–Obukhov surface-layer scheme (Jiménez et al., 2012), and the Noah-MP land-surface scheme (Niu et al., 2011) coupled with SLUCM/BEP are used in this study. Several configurations are made for the adaptation of the urban finescale forecasting, including initializing the inner temperature of the buildings from the observation site (307 K in this case), choosing the Smagorinsky first-order closure (3D) option in the finescale grids (d03 and d04), and the horizontal Smagorinsky first-order closure option in coarse grids (d01 and d02). With this methodology, the largest (flux carrying) eddies are expected to be resolved by the modeling system in the finescale grids (Talbot et al., 2012), and the vertical turbulent diffusion is parameterized according the PBL scheme in the coarse grids. The initial and boundary condition were derived from the ERA-Interim reanalysis data at a horizontal resolution about 79 km (Dee et al., 2011).

A typical heat wave with all-sky clear days from 0000 UTC July 21 to 0000 UTC July 25, 2016 was selected for this study. This heat wave was caused by the large-scale synoptic condition of the strong west Pacific subtropical high system over Hangzhou (Xu et al., 2009; Peng, 2014). Under the control of such a strong subtropical high, a calm weather with low wind speeds and humidity, as well as limited cloud and no precipitation, are long time continued, which led to a heat wave event in Hangzhou (Lin et al., 2013; Quan and He, 2016).

As show in Table 2, seven numerical experiments were conducted and divided into two groups. Group I contains four runs. The SDNN and BDNN runs with the default MODIS 20-category data are called the “default” cases, while the SNUA and BNUA runs with the updated land use data and high-resolution UCPs is called the “new” cases. The contrast of the “new” cases and the “default” cases were conducted to investigate the improvements of urban finescale forecasting by using high-resolution UCPs. The other five experiments (SDNN, SNNN, SNNA, SNUN, and SNUA runs) in Group 2 with different settings of land use datasets, UCPs, and AHR were compared to investigate the mechanism of these factors on the forecasting of this heat wave event. All these experiments were conducted from 0000 UTC July 21 to 0000 UTC July 25, 2016, and the last 3 days were used to evaluate the performance of the integrated WRF/urban modeling system.

To evaluate the improvement of the WRF/urban finescale forecasting system, in two category results of simulations with (“new” cases)/without (“default” cases), the high-resolution UCPs were compared against the observations in Table 3. The observational data are from the Automatic Weather Stations (AWS) maintained by Zhejiang Meteorological Bureau of China Meteorological Administration, including hourly 2-m temperature, 2-m relative humidity, and 10-m wind speed during the heat wave event. Overall, the “new” cases (SNUA and BNUA) yield better results than the “default” cases (SDNN and BDNN) (Table 3). Either the SLUCM or the BEP model with the high-resolution UCPs produced smaller BIAS and RMSE than that without the high-resolution UCPs, especially in the urban area.

Figure 4 shows the comparison of the averaged daily mean temperature at 2 m in the metropolitan region of Hangzhou during 0000 UTC July 22, 2016 to 0000 UTC July 25, 2016. The observed temperature was well consistent with the urban area illustrated in Figures 1b, 2B, which shows the development of UHI effect in downtown Hangzhou, Bingjiang, and Xiaoshan District (Figure 4E). It is shown that the “default” cases underestimated the near-surface temperature and with the position slightly northeastward in Figures 4A,C for the SLUCM and BEP models, respectively. These underestimation and position shift were reduced when the updated land use data and UCPs were adopted in SLUCM and BEP models, which reproduced the UHI center located in a similar region as observed (Figures 4B,D). Further statistical analysis also shows that the SLUCM/BEP model with the default land use data underestimates the near-surface temperature with the RMSE of 0.85/0.69, 1.31/1.07, and 0.96/1.18°C for the daily mean, minimum, and maximum temperatures, respectively, and these RMSEs reduced to 0.80/0.65, 1.20/0.96, and 0.63/0.85°C when using the high-resolution UCPs. The spatial correlation coefficients were also improved, especially in downtown Hangzhou.

The simulated and observed daily mean relative humidity at 2 m are compared in Figure 5. The observed “dry island” appears mainly in the downtown Hangzhou (Figure 5E), which is also reported in cities in East China in previous studies (Hao et al., 2018; Luo and Lau, 2019). Similar to the 2-m temperature, the relative humidity at 2 m simulated by the “default” cases (SDNN and BDNN runs) had a northeastward shift against the observation, which was corrected in the “new” cases (SNUA and BNUA runs). The underestimate of the “dry island” in the “default” cases were also reduced in the “new” cases, especially by the BEP model.

Under the low wind speed condition, all the cases overestimated the wind speed (Figure 6), especially by the SLUCM model (i.e., SDNN and SNUA runs), which cannot reproduce the slowing of wind speeds in urban areas due to building barrier effects (Kristovich et al., 2019). The “new” cases with the high-resolution UCPs reduced these overestimations markedly. For BNUA vs. BDNN run, the BIAS/RMSE of 10-m wind speed significantly decreased from 0.98/1.06 to 0.73/0.88 m s⁻¹, and the correlation coefficient between simulation and observations increased from 0.04 to 0.13.

To comprehensively evaluate the improvements of the urban finescale forecasting by using the high-resolution UCPs, the BIAS, RMSE, and spatial correlation coefficient for the simulations in the whole domain and each urban center were painted in the Taylor diagrams (Figure 7). It can be found that the correlation coefficients and the RMSE for the “new” cases (blue for SNUA and black for BNUA) improved markedly when compared with the “default” cases (red for SDNN and green for BDNN). For the daily mean temperature at 2 m, the blue/black points were located closer to the x-axis and the zero point than the red/green points, which means better performance for the “new” cases.

Figure 8 shows the simulated and observed diurnal variation of the 2-m temperature, 2-m relative humidity, and 10-m wind speed. During daytime (from 0700 BJT to 1700 BJT), all the simulations overestimated the near-surface temperature, especially for the BEP model. During nighttime (from 1800 BJT to 0500 BJT), the BNUA case shows the best performance, while the other three cases underestimated the 2-m temperature. In general, compared with the “default” cases, the “new” cases show improvement in simulating near-surface temperature during nighttime, which may benefit from better representation of nighttime UHI due to the application of high-resolution UCPs in the UCM models. Due to the complex building environment, the urban surface traps more shortwave radiation, which is kept in the urban canopy in the daytime, and releases more heat to the overlayer air at night, which caused more obvious UHI effect. Accompanied by the more obvious UHI effect, the lower 2-m relative humidity was derived from the “new” cases, which are closer to the observations than those from the “default” cases, especially at night (Figure 8B). As show in Figure 8C, there is no obvious diurnal change in the observed wind speed. The “default” cases overestimated the wind speed, especially by the SLUCM model in SDNN run. The “new” cases reduced this error markedly, which is probably associated with the better description of the windward coefficient in the urban area by using the frontal area density parameters, especially for BEP model.

The RMSE of WRF forecasts derived from SNUA run at different resolutions are compared in Figure 9. It is worth noting that the RMSE decreased when refining the computation domain, particularly for the finest domains with grid sizes of 200 m. At such a fine resolution, the detail of the buildings can almost be solved by the model, which represents the geometrical characteristics of urban morphology well by the given UCPs and achieves better results.

Five different urbanization scenarios were designed to estimate the relative effects of updating the land use using the high-resolution UCPs and considering the AHR on the UHI intensity. The SNUA run reproduced the UHI in the downtown Hangzhou with the maximum intensity at about 1.2°C, which matched the observation quite well (Figures 10A,B). The difference between the SNUA and SNUN runs indicate the effect of considering the AHR on the UHI intensity (Figure 10C). The difference between the SNUA and SNNA runs represent the effect of using the UCPs on the UHI intensity (Figure 10D). The combined effects of considering the UCPs and AHR could be obtained from the difference between the SNUA and SNNN runs (Figure 10E). The differences between SNUA and SDNN runs could be attributable to the composite effect of updating the land use using the high-resolution UCPs and considering the AHR on the UHI intensity (Figure 10F). The simulations suggested that the contribution of considering the AHR, using the high-resolution UCPs, and their combination to the UHI intensity average in the downtown Hangzhou was about 0.01, 0.3, and 0.5°C, respectively, while another 0.5°C is contributed from the updating of land use data. These numbers indicated that the land use change has the largest impact, but using UCPs is a non-negligible factor in mesoscale simulations of urban climate.

Figure 11 shows the differences of impervious fraction, zero plane displacement height, and roughness length above canyon between the SNUA run and the SNNA/SDNN run. It can be found that the real impervious fraction from the new updating land use data was not so large as the corresponding values in the look-up tables (Figures 9A,B). Moreover, the roughness length and zero plane displacement height increased notably when using the UCPs in the simulation. This can be explained by the complex building data that represents more reality urban environment, which produced larger roughness than the homogeneous surface in the simulations without the UCPs.

Accompanied by the increase in roughness, more shortwave radiation was trapped by the urban canopy and the ground in the daytime, and the surface skin temperature (TSK) increased clearly in the urban area (Figures 12A1–D1). At night, although the latent heat flux did not change much (Figures 12A3–D3), the sensible heat flux increased obviously (Figures 12A2-D2), and more heat flux was received from the ground in the urban area (Figures 12A4–D4). These additional energies are heating the near surface and, thus, resulting in a notable UHI effect (Figures 12A5–D5). These additional energies are heating the near surface and, thus, resulting in a notable UHI effect (Figures 12A4–D4). It is notable that the high-resolution UCPs played a significant role in affecting the UHI effect than the AHR, and it can be comparable with the effect of land use change.