Assessment of Climate Change Impacts on River High Flows through Comparative Use of GR4J, HBV and Xinanjiang Models

Water Resour Manage (2013) 27:2871–2888 DOI 10.1007/s11269-013-0321-4 Assessment of Climate Change Impacts on River High Flows through Comparative Use of GR4J, HBV and Xinanjiang Models Ye Tian & Yue-Ping Xu & Xu-Jie Zhang Received: 3 January 2012 / Accepted: 26 February 2013 / Published online: 15 March 2013 # Springer Science+Business Media Dordrecht 2013 Abstract This study analyses the extreme high flows in Jinhua River basin under the impact of climate change for the near future 2011–2040. The objective of this study is to investigate the effect of using the bias corrected RCM outputs as input on the extreme flows by hydrological models. The future projections are obtained through the PRECIS model with resolution of 50 km×50 km under climate scenario A1B. The daily precipitation from the PRECIS is bias corrected by distribution based scaling method. Afterwards, three hydrological models (GR4J, HBV and Xinanjiang) are calibrated and applied to simulate the daily discharge in the future. The hydrological models are driven with both bias corrected precipitation and raw precipitation from the PRECIS model for 2011–2040. It is found that after bias correction, the amount, frequency, intensity and variance of the precipitation from the regional climate model resemble the observation better. For the three hydrological models, the simulated annual maximum discharges are higher by using the raw precipitation from PRECIS than by bias corrected precipitation at any return period. Meanwhile, the uncertainties from different models cannot be neglected. The largest difference between three models is about 2,100 m3/s. Keywords Climate change . Hydrological models . Uncertainty . Extreme flows . East China 1 Introduction In the past decades, the world has witnessed an obvious changing climate, particularly temperature increased. Other effects on hydrological systems are also occurring, namely, increased runoff and earlier spring peak discharge in many glacier and snow fed rivers (IPCC, 2007). According to Zhang et al. (2011), north China was dominated by decreasing precipitation and significant decreasing precipitation was found in the Yellow River basin and Huaihe River basin from 1960 to 2000. In southwest and east China, the annual mean Y. Tian : Y.-P. Xu (*) : X.-J. Zhang Department of Civil Engineering, Institute of Hydrology and Water Resources, Zhejiang University, Hangzhou, Zhejiang, China 310058 e-mail: yuepingxu@zju.edu.cn 2872 Y. Tian et al. precipitation increased significantly. In Yangtze River Basin, the extreme precipitation events increased by 10 %–20 % per decades in summer (Wang and Zhou 2005). The Poyang Lake basin in south China, the annual and seasonal streamflows have increased in the past 50 years (Zhao et al. 2010). Incorporated with climate change, human activities have also exerted a huge impact on the water cycles. For example, in Huifa River Basin in northeast China, human activities like regulation and storage of the water projects made a larger contribution to the runoff decrease in wet years than the climate change itself (Zhang et al. 2012a). Global climate models (GCMs) are the most common tools to estimate the climate change at a global scale. However, the resolution of GCMs is too coarse. They can provide predictions of changes in climate of a few kilometers or so at best. In the mountainous and coastal areas the GCM will fail to capture the local detail. However, hydrologic modeling at basin scale requires more detailed information on smaller scales. To solve the spatial resolution problem, there are mainly two kinds of methods commonly used to add details to the coarse information of GCMs at present. One way is through statistical downscaling methods that establish empirical relationships between circulation indices at large scale and predictive variables at local scale (Wilby et al. 1999). Another way is through dynamical downscaling by using a regional climate model (RCMs) for a particular area with the boundary condition from a GCM. Statistical downscaling methods use a ‘weather generator’ to generate weather time series based on an appropriate and fixed unknown model coefficient (Wilby et al. 1998). Ghosh and Katkar (2012) has studied the uncertainties of several downscaling methods including Linear Regression, Artificial Neural Network and Support Vector Machine in climate change impacts assessment. The statistical downscaling methods are cheap, flexible, computationally undemanding and widely used in hydrological studies (Chu et al. 2010; Huang et al. 2011; Willems and Vrac 2011; Xu et al. 2012). But there are also limitations. For example, the predictor–predictand relationships are often nonstationary and the weather series may not always be meteorologically consistent (Wilby and Wigley 1997). RCMs are able to resolve the atmospheric processes and are consistent with GCMs (Wilby et al. 2002). Though some parameterization in RCMs may have an empirical basis, RCMs are more physically based than statistical downscaling and are more acceptably transferable from current to future (Hay et al. 2002). There are many applications of applying different RCMs to assess the future climate. For example, heavy rainfall events over Vietnam are expected to decrease in most areas in 2001–2050 based on the IPCC scenario A1B and A2 by using the Theoretical Physics regional climate model version 3 (RegCM3) (Ho et al. 2011). In the Crati River Basin, average annual temperature will increase between 3.5 and 3.9 °C and cumulative annual precipitation will decrease between 9 % and 21 % in 2070–2099 under scenario A1B and A2 by applying the outputs of RegCM, HIRHAM and COSMO-CLM (Senatore et al. 2011). However, using the outputs of RCM results without bias correction in hydrologic models, produced unacceptably biased hydrological results (Wood et al. 2004). The biases commonly come from boundary conditions and parameterization. Kotlarski et al. (2005) found out that the simulation bias range from −1.1 o C to +0.9 o C for temperature and −31 mm/year to 108 mm/year for precipitation over Germany. After the precipitation and temperature biases are corrected, the accuracy of daily runoff simulation improved (Hay et al. 2002). So it is necessary to carry out bias correction to assure meaningful results before applying it to the hydrological models. There are many bias correction methods to adjust the results from RCM. Linear scaling method is a convenient statistical method to correct the mean value of climate variables (Lenderink et al. 2007). Correction factors are obtained by comparing the mean value of RCM output with historical data sets and then applied to future daily precipitation value and Assessment of Climate Change Impacts on High Flows 2873 temperatures. The major limitation of this method is that it is only sensible for mean values. For the reason that the biases of extreme value of precipitation are larger or smaller than those of mean value, non-linear bias correction methods are developed for simulations of high flows(Leander and Buishand 2007). Non-linear bias correction method not only considers the mean value but also the standard deviation. However such method fails to consider the frequency of the precipitation. Distribution based scaling (DBS) adjusts the RCM precipitation to approximate the long-term observed frequency and intensity distribution by mapping the RCM data onto distribution of observed data. Such methods have been used by many researchers when the output from RCM are used to do some assessment (Wood et al. 2002; Payne et al. 2004; Ines and Hansen 2006; Li et al. 2010; Ueyama et al. 2010). Since climate change is interactive with the hydrological cycle, to investigate the impact of climate change on the water resources, the hydrological model is an effective and widely used tool to simulate the hydrological processes. Based on different recognition and expression of the rainfall-runoff processes, there are a number of hydrological models. They are different in concept, physical equations and therefore parameters and structures. To assess the impact of projected climate change on the discharges, there are uncertainties originated from green house gas emission scenarios, GCMs, downscaling methods, hydrological model structures and parameters (Wilby and Harris 2006). The model structure is one of the most important sources of uncertainty (Chen et al. 2012b). Like the example of finding out the aquifer’s vulnerability towards pollution in Refsgaard’s study (Refsgaard et al. 2006), five consultants have five different results by using same input but five different models and the model structure is a major source of uncertainty in model prediction. Until now, some researchers have used several models in studying the impact of climate change. Najafi et al. (2011) used four different hydrological models to analyze the runoff under two emission scenarios in Tualatin River Basin in the United States, and the uncertainty in the dry season are higher than that in the wet season. Kay et al. (2009) investigated the uncertainty in the impact of climate change on flood frequency in England by using two hydrological models and the results show the same pattern in flood frequency change but different ranges for two models. Jiang et al. (2007) used six rainfall runoff models to compare the impact of climate change on monthly runoff and shows that greater differences in the model results occur when the models are used to simulate the winter flows. Fung et al. (2012) used large ensembles of climate scenarios to explore the adaption of the hydrologic impacts on the fresh water environment in south east England. Much work above has been done related to bias correction of RCM data and the uncertainty of the climate change impact on the water resources by using different methods. The results vary with the models applied in the research and the study areas. As we know, the floods are closely related to the safety of human’s daily life and properties. Especially in recent years, the extreme climate events happened with increasing frequency and intensity. Thus, this study mainly focused on the extreme high flows in Jinhua River basin in the east China. We are curious about the extent of effect in extreme high flows by using bias corrected and raw RCM data as input, and under the projected changing climate how much differences are in the extreme high flows by using different models and different input. The major objectives of this paper are 1) to assess the effect of the bias correction on the precipitation by comparison of the amount, frequency, intensity and variance of the bias corrected precipitation and raw precipitation from RCM; 2) to investigate the indirect effect of the bias correction method on the extreme flows, because the bias in the precipitation may be propagated into the discharges, therefore affects the extreme flows; 3) to find out the differences in extreme runoff under scenarios A1B in the future by using three hydrological models. 2874 Y. Tian et al. 2 Data and Methods 2.1 Study Area and Observed Data The study area is Jinhua River basin, which is the upper catchment of Qiantang River Basin, located in Zhejiang Province in the east China (see Fig. 1). The river basin stretches from latitude of 28.25o–29.75oN and longitude of 119.25o–120.75oE and the size is 5,996 km2. It is surrounded by mountains and hills and the elevation is higher in the southern and northern parts than in the middle part. It has a typical sub-tropical monsoon climate with hot rainy summer and cold dry winter. The annual temperature is about 17.5 °C. The highest temperature reaches around 40 °C and the lowest temperature seldom drops to below 0 °C. The annual precipitation for the period of 1981–1995 is 1,630 mm. However the precipitation is not evenly distributed in the whole year. About 50 % precipitation occurs during May to July. The flood occurred almost every year and the water resources cannot be utilized properly. For example, one of the most serious floods occurred in the June of 1995. The cumulative precipitation of 7 days reached 328.7 mm. The population in Jinhua River Basin is more than five million and is in the growth. The frequent floods have caused a great damage and loss of lives. Therefore, studying the extreme high flows under the changing climate in Jinhua River basin is a meaningful and important issue. Daily observed precipitation from five precipitation stations, daily potential evaportranspiration (PET) from a precipitation station (see Table 1) and daily discharge from a discharge station for the period 1981–1995 are used for model calibration and verification. The daily discharges for the first 10 years, namely 1981–1990, are used for the model calibration and those of the years 1991–1995 are for the model verification. The observed discharge used in the calibration and verification are offered by the Hydrological Bureau of Zhejiang Province, China. The discharge data are sampled at the Jinhua station each day at 8 am and the quality of the data has been checked. The historical observed discharge of 1961–1990 (as baseline) is used for the comparison with the future discharge. The average areal precipitation for the whole basin is calculated by Thiessen pologon method. The PET was calculated by Hamon’s equation (Hamon 1961; Haith and Shoemaker 1987). Fig. 1 Location of the Jinhua River basin and the distribution of the hydrological and meteorological stations. The crosses represent the grid point in the PRECIS model Assessment of Climate Change Impacts on High Flows Table 1 Locations of the hydrological and meteorological stations 2875 Latitude (oN) Longitude (oE) Jinhua Precipitation stations: 29.08 119.62 Jinhua 29.08 119.62 Bada 29.20 120.50 Yiwu 29.30 120.07 Yongkang 28.90 120.02 Zhengzhai 28.90 119.63 Discharge stations: 2.2 Regional Climate Model PRECIS (Providing REgional Climates for Impacts Studies) regional climate model is an atmospheric and land surface model that is able to be applied to any part of the world with the resolution of 50 km×50 km and 19 vertical layers. The model was developed by Hadley Centre of the UK Meteorological Office based on the modified atmospheric component of HadCM3 (Gordon et al. 2000). For this study, the output of HadCM3 was used as the initial and lateral boundary conditions to drive the PRECIS model. The PRECIS model has been used for the study related to the climate change worldwide, for example, China (Yuan et al. 2005; Xu et al. 2006b), India (Gosain et al. 2006; Gosain et al. 2011), South America (Marengo et al. 2009), Africa (Mileham et al. 2009) and Europe (Bartholy et al. 2009; Chenoweth et al. 2011). Xu et al. (2006b) has applied the PRECIS model to the whole China and found that under scenario B2 the precipitation over east China would increase largely in summer but not so much in winter, and in the southern China the precipitation would obviously decrease during 2071–2100. Under the scenario A2 the temperature would increase about 4 °C and precipitation would increase with 12.9 % in 2071–2080 (Xu et al. 2006a). The domain of the grid points in Xu’s study covers whole China by 145 grid points in longitude and 112 grid points in latitude. For our study, we choose the area of latitude 28.25o–29.75oN and longitude 119.25o–120.75oE. Simulated daily temperature and daily precipitation from the PRECIS model for the A1B emission scenario (It describes a future world of very rapid economic growth, global population that peaks in the mid-century and declines thereafter, and the rapid introduction of new and more efficient technologies with a balance across all energy sources.) from baseline 1961–1990 and near term 2011–2040 have been used. The potential impacts of climate change on discharges are assessed by hydrological models. 2.3 Bias Correction Climate data downscaled by dynamical downscaling includes model bias error from the RCM and such error introduces uncertainty when climate data are used in some assessment models (Ueyama et al. 2010). Therefore a bias correction is necessary to match the RCM data with observed data. In our study, distribution based scaling (DBS) is applied to adjust the future daily precipitation of PRECIS (Wood et al. 2002). We do it separately for 12 months considering that the precipitation intensity is different for each month. It involves two steps: 2876 (1) Y. Tian et al. to correct the precipitation frequency. Daily precipitation from the PRECIS model was truncated at a threshold based on its distribution. A threshold value is calculated for each subbasin based on the empirical observed and RCM cumulative precipitation distribution as formula (1), 1 ðFobs ðPobs ÞÞ Pthr ¼ FRCM ð1Þ Pthr is the threshold value; Pobs is the minimum observed precipitation amount considered as a wet day and here we use 0.1 mm; Fobs (…) is empirical cumulative distribution function of observed daily precipitation and F1 RCM ð. . .Þ is inverse cumulative distribution function of RCM. (2) to correct the precipitation intensity. There are many theoretical distributions available to describe the cumulative distribution function of precipitation intensities. One of the most commonly used distributions is two-parameter gamma distribution. Here we use gamma distribution to represent both observed and simulated precipitation intensity. Firstly RCM data are fitted to two-parameter gamma distribution. And then daily precipitation intensity are corrected by mapping it onto observed intensity distribution, which means taking inverse gamma cumulative distribution function of observed data to get the corrected value. The formula is as follow: 1 ðFRCM ðPRCM ÞÞ Pcor ¼ Fobs ð2Þ Pcor is corrected daily precipitation; PRCM is truncated RCM precipitation; FRCM (…) is two-parameter cumulative distribution of truncated RCM precipitation and F1 obs ð. . .Þ is inverse gamma cumulative distribution of observed data. 2.4 Hydrological Models There are hundreds of hydrological models in the world. Here three hydrological models, namely GR4J, HBV and Xinanjiang, are used to simulate the river discharge at Jinhua hydrological station under the impact of potential climate change for the period from 2011 to 2040. All the three models are widely used for various purposes (Oudin et al. 2004; Akhtar et al. 2008; Li et al. 2009b; Engeland et al. 2010; Peng and Xu 2010; Wu et al. 2010). Three models have different number of parameters, different model structures and different physical meaning in simulating the real rainfall-runoff processes. Therefore they are applied in this study. The GR4J is a four-parameter lumped rainfall-runoff model developed by Perrin in 2003 based on the GR3J model (Edijatno et al. 1999; Perrin et al. 2003). The GR4J model has significant improvement in simulating the low flow compared with the GR3J. The GR4J model firstly calculates the net rainfall and PET by subtracting the PET from precipitation. Then through interception a portion of the precipitation goes into the production store, in which the actual evaporation is calculated and percolation occurs. The percolated water reaches the flow routing through leakage. Another portion of precipitation goes directly to the flow routing. The two flow components together are split into 90 % runoff routed by a unit hydrograph and then a non-linear routing store and 10 % runoff routed by a single unit hydrograph. The total runoff is Assessment of Climate Change Impacts on High Flows 2877 finally obtained by adding these two parts together. With those two hydrographs, the time lag between the precipitation event and the corresponding peak discharge can be simulated. The HBV model is a semi-distributed conceptual rainfall-runoff model originally developed by Swedish Meteorological and Hydrological Institute (SMHI) (Bergström 1976, 1992; Lindström et al. 1997). It has been applied widely in climate change, uncertainty analysis and extreme flows estimation (Engeland and Hisdal 2009; Deckers et al. 2010; Chen et al. 2012a; Dakhlaoui et al. 2012; Kriauciuniene et al. 2012). The HBV model is composed of a precipitation and snow accumulation routine, a soil moisture routine, a quick runoff routine and a baseflow routine and a transform function. The HBV model takes into account the effect of snow melting and accumulation. But in the study area it seldom snows and therefore the snow accumulation and melting is not included in the HBV model used for our study. The actual evaporation is obtained by a linear function which decreases as the soil moisture drops. There are two kinds of runoff reservoirs in HBV. One is the upper reservoir which generates the quick flow expressed by a non-linear function and another one is the lower reservoir which generates the baseflow expressed by a linear function. At last the runoff generated from these two reservoirs is routed through a transformation function. The Xinanjiang model is also a semi-distributed rainfall runoff model especially for humid and semi-humid regions developed by Zhao (1992). Many studies have been carried out using this model (Shi et al. 2011; Jiang et al. 2012; Li et al. 2012; Zhang et al. 2012b). The input of three models is daily precipitation and PET. The output is daily discharge. The Xinanjiang model consists of an evapotranspiration component represented by a model of three soil layers, a runoff generation component, a runoff production component, separation of runoff component separating the runoff into surface water, interflow and ground water and flow routing. This model was initially developed for Qiantang River Basin. To assess the performance of the models, the Nash-Sutcliffe efficiency coefficient (NS) is chosen as objective function (Nash and Sutcliffe 1970). The parameters of the models are calibrated by the GLUE method using the observed daily precipitation and PET from 1981 to 1990. The GLUE method is a Bayesian analysis based Monte Carlo method for model calibration and uncertainty analysis (Beven 2006). The verification of three models is carried out for the period from 1991 to 1995. Both the bias corrected precipitation of PRECIS and the one without bias correction are used to drive the hydrological models to calculate the future discharges at Jinhua Station from 2011 to 2040. The corresponding two sets of results are compared. The extreme index, annual maximum discharge, is selected to analyze the potential impact of the climate change on the extreme discharges. 3 Results 3.1 Bias Corrected Precipitation Figure 2 shows the monthly mean amount, frequency, intensity and variance of the precipitation. The results of bias corrected precipitation from the PRECIS model are compared with precipitation data of the PRECIS without bias correction and observation precipitation from 1981 to 1995. 2878 Y. Tian et al. Fig. 2 Monthly mean precipitation a amount, b frequency and c intensity d variance for Jinhua River basin. Solid with squares represents observations; Dashes with circles represent bias corrected data from PRECIS simulation; Dots with diamond represent raw data from the PRECIS simulation without bias correction Without bias correction, there is continuous underestimation of the precipitation amount of the observed one by the PRECIS climate model for the beginning of a year from January to April. For the months from October to December the PRECIS climate model overestimates the observed monthly precipitation. When the precipitation amount is high in June, the PRECIS underestimates the high precipitation amount. Fig. 2(b) shows that the PRECIS simulates less precipitation events than the real situation in 7 months out of a year obviously. As to the precipitation intensity, the PRECIS climate model does not catch the changing regularity of the observed data very well. The observed precipitation intensity goes up from January to the end of May, followed by a sudden decrease in June. A sudden increase can be observed in September, and drops down to the end of year with fluctuation. However, large overestimation of precipitation intensity is found in May and from September to November. The variance of the simulated monthly precipitation is obviously larger than observation in September to November which is similar to the intensity. The four figures indicate that the PRECIS have a good simulation for the amount, but it doesn’t simulate the extreme high precipitation in July very well as it underestimates the intensity in July which is shown in Fig. 2(c). Zhang et al. (2006) found that the simulated bias in the extreme precipitation in the south China was linked to the changes of the Subtropical High in the West Pacific. Besides, it simulates less rainy days and more precipitation amount in September and October, which caused the obvious overestimation in the intensity and variance. After bias correction, the monthly mean precipitation amount from PRECIS is closer to the observed monthly mean precipitation amount for most of the time although there are small differences. Besides, the frequency of precipitation is Assessment of Climate Change Impacts on High Flows 2879 improved obviously compared to that of PRECIS climate model without bias correction. The most significant change is in the precipitation intensity and the variance, and the bias correction makes the trend of precipitation intensity close to the observations. The overestimation from PRECIS in May, September, October and November are corrected and the results match the observed precipitation intensity much better than the one without bias correction. There is still some deviation from the observed data even after bias correction, such as the underestimation of the precipitation frequency from January to April and August. Relatively low frequency is probably because of the difference in the shape of distribution of RCM data and that of observation data. The threshold value is obtained by mapping the position of 1 mm in observed precipitation distribution onto the distribution of daily precipitation of PRECIS. For PRECIS simulations, there are more days with daily precipitation amounts under the threshold values than observed data. The under-threshold RCM data are rejected, so less portion of the PRECIS data are retained. When there is high intense precipitation, bias corrected PRECIS precipitation intensity is still lower than the observed one in July. 3.2 Calibration and Verification of the Hydrological Models Using the observed daily precipitation and PET as the input, three models are calibrated for the period 1981–1990 and verified for the period 1991–1995. Table 2 presents the NS (Nash-Sutcliffe) value during the calibration and verification periods for GR4J, HBV and Xinanjiang models. The maximum NS value is one. If the NS value equals to one, it means the simulation totally matches observations. So the closer the NS value is to one, the better performance of the model simulation is. The results show that the NS values for both calibration and verification periods for all three models are above 0.8. Therefore, the performance of all three models is satisfactory. The NS value show that during the calibration and verification period, the performance of the GR4J model and the HBV model is somewhat better than the Xinanjiang model. Besides, the performance for the verification period is a little bit better than for the calibration period. Figure 3 compares the simulated and observed daily discharges for 1981–1995. The horizontal axis represents the observation and the vertical axis represents the simulation. The black solid line represents the perfect situation when the simulation is the same as the observation. The squares represent the simulated discharge for the GR4J model, the HBV model and the Xinanjiang model in Fig. 3a, b and c respectively. Although there are biases from the daily observed discharge by the simulation, the symbols are around the solid line. For the high discharges, most symbols are below the solid line which indicates the underestimation of the high discharge from the hydrological models. Table 2 NS value for the GR4J model, the HBV model and the Xinanjiang model during the calibration and verification periods GR4J HBV Xinanjiang Calibration 0.91 0.91 0.88 Verification 0.93 0.91 0.89 2880 Y. Tian et al. Fig. 3 Daily discharge simulated by a GR4J, b HBV and c Xinanjiang for 1981–1995 3.3 Impact of Bias Correction on High Flows For the simulation of the future high flows, a comparison is carried out to investigate the discrepancy of using bias corrected data and raw data from PRECIS. In this paper, high flows are represented by annual maximum discharge. Fig. 4 shows the annual maximum discharge by using bias corrected data and raw data for 2011–2040 for the GR4J model, HBV model and the Xinanjiang model. The line with circles represents the annual maximum discharges from bias corrected data and the squares represent the results from raw PRECIS data. For all three models, the largest annual maximum discharge reaches around 10,000 m3/s and the smallest value is less than 1,000 m3/s by using the raw PRECIS data. After bias correction, the annual maximum discharges are lower for most of the time from 2011 to 2040 and the largest annual maximum discharge ranges between 4,000 m3/s and around 6,000 m3/s due to different models. Such difference is caused by the different precipitation input for the models. Bias correction changes the temporal distribution of original PRECIS precipitation intensity, which causes the decrease in the annual maximum discharge. Therefore, attention should be paid when applying the RCM data, for without bias correction, the overestimation of the precipitation intensity may result in the overestimation of the discharge. Besides, the differences of the discharge by using raw and bias corrected input are dependent on the hydrological models. It’s shown that such difference is smallest in the Xinanjiang model, and largest in the GR4J model, which indicates the sensitivity of the models to the input data. Figure 5 shows the annual maximum discharge simulated by GR4J, HBV and Xiannjiang. Generally, there is no obvious trend for the annual maximum discharge, but a three-to-five year interannual discharge cycle with peaks of more than 2,000 m3/s and valleys of less than 1,000 m3/s is shown for the future climate. Though each model simulates different value of annual maximum discharge under the same scenario and by the same input, the fluctuation of the discharge is in consistency. When the value of Assessment of Climate Change Impacts on High Flows 2881 Fig. 4 comparison of the annual maximum discharge from 2011 to 2040 by using bias corrected data and raw data from PRECIS as two sets of input for a the GR4J model, b the HBV model and c the Xinanjiang model annual maximum discharge is below 1,000 m3/s, the difference of simulated discharge from three models is within 100 m3/s. When the value of the annual maximum discharge is large, the differences from the three models tend to be more obvious. For the year 2019, the annual maximum discharge ranges from around 400 m3/s to 5,000 m3/s due to different models. In most of the time, the HBV model predicts the highest annual maximum discharge, while the Xinanjiang model predicts the lowest annual maximum discharge. The HBV is in between. It is likely to be related to the different model structures, such as the division of the soil layers for reserving water and the mechanism of the runoff generation, etc. 2882 Y. Tian et al. Fig. 5 simulated annual maximum discharge by GR4J, HBV and Xinanjiang for the period of 2011–2040 under scenario A1B for the Jinhua River basin 3.4 Extreme Value Analysis of High Flows To compare the impact of bias correction on high flows, the design discharge versus return periods based on simulated high flows by GR4J, HBV and Xinanjiang respectively for 2011–2040 with bias corrected input and raw PRECIS input are presented in Fig. 6. Here the GEV (Generalized Extreme Distribution) distribution is used for extreme value analysis. It is clear in the Fig. 6 that for all the three models the design discharges from raw PRECIS input is larger than those from bias corrected input at any return period. And the differences increase as the return periods increase. For 2 years return period, the difference of the design discharge with bias correction and without bias correction is in the range of 1,400–2,500 m3/s. While for 100 years return period, the difference is 4,000–7,000 m3/s. Such difference is due to the combination of many factors. First of all, the systematic error of the RCM may contribute to it. Though the parameters in the RCM are calibrated, it is impossible to completely remove bias using parameterization for the difficulty in parameterizing all factors that affect the RCM bias (Ueyama et al. 2010). Meanwhile, the bias correction method is based on the assumption that the correction function is constant in time, which means it is the same for the present and the future. And the bias correction method is not a result of impeccable physically based theory and it just represents a curve fitting exercise of convenience (Kundzewicz and Stakhiv 2010). Furthermore, the design discharge of different return periods simulated by three models using bias corrected input for 2011–2040 are compared with those of the observations for the baseline 1961–1990. The results are shown in Fig. 7. The design discharge with 5 years return period is about 2,600 m3/s for observation, and with 100 years return period it increases into almost 5,000 m3/s. Although the NS value from calibration and verification shows that the performances of three models are all good, the simulated design discharge from three models are very different in the changing climate. Under the scenario A1B, the GR4J model shows a decrease of the extreme flows when it is below 3,500 m3/s and an increase when it is above 3,500 m3/s, which indicates a polarization over the extreme flows. The large floods would be even larger in the future and the small ones would become smaller. The Xinanjiang model displays a lower parallel line with the observation. It predicts a decrease in the extreme high flows for all the return periods. As to the HBV model, it remains petty much similar as the current situation within the return period of 3 years. The extreme flows are predicted to increase in the future when it is larger than 3 years return period. That is, the flood risk in the Assessment of Climate Change Impacts on High Flows 2883 Fig. 6 Design discharge versus return periods with bias corrected input and raw input from PRECIS, based on annual maximum discharge from a the GR4J model, b the HBV model and c the Xinanjiang model future under scenario A1B tends to be larger estimated by GR4J and HBV model but smaller by Xinanjiang model. The range of such change between the future projection and historical data is closely related with the return period. When the return period is 2 years, the change of the design discharge ranges from −500 m3/s to −100 m3/s. As the return period increases, the range is −900 m3/s to +1,200 m3/s with 100 years return period. In another word, the range for 100 years return period is five times as much as that for 2 years return period due to different models, which indicates the uncertainty of the high flows from hydrological models becomes large as the increase of the volume of the discharge. For this study, the uncertainty in the extreme high flows in the future is due to (1) the parameters of the hydrological models. Even the parameters of equally good performance may produce different results. Such phenomenon of the equifinality of the parameters in calibration has been illustrated by Beven (2006); (2) the different models applied for simulation. Different models have different interpretation and expression of the real physical processes and therefore may cause varied results; (3) the probability model used for extreme-value analysis of the high flows. 2884 Y. Tian et al. Fig. 7 Design discharge versus return periods from the observation, the GR4J model, the HBV model and the Xinanjiang model for the Jinhua River Basin 4 Discussion This study was initially to assess the direct effect of the bias correction on the precipitation from the PRECIS model and the indirect effect on the future extreme flows, and meanwhile investigate the uncertainty of the extreme high flows using different models. In the bias correction process, although the amount of the monthly precipitation gave good results but the intensity and the variance of the precipitation from the PRECIS had a large bias from the observation in July and from September to November. Zhang et al. (2006) pointed out that the simulated precipitation of the PRECIS was better in the north China than that in southwest coastal area. It was related to the position and intensity of the Subtropical High in the West Pacific which caused the simulated bias in extreme precipitation events. If the bias of the precipitation is not corrected in the projection of the future discharge, it would be propagated to the discharge resulting in the overestimation in the extreme discharge in this study. Meanwhile, the changes of the future extreme discharges simulation by three models were different. The GR4J model and the HBV model predicted the increase in the extreme flows in the future but the Xinanjiang model predicted a decrease. The GR4J model outperformed the other two models. The GR4J model had two soil layers, four parameters. The HBV model has two soil layers and eight parameters. The Xinanjiang model had three soil layers and 14 parameters. It seemed that the more complex and specific hydrological model does not improve the simulation results. The results were different from Li’s study (2009a), which showed that runoff prediction in ungauged catchments are improved by using more complex, specific model structures defined by vegetation and other information. It was likely due to the different study areas and the soil layers in this area were not the key factors of the formation of the discharge. Besides, fewer parameters may be easier to identify while more redundant parameters would add more uncertainty and reduce the chance of get a better result. The ‘best’ model in this study area was the GR4J model in terms of the highest NS value in the calibration and verification. But it only illustrated that the results of the GR4J model fitted the observation best when using the NS as the objective function. However, only using the model with the highest NS value was too risky and not recommended, because the simulation results were affected by many other factors like the objective function and the time interval of the observed data. The uncertainty of the models cannot be neglected, as it was stressed in other researches as well (Beven 2006). Therefore, attention should be paid in Assessment of Climate Change Impacts on High Flows 2885 dealing with the simulated discharges. A way to deal with the uncertainty was the BMA method which gave the weight to the models according to their performance (Madigan et al. 1996; Rojas et al. 2008; Parrish et al. 2012). 5 Conclusion This study analyzed the impact of climate change on the extreme high flows in Jinhua River basin for the near future 2011–2040. Under scenario A1B, the PRECIS RCM was applied to downscale the GCM grids so as to get more detailed information for the catchment scale. The daily precipitation from the PRECIS RCM was bias corrected by distribution based scaling method. Annual maximum discharge was chosen as the index to assess the impact on extreme high flows with three different models, namely GR4J, HBV and Xinanjiang. The results showed that compared to the observed precipitation, the precipitation amount from RCM is higher in the end of the year but lower in the beginning of the year, the precipitation frequency is less and the intensity of annual distribution is not in accordance with the observation. After bias correction, though there are differences between observation and RCM simulation, the bias of precipitation amount is reduced. Besides, there is obvious improvement for the precipitation frequency and intensity with the annual distribution closer to the observation. However the precipitation intensity is a bit lower than observation. It is related to the different shape of distribution between observation and precipitation from PRECIS and the threshold we chose to truncate the precipitation from PRECIS. Comparing the results using raw precipitation from PRECIS and bias corrected precipitation, we found that there is a large difference for the annual maximum discharge for the period of 2011 to 2040. For all three models, the simulated high flows are higher by using the raw precipitation from PRECIS than those by using bias corrected precipitation. The largest difference of the extreme high flows reaches about 7,000 m3/s. It indicates the importance of carrying out the bias correction for the RCM data when using them to assess the effect of climate change on discharge. Though each model presents different annual maximum discharge, they show similar three-to-five year interannual discharge cycle and the fluctuation of the high flows is in consistency. The difference of the design discharge between three models increases with the value of return periods. The largest difference between three models reaches about 2,100 m3/s. In most of the time, the HBV model predicts the highest annual maximum discharge; the lowest is for Xinanjiang and GR4J is in between. For all the three models the annual maximum discharges from raw PRECIS input is larger than those from bias corrected input at any return period. And the difference is increasing as the return periods increase. Compared to the design discharge calculated based on the observations, three models show different changing trends. The HBV model shows an increase in the extreme discharges for the return periods larger than 3 years while the Xinanjiang model displays a decrease in extreme discharges for all the return periods. As to the GR4J model, the extreme floods will become more extreme. The flood risk in the future under scenario A1B tends to be larger estimated by GR4J and HBV, but smaller by Xinanjiang. The magnitude of such change is related to the models and return period. Acknowledgments This study is financially supported by the International Science and Technology Cooperation Program of China (Project No. 2010DFA24320) and the Nature Science Foundation of China (Project No. 50809058). Other supports from Met Office Hadley Centre, UK, Bureau of Hydrology, Zhejiang Province, and 2886 Y. Tian et al. Nanjing Hydraulic Research Institute are highly acknowledged. Finally, many thanks are given to two anonymous reviewers for their valuable comments. References Akhtar M, Ahmad N, Booij MJ (2008) The impact of climate change on the water resources of HindukushKarakorum-Himalaya region under different glacier coverage scenarios. J Hydrol 355(1–4):148–163. doi:10.1016/j.jhydrol.2008.03.015 Bartholy J, Pongracz R, Torma C, Pieczka I, Kardos P, Hunyady A (2009) Analysis of regional climate change modelling experiments for the Carpathian basin. Int J Global Warm 1(1–3):238–252 Bergström S (1976) Development and application of a conceptual runoff model for Scandinavian catchments. SMHI Report RHO NO. 7, Norrköping, Sweden Bergström S (1992) The HBV model - its structure and applications. SMHI Reports RH NO. 4, Norrköping, Sweden Beven K (2006) A manifesto for the equifinality thesis. J Hydrol 320(1–2):18–36 Chen H, Xiang T, Zhou X, Xu CY (2012a) Impacts of climate change on the Qingjiang Watershed’s runoff change trend in China. Stoch Env Res Risk A 26(6):847–858. doi:10.1007/s00477-011-0524-2 Chen X, Yang T, Wang X, Xu CY, Yu Z (2012b) Uncertainty Intercomparison of Different Hydrological Models in Simulating Extreme Flows. Water Resour Manag: 1–17. doi:10.1007/s11269-012-0244-5 Chenoweth J, Hadjinicolaou P, Bruggeman A, Lelieveld J, Levin Z, Lange MA, Xoplaki E, Hadjikakou M (2011) Impact of climate change on the water resources of the eastern Mediterranean and Middle East region: modeled 21st century changes and implications. WRR 47 Chu JT, Xia J, Xu CY, Singh VP (2010) Statistical downscaling of daily mean temperature, pan evaporation and precipitation for climate change scenarios in Haihe River, China. Theor Appl Climatol 99(1–2):149– 161. doi:10.1007/s00704-009-0129-6 Dakhlaoui H, Bargaoui Z, Bárdossy A (2012) Toward a more efficient calibration schema for HBV rainfallrunoff model. J Hydrol 444–445:161–179. doi:10.1016/j.jhydrol.2012.04.015 Deckers DLEH, Booij MJ, Rientjes THM, Krol MS (2010) Catchment variability and parameter estimation in multi-objective regionalisation of a rainfall-runoff model. Water Resour Manag 24(14):3961–3985. doi:10.1007/s11269-010-9642-8 Edijatno, Nascimento ND, Yang XL, Makhlouf Z, Michel C (1999) GR3J: a daily watershed model with three free parameters. Hydrolog Sci J 44(2):263–277 Engeland K, Hisdal H (2009) A comparison of low flow estimates in ungauged catchments using regional regression and the HBV-model. Water Resour Manag 23(12):2567–2586. doi:10.1007/ s11269-008-9397-7 Engeland K, Renard B, Steinsland I, Kolberg S (2010) Evaluation of statistical models for forecast errors from the HBV model. J Hydrol 384(1–2):142–155. doi:10.1016/j.jhydrol.2010.01.018 Fung F, Watts G, Lopez A, Orr HG, New M, Extence C (2012) Using Large Climate Ensembles to Plan for the Hydrological Impact of Climate Change in the Freshwater Environment. Water Resour Manag:1–22. doi:10.1007/s11269-012-0080-7 Ghosh S, Katkar S (2012) Modeling uncertainty resulting from multiple downscaling methods in assessing hydrological impacts of climate change. Water Resour Manag 26(12):3559–3579. doi:10.1007/s11269-012-0090-5471 Gordon C, Cooper C, Senior CA, Banks H, Gregory JM, Johns TC, Mitchell JFB, Wood RA (2000) The simulation of SST, sea ice extents and ocean heat transports in a version of the Hadley centre coupled model without flux adjustments. ClDy 16(2–3):147–168 Gosain AK, Rao S, Basuray D (2006) Climate change impact assessment on hydrology of Indian River basins. CSci 90(3):346–353 Gosain AK, Rao S, Arora A (2011) Climate change impact assessment of water resources of India. CSci 101(3):356–371 Haith DA, Shoemaker LL (1987) Generalized watershed loading functions for stream-flow nutrients. Water Resour Bull 23(3):471–478 Hamon WR (1961) Estimating Potential Evapotranspiration. J Hydraul Eng-asce 87 (HY3):107–120 Hay LE, Clark MP, Wilby RL, Gutowski WJ, Leavesley GH, Pan Z, Arritt RW, Takle ES (2002) Use of regional climate model output for hydrologic simulations. J Hydrometeorol 3(5):571–590 Ho TMH, Phan VT, Le NQ, Nguyen QT (2011) Extreme climatic events over Vietnam from observational data and RegCM3 projections. Clim Res 49(2):87–100. doi:10.3354/Cr01021 Huang J, Zhang JC, Zhang ZX, Xu CY, Wang BL, Yao J (2011) Estimation of future precipitation change in the Yangtze River basin by using statistical downscaling method. Stoch Env Res Risk A 25(6):781–792. doi:10.1007/s00477-010-0441-9 Assessment of Climate Change Impacts on High Flows 2887 Ines AVM, Hansen JW (2006) Bias correction of daily GCM rainfall for crop simulation studies. Agr Forest Meteorol 138(1–4):44–53. doi:10.1016/j.agrformet.2006.03.009 IPCC (2007) Climate Change 2007: Synthesis Report. Contribution of Working Groups I, II and III to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change, Core Writing Team, Pachauri RK and Reisinger A, Geneva, Switzerland Jiang T, Chen YQD, Xu CYY, Chen XH, Chen X, Singh VP (2007) Comparison of hydrological impacts of climate change simulated by six hydrological models in the Dongjiang basin, South China. J Hydrol 336(3–4):316–333 Jiang S, Ren L, Hong Y, Yong B, Yang X, Yuan F, Ma M (2012) Comprehensive evaluation of multi-satellite precipitation products with a dense rain gauge network and optimally merging their simulated hydrological flows using the Bayesian model averaging method. J Hydrol 452–453:213–225. doi:10.1016/ j.jhydrol.2012.05.055 Kay AL, Davies HN, Bell VA, Jones RG (2009) Comparison of uncertainty sources for climate change impacts: flood frequency in England. Clim Change 92(1–2):41–63. doi:10.1007/s10584-008-9471-4 Kotlarski S, Block A, BAhm U, Jacob D, Keuler K, Knoche R, Rechid D, Walter A (2005) Regional climate model simulations as input for hydrological applications: evaluation of uncertainties. AdG 5:119–125 Kriauciuniene J, Jakimavicius D, Sarauskiene D, Kaliatka T (2012) Estimation of uncertainty sources in the projections of Lithuanian river runoff. Stoch Env Res Risk A:1–16. doi:10.1007/s00477-012-0608-7 Kundzewicz ZW, Stakhiv EZ (2010) Are climate models “ready for prime time” in water resources management applications, or is more research needed? Hydrolog Sci J 55(7):1085–1089. doi:10.1080/ 02626667.2010.513211 Leander R, Buishand TA (2007) Resampling of regional climate model output for the simulation of extreme river flows. J Hydrol 332(3–4):487–496 Lenderink G, Buishand A, van Deursen W (2007) Estimates of future discharges of the river Rhine using two scenario methodologies: direct versus delta approach. Hydrol Earth Syst Sc 11(3):1143–1159 Li H, Zhang Y, Chiew FHS, Xu S (2009a) Predicting runoff in ungauged catchments by using Xinanjiang model with MODIS leaf area index. J Hydrol 370(1–4):155–162 Li L, Hong Y, Wang JH, Adler RF, Policelli FS, Habib S, Irwn D, Korme T, Okello L (2009b) Evaluation of the real-time TRMM-based multi-satellite precipitation analysis for an operational flood prediction system in Nzoia Basin, lake Victoria, Africa. Nat Hazards 50(1):109–123 Li HB, Sheffield J, Wood EF (2010) Bias correction of monthly precipitation and temperature fields from intergovernmental panel on climate change AR4 models using equidistant quantile matching. JGRD 115(D10). doi:10.1029/2009JD012882 Li H, Zhang Y, Vaze J, Wang B (2012) Separating effects of vegetation change and climate variability using hydrological modelling and sensitivity-based approaches. J Hydrol 420–421:403–418. doi:10.1016/ j.jhydrol.2011.12.0331029/2007wr006665 Lindström G, Johansson B, Persson M, Gardelin M, Bergström S (1997) Development and test of the distributed HBV-96 hydrological model. J Hydrol 201(1–4):272–288 Madigan D, Raftery AE, Volinsky C, Hoeting J Bayesian model averaging. In, 1996. pp 77–83 Marengo JA, Jones R, Alves LM, Valverde MC (2009) Future change of temperature and precipitation extremes in South America as derived from the PRECIS regional climate modeling system. Int J Climatol 29(15):2241–2255 Mileham L, Taylor RG, Todd M, Tindimugaya C, Thompson J (2009) The impact of climate change on groundwater recharge and runoff in a humid, equatorial catchment: sensitivity of projections to rainfall intensity. Hydrolog Sci J 54(4):727–738 Najafi MR, Moradkhani H, Jung IW (2011) Assessing the uncertainties of hydrologic model selection in climate change impact studies. HyPr 25(18):2814–2826. doi:10.1002/Hyp.8043 Nash JE, Sutcliffe JV (1970) River flow forecasting through conceptual models part I - a discussion of principles. J Hydrol 10(3):282–290 Oudin L, Andreassian V, Perrin C, Anctil F (2004) Locating the sources of low-pass behavior within rainfallrunoff models. WRR 40(11). doi:10.1029/2004wr003291 Parrish MA, Moradkhani H, DeChant CM (2012) Toward reduction of model uncertainty: integration of Bayesian model averaging and data assimilation. WRR 48(3):W03519. doi:10.1029/2011WR011116 Payne JT, Wood AW, Hamlet AF, Palmer RN, Lettenmaier DP (2004) Mitigating the effects of climate change on the water resources of the Columbia river basin. Clim Change 62(1–3):233–256 Peng DZ, Xu ZX (2010) Simulating the impact of climate change on streamflow in the Tarim river basin by using a modified semi-distributed monthly water balance model. HyPr 24(2):209–216 Perrin C, Michel C, Andreassian V (2003) Improvement of a parsimonious model for streamflow simulation. J Hydrol 279(1–4):275–289 2888 Y. Tian et al. Refsgaard JC, van der Sluijs JP, Brown J, van der Keur P (2006) A framework for dealing with uncertainty due to model structure error. Adv Water Resour 29(11):1586–1597. doi:10.1016/j.advwatres.2005.11.013 Rojas R, Feyen L, Dassargues A (2008) Conceptual model uncertainty in groundwater modeling: combining generalized likelihood uncertainty estimation and Bayesian model averaging. WRR 44(12):W12418 Senatore A, Mendicino G, Smiatek G, Kunstmann H (2011) Regional climate change projections and hydrological impact analysis for a Mediterranean basin in Southern Italy. J Hydrol 399(1–2):70–92. doi:10.1016/j.jhydrol.2010.12.035 Shi P, Chen C, Srinivasan R, Zhang X, Cai T, Fang X, Qu S, Chen X, Li Q (2011) Evaluating the SWAT model for hydrological modeling in the Xixian watershed and a comparison with the XAJ model. Water Resour Manag 25(10):2595–2612. doi:10.1007/s11269-011-9828-810.1002/hyp.8058 Ueyama H, Adachi S, Kimura F (2010) Compilation method for 1 km grid data of monthly mean air temperature for quantitative assessments of climate change impacts. Theor Appl Climatol 101(3–4):421–431 Wang YQ, Zhou L (2005) Observed trends in extreme precipitation events in China during 1961–2001 and the associated changes in large-scale circulation (vol 32, art no L09707, 2005). GeoRL 32(17). doi:10.1029/ 2005gl023769 Wilby RL, Harris I (2006) A framework for assessing uncertainties in climate change impacts: Low-flow scenarios for the river Thames, UK. WRR 42(2) Wilby RL, Wigley TML (1997) Downscaling general circulation model output: a review of methods and limitations. Prog Phys Geog 21(4):530–548 Wilby RL, Hassan H, Hanaki K (1998) Statistical downscaling of hydrometeorological variables using general circulation model output. J Hydrol 205(1–2):1–19 Wilby RL, Hay LE, Leavesley GH (1999) A comparison of downscaled and raw GCM output: implications for climate change scenarios in the San Juan river basin, Colorado. J Hydrol 225(1–2):67–91 Wilby RL, Dawson CW, Barrow EM (2002) SDSM - a decision support tool for the assessment of regional climate change impacts. Environ Modell Softw 17(2):147–159 Willems P, Vrac M (2011) Statistical precipitation downscaling for small-scale hydrological impact investigations of climate change. J Hydrol 402(3–4):193–205. doi:10.1016/j.jhydrol.2011.02.030 Wood AW, Maurer EP, Kumar A, Lettenmaier DP (2002) Long-range experimental hydrologic forecasting for the eastern United States. JGRD 107(D20). doi:10.1029/2001jd000659 Wood AW, Leung LR, Sridhar V, Lettenmaier DP (2004) Hydrologic implications of dynamical and statistical approaches to downscaling climate model outputs. Clim Change 62(1–3):189–216 Wu W, Clark JS, Vose JM (2010) Assimilating multi-source uncertainties of a parsimonious conceptual hydrological model using hierarchical Bayesian modeling. J Hydrol 394(3–4):436–446. doi:10.1016/ j.jhydrol.2010.09.017 Xu YL, Huang XY, Zhang Y, Lin WT, Lin ED (2006a) Statistical analyses of climate change scenarios over China in the 21st century. Adv Clim Chang Res 2(1):50–53 Xu YL, Zhang Y, Lin ED, Lin WT, Dong WJ, Jones R, Hassell D, Wilson S (2006b) Analyses on the climate change responses over China under SRES B2 scenario using PRECIS. Chin Sci Bull 51(18):2260–2267 Xu YP, Zhang X, Tian Y (2012) Impact of climate change on 24-h design rainfall depth estimation in Qiantang river basin, East China. Hydrol Processes 26:4067–4077 Yuan F, Xie ZH, Liu Q, Xia J (2005) Simulating hydrologic changes with climate change scenarios in the Haihe river basin. Pedosphere 15(5):595–600 Zhang Y, Xu Y, Dong W, Cao L, Sparrow M (2006) A future climate scenario of regional changes in extreme climate events over China using the PRECIS climate model. GeoRL 33(24):L24702 Zhang Q, Singh VP, Sun P, Chen X, Zhang Z, Li J (2011) Precipitation and streamflow changes in China: changing patterns, causes and implications. J Hydrol 410:204–216 Zhang A, Zhang C, Fu G, Wang B, Bao Z, Zheng H (2012a) Assessments of impacts of climate change and human activities on runoff with SWAT for the Huifa River Basin, Northeast China. Water Resour Manag 26(8):2199–2217. doi:10.1007/s11269-012-0010-8 Zhang D, Zhang L, Guan Y, Chen X (2012b) Sensitivity analysis of Xinanjiang rainfall-runoff model parameters: A case study in Lianghui, Zhejiang province, China. Hydrol Res 43(1–2):123–134. doi:10.2166/nh.2011.131 Zhao RJ (1992) The Xinanjiang model applied in China. J Hydrol 135(1–4):371–381 Zhao G, Hörmann G, Fohrer N, Zhang Z, Zhai J (2010) Streamflow trends and climate variability impacts in Poyang Lake basin, China. Water Resour Manag 24(4):689–706. doi:10.1007/s11269-009-9465-7