A comparison of CMIP5 and CMIP6 climate model projections for hydrological impacts in China

The Fifth and Sixth Assessment Report (AR5 and AR6) of the Intergovernmental Panel on Climate Change (IPCC) (Stocker et al. 2013; Pörtner et al. 2022) both stated that the global temperature has a significant increase during the last century and the global warming increases the likelihood of severe, pervasive, and irreversible impacts, especially for the terrestrial hydrology. This is because the rising of air temperature increases the water holding capacity in the atmosphere and changes the spatial distribution of precipitation on the earth. In other words, the global water cycle process would be affected by the increase of air temperature (Syukuro 2019), which may result in more hydrological extremes such as waterlog and drought (Greene et al. 2009; Hirabayashi et al. 2013; AghaKouchak et al. 2014; Huang et al. 2017; Lin et al. 2020).

The quantification of climate change impacts on hydrology is usually achieved by driving hydrological models using global climate model (GCM) simulations with downscaling or bias correction techniques (Velázquez et al. 2013; Papadimitriou et al. 2016; Paulo et al. 2020). Various GCMs have been developed by different climate centers in many countries. However, due to different parameterizations of some unknown physical processes (Choi et al. 2017), large variability has been found related to climate projections in hydrological impact studies, especially for specific regions (Jiang et al. 2007; Her et al. 2019; Lehner et al. 2019; Zhou et al. 2020). A total of 41 GCMs in Coupled Model Comparison Project Phase 6 were evaluated in simulating monthly precipitation and found that the selection of appropriate models was a critical factor for climate change impact studies (Li et al. 2022). The multiple climate model ensemble (MME) is usually used to represent the variability related to climate models (Rasmussen et al. 2016; Ishizaki et al. 2017; Fathalli et al. 2019; Christiansen 2020).

Since the AR5 of IPCC, GCMs from Coupled Model Inter-comparison Project Phase 5 (CMIP5) were commonly used to provide climate projections for hydrological impact studies. For example, Papadimitriou et al. (2016) used five GCM simulations from CMIP5 to study the high-end climate change impact on European runoff and low flows and found an intensification of the water cycle at higher levels of warming. Asadieh & Krakauer (2017) investigated the impact of 21st century anthropogenic forcing under Representative Concentration Pathway 2.6 (RCP2.6) and RCP8.5 on streamflow extremes and found increased flow in the Arctic Ocean and reduced flow in subtropical arid areas under both forcing scenarios. Pokhrel et al. (2021) used an ensemble of hydrological simulations driven by four GCM simulations from CMIP5 and found that climate change could reduce terrestrial water storage in many regions globally. Yin et al. (2021) predicted the streamflow with 6 GCM simulations under RCP8.5 scenario for 151 catchments over the mainland of China and found that earlier flood timing and larger flood quantiles is likely to occur for most catchments in the future period than in the historical period.

More recently, the GCMs with higher spatial resolution and improvements in physical processes participating in CMIP6 have been available, which use the new shared socioeconomic pathway (SSP)/RCP-based emission scenarios for future simulations (Eyring et al. 2016; O'Neill et al. 2016). The variables of precipitation and temperature from these GCMs were used instead of those in CMIP5 archive for hydrological impact studies. For example, Cook et al. (2020) investigated changes in drought with the variables of precipitation, soil moisture, and runoff projected by 13 GCM simulations under 4 SSPs in CMIP6 and found that the drying in the mean state and risk of the historically most extreme events would increase. Bian et al. (2021) investigated variations in precipitation, runoff, and flood by using four GCMs in CMIP6 and found that the probability of floods would increase in the future over the upper Huai River basin. The evaluation of simulated runoff from 12 GCMs from CMIP6, 6 global hydrological models from the inter-sectoral impact model inter-comparison project and 3 land surface models from the global land data assimilation system show that projections of runoff changes based on these global outputs contain great uncertainties and should be interpreted with caution (Hou et al. 2022).

Since the climate models in CMIP6 provides a new opportunity to obtain a more credible understanding of climate change, it is necessary to compare the performance of GCM in CMIP6 and CMIP5 in simulating climate variables, especially for precipitation and temperature which are the most important variables for hydrological impact studies. Previous studies have focused on the performance of climate simulations, which compared the difference in climate simulations between CMIP5 and CMIP6 GCMs. For example, Lehner et al. (2020) found that simulated temperature and precipitation in CMIP6 showed greater model response variability than those in CMIP5, using 28 GCMs under RCP2.6, RCP4.5, and RCP8.5 from CMIP5 and 21 GCMs under SSP1-26, SSP2-45, SSP3-70, and SSP5-85 from CMIP6. Moreover, Zhang (2021) compared the uncertainties in projections of precipitation and temperature extremes between CMIP5 and CMIP6 archives at the global scale using 24 GCMs from each archive and found that more GCMs are needed to ensure the robustness of climate projections in CMIP6 than CMIP5. However, Fan et al. (2020) found that some individual CMIP6 models performed better than CMIP5 model in simulating summer days, tropical nights, cold spell duration, and diurnal temperature range in spatial pattern scores with 24 GCMs from CMIP6 and 28 GCMs from CMIP5.

GCMs in both CMIP5 and CMIP6 have been used to project climate change and its impact on hydrology in China. For example, the temporal distribution of precipitation was projected to become more uneven in the future than the historical period when using 20 Atmosphere-Ocean General Circulation Models (AOGCMs) in CMIP6 under four SSPs (Zhu et al. 2021). A larger flood is likely to occur in the future period than in the historical period in the mainland of China under the simulations of hydrological models when using GCMs from CMIP5 or CMIP6 (Bian et al. 2021; Yin et al. 2021). In addition, the duration and intensity of drought were projected to increase in some regions of China when using eight GCMs in CMIP6 under three SSPs (Yin et al. 2022).

Even though there were many studies comparing the performance of GCMs between CMIP6 and CMIP5 in simulating climate variables, the comparison of GCM simulations between CMIP6 and CMIP5 in future hydrological impacts is not found in the literatures, especially for the region of China. Given the transformation from climate to hydrology is highly nonlinear, this comparison is necessary. Thus, the objective of this study is to compare GCMs simulations between CMIP5 and CMIP6 in projecting future streamflow changes and their variability related to the choice of climate models. Specifically, changes and inter-model variability of mean flow and annual peak streamflow simulated by 2 hydrological models using bias corrected climate simulations (2 bias correction methods) from CMIP5 and CMIP6 as inputs are compared over 343 catchments in the mainland of China. The rest of the paper is organized as follows. Section 2 presents a brief introduction to the study area and datasets, and Section 3 presents the methodology, including bias correction methods, hydrological models, and data analysis methods. Section 4 presents the results, followed by the discussion and conclusion in Section 5.

This study used a gridded meteorological dataset (0.5° × 0.5°) over China for the period of 1961–2016 to represent observed data. This dataset contains three climate variables including daily precipitation, and daily minimum and maximum temperatures, which were downloaded from the China Meteorological Data Sharing Service System (http://www.cma.gov.cn/). The gridded dataset was generated from 2472 in situ observation gauge stations across China by the thin plate spline interpolation method and GTOPO30 (Global 30 Arc-Second Elevation) data sampling, which has been commonly used in many hydro-climatological studies in China (Xiao et al. 2013; Zhang et al. 2015; Fan et al. 2019; Zheng et al. 2020).

For the climate simulations, daily precipitation and temperatures (maximum and minimum temperatures) derived from the CMIP5 and CMIP6 archives (21 GCMs from CMIP5 and 21 GCMs from CMIP6) were selected (Table 1). The CMIP5 and CMIP6 outputs were downloaded from the Earth System Grid Federation. Even though they are not exactly from the same climate centers, most of them are from the same climate centers with advanced versions in CMIP6. These 42 GCMs provide the climate variables under the historical forcing for the reference period (1976–2005), RCP8.5 (CMIP5) and SSP5-85 (CMIP6) forcing for the 2071–2100 period. The use of a large number of GCM simulations is to adequately cover the inter-model variability.

Since GCM simulations are too biased to be directly used as inputs in hydrological models for impact studies, two bias correction methods were applied to climate model simulations before driving hydrological models. These two methods include a univariate method (daily bias correction method, DBC) and a multivariate method (two-stage quantile mapping, TSQM).

DBC (Chen et al. 2013) and TSQM (Guo et al. 2019) are both quantile mapping-based methods. They correct the distribution of daily precipitation and temperature with two steps. In the first step, the frequency of simulated precipitation occurrence is corrected by determining a precipitation threshold ensuring that the frequency of simulated precipitation is equal to that of the observed one for the reference period. In the second step, the wet-day precipitation amounts and temperature are corrected by a quantile mapping method using 100 percentiles from the 1st to 100th to represent their distribution. Both methods first correct precipitation and temperature individually, while TSQM further introduces inter-variable correlations between daily precipitation TSQM shows better performance in simulating the mean temperature for wet and dry months than DBC (Guo et al. 2019).

Two lumped hydrological models (HMETS and XAJ) with different structures and a number of parameters were employed for runoff simulations. HMETS (Hydrological Model of École de technologie supérieure) is a 21-parameter conceptual hydrological model, including four components that are snow accumulation and snowmelt module, evapotranspiration module, vertical water balance, and horizontal transportation (Martel et al. 2017). These four modules involve two connected reservoirs for representing unsaturated and saturated zones and captures basic hydrological processes. The snow accumulation and snowmelt module not only simulates snowmelt and refreezing processes, but also considers finer processes with 10 parameters in the snowmelt module. The actual evapotranspiration is estimated by a fraction (a parameter to be calibrated) of potential evapotranspiration. The vertical water balance module divides the effective rainfall into four sources of runoff (surface runoff, delayed runoff, hypodermic flow, and groundwater flow) based on two linear reservoirs with six free parameters. The first two runoff components which have four free parameters are routed through two gamma distributions in the horizontal transport.

XAJ (Xinanjiang) is a conceptual rainfall-runoff model with 15 parameters (Zhao Ren-Jun et al. 1980), which divides the watershed area into pervious and impervious areas and considers a 3-layer soil moisture store to simulate actual evapotranspiration. This model is physically based on the theory that runoff generation occurs since the repletion of storage, which means that runoff is not generated until the soil water deficit is replenished, that is impervious areas. All of the effective rainfall transfers to surface runoff in impervious areas. However, effective rainfall is separated into the surface runoff, interflow, and groundwater runoff in the previous area. Surface runoff routes to the outlet of catchments through the unit hydrograph. This model has been widely employed for streamflow simulation and flood forecasting, especially over the humid and semi-humid climate regions in China (Bárdossy et al. 2016; Li et al. 2017; Zhuo & Han 2017; Burek et al. 2020). In addition, the CemaNeige module (Valéry et al. 2014), which is used to simulate the snow accumulation and snowmelt processes in XAJ, has two parameters to be calibrated, divides precipitation into rainfall and snowfall, and calculates the snowmelt using a degree-day method.

The hydrological model parameters were calibrated using observed daily precipitation and potential evaporation, along with the daily streamflow time series. The potential evaporation was calculated by the Oudin equation (Oudin et al. 2005) based on temperatures. The observed data used for the calibration cover a length of 22–52 years depending on the length of the observed streamflow time series, and the odd years were used for calibration and the even years were used for validation. The optimal values of the free parameters were obtained by the algorithm of evolution strategy with covariance matrix adaptation (CMA-ES) (Merelo-Guervós & Castillo-Valdivieso 2004), which outperformed many other algorithms in calibrating hydrological models and it is especially efficient in handling higher-dimension parameter space (Arsenault et al. 2014). The Kling–Gupta efficiency (KGE) that considers the mean, variability, and dynamics of model errors was selected as the objective function for model calibration.

GCM simulations in CMIP5 and CMIP6 were compared in terms of multiple hydro-meteorological indices. Firstly, the mean value of daily mean temperature (tas, the mean value of daily maximum and minimum temperature) and mean annual precipitation (prcptot) were calculated using climate time series, and the mean daily streamflow (Q_mean) and annual maximum daily streamflow (Q_max) were then calculated for hydrological simulations. The mean values were calculated for both temperature and precipitation. Since the precipitation at the daily time scale consists of nonzero and zero values, the annual mean is used. Secondly, the change of hydro-meteorological indices between future and historical periods was calculated. Specifically, the absolute change (future-historical) was calculated for the temperature index, and relative change [(future-historical)/historical] was calculated for precipitation and streamflow indices. In addition, the inter-model variability represented by the standard deviation (std) of hydro-meteorological indices across multiple GCM simulations was calculated for all indices to reflect the variability related to the choice of a climate model simulation. Since the results might slightly differ when using different methods, due to uncertainty related to the choice of bias correction methods and hydrological models, the ensembles mean was used to better represent the results. To obtain robust results, the ensemble means across two bias correction methods and two hydrological models were computed for all indices when calculating their change and inter-model variability.

The increase in mean annual precipitation ranges between 6.3 and 34.4% for CMIP6 and ranges between 4.1 and 20.9% for CMIP5 with the mean value of 16.5 and 11.4% across all watersheds, respectively. Similarly, GCMs in CMIP6 project larger increases in mean annual precipitation by up to 16% than those in CMIP5 for most watersheds, especially for northern China (Figure 2(f)). Only in southwestern China, GCMs in CMIP5 project a slightly larger increase in mean annual precipitation than those in CMIP6.

For both historical and future periods, inter-model standard deviations of mean annual precipitation simulated by GCMs in CMIP6 is larger than those simulated by GCMs in CMIP5 for southern and northern China (Figure 4(e) and 4(f)). However, for central China, the inter-model standard deviations are larger for CMIP5 than CMIP6 with a mean difference of −13.8% for the historical period and −16.8% for the future period. The variation of mean annual precipitation values is shown in Appendix A in the Supplementary Material.

This study compares the GCM-simulated precipitation and temperature in CMIP5 and CMIP6 in terms of climate change and inter-model variability. More importantly, the change and inter-model variability of simulated mean and peak flows are also compared between these two archives after driving hydrological models using bias corrected climate model simulations. The results show that the daily mean temperature and mean annual precipitation are presented to increase in the future period, while GCMs in CMIP6 project a larger increase than those in CMIP5 for almost all watersheds. For hydrological simulations, an increase in mean and peak flows are observed for both CMIP5 and CMIP6, while the difference between CMIP5 and CMIP6 is very limited, especially for southern China. This implies that the change of precipitation dominates the change of streamflow for almost all watersheds in China, because the bias correction method used in this study does not alter the change of precipitation and temperature. Even though the increase in temperature may result in the rise of evapotranspiration, its contribution to streamflow is less important than that of precipitation. However, for most of the southern watersheds, the difference between CMIP5 and CMIP6 does not emerge, which means that the limited difference in the increase of precipitation is offset by the increase of evapotranspiration caused by the rise in temperature.

In terms of the inter-model variability, GCMs in CMIP6 project less variability than those in CMIP5 for most watersheds. This implies that the former is more reliable than the latter in the projection of temperature for the future period. Especially taking into account the fact that the inter-model variability is larger for CMIP6 than CMIP5 for the historical period. However, this is not the case for precipitation, as GCMs in CMIP6 project larger inter-model variability of precipitation for watersheds in southern and northeastern China for both historical and future periods. For central China, the inter-model variability of precipitation is larger for CMIP5 than CMIP6. However, different results are observed for mean and peak flows. For most watersheds in southern and northeastern China, GCMs in CMIP6 project less inter-model variability in both mean and peak flows than those in CMIP5, while an opposite pattern is observed in central China. Apparently, the streamflow results are contradictory to precipitation and temperature. This is because the bias correction method changes the inter-model variability of precipitation for the future period. Moreover, the various hydrological conditions between southeast and northwest China also result in different results. Affected by the East Asian monsoon, precipitation gradually decreases from the southeast coast to the northwest inland, which leads to drought in the north and moisture in the south.

This study uses the change and inter-model variability as the indices to compare the performance of GCM-simulated precipitation, temperature, and their hydrological impacts in CMIP5 and CMIP6. Following conclusions were drawn from this study.

• 1.

  Daily mean temperature, and mean annual precipitation are projected to increase from historical to a future period, while GCMs in CMIP6 project a larger increase than those in CMIP5 for almost all watersheds.

• 2.

  An increase in mean and peak flows are observed for both CMIP5 and CMIP6, while the difference between CMIP5 and CMIP6 is very limited, especially for southern China. The change in precipitation dominates the change of streamflow for almost all watersheds in China.

• 3.

  GCMs in CMIP6 project less variability than those in CMIP5 in temperature for most watersheds, which shows that the former is more reliable than the latter in projection temperature for the future period.

• 4.

  GCMs in CMIP6 project less inter-model variability in both mean and peak flows than those in CMIP5 for most watersheds in southern and northeastern China, while an opposite pattern is observed in central China.

Overall, this study indicates that GCMs in CMIP6 are more reliable than those in CMIP5 in simulating temperature. However, the reliability in projecting precipitation is dependent on regions. When applying a bias correction method to climate model simulations for hydrological impact studies, GCMs in CMIP6 are also more reliable in projecting future hydrological regimes.

This work was partially supported by the National Natural Science Foundation of China (Grant Nos. U2240201, 52079093), the Hubei Provincial Natural Science Foundation of China (Grant No. 2020CFA100), and the Overseas Expertise Introduction Project for Discipline Innovation (111 Project) funded by Ministry of Education and State Administration of Foreign Experts Affairs P.R. China (Grant B18037).

All relevant data are available from an online repository or repositories. The daily precipitation, maximum and minimum temperature of GCMs in CMIP6 is downloadable from http://esgf-node.llnl.gov/search/cmip6/, and that of GCMs in CMIP5 is downloadable from http://esgf-node.llnl.gov/search/cmip5/, and that of reference can be downloaded from http://data.cma.cn/).

The authors declare there is no conflict.