c. Experimental design and model configuration
To capture the climate context of different scenarios and stages in the 21st century, two sets of time-slice simulations were conducted for the 2040s (2040–2049) and 2090s (2090–2099) in each of the three SSP emission scenarios, namely SSP1-2.6 (sustainability), SSP2-4.5 (middle of the road), and SSP5-8.5 (fossil-fueled development). To compare the present and future trends, baseline simulations were also performed for the 2010s (2011–2020) in the SSP2-4.5 scenario because it was assumed to be the closest pathway to reality for the 2010s. Additionally, simulations driven by the European Centre for Medium-Range Weather Forecasts Reanalysis data version 5 (ERA5) for 2011–2020 were performed for bias correction.
The WRF model version 3.9.1 was initiated on May 1 for each year of the two decades and integrated from May 1 to September 30, covering the complete warm season in southern China. The period from May 1 to May 31 was treated as the soil temperature and moisture spin-up stage, and the analysis period was from 1 June to 30 September. Vertical layers were implemented in 39 eta coordinates from the surface to the stratopause, with the planetary boundary layer corresponding to dense layers. Specifically, 20 layers were assigned from the surface to an altitude of approximately 2000 m above the terrain. The WRF model was utilized to simulate atmospheric conditions at multiple spatial scales. The outermost domain, WRF D01, had a 27 km × 27 km horizontal resolution and covered a vast region spanning East and Southeast Asia, the South China Sea, and the tropical western Pacific. Nested within D01 was WRF D02, with a higher resolution of 9 km × 9 km, focused on southern China. WRF D03 covered the Guangdong province, while WRF D04, with a resolution of 1 km × 1 km, was used to capture fine-scale features over the PRD. To ensure accurate representation of the large-scale mean flow from MPI-ESM-1-2-HR, grid nudging was implemented in WRF D01 every six hours during model integration. The domain configuration is shown in Fig. 1 (a).
The planetary boundary layer physics was reflected by the Asymmetric Convective Model version 2 (ACM2) (Pleim 2007). The Unified Noah Land Surface Model (Noah LSM) was responsible for land surface. These two schemes are essential in modeling 2 m temperature for providing a vertical profile of the turbulent kinetic energy and its dissipation rate, and the surface energy budget, respectively. Xie et al. (2012), who quantified the sensitivity of surface variables (2 m temperature and 10m wind speed) to four PBL schemes in the WRF (ACM2, YSU, MYJ, and Boulac) in the Pearl River Delta show ACM2 produces the best simulation of 2 m temperature and 10m wind speed compared to observations. The cloud microphysics was reflected by the WRF single-moment 3-class scheme (Hong et al. 2004). The Kain–Fritsch scheme was implemented in WRF D01 and D02 to parameterize the cumulus convections (Kain 2004). The Rapid Radiative Transfer Model for General Circulation Models was the radiation transfer scheme for both longwave and shortwave processes (Mlawer et al. 1997). This set of WRF physics schemes was also employed in many other studies focusing on the Pearl River Delta (Yeung et al. 2020; Bhautmage et al. 2022). The WRF output was configured to have an hourly frequency to effectively archive the diurnal cycle feature over the PRD.