Urban Impact on Landfalling Tropical Cyclone Precipitation: A Numerical Study of Typhoon Rumbia (2018)

Landfalling tropical cyclones (TCs) expose densely populated and highly developed coastal urban agglomerations to severe natural hazards such as strong wind, extreme precipitation, and pluvial flood (Zhu et al., 2021; Li and Zhao, 2022). For example, Hurricane Harvey, which made landfall in Texas in 2017, impacted Houston with record-breaking rainfall and flooding, causing over 80 fatalities and $125 billion in losses (Wang et al., 2018b). Studies have suggested that the urbanization in Houston exacerbated the hurricane-induced extreme flooding (van Oldenborgh et al., 2017; Zhang et al., 2018). Meanwhile, the rapidly urbanizing Yangtze River Delta (YRD) and Pearl River Delta (PRD) in China are prone to landfalling TCs from the western North Pacific, which are more frequent and stronger compared with those in Atlantic coastal areas (Mei and Xie, 2016; Liu and Wang, 2020; Liu and Chan, 2022). In terms of socio-economic impacts, the urban agglomerations of the YRD and PRD have suffered tremendously from landfalling TCs, such as the severe Typhoon In-Fa (2021), super Typhoon Lekima (2019), and super Typhoon Mangkhut (2018). Thus, landfalling TCs and the related natural disasters affecting coastal urban areas warrant urgent attention and in-depth studies (Yang et al., 2020).

Land surface processes are a pivotal factor influencing landfalling TCs and the related rainfall. After TCs make landfall, the rougher land surface acts as a momentum sink and increases surface friction, thus reducing surface wind speed and weakening the low-level circulation (Li et al., 2014; Chen and Chavas, 2020; Hlywiak and Nolan, 2021). Meanwhile, since the thermal inertia is lower over land than the ocean, the post-landfall surface enthalpy flux is reduced. Consequently, a post-landfall TC, which can be viewed as a Carnot heat engine balanced by surface enthalpy flux and surface frictional dissipation, can hardly be supported (Emanuel, 2007; Wang, 2012). Therefore, TCs usually decay quickly after landfall (Tuleya, 1994; Shen et al., 2002). However, in some cases, the response in landfalling TC precipitation to the land surface process can be quite complicated (Deng and Ritchie, 2020). The aforementioned surface friction and surface heat flux processes may cause divergent landfalling TC precipitation responses on different timescales. On a long timescale, the strong surface friction and reduced surface heat flux make TC rainfall decay with TC intensity. However, within 12 hours after landfall, TC rainfall can be transiently strengthened in reaction to the sudden change in surface roughness and frictional convergence (Chen and Chavas, 2020; Hlywiak and Nolan, 2021). Furthermore, case studies have shown that for some TCs, the post-landfall rainfall was maintained or re-intensified over land by a sufficient surface heat flux from warm and wet soil, which is referred to as the "Brown Ocean Effect" (Emanuel et al., 2008; Andersen and Shepherd, 2014; Wakefield et al., 2021). This phenomenon is usually caused by a positive soil moisture anomaly associated with an antecedent precipitation anomaly or special soil texture (Arndt et al., 2009; Evans et al., 2011; Galarneau and Zeng, 2020). The warm and wet land surface can promote low-level instability and provide water vapor to support the inner-core deep convection and circulation of a landfalling TC (Yoo et al., 2020). Liu et al. (2019) found that surface evaporation contributed about 15%–20% to the post-landfall rainfall in the inner core of Typhoon Utor (2013).

The aforementioned studies focus on the impacts of large-scale land surface properties on landfalling TC precipitation but oversimplify the spatial patterns of land use and land cover. In reality, the underlying surface with spatial heterogeneity can generate complicated dynamic and thermodynamic forcings on landfalling TC rainfall. One example is that strong rainfall can occur over windward slopes when landfalling TCs approach mountainous areas, known as the so-called “Phase Lock by Terrain” (Wu and Kuo, 1999; Wu et al., 2002; Fang et al., 2011). Several researchers have found that inland lakes, wetlands, and other saturated surfaces can fuel landfalling TC rainfall by providing moisture and instability in the lower troposphere (Li and Chen, 2007; Wei and Li, 2013; Nair et al., 2019).

The urban surface is a critical component of the land surface with its unique dynamic and thermodynamic properties that can tremendously regulate convective precipitation, although its spatial scale is relatively small compared with wetlands, forests, and lakes (Liang et al., 2018; Zhang, 2020; Qian et al., 2022). For instance, the stronger surface friction over urban areas can decelerate the low-level wind speed and promote convergence to modify the spatiotemporal evolution of convection (Yu et al., 2018; Li et al., 2021; Yang et al., 2021). On the other hand, the differences in albedo, conductivity, and other thermodynamic properties between urban and natural surfaces can contribute to the urban-rural contrast in surface temperature known as the Urban Heat Island (UHI) (Han and Baik, 2008). Excessive sensible heat flux and anthropogenic heat flux emitted by cities can render the planetary boundary layer (PBL) unstable, which favors convection initiation (Holst et al., 2016; Nie et al., 2017). Conversely, the impervious urban surface militates against evaporation, inhibiting low-level convective available potential energy and moisture (Wang et al., 2018a). Additionally, the interaction between UHI circulations and land-sea breezes has been considered an important trigger for extreme rainfall events in coastal cities (Yin et al., 2020; Gao et al., 2021; Sun et al., 2021).

Despite urban vulnerability to the extreme precipitation and flooding brought about by landfalling TCs, current research has paid little attention to how landfalling TC precipitation responds to the urban impact. Kimball (2008) analyzed urban impacts on TC rainfall regardless of the effect of land cover heterogeneity. Recent case studies by Zhang et al. (2018) and Ao et al. (2022) highlighted the important role of urban friction by comparing low-level kinematic features. However, the physical processes governing TC rainfall are quite complicated, and the contributions from urban dynamic and thermodynamic forcings need to be quantified to identify the dominant physical mechanisms for landfalling TC precipitation evolution.

Typhoon Rumbia (2018) was selected for this study as it caused the largest loss among the four landfalling TCs affecting the YRD urban agglomeration in 2018, with its track directly crossing this area. It made landfall south of Shanghai at 0400 Local Standard Time (LST; UTC + 8 h) on 17 August with a maximum sustained surface wind speed of 23 m s^–1 and a minimum sea-level pressure of 982 hPa. Many prior studies focused on the precipitation activity of intense landfalling TCs, e.g., Hurricane Harvey (2017) with a maximum wind speed of over 60 m s^–1 in Zhang et al. (2018) and Typhoon Lekima (2018) with a maximum wind speed of about 52 m s^–1 in Ao et al. (2022). Despite being only a severe tropical storm, Rumbia (2018) still induced severe flooding and large socio-economic losses (Lei et al., 2020). The objective of this study is to explore whether the urban impact can also contribute to the occurrence of extreme precipitation associated with a weak landfalling TC. Furthermore, this paper aims to disentangle the contributions from different physical processes and reveal the dominant mechanism using physical models of TC precipitation. The remainder of this paper is organized as follows. Section 2 describes the numerical model, the experimental settings, and the physical models of TC precipitation. The simulation results, diagnosis, and specific physical processes are analyzed in section 3. Section 4 discusses the potential urban thermodynamic forcing on TC rainfall. Finally, the findings are summarized in section 5.

The best track data at 3-h intervals from the Shanghai Typhoon Institute of China Meteorological Administration were used as the observed TC track and intensity data (Ying et al., 2014). The observed precipitation dataset was the hourly 0.1° × 0.1° merged precipitation product, which was generated by merging gauge observations from more than 30,000 automatic weather stations in China and the Climate Precipitation Center Morphing precipitation data with the probability density function–optimal interpolation methods (Shen et al., 2014).

The Global Forecast System Final Analysis (FNL) data of the National Centers for Environmental Prediction with a horizontal grid spacing of 0.25° × 0.25° at a 6-h interval were used as the initial and lateral boundary conditions of the numerical simulation. The FNL reanalysis data were also used to verify the WRF simulations.

To reflect the spatial distribution of land use and land cover more accurately, the land-use data at a 1-km resolution in 2018 from the Resources and Environment Scientific Data Center (RESDC), Chinese Academy of Sciences (http://www.resdc.cn/data.aspx), which are based on Landsat TM/ETM satellite data, were used to update the spatial distribution of the YRD urban agglomeration.

Typhoon Rumbia (2018) was simulated from 1400 LST 15 August to 2000 LST 17 August by the Advanced Research version of the Weather Research and Forecasting Model (WRF-ARW), version 4.0 with three nested two-way interactive domains (d01, d02, and d03), and the first 12 hours were treated as the spin-up period. The horizontal grid resolutions for each domain were set at 18 km, 6 km, and 2 km, respectively, with the grid dimensions of 210 × 205, 406 × 385, and 730 × 703 (Fig. 1a). There were 55 vertical levels with the model top at 50 hPa and the model bottom at about 26 m, and 15 of these levels were within the lowest 1-km height. The physical parameterization schemes used in this study are shown in Table 1. The Kain–Fritsch cumulus convection parameterization was only applied in d01 and the time step for d01 was 90 seconds.

Figure 1.  (a) Model nested domains and terrain height (shaded; units: m). The outermost box, the black box, and the white box denote domains d01, d02, and d03, respectively. (b-c) Land use and land cover in CTL and No_Urban. Numbers 12/13 represent cropland/urban areas. The locations of cities in the YRD, including Shanghai, Suzhou, Wuxi, Changzhou, Nanjing, Hangzhou, and Ningbo, are marked as SH, SZ, WX, CZ, NJ, HZ, and NB in (b), respectively.

Two sets of experiments were conducted to analyze the urban impact on the rainfall related to the landfalling TC. In the control experiment (CTL), a single-layer Urban Canopy Model (SLUCM) that considers anthropogenic heat was used to represent the urban physical processes (Kusaka et al., 2001; Kusaka and Kimura, 2004; Chen et al., 2011). The SLUCM has been widely used and has performed well in simulating various convective events, including landfalling TCs. The SLUCM considers a two-dimensional symmetrical street canyon with infinite length and estimates heat and momentum exchanges between the atmosphere and urban road, wall, and roof surfaces. Multiple urban canopy parameters, such as building height, heat capacity, thermal conductivity, and anthropogenic heat, can be defined in the model to realistically characterize urban processes. In our simulations, the SLUCM was coupled with the Noah-MP Land Surface Model. The land-use map was from the Moderate Resolution Imaging Spectroradiometer (MODIS). The distributions of urban areas in YRD were updated according to the urban distributions in the RESDC land-use data mentioned in section 2.1 (Fig. 1b). A sensitivity experiment (No_Urban) was conducted without the SLUCM and the urban areas in YRD were replaced by croplands to exclude any potential urban impacts (Fig. 1c). In the current study, the default urban canopy parameters were used, and the urban type was simply treated as the default High Intensity Residential due to the limited urban data, but the possible urban impacts on TC precipitation could still be analyzed by comparing CTL and the sensitivity experiments.

Following Liu et al. (2019), six ensemble runs were conducted for each experiment with a perturbed initial water vapor mixing ratio in the innermost mesh to reduce the uncertainty from internal variability on the physical sensitivity. The water vapor mixing ratio at the lowest model level was increased or reduced by 3‰, 2‰, and 1‰ relative to the corresponding reference run, respectively. Therefore, there were seven members in each experiment. All results discussed below are based on the ensemble mean for each experiment.

The physical models of TC precipitation used in this study are based on the assumption that the TC precipitation rate ($P_{{\rm{rate}}}$) can be simplified as a function of upward vapor flux:

where ${\varepsilon _{\rm{p}}}$ is precipitation efficiency; ${q_{\rm{s}}}$ is saturated specific humidity; ${\rho _{{\rm{air}}}}$ and ${\rho _{{\rm{liquid}}}}$ are the density of water vapor and liquid water, respectively, with their ratio set to 0.0012; $ w $ is the vertical velocity. The saturated specific humidity ${q_{\rm{s}}}$ is quite close to specific humidity $ q $ in terms of TC. In studies on TC precipitation and its associated physical mechanisms, Eq. (1) has been widely applied (Langousis and Veneziano, 2009; Lu et al., 2018; Chen and Chavas, 2020), and it can be further simplified as:

The relationship defined in Eq. (2) represents the dominant mechanism of heavy precipitation—vertical moisture advection—while neglecting the negligible contribution from local evaporation and a relatively small contribution from change in the total column atmospheric water content (Wilson and Toumi, 2005). Lu et al. (2018) and Langousis and Veneziano (2009) analyzed the relationship between TC precipitation rate and upward vapor flux and found that the highest correlation is obtained at a height of 3 km (about 700 hPa). Thus, the upward vapor flux was calculated at the 3-km level in this paper.

Equation (2) can be linearly decomposed into Eq. (3) to quantify the differences in precipitation, dynamical process, and moist process between the two sets of experiments:

where $ \Delta $ denotes the differences between CTL and No_Urban and $\mathop {\;}\limits^ -$ denotes the area mean.

Equation (3) can be further normalized into:

where the first term on the right-hand side (RHS) is the normalized linear response in $ \overline P $ due to changes in the dynamic component $ \overline w $, the second term on the RHS represents the normalized linear response in $ \overline P $ due to changes in the moist component $ \overline q $, and the final term is the nonlinear residual. Equation (4) is a simple but effective way to quantify the contributions of dynamic and moist processes to TC rainfall (Chen and Chavas, 2020).

The Tropical Cyclone Rainfall Model (TCRM), a recently developed model to quantify the contributions of different physical mechanisms (Zhu et al., 2013; Lu et al., 2018), has been proven to be effective in estimating TC rainfall and elucidating the dominant process (Feldmann et al., 2019; Gori et al., 2020; Xi et al., 2020; Zhu et al., 2021). The TCRM is based on Eq. (1) and can decompose TC rainfall into five main physical processes: the frictional effect, topographic effect, vortex stretching effect, baroclinic effect, and radiative cooling (Lu et al., 2018).

A brief introduction to TCRM is given as follows. Different from the WRF, the TCRM estimates TC rainfall by calculating the upward motion $ w $, which can be decomposed into five components:

where $ {w_f} $ is the frictional component, $ {w_h} $ is the topographic component, $ {w_t} $ is the stretching component, $ {w_s} $ is the baroclinic component, and$ {w_r} $is the radiative cooling component. The upward motion impacts the upward vapor flux directly so that it determines the intensity, evolution, and distribution of TC rainfall (Zhu et al., 2013).

The frictional component can be written as:

where $ {\tau _{\theta s}} $ is the azimuthal surface stress, $M$ is the absolute angular momentum per unit mass, and $r$ is the radius. Furthermore,

and

where $ {C_d} $ is the drag coefficient, $ \left| {\mathbf{V}} \right| $ is the absolute value of surface wind speed, $ V $ is the azimuthal gradient wind, and $ f $ is the Coriolis parameter. In the original TCRM, $ {C_d} $ is simply represented by different constants for ocean and land (Lu et al., 2018). To distinguish between the frictional impacts of differing land coverage, the output drag coefficient from WRF was used to calculate surface stress. The mean drag coefficients for urban, cropland, and water were 0.0095, 0.0084, and 0.0013, respectively. Due to the larger drag effect, the mean surface stress over urban areas in CTL was enhanced by nearly 29% compared to that in No_Urban.

The topographic component ${w_h}$ is given by:

where $ h $ is the terrain height. The stretching component, ${w_t}$, is associated with changes in TC vorticity:

where $ {H_b} $ is a representative depth scale of the lower troposphere. The baroclinic component ${w_s}$ can be estimated as:

where ${{\mathbf{V}}_{g,200\;{\rm{hPa}}}} - {{\mathbf{V}}_{g,850\;{\rm{hPa}}}}$ is the difference of the geostrophic winds at 200 and 850 hPa, and $ {\mathbf{j}} $ is the unit vector pointing radially outward from the TC center.

The radiative cooling component $ {w_r} $ is treated as a relatively small constant (set to –0.005 m s^–1 in Lu et al., 2018). Thus, the contribution from this term was neglected in our study. The input data to drive TCRM includes the wind field, drag coefficient, topography, and the specific humidity along the TC track from WRF output files. The TCRM was initialized at 1800 LST 16 August (10 hours before TC made landfall).

Using linear decomposition and the TCRM, we can quantify the attributes of different physical processes associated with TC rainfall to elucidate the main physical mechanisms behind the urban impact on TC precipitation.

The simulated track and intensity are compared with those from the best track data in Fig. 2. Typhoon Rumbia (2018) made landfall in the southern coastal area of Shanghai and then moved westward across the YRD urban agglomeration, and the CTL run well reproduced the track except for the location of landfall, which was about 25 km to the south of the actual landfall (Fig. 2a). Although the CTL modeled intensity deviated from the observation with a sea-level pressure (the maximum near-surface wind speed) that was 4 hPa (4 m s^–1) higher at landfall, CTL was able to capture the intensification before landfall and the post-landfall decay (Fig. 2b). The mean error of the track forecast in CTL was 20.52 km, and those of the maximum near-surface wind speed and sea-level pressure were 2.63 m s^–1 and 3.27 hPa, respectively.

Figure 2.  (a) The track of Typhoon Rumbia (2018) and (b) its intensity in terms of maximum sustained surface wind (solid; units: m s^–1) and minimum central sea-level pressure (dashed; units: hPa) from the best-track data (BST; black), CTL (red), and No_Urban (blue). The position of TC is marked by dots every three hours. The thinner tracks in (a) denote the tracks from seven ensemble experiments, and the shading in (b) shows the deviation in the simulated intensity among seven experiments, while the thick lines are their ensemble mean. The gray shaded contour in (a) denotes urban areas in CTL. The vertical gray dashed line in (b) denotes the time of landfall (0400 LST 17 August).

The simulated track and intensity in the CTL and No_Urban simulations differed only slightly during landfall (Fig. 2b), and so did the evolution of the low-level structure (Figs. 3a-b). The maximum difference in the landfall location was 16.92 km, and those in the maximum surface wind and minimum sea-level pressure were 1.00 m s^–1 and 0.27 hPa, respectively. The radii of the maximum tangential wind in CTL and No_Urban were both about 100 km from 1800 LST 16 August to 0800 LST 17 August. Although there were no significant differences in the structure of sea-level pressure and tangential wind (Figs. 3a-b), the latter was slower by about 0.5 m s^–1 during landfall, which was probably related to the greater friction in CTL (Fig. 3c). In the same way, the evolution of the total surface wind was analyzed, and similar results were obtained (figure not shown).

Figure 3.  Time-radius Hovmöller diagrams of azimuthally averaged tangential wind speed at the lowest level (shaded; units: m s^–1) and sea-level pressure (black contour; units: hPa) for (a) CTL, (b) No_Urban, and (c) differences between CTL and No_Urban (CTL – No_Urban). The black line denotes the time of landfall, and the black dotted areas in (c) are the wind speed differences that are significant at the 10% level.

The observed and simulated spatial distribution of accumulated precipitation are compared in Fig. 4. Before and during landfall (1800 LST 16 August–0600 LST 17 August), heavy rainfall induced by Rumbia (2018) struck coastal cities such as Shanghai, Hangzhou, and Ningbo with a maximum amount of over 100 mm in 12 hours (Fig. 4a). The CTL run reproduced the spatial distribution of rainfall reasonably well, albeit with an overestimate over land (Fig. 4b). The spatial distribution of rainfall in No_Urban was similar to that in CTL, probably owing to similar tracks (Fig. 4c). However, Fig. 4d indicates that the principal rainband that hit coastal areas in CTL was stronger than that in No_Urban, showing a surplus of over 15 mm concentrated in coastal areas. Significant differences did exist in the outer regions of the TC due to the rainband activity, although these differences were less than 10 mm.

Figure 4.  Spatial distribution of the accumulated rainfall in 12 hours (units: mm) from the observations (a and e), CTL (b and f), and No_Urban (c and g), and the difference between CTL and No_Urban (CTL – No_Urban) (d and h) from 1800 LST 16 August to 0600 LST 17 August (a–d) and from 0600 LST 17 August to 1800 LST 17 August (e–h), respectively. The dashed areas in (b), (c), (f), and (g) denote the domains for calculation in Figs. 8 and 10. The dotted areas in (d) and (h) denote those differences that are significant at the 10% level, and the gray-shaded contours denote the urban areas in CTL.

From 0600 LST 17 August to 1800 LST 17 August, Typhoon Rumbia (2018) moved westward and brought more rainfall to the YRD (Fig. 4e). After landfall, the pronounced heavy rainfall was concentrated to the right of the storm track with a maximum of over 150 mm. The result in CTL bore a spatial resemblance to the observation but exaggerated the main rainband precipitation by near 25 mm in 12 hours (Fig. 4f). Meanwhile, both CTL and No_Urban underestimated rainfall near the TC center and the rainfall located near 30°N. Noticeable differences could also be found in the spatial distribution of rainfall between CTL and No_Urban during this period (Figs. 4g, h).

The simulated low-level wind field, temperature, and relative humidity before and after landfall were further compared with observations from FNL in Fig. 5. The simulations well captured the spatial pattern of the low-level wind field before and after Rumbia (2018) made landfall. The surface wind over the cities affected by Rumbia (2018) decelerated due to urban friction (Figs. 3c and 5). Meanwhile, CTL with SLUCM could reproduce the warming in the afternoon over the coastal cities, e.g., Shanghai, Hangzhou, and Ningbo. However, both CTL and No_Urban overestimated the 2-m relative humidity. The temperature in the southeastern quadrant of the TC was overestimated by about 1°C at 1400 LST on 17 August (Figs. 5d, e, and f). In general, CTL with the urban canopy model could well capture the spatial patterns of the low-level variables and reproduce the impact of the urban friction during landfall and the UHI effect during the afternoon of 17 August.

Figure 5.  Spatial distribution of 2-m air temperature (shaded; units: °C), 10-m wind field (vector; units: m s^–1), and 2-m relative humidity (black contour; over 90%) from observations (a and d), CTL (b and e), and No_Urban (c and f) at 0200 LST on 17 August (a–c) and 1400 LST on 17 August (d–f), respectively. The red TC symbols denote the location of the TC.

Figure 6 depicts the time-radius Hovmöller diagrams of the azimuthally averaged rainfall rate, which can help distinguish the urban impact on the temporal evolution of TC rainfall from a Lagrangian perspective. We first compared the simulations to the observed precipitation. During landfall, the evolution of rainbands simulated in CTL was similar to the observation (Figs. 6a-b). However, CTL underestimated precipitation amounts near the TC center at about 1200 LST 17 August but overestimated rainfall in the outer core of TC (radius >150 km). As a result, biases in accumulated rainfall were also found in Figs. 4e-f. Nevertheless, CTL generally reproduced the evolution of TC rainfall during landfall.

Figure 6.  Same as Fig. 3, but for the azimuthally averaged rain rate (units: mm h^–1) from (a) the observation, (b and d) CTL, (c and e) No_Urban, and (f) the differences between CTL and No_Urban (CTL – No_Urban). (a–c) are only averaged over land.

In the simulations (Figs. 6d-e), TC precipitation was concentrated within a radius of 250 km, and strong rain bands (>6 mm h^–1) were situated in the inner-core region within a radius of 50–100 km from the TC center. The heavy rainfall process (>6 mm h^–1) persisted from 8 hours before landfall to 5 hours after landfall (2000 LST 16 August–0900 LST 17 August). It is noteworthy that precipitation temporarily increased just after Rumbia (2018) made landfall (0400 LST 17 August–0900 LST 17 August), with the maximum rainfall rate being over 10 mm h^–1, which might have been caused by the rapid enhancement of surface roughness and the related frictional convergence within the several hours after landfall (Chen and Chavas, 2020; Hlywiak and Nolan, 2021). After 0900 LST on 17 August, the rainfall rate gradually decreased with time. In the afternoon of 17 August, strong precipitation (>6 mm h^–1) again developed in the inner core. This diurnal variation conforms to the results of previous studies on the diurnal cycle of landfalling TC precipitation, namely, two peaks in the early morning and afternoon (Tang et al., 2019a, b).

The evolution in No_Urban was similar to that in CTL, except for several pronounced differences near landfall. The rainfall rate in the inner core was greater than that in CTL at certain times (0000 LST 17 August). Still, at most times during landfall (0200 LST 17 August–0900 LST 17 August), it was weaker (Fig. 6e). The difference in the azimuthal-mean rain rate that could reach 2 mm h^–1 was mainly concentrated in the inner-core region (Fig. 6f). However, there were few significant differences in the outer region and after landfall. The temporal evolution illustrates that the YRD urban agglomeration promoted TC rainfall, especially in the inner-core region near landfall, while exerting only a minor impact on the outer rainbands and only after Rumbia (2018) moved further inland. Analysis of the evolution of the 2-km height radar reflectivity also shows a similar result (figure not shown).

The TC precipitation after landfall demonstrated strong asymmetry under the influence of surface friction and vertical wind shear (Li et al., 2014). To reveal the patterns of the evolution of TC rainbands in different azimuths, the time–azimuth Hovmöller diagrams of 100 km radius–averaged rainfall rates are depicted in Fig. 7.

Figure 7.  Time–azimuth Hovmöller diagrams of 100 km radius–averaged rain rate (units: mm h^–1) for (a) CTL, (b) No_Urban, and (c) the differences between CTL and No_Urban (CTL – No_Urban). The black line denotes the time of landfall, and the black dotted areas in (c) indicate significance at the 10% level.

Two main rainbands were present during landfall, termed here as the western rainband (2100 LST 16 August–0900 LST 17 August) and the northern rainband (1000 LST 17 August–2000 LST 17 August) (Fig. 7a). The western one impacted the coastal urban agglomeration 8 hours before landfall, with its intensity gradually increasing and reaching its maximum of over 10 mm h^–1 at 0500 LST 17 August. After the entire system moved inland, the western rainband gradually decayed, and its azimuth turned from the northwest into the southeast. The northern rainband developed during the late afternoon of 17 August, and its azimuth was maintained in the northeast.

The western rainband in the CTL run had a much stronger intensity, a wider range, and a larger peak than that in No_Urban (Fig. 7b). The maximum difference reached over 1.5 mm h^–1 (Fig. 7c). On the contrary, the difference in the northern rainband was negligible. Generally speaking, the characteristics of the impact of the urban areas on the two rainbands agreed with the azimuthal-mean result that the urban forcing could strengthen the rainfall near landfall. Yet, this impact dwindled after Rumbia (2018) moved further inland.

The above analysis demonstrates that urban agglomerations could enhance the inner-core rainfall near landfall. To reveal the dominant physical process, the physical models of TC precipitation mentioned in section 2.3 were utilized to quantify the contributions from different processes related to the urban impact on TC precipitation.

Based on Eq. (4), Fig. 8 analyzes the temporal evolution of the normalized responses in terms of precipitation, dynamic processes, and moist processes during landfall. The response in precipitation shows that the urban areas enhanced the inner-core rainfall by nearly 10%, which is consistent with the result of Fig. 6f. The response reached over 10% five hours before landfall and fluctuated around 10% with a peak at 12% during the period of landfall (0100 LST 17 August–0700 LST 17 August). After Rumbia (2018) moved inland, it gradually decreased with an amplitude lower than 5%.

Figure 8.  Temporal evolution of normalized responses (units: %) in precipitation ${{\Delta \overline P }}/{{{{\overline P }_{{\rm{CTL}}}}}}$ (red), dynamic processes ${{\Delta \overline w }}/{{{{\overline w }_{{\rm{CTL}}}}}}$ (yellow), moist processes ${{\Delta \overline q }}/{{{{\overline q }_{{\rm{CTL}}}}}}$ (blue), and nonlinear residual processes ${{(\Delta \overline w )(\Delta \overline q )}}/{{{{\overline P }_{{\rm{CTL}}}}}}$ (green) within a radius of 100 km. The gray-shaded areas are significant at the 10% level in terms of ${{\Delta \overline P }}/{{{{\overline P }_{{\rm{CTL}}}}}}$.

The contributions from dynamic and moist processes were analyzed by the evolution of their responses (Fig. 8). The response in the dynamic process varied with the precipitation response with a similar amplitude. The correlation coefficient between these two terms reached 0.87 (p<0.01). However, the responses in the moist processes and the residual were maintained below 1%, which were quite small. This indicates that the dynamic process featured by low-level upward motion played a more crucial role compared with the moist condition. We also conducted decomposition at other levels and found similar results.

The linear decomposition highlights the importance of upward motion to the response in TC rainfall subjected to urban impacts. Meanwhile, the TCRM can further decompose the upward motion into components related to the four main mechanisms of TC precipitation. Before analysis, the simulated result from TCRM was evaluated first. The radial distributions of azimuthally averaged rainfall during landfall (1800 LST 16 August–2000 LST 17 August) from the observation, WRF modeling, and TCRM estimates are shown in Fig. 9a. Although the inner-core rainfall (radius <50 km) was underestimated in both the WRF and TCRM, the maximum rainfall rate was about 6 mm h^–1 at a radius of 80 km and the rainfall rate in the outer-core region decreased with increasing radius, which was similar to observations. Compared to the WRF results, the radius of the maximum rainfall rate analyzed in TCRM was 4 km larger than that in WRF. Besides, the TCRM underestimated the inner-core rainfall (radius <100 km) while it overestimated the outer-core rainfall (radius >130 km). In general, the TCRM successfully captured the radial distribution of TC precipitation and can be used for further diagnosis.

Figure 9.  (a) Comparison of the 26-h (1800 LST 16 August–2000 LST 17 August) azimuthally averaged rainfall rate (units: mm h^–1) profile from observation (black), WRF (red), and TCRM (blue). (b–d) Same as Figs. 6d–f, but for the rainfall rate (units: mm h^–1) from TCRM.

Regarding the evolution of azimuthal-mean rainfall (Figs. 9b-c), the rainbands in TCRM and WRF strongly resembled each other. However, the strong rainfall in TCRM occurred about 10 km further out, with its peak intensity underestimated by 1 mm h^–1 (Fig. 6). Rainfall before landfall was weaker than that in the WRF, especially from 0300 LST 17 August to 0600 LST 17 August, and the intensification of rainfall near landfall was about 2 hours later than that in the WRF.

Despite simulation biases, the TCRM has proven to be capable of reproducing the evolution of TC rainfall and the rainfall enhancement in the CTL run (Figs. 9c and 10). Thus, the results from the TCRM were further analyzed to elucidate the dominant physical process related to the urban impact on TC rainfall.

It is apparent from the temporal evolution of responses among the four components of TCRM (Fig. 10) that the frictional component has played a dominant role. Its absolute value was 3 to 4 times larger than those of the other three components (figure not shown), and its temporal evolution was highly consistent with that of precipitation, as evidenced by a correlation coefficient of over 0.98 (p<0.01). However, the responses of the other three components were maintained at less than 2% during landfall and showed no significant relationship with the precipitation response. This diagnostic result suggests that the response of the frictional component has dominated the precipitation response, while the urban surface could hardly affect low-level upward motion by the other three physical processes.

Figure 10.  As in Fig. 8, but for the responses (%) in the four components in TCRM. The red, yellow, green, blue, purple, and black lines denote the responses in precipitation from TCRM, frictional component, topographic component, stretching component, baroclinic component, and precipitation from WRF, respectively. The gray-shaded areas are significant at the 10% level in terms of the responses in precipitation from WRF.

The diagnosis based on physical models of TC rainfall has preliminarily revealed the crucial roles of upward motion and frictional processes in the TC precipitation response. However, the specific physical processes remain unclear. Thus, the temporal evolution and vertical structures of low-level wind fields during landfall are analyzed below.

The temporal evolutions of the responses in low-level variables are shown in Fig. 11. The urban surface, with greater roughness, kept decelerating the surface wind so that the near-surface tangential wind maintained a negative response with an amplitude of less than –5 % (Figs. 3c and 11a). Meanwhile, there was clear consistency among the evolution of the surface stress, radial wind, and frictional component (Figs. 10 and 11a). This consistency suggests that when surface stress is strong, it weakens not only the surface wind speed but also the Coriolis and centrifugal forces, possibly resulting in an inward balance and stronger inflow within PBL (Anthes, 1974). The inner-core overturning circulation, represented by the mass streamfunction, was promoted by the aforementioned Ekman adjustment (Fig. 11b). Consequently, the overturning circulation further transported more angular momentum to the top of PBL, causing an enhancement of vertical motion (Chen and Chavas, 2020; Hlywiak and Nolan, 2021). As shown in Fig. 11, this physical process fit well when the precipitation response was significant (1800 LST 16 August, 0300 LST 17 August, 0600 LST 17 August, and 0900 LST 17 August), despite the tendency for the momentum response having a 1-h delay in the late stage of the simulation.

Figure 11.  As in Fig. 8, but for responses in the (a) tangential wind speed at the lowest level (black), radial wind speed at the lowest level (red), and surface stress (blue); (b) mass streamfunction at 1-km height (green) and absolute angular momentum at 1-km height (purple). The surface stress is calculated by total surface wind speed and surface azimuthal wind, which is different from the surface stress in TCRM calculated by the total surface wind speed and 1-km gradient wind.

To verify the abovementioned physical process, vertical cross-sections of the low-level wind fields near landfall (0300 LST 17 August and 0600 LST 17 August) in Fig. 12 were analyzed. The unbalanced process within PBL mentioned above can be quantified by the agradient force (AF), which has been defined as the sum of the azimuthally averaged radial pressure gradient force, the Coriolis force, and the centrifugal force (Huang et al., 2012), as given by Eq. (12):

Figure 12.  Vertical cross-sections of the azimuthal-mean vertical velocity (shaded; units: m s^–1) and radial winds (black contour; units: m s^–1) in the CTL (a and c) and No_Urban (b and d) simulations, and the agradient force (shaded; units: m s^–1 h^–1) and tangential winds (black contour; units: m s^–1) in the CTL (e and g) and No_Urban (f and h) simulations at 0300 LST 17 August (a-b and e–f) and 0600 LST 17 August (c–d and g–h).

At 0300 LST 17 August, the low-level tangential winds were reduced by about 5% due to the greater surface friction in CTL (Figs. 3c and 11a). But this reduction was not pronounced above 0.5 km (Figs. 12e-f). Notably, the AF below 0.5 km in CTL was much stronger than that in No_Urban (Figs. 12e-f), indicating that the inflow and overturning circulation were enhanced by the unbalanced process within the PBL, thus promoting the vertical motion at the height of approximately 1 km (Figs. 12a-b). Three hours later, the reduction in tangential winds was rather small in CTL, but the surface stress, low-level inflow, and overturning circulation were still stronger than those in No_Urban (Figs. 11a, 12g, and 12h). This contributed to the strong vertical motion at heights of 0.5–2 km (Figs. 12c-d).

The above analysis mainly focuses on the effects of dynamic urban forcing on TC rainfall. We further discuss whether thermodynamic urban forcing, such as surface sensible and latent heat fluxes, can influence TC rainfall via a UHI effect or local evaporation.

Figure 13 shows the time-radius Hovmöller diagrams of the azimuthally averaged surface heat flux and temperature near the surface. Consistent with the diurnal cycle of surface enthalpy fluxes, both the sensible heat flux (Figs. 13a-b) and latent heat flux (Figs. 13d-e) were weak near landfall but became much stronger after 1000 LST 17 August. The urban areas enhanced the surface sensible heat flux but suppressed the latent heat flux during the daytime of 17 August (Figs. 13c, f). However, the surface heat flux maintained relatively low values near landfall, and the differences in surface sensible heat flux and surface latent heat flux between CTL and No_Urban were rather small when the precipitation responses were significant. This suggests that none of these forcings could significantly impact surface temperature or low-level humidity (Figs. 8 and 13i).

Figure 13.  Same as Fig. 3, but for surface sensible heat flux (a–c) (units: W m^–2), surface latent heat flux (d–f) (units: W m^–2), and temperature at 2-m height (g– i) (units: °C).

Thus, the contribution of the urban thermodynamic forcing to TC rainfall was rather negligible during landfall. Our conclusion is consistent with Ao et al. (2022), who found that the UHI effect was quite weak before landfall and completely disappeared with the start of the TC rainfall process.

The coastal urban agglomerations face growing risks of extreme precipitation and flooding induced by landfalling TCs. However, current research has paid limited attention to whether and how the precipitation in landfalling TCs responds to the fine-scale urban surface, despite the significant influence of those issues on mesoscale convective precipitation. Meanwhile, considering the intricate physical mechanisms of TC rainfall, it is highly advisable to conduct quantitative attributions to reveal the dominant physical process.

High-resolution ensemble numerical simulations of Typhoon Rumbia (2018), including the CTL and No_Urban experiments, were performed by WRF-ARW in this study. The CTL simulation which considered the actual urban land-use conditions and adopted an Urban Canopy Model, successfully reproduced the evolution of TC track, intensity, and precipitation during landfall. The No_Urban simulation, with urban areas replaced by croplands, was conducted to exclude any possible urban impact. The differences between the two experiments reveal whether and how TC precipitation responds to the urban impact. Furthermore, two physical models of TC precipitation—linear decomposition and TCRM—were utilized to quantify the contributions of different physical processes and reveal the dominant mechanism.

Under the urban impact, slight differences existed in TC track and intensity. Still, the results from the temporal evolution show that inner-core TC rainfall within several hours before and after landfall could be significantly strengthened by nearly 10%. However, the YRD urban agglomeration exerted a minor impact on rainfall in the outer-core region and after Rumbia (2018) moved further inland. The linear decomposition of TC rainfall suggests that the low-level upward motion is a crucial factor in TC precipitation, as a high consistency could be found in the temporal evolution of these two terms. Furthermore, the diagnosis based on TCRM reveals that the frictional component of the vertical motion played a decisive role as its evolution dominated the precipitation response to the urban impact. By contrast, components related to the other three physical processes only accounted for a small portion of the precipitation response. To illustrate the specific physical mechanism of how urban forcing regulates TC precipitation, a more detailed analysis of the temporal evolution and vertical structures of low-level wind fields was conducted. It has been confirmed that the rougher urban surface could strengthen surface friction and further enhance low-level inflow and the inner-core overturning circulation via the unbalanced process within PBL. Consequently, more angular momentum was transported to the top of PBL, contributing to the enhancement of vertical motion and eventually inducing stronger rainfall in the inner-core region.

This study demonstrates that although the coastal urban areas only account for a relatively small proportion of the underlying surface, their unique dynamic forcing could significantly strengthen the inner-core rainfall of TCs, implying that urbanization has the potential to increase the risk of heavy rainfall caused by landfalling TCs over the coastal areas. Furthermore, the quantitative diagnosis by physical models of TC precipitation and the specific analysis of low-level structure also highlights the urban frictional effect and the related unbalanced processes within the PBL, which could be a major influential factor contributing to the intensification of TC rainfall impacted by urban areas. These findings provide new insights into how land-surface processes regulate high-impact weather events.

Based on the sensitivity experiments of a weak landfalling TC case, this paper mainly focuses on the responses in the temporal evolution of TC precipitation. Therefore, further studies should address how urban forcing affects landfalling TCs with various intensities or spatial scales and evaluate the response in the spatial distribution of TC rainfall. Additionally, several other factors, like anthropogenic heat, surface evaporation, and aerosol emission, may contribute to TC precipitation. The investigation here is only a case study near the YRD urban agglomeration, and TC precipitation over other urban agglomerations should be further examined. Since this study only provides a preliminary analysis of the role of urban thermodynamic forcing, more in-depth investigations are warranted to clarify its complex physical processes. Besides, the Urban Canopy Model and the land-use properties employed in the simulations are relatively simple. Urban surface data with greater refinement have recently been developed (Ren et al., 2019; Yang et al., 2021), allowing for a more realistic characterization of urban processes. Therefore, future efforts can be made to evaluate the urban impacts on TC rainfall using better-performing urban models such as multi-layer urban schemes and more refined urban surface datasets.

Acknowledgements. This study was supported by the National Science Foundation of China (Grant Nos. 42088101 and 42175005) and by the Postgraduate Research and Practice Innovation Program of Jiangsu Province (KYCX22_1137).