2.1. CORDEX-SEA phase 2 simulations

The 5 km × 5 km downscaling simulations over Peninsular Malaysia conducted under the CORDEX-SEA phase 2 programme (Figure 1) were carried out by downscaling the three best performing 25 km × 25 km RegCM4 downscaled simulations from the first phase of SEACLID/CORDEX-SEA programme (Tangang et al., 2020). These simulations are the runs driven by EC-Earth, HadGEM2-ES and MPI-ESM-MR.

The regional climate model (RCM) used in the 5 km downscaling simulation is the newer non-hydrostatic RegCM4.7 (RegCM4.7-NH) (Coppola et al., 2021) developed by International Centre for Theoretical Physics (ICTP). The conducted 5 km downscaling simulations cover the historical period of 1971–2005 and three future time slices, i.e., the early (EC; 2012–2039), mid (MC; 2041–2069) and late 21^st century (LC; 2071–2099) considering both RCP 4.5 and RCP 8.5 future scenarios (Table 1). Initial and boundary conditions (ICBC) for the 5 km simulations were interpolated from the 6-hourly output of the first phase CORDEX-SEA 25 km simulations.

The RCM and the model physical settings used in the CORDEX SEA phase 2 simulations and their forcing 25 km CORDEX-SEA phase 1 simulation are summarised in Table 2.

2.2. Model performance and projected future precipitation extreme analysis conducted

The model performances were first analysed by comparing the simulated precipitation climatology to that of reference data to quantify the model’s biases. In this assessment, the 25 km simulations will also be included to visualise the improvement the 5 km simulations would have over their respective simulation forcings. Aside from model biases, the annual precipitation cycle simulated by the 5 km simulations, the 25 km forcing simulations and observed by reference data will also be compared. The reference data of choice in this study is that of Climate Hazards Group InfraRed Precipitation with station data 2.0 at 5 km resolution (CHIRPS05) (Funk et al., 2015), which has been shown to perform reasonably well over the study area (Ayoub, Tangang, Juneng, Tan, & Chung, 2020). Another reason CHIRPS05 data was chosen as the preferred reference data is that its 5 km resolution matches that of the CORDEX-SEA phase 2 simulations.

Analysis of future projections of precipitation extremes over Peninsular Malaysia was then carried out by evaluating future changes for near, middle and end of the 21^st century for each of the emission scenarios using an ensemble created from all the 5 km simulations following the approach used by Supari et al. (2020) and Tangang et al. (2020). Under this approach, all ensemble members will have an equal weightage in the ensemble created. This analysis used the extreme climate indices based on the Expert Team on Climate Change Detection and Indices (ETCCDI) (Zhang et al., 2011). Table 3 indicates the list of these ETCCDI indices used. Maps of projected changes in extreme precipitation relative to the historical period are made.

3.1. Model evaluation and comparison between CORDEX-SEA phase 1 and 2 simulations

This section describes the performances of CORDEX-SEA phase 2 5 km simulations compared to those of CORDEX-SEA phase 1 25 km in simulating seasonal precipitation over Peninsular Malaysia. Figure 2 shows the biases of 25 km and 5 km simulations for DJF (wet season) and JJA (dry season) over Peninsular Malaysia during the historical period.

The comparison between precipitation biases in the 5 km and 25 km simulations shows that generally, for both seasons, the biases have been significantly reduced in the 5 km simulations compared to that of the 25 km. This reduction in biases in the 5 km simulations can be attributed to the newer combination of model physics schemes used: the cumulus parameterisation scheme (CPS) was changed from MIT-Emanuel to that of MIT-Emanuel over land and Kain-Fritsch over the ocean, and land surface scheme from Biosphere–Atmosphere Transfer Scheme version 1e (BATS1e) to that of Community Land Model version 4.5. The change in the model physics schemes was made because the CPS MIT-Emanuel and land surface scheme BATS1e combination used in the CORDEX-SEA phase 1 tend to cause the RegCM4 to consistently produce strong precipitation in the Southeast Asia region (Chung, Juneng, Tangang, & Jamaluddin, 2018). While general improvements have been made in the 5 km simulations after replacing the model physics schemes combination, wet biases can still be seen along the coast and the mountainous areas. These imperfections can be attributed to these in the CHIRPS05 itself.

Figure 3 depicts the observed and simulated precipitation annual cycle by CHIRPS05, 5 km and 25 km simulations over two sub-regions in Peninsular Malaysia. The two sub-regions, R1: west coast of Peninsular Malaysia, and R2: east coast of Peninsular Malaysia, were divided based on the distinct precipitation annual cycle pattern shown (Wong et al., 2016). Whereby the precipitation annual cycle pattern over R1 shows a bimodal pattern while that of R2 shows a unimodal pattern, as demonstrated by CHIRPS05 (Figure 3).

From the plots shown in Figure 3, it can be seen that both the 25 km and 5 km simulations accurately captured the bimodal precipitation annual cycle of sub-region R1. Of course, due to the 5 km simulations having much smaller precipitation biases than 25 km, the amplitude of the annual cycle simulated matches CHIRPS05 better. The most significant discrepancy between the 25 km and 5 km simulations in reproducing the annual precipitation cycle lies in the R2 sub-region. It can be seen that all the 25 km simulations could not faithfully capture the unimodal precipitation annual cycle pattern in this region, unlike that of the 5 km simulations. Thus, by increasing the model horizontal resolution from 25 km to 5 km, the originally missing east-west precipitation contrast of Peninsular Malaysia can be correctly simulated by the model. This could be due to the reason that the 5 km simulations can more faithfully represent the complex mountainous range that separates Peninsular Malaysia into east and west coasts than the 25 km simulations. Hence, the comparison suggests that increasing the horizontal resolution from 25 km to 5 km can improve the precipitation simulations, consistent with other studies (Kim, Kim, & Cha, 2022; Ngai et al., 2020; Tuju, Ferrari, Casciaro, & Mazzino, 2022).

3.2. Projected changes in mean and extreme precipitation

The changes in mean and extreme precipitation (which includes CDD, CWD, R20mm and Rx1day) over Peninsular Malaysia projected by the ensemble of ECE05, Had05 and MPI05 for each of the 21^st century future periods were plotted and shown as Figures 4 to 8. Figure 4 shows the projected mean DJF and JJA precipitation changes under RCP 4.5 and RCP 8.5 for EC, MC and LC periods.

In terms of seasonal mean precipitation (both DJF and JJA), the projections show that there will be a general tendency for gradual dryness over Peninsular Malaysia in the coming future (Figure 4). This dryness tendency will occur under RCP 4.5 and RCP 8.5 futures, which is stronger under the RCP 8.5 scenario. Such results are consistent with those of Kang, Im, and Eltahir (2019) and Tangang et al. (2020). In the EC period, decreases in mean JJA rainfall can be up to 10–15%, especially under RCP 8.5. The level of dryness appears to be higher over the northwestern region of Peninsular Malaysia. During DJF, however, some areas are projected to have increased rainfall for RCP 4.5. Increased mean rainfall is also projected in southern and interior parts of Peninsular Malaysia during the MC period, including RCP 8.5. By the LC period, for both RCPs, the mean rainfall is projected to show a decreasing pattern.

The drying tendency of Peninsular Malaysia in the RCP 4.5 and RCP 8.5 futures is also reflected in the precipitation extremes (Figures 5–8). Alongside the decrease in mean precipitation, an increase in the future CDD is projected (Figure 5). This indicates that the future periods will have longer dry days, especially by the 21^st-century MC and LC periods. These projected CDDs are consistent with those of Supari et al. (2020). This general increase in CDD in the future matches well with the projected decrease in CWD (Figure 6). This reduction in CWD is pronounced over the southern region of Peninsular Malaysia during the JJA season in the LC period of RCP 8.5. Consequently, this increase in CDD while a decrease in CWD explains the decrease in the mean precipitation. Nevertheless, a slight increase in CWD during the EC period can also be seen for the DJF season under RCP 4.5.

 

Similar to CDD and CWD, the projected changes in R20mm show a general reduction in frequency in the future periods of both RCP 4.5 and RCP 8.5 futures when compared with the historical periods, particularly during MC and LC of the 21^st century (Figure 7). This indicates that Peninsular Malaysia, especially the west coast, will be experiencing fewer days with very heavy precipitation.

Unlike other indices, the intensity index, Rx1day, shows mixed projected changes in future periods. In the EC period, increased intensity of extreme precipitation during DJF is projected under the RCP 4.5 scenario. A general increase in Rx1day can be seen throughout Peninsular Malaysia. This increase of Rx1day slowly reduced as the 21^st century progressed. By the MC period, this increase of Rx1day will only be confined to the southern region of Peninsular Malaysia, and by the LC period, Rx1day generally shows a decreasing pattern. For the RCP 8.5 scenario, this increasing-decreasing tendency of Rx1day during DJF is relatively consistent through the 21^st century, where a decreasing tendency will generally be found over the northern region of Peninsular Malaysia, while over the southern region of Peninsular Malaysia, the Rx1day typically showing an increasing tendency. Unlike DJF, JJA Rx1day showed a gradual decreasing tendency throughout the century, especially over the west coast of Peninsular Malaysia. This decrease in Rx1day appeared stronger under the RCP 8.5 scenario than that of RCP 4.5.