1. Introduction
Global climate change and its impact on hydrology and water resources have received special attention due to its effects on land use and development [
1]. The hydrologic cycle in watersheds is changing greatly under the influence of global climate change. According to the fifth assessment report of Intergovernmental Panel on Climate Change (IPCC5), the average global surface temperature has risen by 1.5 °C since the industrial revolution, confirming that climate change is happening all over the world [
2,
3] and brings out more obvious fluctuation of precipitation and evaporation at both annual and interannual scales. This changing climate will eventually influence hydro systems, including spatial and temporal runoff distribution as well as available water resources [
4]. According to the 4th version of the World Water Development Report (WWDR), the availability of water resources will decrease as the human demand for water increases continuously. Under the dual effects of climate change and social development, some river basins are facing problems such as the frequent occurrence of extreme hydrological events including drought and flood [
5]. Climate is therefore becoming a crucial factor of vulnerable changes in global water circulation.
One of the ways to analyze climate effects on runoff, evapotranspiration, and water resources are Global Circulation Models (GCMs). GCMs provide insights of both historical and future climate scenarios. It can simulate the evolution of the earth’s climate system and its state changes over time, including the atmosphere, land surface conditions, sea, and ice [
6]. The Coupled Model Intercomparison Project 5 (CMIP5) developed new future climate scenarios called representative concentration pathways (RCPs) [
7] and gives many possibilities for future climate scenarios. Many countries and institutions have made their own GCMs to provide convenience for climatic and hydrological researchers [
8,
9,
10]. Several simulated GCMs have been used as an important input of Soil and Water Assessment Tool (SWAT) models to assess the hydrological responses to climate change in many watersheds [
3,
4,
11,
12]. However, the direct use of GCM outputs in studies of hydrological impacts still remains a challenge as the GCM output usually shows errors and uncertainties with observed data [
1,
13]. Thus, GCM output should be either downscaled to match with the basin scale [
14] or corrected to decrease the systematic bias between simulated and observed data to increase model precision and accuracy before being used in any climate and hydrological analysis.
For bias correction methods, correction techniques can be mainly classified into two categories: simple scaling technique (mainly containing linear scaling (LS) and power transformation method (PT)) and sophisticated distribution mapping methods (with Empirical cumulation distribution function (ECDF) of the most typical) [
15]. Many researchers have evaluated the performances of different bias correction methods. For example, Luo et al. [
16] compared the effects of LS, DT (Daily Transition), LOCI (Local Intensity Scaling), PT, VARI (Variance Scaling), and ECDF methods of either precipitation or temperature in the Kaidu River Basin in Xinjiang, China, and found that ECDF performs better than other methods. Teutschbein et al. [
17] also made the introduction and comparison of these different correction methods in Sweden, and also found that all the methods are effective, while distribution mapping is of relatively more success.
Beside GCMs, hydrological modeling is also a powerful tool in the analysis of climate change as it is responsible for providing information on the impacts of future scenarios in the availability of water resources based on land use. There are numerous hydrological models developed by many researchers [
18,
19,
20,
21]. In general terms, they can be classified into two categories: lumped and distributed. Lumped hydrological modeling, such as the SIMHYD model [
22] and
Génie Rural à 4 paramètres Journalier (GR4J) model [
23], places emphasis on physical principles, aiming at reproducing the non-linear water balance occurring at a finite scale in the soil [
24]. For example, Li et al. [
3] simulated and predicted the future runoff of the Tibetan Plateau by using a combination of the SIMHYD and GR4J models. Distributed hydrological models, with Shertan and the variable infiltration capacity (VIC) model being relatively typical, considered the spatial uneven distribution of environmental variables, such as precipitation and different land uses, as compared with a lumped model [
25]. It provides many simulation functions and can expand runoff simulation to water resources and environmental management [
26]. Birkinshaw et al. [
14] predicted the outflow of the Three Gorges Reservoir using the Shertan hydrological model under climate change. The semi-distributed hydrological model is another category that usually separates a large watershed into several sub-watersheds with simple structure and higher accuracy [
27]. The Soil and Water Assessment Tool (SWAT), a basin-scale and physical-based hydrological model, is one of the most used semi-distributed models for hydrological applications. For example, Muhammad et al. [
12] used global climate data to drive the SWAT model to analyze the future trends of temperature, rainfall, and runoff in different climate scenarios in northwestern Pakistan. Luo et al. [
1] constructed a harmony control model based on the coupling of the SWAT hydrological model, water quality model, and ecological model based on the harmony theory. However, the simulation results of the hydrological models often contain uncertainties including parameter calibration and selection of hydrological models, but the main source is still from the bias and low resolution of GCM outputs [
28,
29]. Although studies have focused on different methods of bias correction of GCM outputs, few studies have presented a combination method of bias correction that may reduce errors more efficiently.
Moreover, there is no doubt that climate changes influence hydrological regimes, especially evapotranspiration and runoff, but the changing hydrological variables would further simultaneously affect the socioeconomic water resources supply and demand system because the amount of water resources is mostly from precipitation. There are also a great number of studies that assess the responses of water resources under climate change. For example, Chattopadhyay et al. [
28] evaluated evapotranspiration and hydrological droughts in the Kentucky River Basin by using SWAT. Fonseca et al. [
11] also assessed the total runoff changes under future RCPs of the Tâmega River Basin in the north of Portugal by using the Hydrological Simulation Program FORTRAN (HSPF) model. Previous studies have obtained abundant results of hydrological changes under climate change in arid and semi-arid areas like northwest China. However, relatively fewer studies have put emphasis on monsoon and humid areas with the same problems. Although arid or semi-arid areas are more likely to encounter extreme hydrological events, especially droughts, monsoon and humid areas are no better than arid areas as they are more likely to encounter flood disasters, which cause numerous economic losses and a threat to human lives. For example, flood events almost happen every summer, especially in Southern China. Therefore, it is still necessary to further analyze the mutual relationship between the changing climate and hydrological processes, because future climate change is still giving greater challenges to regional water supply and demand balance in more areas, which is of practical significance to hydrology, water allocation, and scientific and sustainable water management.
In this paper, the links between climate and hydrology are studied to better understand the impacts of hydrological variables on climate change and their responses to the water resources system. The selection and analysis of GCM outputs are examined. The SWAT model is applied to simulate both the historical and future runoff based on changing climates. The main objectives of this study are to (1) analysis the historical and future GCMs outputs using several bias correction methods and their hybrid method; (2) to explore the runoff and evapotranspiration response to future climate projections; and (3) to give different strategies according to future scenarios to provide references for water management
5. Conclusions
In this study, we analyzed several GCM outputs by using both single and combined bias correction methods and selected the most suitable corrected GCM as input for the SWAT model. The historical (1964–2005) and future scenarios (2016–2055) with three RCPs were used to simulate the precipitation, temperature, evapotranspiration, and runoff. The hydrological variables indicated great heterogeneity, especially in monthly distribution. The following conclusions can be drawn in this study:
(1) All the correction methods have a positive effect compared with the original GCM. For precipitation, the ECDF-corrected method performs better than LOCI and LS, and the effect of the hybrid method containing ECDF is also more obvious than that containing LOCI. The correcting method except ECDF is less obvious in the BNU-ESM and IPSL-CM5A-MR model in terms of RMSE, but it is better in the MIROC5 model, which demonstrates the GCM itself has uncertainty and complicated determinants. For temperature, the simulating result is better than precipitation as it has less uncertainty, and VARI performs the best among the single methods.
(2) The hybrid method of correcting precipitation and temperature has the superimposed effect in terms of NSE in most cases of the study area, which means the effect of the hybrid method is more obvious than either of the single methods to some extent. Therefore, choosing a proper GCM and correcting method is also a key procedure, and this paper can provide a reference for the method of bias correction.
(3) Both the trend of temperature changes in multiyear and its annual average value are increasing and the maximum rate occurs in the RCP8.5 scenario. The monthly distribution of temperature also increased in each month. The trend of precipitation changes in multiyear showed an increasing trend in RCP4.5, while other future scenarios showed a decreasing trend, but the annual average precipitation is increasing relative to baseline, and the maximum changing rate is under RCP8.5. The monthly distribution trend of precipitation is uneven, reflected by the increase in mid-flood season, decrease in other seasons, increase in peak flow, and delay of the peak occurrence time.
(4) The annual evapotranspiration increased in all three future scenarios, but the multiyear trend increase only occurred in RCP4.5. Its positive changes also occurred in flood seasons and kept pace with historical periods in non-flood seasons. The multiyear trend of runoff increased only in RCP4.5 and the annual average of runoff decreased slightly compared with baseline, but in future scenarios, the multiyear average runoff has increased compared with baseline and the maximum runoff occurs under RCP8.5. The monthly distribution of runoff is similar to precipitation and also showed the uneven trend in the future. As the precipitation continues increasing compared with baseline and raising the temperature can accelerate the streamflow, the annual average runoff in the further future may increase.
Although this research determined the hydrological response to climate change for the Lijiang River Basin, there are some issues that require further discussion; for example, the uncertainty of GCM outputs analysis and the performance of several GCM outputs that best fit the study area. The standard should be various and comprehensive and not limited to R2 or NSE. Another valuable endeavor would be using more samples of GCM outputs. The best bias correction methods may differ from multiple GCM outputs as GCM outputs are generated based on several assumptions that cause uncertainties, and the uncertainties analysis based on more samples can be conducted in our further studies.