2.1 Experimental design

The simulations were carried out in one nested horizontal domain in this study. The domain covered the Southeast Asia region, with 1-hour temporal resolution and 30 km x 30 km spatial resolution. The year 2013 was a neutral year for the El Niño–Southern Oscillation (ENSO) and was selected as the base year of the present-day simulation, and the future projection was for the years 2030, 2050, 2070, and 2100. The simulation was carried out for January and July of the selected years. For January, the projection started at 0000 UTC on the first of January and ended at 0000 UTC on the first of February. While for July, the projection time began at 0000 UTC on the first of July and ended at 0000 UTC on the first of August.

2.2 Dataset input: Initial boundary conditions

This study applied two time-dependent meteorological fields as initial and boundary conditions for the WRF model. For the present-day simulation, the time-dependent meteorological field was obtained from the global NCEP FNL (https://rda.ucar.edu/datasets/ds083.2), as mentioned. The NCEP FNL dataset consists of surface information with 26 mandatory levels (1000 millibars–10 millibars) of the surface boundary level. The meteorological parameters include the surface temperature, sea surface temperature, sea-level pressure, geopotential height, relative humidity, ice cover, vertical motion, vorticity, and winds (National Centers for Environmental Prediction, 1994). While for the future simulations, the NCAR’s Community Earth System Model (CESM) from global bias-corrected climate model output datasets (Hurrell et al., 2013) was used and was obtained from (https://rda.ucar.edu/datasets/ds316.1). These future initial and boundary condition information were then downscaled into 30 km x 30 km resolution to fit for regional projection using the WRF model. In support of Coupled Model Intercomparison Experiment Phase 5 (CMIP5) (Taylor, Stouffer, & Meehl, 2012) and the Intergovernmental Panel on Climate Change Fifth Assessment Report (Flato et al., 2013), the CESM simulations, therefore, were utilized to produce future-day simulation, which has a better agreement in simulating temperature and precipitation globally compared with real-time observations (Knutti, Masson, & Gettelman, 2013).

2.3 Climate change scenarios

This study utilizes the climate change scenario that the IPCC developed in their Fifth Assessment Report (Flato et al., 2013). The climate change scenario provides more comprehensive external forcing scenarios and higher resolution than CMIP3 from IPCC AR4 (Knutti & Sedlacek, 2013). In this study, only two climate scenarios were used and discussed, namely the RCP4.5 and RCP8.5. The RCPs are input for climate and atmospheric chemistry modelling that documents the emissions, concentrations, and land-cover change projections. The four RCPs (RCP2.6, 4.5, 6.0, and 8.5) reflect the year 2100 greenhouse gas radiative forcing values from 2.6 to 8.5 Wm^-2 (Nazarenko et al., 2015). The RCPs include the lowest forcing level scenario RCP2.6 (Vuuren et al., 2011), two median range or stabilization scenarios RCP4.5 (Thomson et al., 2011) and RCP6.0 (Masui et al., 2011), and the high-end or business-as-usual scenario RCP8.5 (Riahi et al., 2011). The future scenario of RCP4.5 applied in this study is a low-to-moderate emission scenario, where the greenhouse gases (GHGs) radiative forcing will reach 4.5 Wm^-2 by the year 2100 (Thomson et al., 2011). It represents a scenario where various adaptive policies have been applied to limit the radiative forcing. At the same time, RCP8.5 indicates a higher emission scenario with GHGs radiative forcing that will reach 8.5 Wm^-2 by the year 2100 (Riahi et al., 2011).

2.4 Climate model evaluation

Most researchers apply dynamical downscaling by RCMs from GCMs due to the uncertainties embedded within the GCMs, especially when resolving complex terrain (IPCC, 2001). Moreover, the error of GCMs with large-scale resolution can be transmitted to the RCM (Noguer, Jones, & Murphy, 1998). As a result, the validation or evaluation process is necessary for RCM, downscaled from GCMs before using it for the climate projection. The evaluation was carried out with observation datasets from the Climate Research Unit (CRU), University of East Anglia. The dataset spaced at 0.5° × 0.5° resolution contained with full-set monthly mean surface climate (Harris, Jones, Osborn, & Lister, 2010). Since the spatial resolution and grid location between model and observation datasets are different, they are bilinearly interpolated into the same 30 km x 30 km grid covering the study area. To evaluate the model’s performance, we used averages of only inland data from observation and model and used the following statistical techniques.

2.1

During the mid-century, there were high-temperature change anomalies projected over the sub-continent region in mainland SEA ranging from 0.88°C to 4.71°C and -3.32°C to 5.74°C under RCP4.5 and RCP8.5, respectively. Meanwhile, the maritime region would experience warmer temperature changes from 0.41°C to 1.16°C and 0.38°C to 0.98°C under RCP4.5 and RCP8.5, respectively. The simulation results also showed a higher temperature increment over the sub-continent region as compared to the earlier finding by Raghavan, Hur, and Liong (2018), which reported increasing between 0.8°C and 1.4°C under both RCP4.5 and RCP8.5 scenarios. However, an earlier study by (Gasparrini et al., 2017) has projected a higher increment of between 3.5°C and 4.5°C at the end of the century for this region. Owing to the uncertainty associated with anticipated temperature changes across climate models and extrapolated climate–response relationships, the estimates of the net surface temperature change over this region might be hampered by poor precision, particularly in places projected to undergo a significant shift in temperature (Almazroui et al., 2020; Dieng et al., 2022) .

3.2 Total precipitation

Under RCP4.5, the simulated monthly mean total precipitation in the SEA region for January 2013, 2030, 2050, 2070, and 2100 were 12.57 mm, 80.30 mm, 13.87 mm, 155.63 mm and 135.66 mm, respectively. In July, the mean surface temperatures were 100.53 mm, 90.40 mm, 169.94 mm, 120.23 mm and 90.76 mm, respectively (Table 3). Precipitation analyses under the RCP4.5 scenario showed that the total precipitation of July was higher than January in 2030. The total precipitation projection under RCP4.5 showed changes in the monthly mean of 1.29 mm to 123.08 mm during the January period and between 69.40 mm to -9.76 mm for the July period during the mid-and end-century, respectively. The spatial distributions of projected total precipitation under RCP4.5 are shown in Figures 5 and 6. The results show that total precipitation increments are concentrated over the insular region during January, with pronounced increments after the mid-century. Yet the total precipitation shifted to show an increment over the mainland SEA region during the July period despite having the signal reduced toward the end of the century. As depicted in Figure 5, Indonesia showed the highest total precipitation changes from 41.3 4 mm–284.24 mm, followed by Malaysia with 6.73 mm–225.24 mm and the Philippines from 3.08 mm–184.22 mm during the mid-century and end-of-century periods, respectively. The mainland SEA region only showed a significant increment at the end of the century, with Myanmar setting the highest total precipitation with 242.24 mm, followed by Cambodia (269.45 mm), Laos (160.81 mm) and Vietnam (146.42 mm).

In July, the total precipitation was simulated to increase over the whole domain area relative to the baseline period. Toward the end of the century, mainland SEA was expected to have a higher amount of precipitation than the insular region during the July period. The entire area receives the most increased total precipitation in 2050, with an increment of 69.40 during the mid-century and a reduction by 9.76 mm at the end of the century. The highest precipitation was observed over Myanmar, with monthly precipitation during July ranging between 131.91 mm–277.39 mm before decreasing to 147.15 mm–177.94 mm at the end of the century. Meanwhile, the rest of the mainland SEA region showed an increment of monthly total precipitation between 62.87 mm–202.92 mm before decreasing by 62.87 mm–105.86 mm at the end of the century (Figure 6). The seasonal total precipitation under the RCP4.5 simulation, particularly during January, was highly varied, especially over Malaysia, Indonesia and the Philippines. The high rainfall over these regions might be caused by the monsoonal winds that transport moisture from the South China Sea and enhance convergence, as discussed in (Tangang, Chung, & Juneng, 2020). Moreover, the high total rainfall, particularly over Malaysia and Indonesia, may also be influenced by the existence of the synoptic Borneo Vortex (Chen, Yen, & Matsumoto, 2013) and the moving oscillation of the Indian Ocean as known as Madden Jullian Oscillation (MJO) (Saragih, Fajarianti, & Winarso, 2018). The MJO and Borneo Vortex phenomena cause this area to become an active area of deep convection associated with heavy rainfall, especially during the wet season.

Under the RCP8.5 scenario, the monthly total precipitation in January for SEA was generally lower toward the end of the century compared with the RCP4.5 scenario. The mean total precipitation for January of 2013, 2030, 2050, 2070 and 2100 was projected at 92.98 mm, 169.68 mm, 84.82 mm, 16.10 mm, and 118.79 mm, respectively. While in July, the monthly total precipitation for the respective years was larger than under RCP4.5 simulation but showed a decreasing trend toward the end of the century with total simulated values of 248.86 mm, 258.96 mm, 245.77 mm, 218.25 mm and 92.42 mm, respectively (Table 4). Toward the end of the century, there was no significant increment of total precipitation over mainland SEA during January. However, from spatial distributions of projected total precipitation under RCP8.5 (Figure 7), there was considerable increment over the insular region, particularly toward the mid-century, with the highest increment of total precipitation observed in Indonesia (from 319.33 mm to 332.53 mm) followed by Malaysia (from 215.73 mm to 238.16 mm) and the Philippines (from 135.74 mm to 158.81 mm). A slightly lower marginal increment was observed over the same region toward the end of the century, with total precipitation of 256.67 mm, 51.57 mm, and 193.41 mm for Indonesia, Malaysia and the Philippines.

The analyses of total precipitation during the July period (Figure 8) revealed a significant increment of simulated precipitation over mainland SEA despite having the signal reduced toward the end of the century. Myanmar received the highest amount of rainfall with 388.71 mm in 2013, 294.04 mm in mid-century and 85.92 mm at the end of the century. Other regions in mainland SEA showed total precipitation ranging between 209.61 mm–256.06 mm in 2013, 236.46 mm–295.16 mm during mid-century and between 47.38 mm–105.56 mm at the end of the century. Although the insular region follows the same pattern, the simulated total precipitation over the Philippines was an all-time high with 316.57 mm in 2013, 273.27 mm in mid-century and 126.11 mm at the end of the century. In contrast, Indonesia and Malaysia are projected to receive a higher rainfall rate during July by the middle of 91.93 mm and 201.25 mm, but a lower rate at the end of the century, with total rainfall of 9.86 mm and 38.08 mm, respectively. The average surface temperature and total precipitation are recorded as higher anomalies under RCP8.5. These results suggest that significant changes, particularly in the surface temperature and precipitation, could potentially increase this region’s climate-related risks and vulnerability.

The simulated total precipitation in this study was significantly different from the finding of (Tangang et al., 2020) using the CORDEX-SEA multi-model simulation. Their study concluded that the robust significance of total precipitation reduction between 10-30% over maritime sub-continent regions, particularly over the Indonesia region during the dry season (June–August) under RCP4.5 and RCP8.5 by the middle and end of the century. However, during the wet season (December–February), there was a robust increment of 10–20% of the total precipitation observed over Indonesia and a 10–20% reduction over the Philippines in both RCPs. (Tangang et al., 2020) further suggested that the mainland SEA region would experience a 10-15% increment in the total precipitation under both RCPs towards the end of the century except for Vietnam and Cambodia. The higher tendency of drying observed could be associated with enhanced divergence and subsidence effect over the Maritime Continent (Giorgi, Raffaele, & Coppola, 2019) caused by the deep tropical squeeze resulting from the equatorward contraction of the rising branch of the Hadley Circulation as the climate continues to warm (Fu, 2015).

3.3 Climate model evaluation

Table 5 shows the evaluation values of the surface temperature from the WRF model under RCP4.5 and RCP8.5 scenarios relative to the CRU observation dataset. The simulated surface temperature was 25.81°C in January and 27.41°C in July, under RCP4.5. Meanwhile, the RCP8.5 simulation showed a slightly lower temperature of 25.57°C in January and a higher temperature of 27.27°C in July compared with the CRU observation dataset. The model has a lower than 1% bias during January but a higher bias during July with 5.38% and 5.77% for both RCP4.5 and RCP8.5 scenarios, respectively. Meanwhile, the values of FB and NMSE were insignificant. The model has the Fa2 value of 1.0 during the January simulation, indicating an ideal simulation but a slightly higher value during the July simulation with a value of 1.1, indicating a slight overprediction within a factor of two of the observed values. The warm bias over this region could be due to a poor representation of land-atmosphere interactions, amplified by unresolved albedo feedback and further alleviated by large cloud biases (Garrido, González-Rouco, & Vivanco, 2020).

Table 6 shows the evaluation values of the total precipitation simulation under both RCPs scenarios relative to the CRU observation dataset. The simulated total precipitations were 29.6 mm and 117.3 mm in January and 159.6 mm and 288.6 mm in July for both RCPs, respectively. There were larger biases for both RCPs, which underestimated the total precipitation during January but overestimated during July. During July, under the RCP4.5 scenario, the values of FB and NMSE were lower than 0.5. Though the model has poor performance in simulating total precipitation during January based on the Fa2 value, it relatively performed well during July with a Fa2 value of 1.3. A similar observation by Kong and Sentian (2015) reported a high bias, especially in the mountainous area and interior region. The larger bias might be caused by poor representation of the convective parameterization and hydrological cycle by the model (Alves & Marengo, 2009; Salimun, Tangang, & Juneng, 2010; Sinha et al., 2013).