The importance of future urban development in hourly extreme rainfall projections- comparing global warming and urbanization forcing over the Pearl River Delta region


 The impacts of future urban development and global warming forcing on hourly extreme rainfall over the Pearl River Delta (PRD) area have been investigated, by dynamically downscaling General Circulation Model (GCM) outputs using the Weather Research and Forecasting Model (WRF) at convection-permitting resolution, coupled with an Urban Canopy Model (UCM). Three downscaling experiments corresponding to different urban land cover (1999 and projected 2030) and climate (1951-to-2000 and 2001-to-2050 GCM simulations) were designed. Near-future climate change (up to 2050) and 1999-to-2030 urban development effects on PRD extreme precipitation were then examined. Results show that climate change and rapid urban development forcing have comparable positive effects on the intensity as well as heavy hourly rainfall probability over the PRD megacity. Global warming tends to increase heavy rainfall probability (from 40 to 60mm/hr) by about 1.3 to 1.8 times, but suppresses the frequency of light rainfall. Urban development increases urban rainfall probability within the whole range of intensity, with frequency for very heavy rainfall (> 90mm/hr) almost doubled. Overall, forcing due to rapid urban development plays an important role for projecting rainfall characteristic over the highly urbanized coastal PRD megacity, with impacts that can be comparable to global warming in the near future.


Introduction
There is now ample evidence that anthropogenic activities can exacerbate meteorological hazards such as heat waves and extreme rainfall. Since the last century, many places have seen rapid urbanization 1 , which modulates climate over cities, leading to changes in surface high temperature 2,3 , precipitation 4,5 and circulation 6,7 . It is well known that over the city area, more shortwave radiation is absorbed, due to modi ed land use, lower surface albedo and urban morphology 8 . Anthropogenic heat (AH) released from buildings, tra c and human populations can also promote the formation of urban heat island (UHI) 9 , which also results in higher temperature 2 and provides an environment conductive to convection 10,11 .
Many studies show that urbanization can enhance rainfall intensity downstream of some cities 12-15 . However, this result might depend on the UHI intensity; recent studies for Beijing demonstrate that strong UHI can increase rainfall directly over the downtown area, while during weak UHI days, precipitation tends to be bifurcated and avoids the city center 16,17 . AH released over megacities also plays a role in intensifying extreme rainfall and local convection [18][19][20][21] . On the other hand, impervious surface within a city means less water permeability and surface evaporation, which results in lower surface moisture and decreased latent heat ux to the atmosphere 22,23 . Studies show that decreased water content can reduce convective available potential energy (CAPE), hence the total precipitation amount over the Beijing urban area 24,25 . For coastal urban locales, enhanced surface temperature induced by urbanization can lead to stronger low-level ow from the ocean, which carries water vapor to the urban area. This is also conducive to stronger extreme precipitation intensity in some coastal cities such as Tokyo, Osaka, and the Pearl River Delta (PRD) megacity 20,21,26−30 . At the same time, climate change induced by anthropogenic emissions increases temperature strongly on the global scale 31 , with some climate projections giving even stronger warming trends in the future 32 .
Global warming is likely to suppress light precipitation over the monsoon area and contiguous US [33][34][35] or even total precipitation in some subtropical area such as Japan 36,37 , due to enhanced thermal stability.
On the other hand, according to the Clausius-Clapeyron (CC) relationship, the saturated vapor pressure increases by ~ 7% per degree near-surface warming, which means the atmosphere can accommodate more water vapor under a warmer climate background. The atmosphere will get moister in the future, and heavy rainfall will become stronger and more frequent 34,38,39 . Indeed, increasing trends of extreme rainfall have already been observed in many regions [40][41][42] . Studies using numerical simulations to examine the relationship between global warming and extreme rainfall also give results consistent with observations, with changes in the intensity of extreme rainfall generally exceeding those in the annual mean in the tropical areas 20,43,44 .
The PRD region is a megacity cluster located in the southern coast of China, and has experienced rapid urbanization since the early 1980s; there is strong precipitation during the summer monsoon season, mainly due to severe thunderstorm systems and also tropical cyclones 45,46 . Observations indicate that both the intensity and frequency of PRD extreme rainfall have increased (more than 5% per decade) over most PRD megacity from 1971 to 2016 47 , concurrent with regional urban expansion 28 . There is in fact robust intensi cation of summertime extreme rainfall by the UHI effect for all rainfall types in the PRD urban area 48 . At the same time, an increasing trend of both the strength and frequency of extreme rainfall can be observed over South China and PRD, while at the same time the number of rainy days has decreased signi cantly 32,49,50 . Results from numerical modeling also show that global warming can strongly enhance both intensity and frequency of extreme rainfall over PRD 20 .
In this work, we investigate: (1) how urban development and global warming together might affect the characteristics of intense precipitation over PRD in the coming decades, and (2) whether the effect of urban development and climate change on PRD urban extreme rainfall are comparable. A recent work utilized a high-resolution model, with various types of urban land use incorporated, for assessing the impacts of both urban development and global warming on thermal comfort within PRD 51 . To our knowledge, this is the rst study using a similar approach to exam their twin impacts on PRD extreme rainfall. In particular, dynamical downscaling of General Circulation Model (GCM) outputs, from historical runs and near-future climate projections, were carried out using a convection-permitting regional model with simulated urban canopy over city areas. Moreover, a predicted 2030 land use dataset over the PRD region is considered, which is derived from a land use prediction model that adopts with "training" based on historical PRD land use 52 . This way, magnitudes of extreme rainfall changes induced by anthropogenic activities (urban development and climate change) over the PRD region can be quanti ed; results based on similar modeling strategies can serve as a scienti c basis for local policy makers to develop climate change adaptation plans in relation to the hydroclimate projections.

Results
In this study, three sets of parallel experiments were carried out: dynamical downscaling with 1999 urban land use under present climate  conditions (referred to as 99LS-HIS), dynamical downscaling with 1999 urban land use under near-future (2001-2050) climate conditions (referred to as 99LS-FUT), and nally dynamical downscaling with 2030 urban land use under near-future climate conditions (referred to as 30LS-FUT). Details of the experimental setup and land surface data prescriptions are given in methodology. We rst investigate the change of surface temperature induced by urban development from 1999 to 2030 over the PRD area, as well as that due to global warming over a comparable period. Figure  Also shown in Fig. 2 gives the 30LS-FUT minus 99LS-HIS, and 30LS-FUT minus 99LS-FUT 2-m temperature, averaged over all extreme cases, in the PRD region. Overall, urban development can lead to surface warming by about 0.6 to 0.8 o C over the city area. When the effect of global warming is also considered, the 2-m temperature is enhanced by about 1 o C in the same region, as compared to 0.6 o C increase over the ocean. Therefore, according to these experiments, global warming and urban development can result in a comparable temperature increase over the PRD mega-urban region in the near future.
In order to investigate the in uence of urban development and climate change on rainfall characteristics, Fig. 3 shows the mean rainfall difference between (a) 30LS-FUT and 99LS-HIS, and that between (b)30LS-FUT and 99LS-FUT, averaged over on all extreme cases considered. In the near future, the mean rainfall amount over the PRD urban area, for these extreme events, is enhanced by about 5-8 mm/day over part of the land area due to global warming, but the difference of rainfall intensity can be small in the coastal area. Future projected urban development also leads to more accumulated rainfall, but only at highly urbanized locations (the PRD city cluster comprising Guangzhou, Foshan, and Dongguan, between  Fig. 3b), with signi cant enhancement by about 8-10 mm/day. Over the more southern part of the domain, rainfall change seems to be insigni cant, which might be related to the weaker change of temperature and vertical motion in the southern part. When considering the whole 2030 PRD urban area, the area averaged accumulated rainfall increased by about 13.5% due to global warming; on the other hand, the increment is about 9.7% due to projected urban development alone. Statistical tests con rm that aforementioned precipitation change over land in the northeast/west part of the domain, due to global warming in near future, passes the 95% signi cance level. Increased total rainfall in north and northwest part of the megacity area, caused by projected urban development, is also found to be statistically signi cant. Moreover, for the rainfall frequency, PDFs of hourly precipitation rates over the mega-urban area are considered. Figure 3c shows rainfall PDFs from 99LS-HIS, 99LS-FUT, and 30LS-FUT experiments within the range of 1 to 110mm/hr. Compared with 99LS-HIS, frequency of light rainfall events (those from 1 to 10mm/hr) for 99LS-FUT is decreased (by about 20% or more). On the other hand, global warming can strongly increase the probability of heavy rainfall (more than 50mm/hr) in the near future, with the likelihood enhanced by ~ 30 to 80%. By comparing 99LS-FUT and 30LS-FUT, it can be inferred that urban development can also increase the frequency of urban precipitation; such enhancement, however, is found for all rain rates (1-110mm/hr), with even stronger effect on heavy rainfall (i.e., rain rate more than 50mm/hr). For hourly rainfall in the range of 50-100mm/hr, the frequency increase is ~ 40 to 80% due to PRD urban development. It is noteworthy that both urban development and global warming can enhance the frequency and intensity of extreme rainfall over the PRD urban area.
To better compare impacts of urban development and global warming on rainfall frequency, Fig. 3d shows the ratio of rainfall probability between 30LS-FUT and 99LS-HIS (black), 99LS-FUT and 99LS-HIS (blue), and 30LS-FUT and 99LS-FUT (red) over the region with strongest signals in the Pearl River Estuary area (see black box in Figs. 3a and 3b). When considering both effects (see 30LS-FUT vs 99LS-HIS curve), it is obvious that frequency of heavy rainfall occurrence is enhanced more than that of light rainfall. It is noteworthy that urban development seems to have stronger in uence on extremely heavy rainfall; the frequency of larger then 90mm/hr precipitation is doubled due to urban development (see 30LS-FUT vs 99LS-FUT). On the other hand, the downscaled global warming effect is most prominent within the range of 40-60mm/hr, with frequency enhanced by ~ 80 to 100% (see 99LS-FUT vs 99LS-FUT).
Under a warmer climate, there is higher surface evaporation (see Figure S1) which can increase the background moisture content. In fact, the precipitable water is increased in 99LS-FUT compared with 99LS-HIS in the whole PRD region ( Figure S2), which is conducive to stronger extreme rainfall in the near future. However, the temperature and relative humidity difference caused by warmer background climate (99LS-FUT vs 99LS-HIS) increases with height, while the average temperature (relative humidity) difference over the urban area is only about 0.4 o C (0.57%) at surface, then increasing to 0.75 o C (1.04%) at 2000m ( gure not shown). Hence, the 99LS-FUT experiment gives higher environment virtual temperature, leading to increased CIN for air parcels under 1400m compared to 99LS-HIS (see Figure S3). This is consistent with the results that global warming can enhance the thermal stability, making atmosphere more stable, and convection more di cult to be triggered [33][34][35][36][37] . The increased precipitable water and atmospheric stability plays an opposite effect on extreme rainfall. For these extreme precipitation events, more intense and frequent extreme rainfall over land area could be due to enhanced moisture content under a warmer climate, while the higher CIN is likely the reason why the frequency of light rainfall (1 to 10mm/hr) is suppressed in the PRD mega-urban area due to climate change.
It is known that urbanization results in lower surface humidity over the urban area due to decreased surface evaporation, which leads to a decrease of convective available potential energy (CAPE) for parcel under 600m of height. However, the CAPE difference is still weak above 600m (no more than 3J/kg) over the 2030 PRD megacity ( gure not shown). Considering that the impact of urban development is not uniform over the whole city area, difference in CAPE for parcels rising at 1000m between 30LS-FUT and 99LS-FUT is calculated (see Figure S4). There is an increase of CAPE of about 15 to 30 J/kg over north and northwest part of the mega urban region (again locations with the greatest urban development, such as Guangzhou, Dongguan, Foshan, and Panyu, indicated by pen green circles in Figure S4), which strengthen local convection on these locations; enhanced CAPE is also consistent with the rainfall increase (about 3-12mm/d) at the same locations. To nd out whether convection and local water vapor content are changed due to urban development, Fig. 4 gives the vertical pro les of speci c humidity and wind difference between 30LS-FUT and 99LS-FUT, along a northeast-southwest cross section (see Fig. 1c). Red and blue bars, at the bottom of the same gure, indicate the projected new urban area in 2030, and existing urban area in 1999, respectively. Due to urban development, more water vapor is found at the height from 300m to 6km, which can be attributed to increased moisture convergence in relation to induced circulation by a stronger UHI effect, especially in the highly urbanized megacity area (such as Dongguan), with speci c humidity increased more than 0.15g/kg from 500m to 5km. Compared with 99LS-FUT experiment, 30LS-FUT gives stronger vertical motion over the most urban area from 1km to 6km. This is especially the case over the highly urbanized region of 113.2 to 114.2 o E, with obvious development in 2030 compared to 1999. Similar results were found for other cross-sectional plots. For the east-west cross-section through Guangzhou (see Figure S5a), enhanced speci c humidity and vertical motion were seen over the Guangzhou and Foshan area, and the anomalous speci c humidity can reach 4km in the highly urbanized area; this is consistent with stronger vertical motion there. As can be inferred from the south-to-north cross-section through Guangzhou (see Figure S5c), urban development and presumably the induced additional UHI can lead to anomalous low-level southerly ow from the ocean towards to the city area, thus advecting moisture into the mega-urban region. Moreover, compared with PRD urban land use in 1999 and 2030, large changes of urban development were found in two regions  Fig. 1b and 1c), while enhanced precipitation and CAPE were only found in former region due to urban development. For the latter region, there are also stronger low-level southerly ow and slightly enhanced water vapor above 500m due to urban development (see Figure S5e). However, the change of vertical wind speed are weak over the Hong Kong and Shenzhen; there are even decreased vertical wind speed above 3km over 22.6-22.8N due to urban development. It appears that due to the strongly enhanced vertical motion over Guangzhou, Foshan, and Dongguan, the sinking branch from the north and northwest part of the domain tends to suppress convection over other urban locations (such as Shenzhen and Hong Kong, 22.5-22.8N, 113.8-114.3E). Hence stronger convection and precipitation are only found in the north and northwest part of the megacity. Overall, for the 30LS-FUT experiment, stronger convection found over the area with strongest urban development is consistent with higher CAPE over the same locations; induced low-level convergence and southerly ow from the ocean act to increase the atmospheric moisture content, which is also conducive to stronger and more frequent extreme rainfall there.

Discussions And Summary
In this study, impact of future urban development on hourly extreme rainfall projection over the PRD mega-urban region has been investigated and compared with global warming effect, based on dynamical downscaling of extreme rainfall cases taken from GCM runs using a convection-permitting regional model, with three types of urban land use and corresponding parameters incorporated in the model's urban canopy. Parallel experiments were designed by varying the urban land use (1999 vs 2030) and background climate conditions (present vs near-future). Results showed that urban development can lead to surface warming by about 0.6 to 0.8 o C over the city area. Near-future global warming effects can enhance surface temperature by ~ 0.3 to 0.8 o C (0.4 to 0.8 o C) over the PRD land (nearby ocean). For precipitation, both urban development and global warming can enhance the intensity as well as the occurrence rate of extreme rainfall. Global warming can increase the extreme rainfall amount (averaged over entire integrations for all selected extreme cases) over part of the PRD land area by about 3-12mm/day, while the increase due to urban development was found to be around 6-12mm/day over the north and northwest part of PRD region (i.e. locations which are projected to become even more urbanized in the near future). Moreover, the intensity of accumulated rainfall averaged over the 2030 PRD urban area increased by about 13.5% (9.7%) due to global warming (urban development). This result is consistent with a recent meta-data analysis which reported that accumulated rainfall increases by about 11 to 21% in the city center (and 14 to 22% in the downwind direction) due to urbanization 53  It was also found that, due to the multi-center nature of PRD, sinking branch from the highly urbanized area could suppress convection over the urban in coastal areas (such as Shenzhen). Hence stronger convection and precipitation was only found in north and northwest part of the domain. Also noteworthy is that global warming has a negative impact on the frequency of light rainfall (by around 20% of reduction) during these extreme precipitation events, which may be attributed to stronger CIN, and more stable atmosphere under a warmer climate. But for heavy rainfall (within the range of 40 to 60mm/hr), global warming would substantially enhance its probability by about 1.3 to 1.8 times, which is supported by the enhanced evaporation and water vapor in the atmosphere under a warmer climate. Urban development, on the other hand, can increase the likelihood of urban rainfall in all ranges (1-110mm/hr). Fractional increase of probability from 99LS-FUT to 30LS-FUT is no more than 5% for light rainfall (1-10mm/hr), but reaching 40 to 80% for heavy rainfall (50-110mm/hr), and especially extremely heavy rainfall (stronger than 90 mm/hr). Signi cant enhancement of extreme rain rates is due to regionally increased CAPE as well as decrease of CIN. The ocean also has an important effect versus urbanization on rainfall in the coastal PRD region; though urban development can reduce near-surface water vapor content in the large scale (30LS-FUT vs 99LS-FUT), due to stronger induced low-level southerly ow (from the ocean to megacity area) and stronger moisture ux convergence, urban development still results in more water vapor above the height of 300m, during extreme rainfall events. Overall, it is found that incorporating future urban development information is of great importance for the PRD region, in view of comparable impacts on temperature, intensity and frequency of extreme rainfall over the PRD megaurban area compared to global warming forcing in the near future.
There are some limitations in this study. It is known that urban-induced aerosol can play a cooling effect and inhibit the formation of rainfall in some situations. However, aerosols and their future changes are not considered due to lack of data; their impacts on urban extreme rainfall will be researched in future studies. Moreover, extreme rainfall cases from only one GCM (namely the GFDL-ESM2M; see methodology) were examined in this study. For urban development, however, we suspect that its effect is not too sensitive to the background climate. Higher surface temperature and more moisture ux convergence induced by urban development can still enhance the intensity and frequency of precipitation over the PRD mega-urban area, under other GCM climate conditions. For global warming signals, further inspection showed that the increase of SST and evaporation given by this GCM are generally consistent with other GCM projections. Still, it will be of interest to dynamically downscale products from other GCMs, using the same regional modeling framework, to further compare the impacts of urbanization vs global warming on severe rainfall over PRD or other coastal megacities.
Finally, the urban canopy model used here can incorporate three types of urban land use only. Such representation might still not be realistic enough for a complex urban environment, such as that of the PRD mega-urban cluster. Details such as heat released by air conditioning and its timing, interaction with various atmospheric layers, etc., are neglected. In our next study, we plan to adopt a multi-layer urban canopy module that also allows more urban land types, with parameters based on the World Urban Database and Access Portal Tools (WUDAPT) dataset 54,55 . Such a modeling system will be ideal for studying the interaction between realistic urban environment and meso-scale systems. Results should be invaluable for mitigation of and adaptation to extreme weather and related hazards happening in highly  Figure S6 shows the monthly mean values of the temperature, geopotential height, Uwind, and V-wind variables from GFDL-ESM2M versus those from ERA-interim, for the months of May, June, July, August, and September, for the 1970-2014 period; climatological summertime (May-to-September) mean distribution of 500hPa temperature, 500 hPa geopotential height, and 850 hPa temperature with u,v wind circulation are shown in Figure S7 117 o E) was rst computed. Days during which the daily rainfall is larger than the 99th percentile (based on rainfall on wet days, when rain rate > 0.1mm/day) were de ned as extreme rain days. 60 extreme cases in the summertime period of May to September, not related to tropical cyclone (TC)-like systems > (hereinafter referred to as non-TC cases) were selected. 30 extreme cases were selected from the historical era , and 30 extreme cases from the near-future era . GCM outputs from these cases were then downscaled by WRF, with integrations starting from 48 hours prior to the extreme rain day and lasting for at least ve days.

Local Climate Zone (LCZ) Data, UCM and Model Experiments
Altogether, three sets of downscaling experiments using WRF-SLUCM were conducted: in 99LS-HIS, extreme events from the GFDL-ESM2M present climate run  were dynamical downscaled, with prescribed 1999 urban land use; in 99LS-FUT, extreme rainfall cases from the near-future  73 . Firstly, a data mining technique, namely the Arti cial Neural Network (ANN), is used to learn the occurrence probability for each land-use from historical LCZ maps, based on geographical factors such as slopes, and distance from city centers. Secondly, the future demands for urban land-use are projected based on trajectories of demographic and socioeconomic developments. Finally, the Cellular Automata (CA) model is used to model the land-uses conversions based on current LCZ maps (2014 LCZ maps), with the occurrence and neighborhood in uence probability be repeated until the future demands are met. And the projected near-future maps are generated by the CA model.
To simplify the prediction scheme, the ten types of urban LCZ were regrouped into three types: "Low Intensity Residence" (type 1) comprises Open Mid Rise, Open Low Rise, Sparsely Built, Open High Rise, Lightweight Low Rise, Large Low Rise; "High Intensity Residence" (type 2) comprise Compact Mid Rise; Compact Low Rise, and nally, "Commercial and Industrial" (type 3) includes both Compact High Rise and Heavy Industry. Table 1 gives the UCM parameters prescribed for these three land use types. Also shown in Fig. 1 are the 1999 and 2030 urban land use distributions (after re-grouping) in the innermost model domain, with yellow, red, and purple indicating type 1, type 2 and types 3 urban land use, respectively.