Hydro-climatic projection provides critical information for promoting disaster mitigation, sustainable urban planning, ecological analysis and water resource and agricultural management under future climate change. Such an operation has been relying on the incorporation of climate simulation from General Circulation Models (GCMs) into hydrological models so that the future hydro-climatic changes are quantified under greenhouse gas emission scenarios [1
]. Dating back to the 1950s, GCMs were initially developed to simulate the large-scale behavior of atmospheric circulation by Phillips [4
]. Since then, GCMs have undergone numerous evolutions from simply capturing realistic regional climate patterns by a few modelers during the 1960s–1970s to solving the major technical features of the climate by a group of international modelling communities in the late 1990s [5
]. These developments have made GCMs a crucial source of information for investigating the changing state of climate systems, particularly for their uses in the recently released sixth assessment report of the Intergovernment Panel in Climate Change (IPCC), providing the latest physical science basic of climate change for policy makers and the public [5
The Coupled Model Intercomparison Project (CMIP), a collaborative modelling framework was formed in 1995 to foster the management and improvement of GCMs. The early phases (1 and 2) of CMIP started with model experiments based on 18 GCMs, and the ensemble size has expanded to 25 GCMs in Phase 3 launched in 2005–2006. In 2010, CMIP Phase 5 was initiated and its GCM ensemble consisted of 40 GCMs. Currently, the latest CMIP Phase 6 (CMIP6) launched in 2013 includes the output of 55 GCMs [7
]. A major difference between CMIP6 and CMIP5 is the scenario of GHG emissions for future anthropogenic forcing of climate change, where CMIP5 projections are simulated based the radiative forcing pathways (RCPs) [9
], while the recent CMIP6 applies socio-economic pathways (SSPs) [10
]. In addition, the CMIP6 GCMs include models with finer spatial resolution, and models with enhanced representation of cloud microphysical and earth system processes [11
]. The significance of HighResMIP is that it allows investigation of the sensitivities of present-day climate representation and future climate projection solely to changes in GCM resolution. The CMIP6 GCMs have been extensively used to evaluate climate extremes such as precipitation, temperature and drought [8
]. To date, the use of the CMIP6 in hydrological studies is relatively rare, but such outputs are likely to increase dramatically in the near future [2
Tropical rivers, especially for those over Southeast Asia, are generally characterized by pronounced streamflow seasonality, with some having two flood peaks observed within a year due to the influence of monsoon-induced rainfall and equatorial convective rainfall [14
]. Jahandideh-Tehrani, et al. [15
] revealed that climate change has a significant impact on tropical streamflow, hence, the responses of tropical streamflow to climate change have been extensively investigated in tropical basins [16
]. As previously noted, GCMs are a major tool to provide precipitation and temperature information to facilitate assessment of potential future tropical streamflow changes. Therefore, application of the latest CMIP6 GCMs can provide new insights into possible tropical hydro-climatic changes under climate change.
Considerable biases have been found in GCMs with coarse resolution when they are used to simulate fields of regional precipitation and convection rainfall for hydrological modelling. For instance, Lebel [19
] found that GCMs with coarse resolutions (1.6° × 3.75°) exhibit biases in representing the seasonal precipitation in West Africa due to the poorly resolved mesoscale convective systems. To address the issue of the coarse GCM resolution, dynamic or statistical downscaling is needed to narrow down the gap between GCMs and hydrological model [3
]. As one of the CMIP6-endorsed MIPs, the recently released HighResMIP provides a suite of GCMs simulations with atmospheric horizontal grid spacing of ~25 km or finer [7
]. The ability to resolve small scale weather features is important for basin scale analysis, and means HighResMIP GCMs will likely be employed by hydrologists to study hydro-climatic changes around the world, including the tropical regions. Tan, et al. [20
] found that high-resolution HighResMIP models projected a higher flow rate during flood season than low-resolution models in the Kelantan River Basin, Malaysia. To date, there has been little investigation of about the role of HighResMIP models resolution on hydro-climatic projections. This study attempts to explore how different model resolutions could affect the tropical streamflow outputs.
The Johor River Basin (JRB) is a tropical river basin closely associated with freshwater supply for Malaysia. It is also a vital water source of Singapore as it provides up to 1157 million litres of freshwater per day [21
]. Historical record indicates a number of severe drought events in 1990, 1997, 2005, 2010, 2010–2014 and 2019, leading to water supply disruption in Johor [22
]. Floods also occur frequently in this region, for example, in 2006–2007 flooding caused 18 deaths, more than 100,000 people were evacuated and about USD 0.5 billion of losses were sustained [24
]. Thus, studying streamflow changes is important to help local authorities to mitigate the impact of these events. Tan, et al. [25
] reported high flows in the JRB will increase significantly, particularly under the RCP4.5 and RCP8.5 scenarios. In addition, the number of meteorological droughts over the JRB is expected to increase in the mid-21st century, based on the Coordinated Regional Climate Downscaling Experiments—Southeast Asia (CORDEX-SEA) projections [26
]. However, investigation of the streamflow changes under the latest SSP scenario is still lacking, so this study provides new insight on potential streamflow changes of the JRB.
The main purpose of this study is to evaluate future hydro-climatic changes of the JRB in the southern Peninsular Malaysia by incorporating the latest CMIP6 HighResMIP climate projections into the Soil and Water Assessment Tool (SWAT), focusing on uncertainty quantification of the streamflow outputs from different model ensemble sub-sets, e.g., all models (Ens_Mean), low resolution (Ens_LR, >1°), medium resolution (Ens_MR, 0.5° to 1°) and high resolution (Ens_HR, ≤0.5°) horizontal resolution models. SWAT is a widely used hydrological model in many parts of the world [27
], including Southeast Asia [29
]. A total of 10 HighResMIP GCMs under the SSP5-8.5 scenario were used to project the differences of hydro-climatic changes between historical (1985–2014) and future (2021–2050) periods. The findings can be used by local authorities to prepare climate adaptation and mitigation strategies based on the latest IPCC suggested socio-economic scenarios. Further, the research framework developed here could be readily duplicated and applied in other river basins, especially those located in the tropical regions.
The SWAT model exhibited a better performance (Figure 2
) during the period prior to reservoir management being introduced in 1993. Restriction of reservoir management data to the public or researchers is a common issue for many transboundary basins in developing and less developed countries [60
], and further information is needed to improve the SWAT modelling in the regulated flow period. This study demonstrates that adding some basic reservoir and water supply information along with reservoir parameter calibration improved the SWAT capability in regulated flow simulation period. Kim and B. Parajuli [61
] found that selection of a suitable reservoir outflow approach is important to accurately simulate the regulated streamflow. Zhang et al. [52
] incorporated water usage information into SWAT to better simulate regulated flow under the absence of detailed reservoir management information. However, the reservoir module within SWAT is relatively simple and it can be difficult to provide reliable estimation of reservoir release [62
]. Recently, reservoir operation functions have been integrated into the SWAT+ version, but this improvement is solely based on the operation of reservoirs within the United States. In future work, more comprehensive reservoir operational and parameter calibration modules need to be developed to cope with tropical conditions.
The analyses of KGE indicate that the SWAT model reasonably simulate both the unregulated and regulated flows (Figure 2
). Noted that the KGE values are higher than the NSE and R2
values, but they cannot be compared directly since there is no unique relationship between them [63
]. Additionally, high NSE values do not directly translate into high KGE values [63
]. In general, NSE analysis leans towards informing the model’s capability in high flows simulation due to the quadratic nature of the analysis [64
]. Figure 2
shows that the SWAT model underestimated some high flow periods during the regulated validation period. A lack of precipitation data in the upstream part of the JRB may lead to the model being unable to accurately capture the precipitation pattern, thus resulting in the underestimation.
In summary, Ens_HR has a slightly better ability to simulate the observed temperature data than the lower resolution model ensembles in the JRB. Consistent with the IPCC’s most recent assessment report [6
], the estimated warming of the JRB is likely to increase around 1 °C by the mid-21st century (Table 1
). We note that both the mean daily minimum and maximum temperatures are projected to increase at a similar rate, which is contrary with the observed historical trends in Malaysia. In fact, the observed rate of warming in minimum temperature actually increased about two-times greater than the maximum temperature from 1985 to 2018 over Malaysia as measured at multiple ground stations [30
The HighResMIP GCMs showed a better capability in capturing temperature than precipitation over the JRB, and similar findings have been reported in other CMIP6 models [65
]. The low-resolution models were unable to capture the seasonal precipitation cycle over the JRB accurately compared to the moderate- and high-resolution models. For example, FGOALS-L and HadGEM-LM underestimated precipitation during the southwest monsoon (May to September) and the second phase of the northeast monsoon (February to March) as shown in Figure 3
a. This shows systematic wet biases are still existing in CMIP6 models in this region, particularly in the low resolution HighResMIP GCMs.
In hydro-climatic modelling, climate models are invariably cited with a larger uncertainty than hydrological models, calibration parameter and bias correction technique [66
]. Streamflow and surface runoff are highly sensitive to changes in precipitation [69
]. High resolution climate models have demonstrated the ability to better represent the large-scale atmospheric circulation, tropical instability waves, storm tracks, and other circulation features [35
]. Mean monthly flows during the flood period (November to December) are projected to increase across all the ensembles, by 2.6 to 10.8% for Ens_HR and 0.2 to 15.6% for Ens_MR compared to −3.7 to 4.9% for Ens_LR (Figure 5
d). Ens_HR suggests a reduction of monthly flows during the southwest monsoon, but Ens_MR and Ens_LR project wetter conditions such that improved hydro-climatic assessment in tropical region will require greater confidence in precipitation projections in the dry season as well as the wet season.
The ability to adapt to future changes in precipitation, water availability and flooding requires hydro-climatic projections at a local scale. This study incorporated the latest CMIP6 HighResMIP GCMs into a calibrated SWAT model to project the future hydro-climatic changes over the JRB, Malaysia, and investigate the role played by the spatial resolution of GCMs. Model projections were classified into different groups of multi-GCM ensemble, including of all simulations (Ens_Mean), low resolution (Ens_LR), medium resolution (Ens_MR) and high resolution (Ens_HR) HighResMIP groups. Changes in hydro-climatic features for both the dry and wet seasons were evaluated between a baseline (1985–2014) and mid-21st century period (2021–2050) using a high level GHG emissions scenario.
In the JRB, SWAT presented a good performance in producing the observed hydrological features, which was presented in terms of both unregulated and regulated (before and after the Linggiu reservoir was constructed) monthly streamflow. The CN2 parameter related to surface runoff was found to be the most sensitive parameter in this basin. We note that addition of reservoir parameters during calibration and water supply information in the model setup could improve SWAT’s performance where reservoir management data is not available. The SWAT model tended to underestimate both the high and low flows in this basin, showing that more studies are needed to improve the SWAT modelling in regulated flow simulations in the absence of reservoir management data.
Assessment of different GCM resolutions in simulating historical climate is required both as part of model development and to help in selecting better models for impact assessment and decision making, e.g., IPCC reports [6
]. This spread in performance indicates that inter-model variability poses a significant challenge to accurate hydro-climatic impact assessment. Historical temperature is simulated somewhat better than precipitation by the HighResMIP GCMs over the JRB. The models simulated the primary precipitation peak one or two months earlier than observations. This bias was evident in the CMIP5 models and persists in the latest CMIP6 models, even in the high-resolution models. However, the moderate- and high-resolution models better capture the seasonal precipitation cycle compared to the low-resolution models. Our findings show that the low-resolution models had a larger uncertainty during the low rainfall period simulations as compared to high precipitation periods in this basin. In addition, the high-resolution models better represent minimum temperatures than the low-resolution models, though there is less spread in their ability to simulate maximum temperature simulations across different resolutions.
In the JRB, annual total precipitation is projected to increase by 0.4 to 3.1% across the variable resolution ensemble evaluated here. The high-resolution models, project increased precipitation during the flood period up to 13.4%, but a decrease of up to 11.9% was found during the dry period, indicating an amplification of the seasonal cycle. Annual mean daily minimum and maximum temperatures are projected to increase up to 0.9 °C and 1.1 °C, respectively, over the JRB. Associated with these changes in precipitation and temperature are projections of increases in annual streamflow and surface runoff from 0.9% to 7.0% and 7.0% to 20.6%, respectively. Similar to precipitation trend, monthly streamflow and surface runoff will likely be increased up to 15.6% and 48.3% during the flood period (November to January). The model experiments analyzed exhibit a certain degree of disagreement between high-resolution and low-resolution GCMs in the projected changes over during the dry period (February to March and May to September) in terms of precipitation, streamflow and surface runoff, while more consistent changes are seen during the wet seasons. The uncertainty is presented under a single GHG emission scenario due to the limited data availability of HighResMIP. Such an uncertainty is worth further investigation in future works using simulations of different SSP scenarios (SSP1-2.6, SSP2-4.5 and SSP3-7.0) and for a longer study period until the end of this century, though this can be strongly constrained by the limited computational resources, especially for high-resolution simulations.
Whilst this study has highlighted some limitations in the ability of state-of-the-art climate models to simulate the current climate in the JRB, it has shown that the models, particularly those with higher resolution, are able to reproduce the key hydrological features of the current climate. When those model outputs are used as inputs to a hydrological model and combined with future climate projections, they show some consistent patterns of hydro-climatic change which provide important information for the government and key stakeholders in the region in the context of adaptation. In particular, this study highlights an analysis framework that could be readily reproduced in other basins and extended to include other global and regional climate models.