Evaluation of future runoff variations in the north–south transect of eastern China: effects of CMIP5 models outputs uncertainty

Runoff, an important water flux, could pose great threats to water security due to changes of its timing, magnitude, and seasonality (Taormina et al. 2015; Berghuijs et al. 2017; Sun et al. 2017; Fotovatikhah et al. 2018). It is difficult and challenging to estimate the range of uncertainty of runoff prediction. Climate change is considered as a key factor to impact variations of runoff (Sankarasubramanian et al. 2001; Chiew et al. 2009; Alkama et al. 2010; Roderick & Farquhar 2011; Wu & Chau 2011; Wang et al. 2013). However, climate change has become increasingly significant in the past decades, and there are many uncertainties about the future variations of climate change. Thus, investigating future runoff change in response to climate change could supply valuable insights into the impacts of climate change on water resources and the hydrological cycle (Cheng & Chau 2004; Maurer & Duffy 2005; Winter et al. 2015; Yang et al. 2018; Zhang et al. 2018).

A wealth of research has been extensively devoted to explore the impact of climate change on the variations of runoff in recent years, both at the regional level and at the global level. It is considered to be one way to capture the variability of the climate change responses and to estimate the uncertainties using the results of several climate simulations. For example, Etchevers et al. (2002) employed six climate scenarios to compare the variations of the total annual runoff to those with the present climate. They found that the increasing temperature may cause the decreasing runoff (−11%), while the increasing temperature and precipitation indeed produced the increasing runoff (39%). They also pointed out that the uncertainties were very difficult to evaluate. Gordon & Famiglietti (2004) used the Canadian Centre for Climate (CCC) Modeling and Analysis (CGCM1 version) and the Hadley Centre for Climate Prediction and Research of the Meteorological Office of the United Kingdom (HADCM2 version) coupled ocean–atmosphere general circulation models (GCMs), and four land surface models (Biome-BioGeochemical Cycles (Biome-BGC), Century, Lund-Potsdam-Jena (LPJ), and MAPSS-CENTURY (MC1)) to predict the future runoff. They found that nearly all models in all watersheds examined produced statistically significant increases in runoff. However, there still were great uncertainties between model and scenarios. In a few watersheds, both positive and negative trends were documented from one GCM scenario. For a comprehensive assessment and to better capture uncertainty, a multi-model approach was taken by Hovenga et al. (2016). Climate change projections that referred to the Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report (AR4) with HadCM3 (HADCM3), IPSL-CM4 (IPCM4), and ECHAM5-OM (MPEH5) (Semenov & Stratonovitch 2010) were used to evaluate monthly average runoff. They indicated that output in terms of runoff and sediment loading showed large distinctions between GCMs when incorporating climate change into the Soil and Water Assessment Tool (SWAT) model. Although the above results indicate that there are huge differences regarding the estimation of runoff, the range of uncertainties of runoff fails to be estimated using more GCMs. Therefore, in this study, the range of uncertainty of estimates of the runoff on the NSTEC was explored based on multiple driving data sets from multiple GCMs.

In this study, the area of the NSTEC was chosen as the study region. It ranges from longitude 108° to 118° E at latitudes less than 40° N and from longitude 118° to 128° E at latitudes equal to or greater than 40° N. The runoff is difficult to simulate and predict in the region because the climate condition is influenced by the East Asian monsoon. Extreme weather and climate events occur frequently in this study. Thus, the response of runoff to future climate change must be demonstrated to better understand the change in the regional terrestrial hydrological cycle.

The original LPJ model (‘LPJ version 1’) is employed to explore the uncertainty of future runoff. The LPJ model is developed to simulate natural vegetations and their carbon dynamics and stock in a process-based manner, spatially explicit (0.5°) and at high temporal resolution (daily), under given climate and land-use patterns based on the BIOME family of biogeographical equilibrium models (Prentice et al. 1992; Sitch et al. 2003). The model includes ten plant functional types (PFTs) used to distinguish different photosynthetic (C3 vs. C4), phenological (deciduous vs. evergreen), and physiognomic (tree vs. grass) features. A more important fact is that the model can describe the hydrological cycle (e.g., evaporation, transpiration, interception, runoff, and discharge), which is coupled to vegetation dynamics, and interacts with the vegetation dynamics. For each grid cell, the annual runoff is affected by the precipitation and temperature, which control the annual actual evapotranspiration. The LPJ model is widely used to discuss the regional terrestrial hydrological cycle. For example, Konzmann et al. (2013) quantified global variations in irrigation demand using climate projections from 19 different GCMs up to the 2080s and the LPJ model. Giuntoli et al. (2018) estimated the uncertainties in projected runoff over the conterminous United States using the LPJ model. These studies imply that the LPJ model can be used to study the impact of climate change on the future runoff.

The LPJ model is driven by monthly precipitation, temperature, wet day frequency, and cloud cover at a 0.5° pixel size. To evaluate the uncertainty of future runoff, climate projections from ten GCMs with higher resolution compared to other GCMs of the Coupled Model Intercomparison Project 5 (CMIP5) from the Representative Concentration Pathway (RCP) 4.5 scenario were used as forcing data for LPJ during 2011–2100 (Table 1; IPCC 2013). The data were spatially interpolated to a 0.5° resolution and bias-corrected (based on 1961–1990 bias) with CRU TS2.1 climate data set (Mitchell & Jones 2005). Additionally, a data set of atmospheric CO₂ concentrations from RCP4.5 during 2011–2100 was also essential (IPCC 2013). Soil texture data were based on the Food and Agriculture Organization (FAO) soil data set (Zobler 1986).

The LPJ model is run over 1,000 years, repeating years 1901–1930 of the CRU TS2.1 climate data set, followed by transient runs for all available climate trajectories for 1901–2000. The predicted ecosystem state was the new initial condition used to run the ten GCMs for 2011–2100 under RCP4.5 scenario. Grid cells with vegetation cover below 10% were considered deserts (Heyder et al. 2011). Finally, model runs with daily output were performed for selected grid cells to demonstrate in more detail the effects of processes underlying the changes in runoff.

The regional pattern of average present runoff (Figure 1) shows that high values occur in Southern China (>600 mm) characterized by high precipitation. In contrast to Southern China, medium or low values (<400 mm) occur in Northern and Northeastern China with low precipitation and high temperature and evapotranspiration, respectively. The regional averaged runoff is 228.20 mm during 1971–2000.

The spatial character of runoff for 2011–2100 is shown in Figure 2, with the amounts of runoff showing a consistent pattern for the different scenarios. For all climate change scenarios, the runoff is high at low latitudes and low at high latitudes. However, the estimated amounts of runoff for the NSTEC for all climate change scenarios were different. The total runoff projected for the NSTEC ranged from 268.9 mm (MRI-CGCM3) to 544.2 mm (MIROC5) for all climate change scenarios that originated from the output of the ten GCMs. These different features were clear for the zonal mean of soil carbon in medium and low latitudes, i.e., from 20° to 30° N (Figure 3). This implies that there are great uncertainties about the prediction of runoff at medium and low latitudes. At high latitudes, the predictions of future runoff show coincident results for all GCMs.

The interannual variations of annual runoff projected for the NSTEC for all climate change scenarios were also evaluated, as shown in Figure 4. All numerical results show that the future runoff will increase. From the above modeling results, annual runoff increases in 2011–2100 compared with that for 1971–2000. However, interannual variations in the soil carbon estimates were different during 2011–2100 for all climate change scenarios. The increasing trend of the future runoff for HadGEM2-AO (1.51 mm year⁻¹, R² = 0.4692) is larger than those for other GCMs. However, the increasing trend of the future runoff for Bcc-csm1-1 (0.52 mm year⁻¹, R² = 0.1495) is smaller than those for other GCMs (Table 2). All estimations about the future runoff also suggest that the decadal growth is consistent during the study period for all GCMs (Table 3). For example, the runoff is 504.43, 560.19, and 567.91 mm year⁻¹ in 2011–2040, 2041–2070, and 2071–2100, respectively, for MIROC5. There are obvious characteristics for different seasons regarding the future runoff (Figures 5–8). In spring and summer, the variations of future runoff play an important role in annual runoff. In these two seasons, the range of uncertainties of future runoff is large in Southern China, while it is small in Northern China and Northeastern China. However, the variations of future runoff are weak in autumn and winter. From the above results, the uncertainties of predictions of future runoff mainly may come from the uncertainty of runoff forecast in spring and summer in Southern China.

During the hydrological cycle, the precipitation and the actual evapotranspiration, which is controlled by the temperature, are two main factors that affect the variation of runoff. In this section, the variations of precipitation, temperature, and actual evapotranspiration are explored. From 2011 to 2100, the temperature and the precipitation show a slight increase for all GCMs (Figure 9). The uncertainties range of prediction of temperature is from 6.99 °C (FGOALS-g2) to 11.96 °C (MIROC5), and that of precipitation is from 702.09 mm (MRI-CGCM3) to 1,228.60 (MIROC5) mm in the whole study area (Table 4). The uncertainties of temperature cause those of actual evapotranspiration. The uncertainty range of actual evapotranspiration is from 466.92 mm year⁻¹ (MRI-CGCM3) to 544.19 mm year⁻¹ (MIROC5) (Table 5). Although there are large uncertainties about the prediction of runoff for different GCMs, the spatial patterns are coincident. The prediction of future runoff shows gradiently increasing trend from north to south in the study region (Figure 10).

Many previous studies have suggested that the variation of runoff may be dependent on precipitation and temperature (Zhang et al. 2014; Fowler et al. 2018). The variation of temperature may be associated with evapotranspiration. From our studies, the uncertainties of future runoff is also rooted in variations of precipitation and actual evapotranspiration. For example, for the MRI-CGCM3 model, the low precipitation and the low actual evapotranspiration finally impact the low runoff. However, for the MIROC5 model, although the high temperature produces high actual evapotranspiration, the supplement of high precipitation also leads to high runoff. These numerical results support the fact that the growing regional precipitation is the dominant driver of positive runoff trends (Frans et al. 2013). Etchevers et al. (2002) demonstrated that the runoff structure matched quite well the precipitation anomaly, while the evaporation anomaly was more homogeneous and less strong. The study by Roderick & Farquhar (2011) also showed the diversity of projections for precipitation with a correspondingly large range in projections for runoff using future climate scenarios (2070–2099) derived using Intergovernmental Panel on Climate Change AR4 climate in the Murray–Darling Basin (MDB). Zhang et al. (2014) emphasized that the runoff elasticity to local temperature change is negative in the northern middle to high latitudes with values of local temperature increase. However, runoff sensitivities to local temperature change are small in the extended tropical area. On the other hand, the initial condition may be other factor causing the uncertainty of runoff prediction. In this study, the initial condition is considered as the same for ten forcing data of GCMs. Despite this, the uncertainty of runoff prediction due to forcing data is still an indisputable fact.

This study aims to investigate the range of uncertainties of runoff to climate change over China using numerical modeling methods. Our findings illustrated that there are considerable uncertainties about the future estimation of runoff (from 268.9 mm to 544.2 mm). The characteristic of uncertainty clearly exists in the south of the study domain; and the physical mechanism also shows a different pattern. Precipitation plays a main role in the variations of runoff for the MIROC5 model although the temperature is high. The low precipitation and temperature are two factors that cause variations of runoff for MRI-CGCM3. Our findings agree well with previous studies (Gordon & Famiglietti 2004; Zhang et al. 2014; Liu et al. 2017).

Although the range of uncertainties of future runoff is estimated, the extent of uncertainties may be underestimated. There are some studies that have studied the maximal uncertainties about simulation and prediction. For example, Sun & Mu (2013) employed the conditional nonlinear optimal perturbation related to parameters (CNOP-P) approach (Mu et al. 2010) to evaluate the variations in soil carbon in response to increases of 2 °C in temperature and 20% in precipitation with changes in the variability of temperature and precipitation. In their studies, these authors attempted to determine the maximal uncertainty of the models of soil carbon. However, that climate change scenario was restricted to only the increase in temperature and precipitation by 2 °C and 20%, respectively, which may be statistical results from the GCMs. The extent of the variation in temperature and precipitation in future projections from the GCMs is not consistent (Sun & Mu 2014). To overcome this shortcoming, Sun & Mu (2017, 2018) combined the outputs of ten GCMs under the RCP 4.5 scenario with the CNOP-P approach to try to evaluate the maximal uncertainties of soil carbon and net primary production. Based on their idea, the maximal uncertainties of future runoff could be used for reference. In addition, the maximal uncertainties about the numerical simulation and prediction could be estimated, however the minimal uncertainties are also estimated. These studies will be discussed in the future; also, the sensitivity of the parameters will be discussed.

Funding was provided by grants from the National Key Research and Development Program of China (No. 2017YFA0604804, 2016YFA0600804), the National Natural Science Foundation of China (Nos. 41675104), and Youth Innovation Promotion Association, Chinese Academy of Sciences (No. 2015060).