Changes in climate have the potential to significantly alter the movement of sediment through watersheds and directly affect dredging needs in rivers and harbors. There are over sixty-three commercial harbors in the Great Lakes and over 600 miles of navigation channel maintained by the U.S. Army Corps of Engineers (USACE). In 2014, an estimated 132 million tons of commodities were transported to and from U.S. ports located on the waterways of the Great Lakes system (USACE, 2014). Many of the harbors are located at the outlets of rivers that can convey large amounts of sediment, necessitating periodic dredging to maintain the navigation channels. In spite of the importance of this system, previous studies have not examined the potential impacts of projected future climate changes on both sediment yield (sediment eroded from the landscape and delivered to the river) and the dredging requirements in this region.
Current climate change projections generally show increasing temperatures and precipitation in the Great Lakes region of the United States, although the magnitude and seasonality of these changes depends on the emissions scenario and climate model (Hayhoe et al., 2010;IPCC, 2014;Pryor et al., 2013). Precipitation is expected to increase in the winter and spring, but decline in the summer; temperatures are projected to increase more in the winter during the early part of the century, with changes in summer temperatures catching up by mid-century (Hayhoe et al., 2010).
The Third National Climate Assessment found that extreme rainfall and flooding events, and their associated erosion, are on an upward trend in the Midwest, including Indiana, Michigan, and Ohio (Pryor et al., 2014).
Numerous studies have examined the potential implications of climate change on streamflow and sediment yield (e.g. Mukundan et al., 2013;Park et al., 2011;Serpa et al., 2015). In the Upper Midwest and Great Lakes regions, O'Neal et al. (2005) found that variability in soil loss would increase due to changes in crops. Two separate studies looked at climate change effects on northern Illinois watersheds and found that streamflows would decrease, based on the projected climate change scenarios (Cherkauer and Sinha, 2010;Chien et al., 2013). Several Soil and Water Assessment Tool (SWAT) models of the Maumee River have examined the potential effects of climate change scenarios. For example, Bosch et al. (2014) modeled four watersheds that drain to Lake Erie and projected that flow and sediment yield would increase, based on climate projections from two emissions scenarios and three General Circulation Models (GCMs).
In contrast, a more narrowly focused study on the Maumee that utilized three GCMs and a single emissions scenario found that annual average flow and sediment loads will decrease by midcentury (2045 to 2055), although there was significant variability in the monthly sediment loads (Verma et al., 2015). As part of a nationwide study of 20 watersheds with SWAT simulations, Johnson et al. (2015) found that five of their six climate change scenarios would likely increase flow and sediment delivery in the Maumee by mid-century (2041 to 2070).
Dredging quantities are imperfectly correlated to sediment discharge (the sediment delivered to the mouth of the river) since they depend on downstream water levels (for example, Lake Michigan levels varied 1.9 m over the last 30 years), the location where the sediment settles out relative to the navigation channels, and on the amount of funding available to conduct the dredging operations. Some studies that discuss dredging in the context of climate change do so through the lens of rising water levels (Schwartz et al., 2004;Smith, 1991) rather than looking at changes in delivery of sediment from rivers. Schwarz et al. ( 2004) used both future projections and an arbitrary scenario of Great Lakes water levels to estimate increased dredging costs at Goderich, Ontario, on Lake Huron, but did not consider the possibility of changing riverine sediment input to the harbor. Other authors consider the dredging as either one component of the overall sediment budget (Morang et al., 2013;Templeton and Jay, 2013) or as a causative effect of increased sediment delivery (Zhang et al., 2010). We are not aware of any studies that directly link projected future riverine sediment delivery to changes in dredging needs.
In this study, we used SWAT models of two large US watersheds draining into the Great Lakes to quantify the likely effects of climate change on the streamflow, sediment yield to the river, and sediment discharge at the mouth of the river. SWAT-calculated sediment loads are then input to two different statistical sediment dredging models calculated from historical dredging costs for each system. We then drive these linked models with both historical climate and future climate simulations based on downscaled scenarios from the 5 th Coupled Model Intercomparison Project (CMIP5) for both "Contemporary" (~2011-2030) and " Mid-Century" (2031-2050) periods. We run the whole suite of 234 climate models ensemble members included in the CMIP5 dataset to better understand how climate forecast uncertainties will propagate through the paired SWAT sediment transport and statistical dredging models. In the body of the paper, we discuss results for Representative Carbon Pathways (RCPs) 6.0 and 8.5, while results for the remaining RCPs (2.6 and 4.5) are in the Supporting Information. These results provide both a more comprehensive view of how climate may impact sediment yield differentially in these neighboring watersheds and a first quantification of how dredging costs may respond to climate changes.
Two large, adjacent watersheds in the southern Great Lakes were selected for this study: the St.
Joseph River and the Maumee River (Figure 1). We chose these two watersheds because of their size, proximity to each other, and dredging requirements at the river mouths in the Great Lakes.
The Soil and Water Assessment Tool (SWAT) is a semi-distributed, lumped parameter hydrologic model developed by researchers at the U.S. Department of Agriculture's Agricultural Research Service (USDA-ARS) (Arnold et al., 2012;Neitsch et al., 2011). It is often used for sediment yield studies (Alighalehbabakhani et al., 2017;Gassman et al., 2014;Krysanova and White, 2015) and is increasingly used to examine climate change impacts (Chaplot, 2007;Chien et al., 2013;Ficklin et al., 2009;Johnson et al., 2015). SWAT models split their domain into subwatersheds and then subset these into Hydrologic Response Units (HRUs). HRUs are the basic computational units of a SWAT model, which represent all of the area within a subwatershed with similar soils, slopes, and land uses.
We developed SWAT models independently for each watershed using the ArcSWAT 2012.10.0.7 plugin for ArcGIS, and used SWAT 2012 rev. 622. Digital elevation models with a resolution of 1 arc-second were obtained from the National Elevation Dataset and used to delineate the watersheds. The 2006 National Landcover Dataset was used to determine land use/land cover types and we used the default crop and harvest management parameters from ArcSWAT. Soil types and soil hydraulic properties were determined using the SSURGO database from the Natural Resources Conservation Service. Information on dams in the watersheds was obtained from the National Inventory of Dams maintained by the U.S. Army Corps of Engineers and those we deemed significant because of size or location were included in the models. We selected dams with storage greater than 1,233,000 m 3 for inclusion in the models. We also included the St. Joseph River Dam, in Fort Wayne, IN, which only has a storage of 1,078,000 m 3 while draining over 16% of the Maumee basin. These datasets were all imported into ArcSWAT and used to determine watersheds, subwatersheds for modeling purposes (shown in Figure 1), and HRUs. The St. Joseph SWAT model consisted of 32 subbasins and 278 HRUs, along with 17 dams. The Maumee SWAT model had 24 subbasins, 307 HRUs, and 5 dams.
The United States Department of Agriculture's Agricultural Research Service (ARS) provides weather data, in SWAT format, for all counties across the US. The daily data covers January 1950-December 2009, with the exception of January 2002. January 2002 was filled using SWAT's weather generation routines that create typical weather time series for the location and time period. We included 2002 in our simulations, but to avoid biasing further analyses due to the weather generation routine excluded January 2002 from goodness-of-fit calculations, and excluded the entire 2002 year from the downscaling bias analyses.
After initial set up of the models in ArcSWAT, we calibrated them using the SWAT Calibration and Uncertainty Programs (SWAT-CUP) tool (Abbaspour, 2015). We ran both models from 1980 through 2009, using a daily time step, with at least a five year spin-up period. Monthly outputs from the models were used for all comparisons. We began our calibration by using the Sequential Uncertainty Fitting version 2 algorithm (SUFI2) in SWAT-CUP to determine the sensitivity of the parameters in the SWAT models, based on the full allowable range of each parameter. We then focused our efforts in succeeding calibration iterations on those parameters that had the most significant effect on the model outputs. Waterville, OH (#04193500) using periods matching the streamflow.
The U.S. Army Corps of Engineers provided us with dredging quantities for St. Joseph Harbor, the lower Maumee River, and the Maumee Bay (M. Mahoney, personal communication, 20-Sep-2013). We created two models for dredging costs: 1) a linear regression, fit to historical dredging data and simulated modeled sediment fluxes, and; 2) a simpler 1:1 correlation between simulated sediment discharge and dredging costs (or percentage change in each). Our use of two models provides an estimate of cost model structural uncertainty, and allows us to evaluate a range of possible outcomes. To fit each model, dredging data from 1989-2009 was used for St.
Joseph Harbor and1990-2009 for the Maumee River and Maumee Bay dredging sites.
We created linear regression models between the annual dredging costs, converted to 2009 dollars using the U.S. Bureau of Labor Statistics Consumer Price Index data, and the modeled sediment discharge from the SWAT models run using the historic gage data. To evaluate possible time-lagged responses between sediment discharge and dredging, regressions were tested using simulated sediment results from the same calendar year as the dredging; the same water year as the dredging; the prior calendar and water years; and one and two year (calendar and water year) moving averages of sediment discharge. As there are two dredging sites in the Maumee Watershed, in the River itself and in the bay at its mouth, we examined regressions to the Maumee River and Bay dredging sites both separately and as a combined amount. We also added long-term average monthly water levels of Lakes Michigan and Erie to the regressions for the St. Joseph and Maumee dredging sites respectively.
The analyses presented in the main paper utilized the World Climate Research Programme's (WCRP's) Coupled Model Intercomparison Project phase 5 (CMIP5; Taylor et al., 2012) multimodel dataset. We acquired bias-corrected, spatially downscaled versions of these datasets from a publicly available archive created by the United States Bureau of Reclamation and others (Brekke et al., 2013). The temperature and precipitation data in this archive are available at a monthly time step and a spatial resolution of 1/8°. This data needed to be further disaggregated for use with the SWAT models, which use daily data at a single weather gage location for each sub-basin. We utilized all 234 CMIP5 model projections available from the archive. The CMIP5 dataset consists of multiple Representative Carbon Pathways (RCPs), run across a large number of individual models. To consider a larger ensemble, additional analyses were run with 112 CMIP3 scenarios (Meehl et al., 2007) and the results are included in the Supporting Information for this paper.
For brevity, analysis in this paper is limited to CMIP5 RCP 6.0 (Masui et al., 2011;37 projections) and RCP 8.5 (Riahi et al., 2011;71 projections), which represent a plausible range of CO2 emissions given no additional conservation efforts. These RCPs are particularly relevant given the 2011 -2050 simulation period of this study. RCP 6.0 is most similar to the older B2 Special Report on Emissions Scenarios (SRES; Nakicenovic and Swart, 2000), while RCP8. The calibration of the Maumee River model had nearly identical goodness-of-fit statistics for monthly flows at both stream gage sites. The calibration Nash-Sutcliffe efficiency was 0.79 at Waterville, OH and 0.80 at Defiance, OH (see Figure 1 for locations). Validation Nash-Sutcliffe efficiencies were 0.79 at Waterville and 0.82 at Defiance. Sediment discharge at Waterville produced a very good percent bias both for the calibration period (+4.6%) and the validation period (+2.5%). 2013) utilized a temperature and precipitation dataset that was scaled to match long-term average statistics (Maurer et al., 2002). The downscaled climate model temperatures have a mean annual bias of +0.02 °C and a standard deviation of 0.01 °C for both RCP 6.0 and 8.5, relative to the gage observations.
The precipitation values for RCP 6.0 and 8.5 have mean annual biases of -49.5 mm/yr (-5.1% of mean observed precipitation) and -48.9 mm/yr (-5.0%), respectively. The standard deviation of the precipitation values is 25.0 and 29.0 mm/yr for RCP 6.0 and 8.5 respectively. It is also important to note that the sample sizes for the two RCPs discussed are different, as there were 37 RCP 6.0 scenarios and 71 RCP 8.5 scenarios available from the archive.
Biases in the downscaled climate inputs have the potential to propagate into the SWAT model outputs. Figure 4 shows the PDFs of the simulated historical streamflow and sediment discharge at the mouth of each river. Generally, the PDFs all follow similar patterns to the observed data.
Potential differences may exist due to the small sample size of the observations. Streamflow and sediment discharge for the St. Joseph River have a slight high bias, while sediment discharge for the Maumee River has a slight low bias. The Maumee streamflow PDFs for the RCP 6.0 and 8.5 scenarios reasonably match the observed PDF. In order to limit the potential influence of this on our analysis, we used anomalies (differences between projected and historical time periods from the same data set) for the remaining analysis. Results for CMIP3 scenarios and additional CMIP5 RCP scenarios are summarized in the Supporting Information of this paper.
The biases in SWAT model outputs, when using the downscaled projections to simulate the historical time period, are most likely due to the biases in the downscaled and disaggregated CMIP precipitation and temperature data. This may be attributed to a combination of the spatial and temporal disaggregation processes used and the climate models themselves. average sediment discharge for the Maumee River. All PDFs are for the historical period (1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2003)(2004)(2005)(2006)(2007)(2008), simulated using observed climate data and downscaled climate data. The actual annual dredging expenditures and the modeled costs are shown in Figure 5. Table II shows the fit and parameters for the best linear model of dredging costs at each location, where QS, WY is the sediment discharge for the water year of interest, QS, WY-1 is the sediment discharge for the preceding water year, QS, CY-1 is the sediment discharge for the preceding calendar year, and DS, WY is the deposition in the downstream reach of the Maumee River for the current water year. The estimate of St. Joseph Harbor dredging costs had an R 2 of 0.48. The cost of dredging the two sites associated with the Maumee River were estimated as the sum of the Maumee Bay (R 2 =0.30) and Maumee River (R 2 =0.15) costs. The multiple linear regressions including water levels showed no significant improvement over the simple linear regression for either the St. Joseph Harbor or the Maumee Bay dredging sites.
Inclusion of Lake Erie water levels did improve the model fit for the Maumee River site (R 2 =0.38). This improved estimate is shown as the dashed blue line in Figure 5c. While climate change will affect future lake levels, it is unclear what the effect will be and we opted not to include it in our estimates of future dredging. S1 andS2, respectively.
The median streamflow values from the summary tables show small differences from the current climate. In contrast, the box plots in Figure 6 illustrate that the changes in median monthly streamflow are small relative to the variability across climate scenarios for the Contemporary period. This is also true for the St. Joseph River in the Mid-Century period. However, the Mid-Century Maumee streamflows have a median increase of 6.1 cms for the RCP 6.0 scenarios and 3.9 cms for the RCP 8.5 scenarios. Sediment discharge follows the same patterns as the sediment yield (Figure 7). The most significant difference is that over 75% of RCP 6.0 scenarios show an increase in sediment discharge for the Maumee relative to the current values. The percentage changes in the sediment discharge for the Maumee are similar to the simulated changes in streamflow The responses of the two adjacent watersheds are similar, with the exception of the Mid-Century period in which median streamflow, sediment yield, and sediment discharge all start to increase in the Maumee River watershed, while they continue to decline in the St. Joseph River watershed. A deeper investigation of model outputs revealed that this difference is due to the much greater proportion of agricultural land in the Maumee (Figure 1). Sediment yield from agricultural land can be significantly affected by the cover practices used, with low or no-till practices and cover crops significantly reducing the soil erosion. This also implies that the timing of large precipitation events that coincide with periods of bare ground can produce a large proportion of the annual sediment yield. The effects of climate change will depend on the coincident timing of these precipitation events and conditions, also suggesting that management will be important to mitigate the effects of climate change on sediment yield in agricultural watersheds.
The higher temperatures in the Mid-Century scenarios lead to simulated faster crop growth, producing earlier and larger harvests. This increase in agricultural production can be seen in Figure 8, which shows the change in harvested yield per hectare. A similar increase in future crop yield due to longer growing seasons has been identified as a potential effect of climate change (Pryor et al., 2014). In the model, once a crop is harvested, the land lays fallow, with little to no transpiration, until the next growing season. This allows small increases in the modeled sediment yield (due to erosion from the bare earth) as well as increased runoff that translates into increased streamflow and sediment discharge. This model phenomenon, as evidenced by a shift in evapotranspiration earlier in the year, was also noted by Ficklin et al. (2009) for a SWAT model of a highly agricultural watershed in California. This example shows the importance of looking closely at both the model results and the underlying processes. The changes in dredging costs vary between the two watersheds, the modeled time periods, across the climate models, and between the two different estimates of costs. Of note is that, for the St. Joseph River, the regression equation estimates show greater changes and variability than the 1:1 sediment discharge:dredging cost relationship, while the opposite is true for the Maumee.
This difference in response between the two approaches to estimating the future dredging costs indicates the potential importance of examining multiple approaches when using empirical models.
The historical dredging was not driven solely by the amount of sediment being delivered by the river. The areas dredged are coastal harbors on the Great Lakes and are affected by longshore transport of sediment, short time period seiche events (over hours to days), and variations in lake levels on seasonal, annual, and decadal time scales (Gronewold & Stow, 2014;Quinn, 2002). In particular, our modeling shows that the Maumee River site dredging appears to be driven by lake level variations on Lake Erie (Figure 5c). Dredged volumes are also affected by the limited budget available to the U.S. Army Corps of Engineers in any given year; there is a backlog of dredging need across the Great Lakes (USACE, 2015).
We modeled future conditions in two large watersheds, the Maumee and St. Joseph Rivers, using 108 different sets of climate change inputs representing a plausible range of CO2 emissions. In general, the median results suggest small decreases in streamflow in both watersheds, with similar decreases in sediment delivery to the river mouths. The exception to this is the Mid-Century scenario (2031-2050) for the Maumee River Watershed, where its managed agricultural landscapes are likely to drive the sediment and streamflow response of the watershed. This implies that the response of farmers to the changing climate will significantly impact the streamflow and sediment yield in agricultural areas.
There is a large amount of variation in the climate change model projections that drive similarly large variations in the predicted sediment yield and sediment discharge response. Even though averages across climate model ensembles tend to show little change, the variance is large. Of note, the differences between RCP 6.0 and RCP 8.5 scenarios are smaller than the variation across models within each scenario. The responses will also vary between watersheds depending on the dominance of agricultural lands, farming practices, soil types, and other factors.
We also estimated dredging costs using two methods and, in general, they decrease slightly at St. This study focused on the aggregated effects of a large number of downscaled climate scenarios but only a single sediment modeling framework using SWAT. Our understanding of the potential impacts of climate change could benefit by extending this research to include other sediment models and to examine the differences and variability within the CMIP5 projections.
There is also a need to explore the likely responses of farmers to lengthening growing seasons and the impacts of climate-induced changes in agricultural and management practices on the sediment regime of watersheds.
Interested readers may view additional model results in the Supporting Information accompanying this paper. Figure S1 shows the PDFs of observed and downscaled temperature and precipitation. Figure S2 shows the PDFs of mean annual temperature and precipitation for both the CMIP3 and CMIP5 scenarios. We have provided the results for all of the CMIP3 scenarios modeled and the A1b scenarios as Figures S3 to S7. CMIP5 results for all RCPs combined, as well as for RCPs 2.6 and 4.5 are presented in Figures S8 to S12. Tables S1 andS2 report the SWAT model outputs for the Contemporary and Mid-Century periods, respectively.
Table S3 reports the estimated change in dredging costs.
Processes, 24(11), 1421Processes, 24(11), -1432Processes, 24(11), . doi:10.1002/hyp.7599 /hyp.7599