Assessing future runoff changes with different potential evapotranspiration inputs based on multi-model ensemble of CMIP5 projections

Highlights

• Xinanjiang model performed well in runoff simulation in North Johnstone catchment.

• Runoff projected with different ETp inputs showed similar changes in the future.

• Runoff in spring and winter would decrease in the future.

• GCMs and RCPs were major factors resulting in uncertainty in runoff projections.

Abstract

Runoff projection under future climate scenarios has been widely studied to investigate the impacts of climate change on regional water availability. However, uncertainty in runoff projection due to different ETp inputs has not been fully assessed. This study firstly adopted the physically-based Penman model, temperature-based Hargreaves model, and radiation-based Abtew, Jensen-Haise, and modified Makkink models to drive Xinanjiang (XAJ) model, thus investigating the influence of different potential evapotranspiration (ETp) inputs on runoff simulation. Then, we used the validated XAJ model to project runoff in North Johnstone catchment, northeast Australia. Lastly, we quantified the uncertainty caused by 34 global climate models (GCMs), different representative concentrative pathway (RCP) scenarios (RCP4.5 & RCP8.5), and different ETp models with the technique of three-way analysis of variance (ANOVA). We found that XAJ model performed well (R² ≥ 0.88, NSE ≥ 0.86) and showed low sensitivity to different ETp inputs in runoff simulation and projection. Under future climate scenarios, spring and winter runoff had a large decrease, which was mainly caused by the decrease in rainfall. The mean decreases in spring and winter runoff were 14.6% – 20.1% and 10.3% – 15.2% respectively by 2090s under RCP8.5. GCMs (50.9% – 67.4%) and their interaction with RCPs (35.4% – 46.6%) were the dominant factors resulting in uncertainty in runoff projection. Our study not only advanced the understanding of the impacts of different ETp inputs on runoff projection but also offered insights on the understanding of the roles different factors played in the uncertainty in runoff projection. We expect such knowledge can provide a way forward to narrow down the uncertainty in runoff projection, thus providing more robust information for policy makers in water management.

Introduction

Runoff is one of the key processes in water transport both for surface water bodies (e.g., rivers, lakes, wetlands, and oceans) and groundwater (Ghasemizade and Schirmer, 2013, Kuchment, 2004). The amount of runoff from each rainfall event has direct or indirect influences on water availability in many aspects of human activities such as agricultural and industrial production, and domestic life (Allan et al., 2020, Devi et al., 2015). Previous studies have shown that climate change with increased temperature and changed rainfall patterns has great impacts on runoff (Arnell and Gosling, 2013, Bosshard et al., 2013, Im et al., 2009). For instance, Arnell and Gosling (2013) found that more than 47% of the land surface would experience increases in mean annual runoff whereas around 36% of that would witness mean annual runoff decrease due to changes in temperature and rainfall. Meanwhile, considerable variation in the impact of climate change on runoff has been found among different regions (Arnell and Gosling, 2013, Do et al., 2017, Shen et al., 2014). Providing a robust projection of regional runoff under climate change plays a significant role in understanding local water resources management and revealing the impacts of a changing climate on local hydrological cycle (Allan et al., 2020, Yan et al., 2020).

Different kinds of methods such as climate elasticity (Xing et al., 2018, Yang and Yang, 2011), Bayesian approach (Freni et al., 2009), and hydrological model (Chiew et al., 2018, Islam et al., 2014) can be used for runoff simulation and projection. Compared with other methods, one of the most important advantages of hydrological models is that they are capable to detect the hydroclimate response to changes by offering a comprehensive and reliable approach at a certain catchment (Guo et al., 2017a, Li et al., 2020). In previous studies, hydrological models were widely used as powerful tools to investigate runoff response to climate change (Fowler et al., 2018, Pechlivanidis et al., 2016, Vaze and Teng, 2011). Specifically, historical observed rainfall and runoff sequences are used to calibrate and validate the performance of hydrological models. Then, runoff can be projected by the calibrated hydrological models forced with the climatic factors derived from global climate models (GCMs) (Arnell, 2011, Chen and Yu, 2015). For instance, Senent-Aparicio et al. (2017) investigated the influence of climate change on runoff in Mediterranean Europe with SWAT model and found that runoff would decrease due to the increase in temperature and decrease in rainfall. In Western Australia, Islam et al. (2014) downscaled climatic data from 11 GCMs under A2 and B1 emission scenarios to drive LUCICAT model for the future rainfall-runoff projection. They found that projected decrease in rainfall would result in a large decrease in runoff for Western Australia in the mid and late of 21st century.

According to closed water balance, runoff for a certain region is roughly the difference between rainfall and actual evapotranspiration (ETa) in a long-term period (Montaldo and Oren, 2018). Therefore, the estimation of ETa is expected to influence the simulation of runoff (Riegger and Tourian, 2014). In the process of runoff simulation with most hydrological models, potential evapotranspiration (ETp) is an essential input to calculate ETa used for simulating runoff (Bai et al., 2016, Li and Zhang, 2017). However, various ETp models generally produce different ETp estimates (Feng et al., 2016, Kumar et al., 1987; Kumar Roy et al., 2020). In this case, which ETp model could produce better runoff simulation is an important question to answer (Dakhlaoui et al., 2020, Oudin et al., 2005, Seiller and Anctil, 2016). In other words, will the difference in ETp estimates result in different runoff simulations/projections? Addressing this question is important in runoff projection under a changing climate as ETp is greatly influenced by future climate (Pan et al., 2015, Zheng et al., 2017).

Oudin et al. (2005) adopted ETp estimated by 27 ETp models to drive four rainfall-runoff hydrological models and investigate the influence of different ETp inputs on historical runoff simulation over 308 catchments across France, Australia, and the United States. They found that these hydrological models showed low sensitivity to ETp inputs but temperature-based and radiation-based ETp models yielded the best runoff simulation. Under future climate scenarios, Seiller and Anctil (2016) assessed the sensitivity of 20 hydrological models in runoff projection to ETp estimated by 24 different equations. They found that the different ETp inputs exerted moderate influence on runoff projection. In Korea, Bae et al. (2011) investigated the sensitivity of three hydrological models to seven ETp methods in runoff projection with downscaled climate data from 13 GCMs. They concluded that the influence of different ETp on runoff projection became larger. On the contrary, Dakhlaoui et al. (2020) found that discharge simulated by three hydrological models was not sensitive to the ETp estimates under different climate conditions. In summary, though studies about the influence of different ETp inputs on future runoff projection are becoming common, there is no consistent conclusion yet.

Another unavoidable challenge in future runoff projection is the uncertainty caused by many factors such as GCMs, hydrological models (Knutti and Sedláček, 2012, Teng et al., 2015), and emission scenarios (Woldemeskel et al., 2016). For instance, Teng et al. (2012) projected runoff based on 15 GCMs with five hydrological models in southeast Australia and found that uncertainty caused by GCMs was much larger than that caused by hydrological models. Vetter et al. (2017) investigated the uncertainty in runoff projection caused by five GCMs, four RCPs, and nine hydrological models across 12 large-scale catchments worldwide. They found that GCMs and RCPs were the main factors resulting in the uncertainty. Similarly, Chegwidden et al. (2019) found that the choice of RCPs or GCMs was the main source influencing the spread in annual streamflow volume and timing. These studies would provide useful information to quantify the dominant source of uncertainty in runoff projection. However, few of them considered the possible contribution of different ETp inputs and their interaction with other factors to the uncertainty in runoff projection. Given that ETp is essential for runoff projection, especially the influence of ETp may become larger under future climate scenarios (Seiller and Anctil, 2016), it is necessary to include ETp in the uncertainty analysis in runoff projection.

Thus, the objectives of this study are dual: 1) to investigate the influence of different ETp inputs both in historical runoff simulation and future runoff projection; 2) to quantify the relative contribution of GCMs, RCPs, ETp models, and their interaction to the uncertainty in runoff projection. To achieve these goals, we calibrated and validated Xinanjiang (XAJ), a rainfall-runoff hydrological model driven by different ETp inputs against observed historical runoff at a humid catchment in northeastern Australia. Then, we used validated XAJ model to project future runoff under RCP4.5 and RCP8.5 with climate data downscaled from 34 GCMs. Based on the projected runoff, we quantified the relative contribution of different factors with the method of analysis of variance. We expect this study can offer further insights into the impacts of different ETp inputs on runoff projection and help to clarify the role of ETp inputs on the related uncertainty. The knowledge from this study will be helpful to guide the ETp model choice in future runoff projection. Results in this study can also provide a way forward to narrow down uncertainty in runoff projection.