Variance decomposition of forecasted sediment transport in a lowland watershed using global climate model ensembles
Introduction
Forecasting sediment transport requires projecting the change in future climate. Global climate model (GCM) ensembles provide an approach to input climate variables to hydrologic modeling for forecasting sediment transport. GCM forecasts physically account for changing climate variables and these variables are useful in sediment transport models for predicting sediment loadings that could disrupt ecosystems or fill reservoirs. Robust forecasts will allow watershed managers to avert problems before they occur. With these applications in mind, we advance sediment transport research by investigating GCM projections through water and sediment models, the variance structure associated with the simulations, and sediment processes and their connectivity in a basin.
Recently, a number of hydrology research studies couple GCM projections with sediment transport modeling (see Table 1). Bussi et al. (2016) used climate scenarios by changing precipitation and temperature to predict sediment loads in the River Thames catchment, United Kingdom. Gould et al. (2016) used climate scenarios and their precipitation and temperature results to study runoff erosion in northern Rocky Mountains, United States. Shrestha et al., 2016, Shrestha et al., 2018 used precipitation and temperature from multiple GCMs, multiple emission scenarios, and sets of model parameters to investigate sediment load predictions and reservoir sedimentation in the Mekong River Basin, Laos. Wallace et al. (2017) used multiple GCMs results of precipitation, temperature, and solar radiation to project sediment transport rates at three spatial scales in Indiana, United States. Yu et al. (2017) used climate scenarios from three GCMs to project sediment discharge due to changing precipitation and temperature in the Upper Yellow River Basin, China. Giri et al. (2019) applied multipole GCMs to project impacts of climate and land use change on the sediment production in Neshanic River Watershed, New Jersey, USA. Loiselle et al. (2020) used two climate scenarios to study the change in sediment and total organic carbon in Elbow River Watershed in Alberta, Canada. Mukundan et al. (2020) applied multiple GCM projections to study the impact of climate on nutrient and sediment loadings in the New York City water supply watershed, United States. Wang et al. (2020) applied GCM projections to study climate change impacts on the riverbed, wetland topography, and ecology in Yarlung Zangbo-Brahmaputra River, China.
The recent studies predict sediment flux in a changing climate throughout the world and advance the technology of model coupling. However, sources of variance impacting projected sediment transport has been understudied despite the fact that the variance structure can be conceptualized (see Fig. 1). Table 1 reviews sources of variance when projecting sediment transport with GCMs, and these sources of variance include: ‘GCM model realizations’, ‘GCM ensemble selection’, ‘hydrology inputs’, and ‘hydrologic model parameterization’. Only GCM model realization has been widely considered in previous studies, and other variance sources were rarely investigated.
All previous studies consider the source of variance termed ‘GCM model realizations’—meaning multiple GCMs with one or more downscaling methods and emission scenarios were input to show sets of sediment modeling results. The realization of GCM and its climate modeling factors (e.g., GCM type and model version, emission scenario, downscaling method, and bias correction technique) are well reported to impact hydrologic predictions due to underlying structure and assumptions of individual GCMs as well as dynamic or statistical methods when post-processing GCM results (Tu, 2009, Sheshukov et al., 2011, Harding et al., 2012, Chen et al., 2013, Al-Mukhtar et al., 2014, Neupane and Kumar, 2015, Al Aamery et al., 2016, Al Aamery et al., 2018). For this reason, ‘GCM model realization’ is considered a prominent source of variance impacting sediment transport predictions.
The set of GCM realizations chosen to include in GCM ensembles is the source of variance termed ‘GCM ensemble selection’. Many choices of GCMs are available, but GCMs may be highly correlated with one another or poorly suited for the basin studied. GCM ensemble selection procedures have been developed by climate scientists for choosing a set of GCMs as a function of model historical performance, future climate uncertainty, and model independency (Mendlik and Gobiet, 2016, Lee and Kim, 2017, Pechlivanidis et al., 2018). ‘GCM ensemble selection’ has not been investigated previously, to our knowledge, as a source of variance in sediment transport forecasts (see column 4, Table 1).
Hydrologic variables that change in a future climate that are output from GCMs and input to hydrology models is the source of variance termed ‘hydrology inputs’. GCM projections suggest precipitation, temperature, relative humidity, net radiation, and wind speed all may change substantially in the future (McVicar et al., 2012, Wild, 2009, Willett et al., 2008, Al Aamery et al., 2018). However, many previous studies only consider precipitation and temperature inputs to sediment transport models because these variables are assumed the main drivers for changing future hydrology (Al Aamery et al., 2016, Mejia et al., 2012, Chien et al., 2013, Brekke et al., 2013). Changes in relative humidity, solar radiation, and wind speed, might also shift future sediment transport, and these sources of variance have not been considered previously, to our knowledge (see column five, Table 1).
The set of parameters chosen or calibrated in hydrologic models is the source of variance termed ‘hydrologic model parameterization’. Parameterization of a hydrology model is well recognized as a source of variance impacting streamflow and sediment transport (e.g., Moriasi et al., 2007). However, the importance of hydrologic model parameterization has been marginalized to some degree when forecasting streamflow with GCMs (Niraula et al., 2015, Al Aamery et al., 2018). Only a few studies consider this source of variance previously (see column six, Table 1).
This paper provides methods and results for decomposing the variance from the four mentioned sources to gain understanding of their relative importance on sediment transport predictions. Knowledge of the most sensitive linkages in projections can provide insight to where river managers should invest their resources when making forecasts. Knowledge of the variance structure can also provide the most important areas of future research collaboration between climate researchers and hydrologic researchers.
Another contribution of this paper is GCM projection through a sediment transport model formulated with sediment connectivity theory. Seven previous studies used the semi-distributed modified universal soil loss equation (MUSLE) embedded in the soil water assessment tool (SWAT) to predict sediment transport with GCMs, and three previous studies used distributed sediment transport models (see column seven, Tab 1). No previous studies, to our knowledge, used sediment connectivity-based models with GCMs. Sediment connectivity is defined as an emergent system property reflecting the continuity and magnitude of sediment linkages within geomorphologic compartments during a given time period (Heckmann et al., 2018). Given the increase in availability of high-resolution geospatial data, the ability to simulate explicit sediment pathways with sediment connectivity approaches has become increasingly plausible and has been shown to improve understanding and simulation of sediment processes (e.g., Vigiak et al., 2012; Heckmann et al., 2018; Mahoney et al., 2020a, Mahoney et al., 2020b, Wu et al., 2020). An attractive feature of the sediment connectivity approach is explicitly representing these pathways, which is underdeveloped in semi-distributed and fully distributed sediment models (Mahoney et al., 2020a).
Several features of sediment connectivity modeling make it a potentially advantageous approach for coupling with GCMs. First, Mahoney et al., (2018) found sediment connectivity modeling overcomes spatial limitations of semi-distributed conceptual models such as MUSLE by grounding the model in physical equations and estimating sediment transport at the raster cell scale. Second, connectivity simulations first predict the spatial domain most likely contributing sediment to the river, and in turn provide reduced computational complexity of fully distributed sediment prediction (Mahoney et al., 2018). Third, several researchers have shown sediment connectivity models can highlight hotspots of sediment production, which ultimately improves decision making when analyzing sediment forecasts (e.g., Vigiak et al., 2012; Heckmann et al., 2018; Mahoney et al., 2018, Husic et al., 2020). Fourth, sediment connectivity captures the morphometry of the basin because parameters associated with watersheds shape, size, and gradient are spatially explicit in connectivity formula (Dikpal et al., 2017, Mahajan and Sivakumar, 2018).
Sediment connectivity modeling allows prediction of the sediment processes occurring in a basin under changing climate conditions. The potential controlling sediment transport process are many (see Fig. 2), including soil erosion and transport across the landscape, erosion of bed and bank sediment in the stream corridor, redistribution, and sediment deposition throughout the basin, and sediment flux leaving the watershed. The relative importance of these processes reflects the system's net behavior as shear, transport, or source limited and the aggrading or degrading condition of the river network. Questions remain as to a system’s net behavior and in turn response under future climates. For example, GCM projections suggest wet, temperate regions of the world will experience higher rain- and snowfall, higher humidity, more cloud cover, and lower wind speed over the next century (Melillo et al., 2014, McVicar et al., 2012, Wild, 2009, Willett et al., 2008, Al Aamery et al., 2016). These climate changes are suggested to in turn shift the hydrologic cycle to produce higher runoff and higher streamflow in wet temperate watersheds (Al Aamery et al., 2018). However, this net impact on a system’s behavior as shear, transport, or source limited remains under-discussed in the hydrology community.
The objectives of this study were to: (i) investigate the uncertainty structure of sediment transport fluxes impacted by numerous sources of variance when projecting with GCMs; (ii) investigate forecasted changes in specific soil erosion and sediment transport processes found from sediment connectivity modeling; and (iii) provide results of a net climate change impact on a system’s behavior as shear, transport, or source limited for a basin located in a wet, temperate region.
Section snippets
Study site and materials
The Upper South Elkhorn Watershed (Fig. 3), located in the Kentucky river basin, Kentucky USA, has been selected for this study. The watershed area is 61.7 km2, the average slope is 5.3%, and the elevation ranges from 255 m to 324 m asl. Urban lands dominate the watershed land use with 47%, followed by agricultural lands with 38.4% and forest with 14.6%. The watershed is characterized as mild sloped with gently rolling hills; the soil is mainly silt loam; the floodplains are long and flat; and
Methods
An overview of the method is shown in Fig. 4. Our approach was to permeate the model selection criterions to select representative subsets of GCMs. Then we used SWAT to simulate the hydrologic conditions of the watershed, where we extracted the results of streamflow, surface runoff depth, and daily curve number. SWAT extracted results have been used in our process-based-connectivity and instream sediment transport models. Surface runoff and daily curve number results are inputs to the upland
Select ensembles of GCMs
Results of the model selection method provided three GCM ensemble designs with different sets of GCMs across statistical downscaling and dynamic downscaling projects (see Table 2 and Fig. 5; and Tables A, B, C, and D and Figures A and B in Supplementary material). Design #2 showed the best choice for a GCM ensemble due to its highest value of the maximum entropy measure while still considering model representativeness, historical performance, and independency.
Results shown in Table 2 tend to
Conclusion
The conclusion of this paper is as follows:
- 1.
Model evaluation results show the GCM model selection procedure based on model representativeness, historical performance, independency and lack of additional clustering for the downscaling approach provided the best ensemble design. Model evaluation of bias correction approaches showed quantile delta mapping outperformed scaled distribution mapping and quantile mapping in this study. Hydrologic and sediment model performance was found adequate for the
CRediT authorship contribution statement
Nabil Al Aamery: Methodology, Software, Validation, Formal analysis, Writing - original draft, Visualization. James F. Fox: Conceptualization, Methodology, Investigation, Resources, Writing - original draft, Writing - review & editing, Supervision, Project administration, Funding acquisition. Tyler Mahoney: Conceptualization, Methodology.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
We gratefully acknowledge the financial support of the this research under National Science Foundation (NSF) Award 163288. We thank the CMIP3, CMIP5, NARCCAP and CORDEX climate projects for making their data publically accessible.
References (79)
- et al.
Modelling hydrology and water quality in the pre-alpine/alpine Thur watershed using SWAT
J. Hydrol.
(2007) - et al.
Quantifying relative uncertainties in the detection and attribution of human-induced climate change on winter streamflow
J. Hydrol.
(2016) - et al.
Evaluation of climate modeling factors impacting the variance of streamflow
J. Hydrol.
(2016) - et al.
Variance analysis of forecasted streamflow maxima in a wet temperate climate
J. Hydrol.
(2018) - et al.
Hydrologic uncertainties in climate change from IPCC AR4 GCM simulations of the Chungju Basin, Korea
J. Hydrol.
(2011) - et al.
Modelling the future impacts of climate and land-use change on suspended sediment transport in the River Thames (UK)
J. Hydrol.
(2016) - et al.
Performance and uncertainty evaluation of empirical downscaling methods in quantifying the climate change impacts on hydrology over two North American river basins
J. Hydrol.
(2013) - et al.
Modeling the potential impacts of climate change on streamflow in agricultural watersheds of the Midwestern United States
J. Hydrol.
(2013) Frequency of precipitation and temperature extremes over France in an anthropogenic scenario: model results and statistical correction according to observed values
Global Planet. Change
(2007)- et al.
Does internal climate variability overwhelm climate change signals in streamflow? The upper Po and Rhone basin case studies
Sci. Total Environ.
(2014)