You are on page 1of 14

Science of the Total Environment 653 (2019) 1557–1570

Contents lists available at ScienceDirect

Science of the Total Environment

journal homepage: www.elsevier.com/locate/scitotenv

Mapping landscape-level hydrological connectivity of headwater


wetlands to downstream waters: A catchment modeling
approach - Part 2
In-Young Yeo a,b,⁎, Sangchul Lee c,d, Megan W. Lang b,1, Omer Yetemen a, Gregory W. McCarty d,
Ali M. Sadeghi d, Grey Evenson e
a
School of Engineering, The University of Newcastle, Callaghan, NSW 2308, Australia
b
Department of Geographical Sciences, University of Maryland, College Park, MD 20742, USA
c
Department of Environmental Science and Technology, University of Maryland, College Park, MD 20742, USA
d
US Department of Agriculture – Agricultural Research Service, Hydrology and Remote Sensing Laboratory, Beltsville, MD 20705, USA
e
Department of Food, Agricultural and Biological Engineering, The Ohio State University, Columbus, OH 43210, USA

H I G H L I G H T S G R A P H I C A L A B S T R A C T

• Geographically isolated wetlands (GIWs)


affect hydrologic functioning of water-
sheds;
• SWAT improved to simulate wetlands in
an agricultural watershed;
• GIWs influenced watershed scale water
budgets and stream flow regimes;
• GIWs affected watershed hydrology at
multiple time scales under changing cli-
mate;
• GIWs induced tightly coupled surface-
groundwater systems.

a r t i c l e i n f o a b s t r a c t

Article history: In Part 1 of this two-part manuscript series, we presented an effective assessment method for mapping inunda-
Received 28 May 2018 tion of geographically isolated wetlands (GIWs) and quantifying their cumulative landscape-scale hydrological
Received in revised form 30 October 2018 connectivity with downstream waters using time series remotely sensed data (Yeo et al., 2018). This study sug-
Accepted 16 November 2018
gested strong hydrological coupling between GIWs and downstream waters at the seasonal timescale via
Available online 20 November 2018
groundwater. This follow-on paper investigates the hydrological connectivity of GIWs with downstream waters
Keywords:
and cumulative watershed-scale hydrological impacts over multiple time scales. Modifications were made to the
Wetlands representation of wetland processes within the Soil and Water Assessment Tool (SWAT). A version of SWAT with
Hydrological connectivity improved wetland function, SWAT-WET, was applied to Greensboro Watershed, which is located in the Mid-
Watershed modeling Atlantic Region of USA, to simulate hydrological processes over 1985–2015 under two contrasting land use sce-
Wetland hydrologic function narios (i.e., presence and absence of GIWs). Comparative analysis of simulation outputs elucidated how GIWs
Remote sensing could influence partitioning of precipitation between evapotranspiration (ET) and terrestrial water storage,
and affect water transport mechanisms and routing processes that generate streamflow. Model results showed
that GIWs influenced the watershed water budget and stream flow generation processes over the long-term

⁎ Corresponding author at: School of Engineering, The University of Newcastle, Callaghan, NSW2308, Australia.
E-mail address: In-Young.Yeo@newcastle.edu.au (I.-Y. Yeo).
1
Currently U.S. Fish and Wildlife Service, National Wetlands Inventory, Falls Church, VA 22041, USA

https://doi.org/10.1016/j.scitotenv.2018.11.237
0048-9697/© 2018 Elsevier B.V. All rights reserved.
1558 I.-Y. Yeo et al. / Science of the Total Environment 653 (2019) 1557–1570

(30 year), inter-annual, and monthly time scales. GIWs in the study watershed increased terrestrial water storage
during the wet season, and buffered the dynamics of shallow groundwater during the dry season. The inter-
annual modeling analysis illustrated that densely distributed GIWs can exert strong hydrological influence on
downstream waters by regulating surface water runoff, while maintaining groundwater recharge and ET under
changing (wetter) climate conditions. The study findings highlight the hydrological connectivity of GIWs with
downstream waters and the cumulative hydrological influence of GIWs as hydrologic sources to downstream
ecosystems through different runoff processes over multiple time scales.
© 2018 Elsevier B.V. All rights reserved.

1. Introduction of GIWs (between wetlands or with streams) along the entire flow path
(Golden et al., 2017) at a landscape scale. As discussed in Part 1 of this
Headwaters are essential elements of a drainage catchment. First two-part manuscript series (Yeo et al., 2018), long-term field-
or second-order headwaters make up nearly two thirds of the total monitoring data are rarely available, particularly for multiple GIWs,
length of river networks (Freeman et al., 2007) and contribute up and interpretation of in-situ measurements from selected sites may
to about half of the mean water volume and nitrogen fluxes for not properly represent the hydrodynamics of most wetlands and conse-
fourth- or higher-order rivers and navigable waters in the US quent wetland-stream surface water connections. Furthermore, there
(Alexander et al., 2007). Understanding hydrological processes in exist few process-based spatially distributed models which can simu-
headwaters is an essential step towards their protection; as sources, late a fully coupled groundwater-surface water-wetland system pro-
flowpaths, and residence time (i.e., lag) of water and other solutes cesses with required spatial details within a hydrologic modeling
throughout landscapes are highly dependent on headwater pro- framework (Lee et al., 2018b; Ameli et al., 2017; Fossey et al., 2016;
cesses (McDonnell and Beven, 2014). Evenson et al., 2015, 2016, 2018; Golden et al., 2014; McLaughlin
In the Coastal Plain of the Chesapeake Bay Watershed (CBW), wet- et al., 2014).
lands are crucial components of headwater landscapes (Lane and Hence, we presented an effective assessment method to map
D'Amico, 2016; McDonough et al., 2014; Lang et al., 2012). In part be- landscape-level hydrological connectivity between wetlands and down-
cause of their abundance and dense distribution, wetlands in this region stream waters using time series inundation products (i.e., subpixel water
illustrate gradients of hydrological connectivity with downstream wa- fraction (SWF) maps derived from Landsat time series records that show
ters, which can vary spatially and temporally from near surface and the proportion of inundated area within a 30-m pixel) and daily stream
groundwater hydrological isolation to strong surface and groundwater flow discharge data in Part 1 of this manuscript series (Yeo et al., 2018).
connection with the river network. The spatial distribution, hydrope- This approach provides implicit (i.e., spatially lumped) connectivity in-
riod, and hydrological coupling of wetlands with surrounding landscape formation (Golden et al., 2017), indicating watershed-scale hydrological
elements including river networks, are highly variable and difficult to connectivity between GIWs and downstream waters at the ‘seasonal’
measure (Ali et al., 2018). However, previous studies demonstrated time scale, during annual high groundwater levels (i.e., precipitation
that geographically isolated wetlands (GIWs), those entirely surrounded (P) N evapotranspiration (ET)). Due to the dominant wetland character-
by uplands without persistent surface water connection with the river istics of the study site (i.e., predominately forested wetlands) and the
network, can exert strong influence on nearby waters (Brooks, 2006; limited capacity of optical remote sensing, time series SWF maps were
Cohen et al., 2016; Lane et al., 2018). They can directly affect headwater only available in early spring, rather sporadically over the last 20+
hydrology, channel network development, and ecological integrity of years (1985–2010). Hence, continuing assessment and monitoring of
downstream waters in this region (Alexander et al., 2018; Calhoun wetland ecosystems from the seasonal to the intra-annual and multi-
et al., 2018; Lee et al., 2017, 2018b; McDonough et al., 2014). GIWs func- decadal time scales using this methodology was not possible.
tion as water sources to downstream ecosystems through different run- In this paper, we present a comprehensive modeling approach
off processes (e.g., overland flow (i.e., fill and spill), near surface or aimed to improve understanding of hydrological functions and connec-
shallow subsurface water, deep groundwater, preferential flow through tivity of GIWs with surrounding landscapes and river networks over
macrospores (developed by wetland vegetation) or shallow perched multiple time scales. We used a modified version of the Soil and
groundwater (formed in dual porosity soil profiles), etc.), with varying Water Assessment Tool (SWAT), a semi-distributed physically based
time lags (Alexander et al., 2018; Ali et al., 2018; Golden et al., 2017). watershed model, to simulate hydrologic processes for an agricultural
The size, density, and relative positioning of GIWs along with other land- watershed with dense distributions of GIWs. SWAT was modified with
scape and climate factors affect the degrees and types of hydrological improved process representation for wetlands (referred to as SWAT-
connectivity between wetlands and downstream waters (Lane et al., WET, SWAT for Wetland, hereafter), so that site-specific onsite charac-
2018). In part because of this landscape-level connectivity, recent stud- teristics, upstream flow contribution to wetlands, and wetland interac-
ies highlight the importance of protecting temporary waterways and tion with stream networks could be better accounted for (Lee et al.,
geographically isolated wetlands, despite the fact that hydrologic cou- 2017, 2018b; Yang et al., 2016; Evenson et al., 2015; Liu et al., 2008).
pling between GIWs and river networks through surface water connec- We then set up SWAT-WET with more spatialized wetland parameters
tions can be episodic and/or intermittent (Lane et al., 2018; Alexander using SWF maps and other geospatial data (Lee et al., 2017, 2018b). The
et al., 2018; Leibowitz et al., 2018; Evenson et al., 2018; Lee et al., simulation was conducted at a daily time step under two contrasting
2018b; Golden et al., 2017; Cohen et al., 2016). land use scenarios (i.e., with the presence and absence of GIWs) over
The hydrological connectivity of wetlands (as measured by fre- the period of 1985–2015. These two scenarios were set to elucidate
quency, duration, timing, type, and magnitude) to the river network how GIWs in aggregate could influence the partitioning of precipitation
and other water bodies is site-specific. There is a great need to quantify between evapotranspiration and terrestrial water storage, and conse-
hydrological connectivity to better inform conservation and restoration quently affect water transport mechanisms and routing processes that
of these ecosystems and the services that they provide (Ali et al., 2018). generate streamflow at the watershed scale. The modeling results
For those watersheds with a dense distribution of GIWs (e.g., coastal showed the types and frequency of the watershed-scale hydrological
watersheds located in CBW), it is extremely difficult and thus impracti- connectivity of GIWs and the cumulative impacts on downstream wa-
cal to measure, model, and represent ‘explicit’ hydrological connectivity ters over multiple time scales.
I.-Y. Yeo et al. / Science of the Total Environment 653 (2019) 1557–1570 1559

2. Data and methods According to the Soil Survey Geographical database (SSURGO; https://
www.nrcs.usda.gov/wps/portal/nrcs/site/soils/home/), these soils be-
2.1. Description on the study area long to Hydrologic Soil Group (HSG) ‘C’ and ‘D’ types, with low infiltra-
tion capacity (b4 mm/h or 0.15 in/h) through the soil profile (Fig. 1d).
This study was conducted on the same site as Part 1 (Yeo et al., However, SSURGO data show that the uppermost soil horizons below
2018), the Greensboro Watershed (GBW) (Fig. 1a). The 292 km2 study GIWs are in fact rather silty (~23%) or sandy (~41%) with less clay con-
watershed includes a high portion of forest land cover (~48.3%), and tent (~6%). The top soil layers where wetlands are located (i.e., wetland
cropland (36.1%) (Fig. 1b). The study area is characterized by relatively bottom), have rather coarse soil texture (e.g., coarse-loamy to fine-
flat topography (0–26 m above sea level; Fig. 1c). A large portion of soils loamy, sandy, or fine-silty); hence, a large portion (N70%) of these
in the study watershed are hydric (i.e., poorly to very poorly drained). soils have very high hydraulic conductivity (K; e.g., K = 2160 mm/h)

Fig. 1. The study watershed showing: (a) the geographical location of Greensboro watershed (located in the state of Maryland and Delaware, USA); (b) the land use distribution;
(c) elevation; (d) soil characteristics as shown by USDA-NRCS Hydrologic Soil Group; (e) saturated hydraulic conductivity of wetland bottoms by SSURGO; and (f) the spatial
distribution of geographically isolated (GIW) and surface water connected (SCW) wetlands and the delineated subwatersheds.
1560 I.-Y. Yeo et al. / Science of the Total Environment 653 (2019) 1557–1570

(Fig. 1e). However, the hydraulic conductivity below these top wetland with a clear upward trend similar to the regional trend. Precipitation
soil layers is much lower (b400 mm/h). This contrasting wetland soil was rather evenly distributed throughout the seasons. Estimated
characteristic is mostly noticeable for poorly drained soils (D type). As mean annual potential evapotranspiration (PET; ~1180 mm) was close
discussed in Part 1 (Yeo et al., 2018), forested wetlands are often classi- to MAP, while monthly PET over the growing season (May to Septem-
fied as forested land cover by the National Land Cover Database (NLCD; ber) exceeded precipitation by ~230 mm (see details in Section 3.4).
https://www.mrlc.gov/), but the U.S. Fish and Wildlife Service National PET was estimated using the Penman-Montheith method. The Palmer
Wetlands Inventory (NWI) geospatial dataset shows ~ 25% of forests Drought Severity indices provided by the National Oceanic and
within the watershed are forested wetlands and most are GIWs (Yeo Atmospheric Administration (NOAA) National Centers for Environmen-
et al., 2018; Lang et al., 2012) (Fig. 1f). tal Informatics (NOAA NCDC, 2015) showed that the study region expe-
Climate in the study region is highly variable, with a strong seasonal rienced highly variable climate conditions during the simulation period
cycle, year to year variations, and a wide range of extreme climate (from very wet to drought conditions), as discussed in Part 1 (Yeo et al.,
events (e.g., floods, droughts, heat waves, hurricanes, and ice storms). 2018).
However, this Northeastern region of the US experienced clearly
increasing temperature and precipitation trends over 1895–2011 2.2. Wetland-watershed modeling approach
(Kunkel et al., 2013). For example, regional precipitation showed statis-
tically significant (at a 95% confidence level) positive anomalies during A schematic overview of the study approach is presented in Fig. 2.
fall (6.1 mm/ decade) and for the annual total (10 mm/decade), when SWAT-WET was applied to simulate hydrological processes at a daily
assessed using the climate record over 1901–1960 as the reference pe- time step over the period of 1985 to 2015. Two land use scenarios
riod (Kunkel et al., 2013). While annual precipitation varied over time, it were set up as: (1) with GIWs as per current land use distribution (re-
showed a clear shift towards greater variability with higher totals since ferred to as WetLU hereafter) and (2) replacing GIWs with upland for-
1970. The historical record (over 1895–2011) showed that the study re- ested land cover (the dominant wetland vegetation type in the study
gion experienced its three driest years prior to the 1970s (i.e., 1930, area as indicated by the NWI geospatial dataset; denoted as NoWetLU,
1941, and 1965), but very wet summers since 2000 (e.g., wettest sum- no wetland land use, hereafter). Under the NoWetLU Scenario, GIWs
mers in 2006 and 2009). (i.e., “sinks”) were filled using the DEM preprocessing tool embedded
Local climate data over the period of 1985–2015 showed that the in ArcSWAT (Winchell et al., 2010). This tool reconditions the DEMs to
study area received mean annual precipitation (MAP) of ~1130 mm be depressionless by raising the elevation values of sinks, so that each

Two Land Use Scenarios (with presence or absence of GIWs)

(a) SWAT-WET (b) GIW parameters (NWI, soil)

DEM, soil, land use & management, Model simulaon


climate input

SWAT-WET Model

Subwatershed

GIW drainage zone


[A]
[B*]
SCW drainage zone
(c) SCW parameters (NWI, SWF map, soil)

Direct drainage zone Model simulaon

Routing to the subwatershed outlet

Routing to the watershed outlet

Multi-temporal statistical analyses of terrestrial water storage (S; dS/dt=P-ET-Qs) and streamflow
(Qs) generation process at the watershed scale

Fig. 2. Overview of the study approach with improved process representation for wetlands within SWAT-WET [adapted from Lee et al., 2018b; Yang et al., 2016; Evenson et al., 2015; and
Neitsch et al., 2011].
I.-Y. Yeo et al. / Science of the Total Environment 653 (2019) 1557–1570 1561

grid cell can be assigned a specific flow direction based on an eight- SWAT simulates soil moisture, surface water runoff, and infiltration
direction (D8) method (Winchell et al., 2010). Then rainfall-runoff char- based on the Soil Conservation Service (SCS) Curve Number (CN)
acteristics were estimated based on the Curve Number (CN value spe- method at the HRU level (Neitsch et al., 2011). The CN value is statisti-
cific to forest land cover and soil type adjusted by slope (see details in cally determined based on soil permeability, land use, and antecedent
Sections 2.2.1 and 2.2.2). soil water conditions and further adjusted by slope. Note that SWAT of-
Using those simulated daily water fluxes under the two contrasting fers another method for soil moisture prediction via the Green & Ampt
land use scenarios, we calculated terrestrial water storage (S; dS/dt = P- method, to estimate infiltration with sub-daily precipitation data
ET-Qs, where Qs denotes streamflow and t time) for the watershed, and (Neitsch et al., 2011). Infiltrated water is redistributed through a soil
investigated the overall mean, seasonal (intra-annual), and inter- profile, using a storage routing technique through each soil layer (up
annual hydrological impacts of GIWs at the watershed scale. In particu- to 12 layers) in the root zone. A soil layer is assumed to be saturated if
lar, we focused on quantifying the role of GIWs in: (1) recharging soils water content of the soil layer exceeds field capacity. When field capac-
and groundwater by seepage after the partitioning of precipitation be- ity of a soil layer is exceeded and the layer below is not saturated, per-
tween atmospheric (AET) and terrestrial water storage (S), and (2) af- colation (i.e., downward flow through soil profile, PERC) occurs with
fecting overall water transport mechanisms and routing processes to flow rate governed by the soil layer's saturated hydraulic conductivity
generate downstream flow discharge over multiple time scales. Note (Fig. 2b). Soil waters between layers are further removed by plant
that terrestrial water storage (dS/dt) and streamflow regimes were esti- water uptake and soil water evaporation. In addition, upward water
mated based on changes in total water fluxes (from surface, soil, and movement within shallow groundwater by capillary rise (to soils) and
groundwater, including water stored in depressional storage) moving plant water uptake directly from the shallow groundwater as a form
through a watershed. The two contrasting land cover scenarios allowed of ET are simulated using ‘revap’ in SWAT (Fig. 2b). Saturated flow is
isolation of GIW hydrological effects, making it possible to evaluate rel- simulated to move laterally, based on slope, hillslope length, and satu-
ative effects cumulatively, by comparing differences in estimated water rated hydraulic conductivity using a kinematic storage model. This lat-
fluxes and flow regime variables at the watershed scale. eral subsurface flow contributes to stream flow.
A series of statistical analyses were conducted to demonstrate the In SWAT, evapotranspiration (ET) is estimated, including evapora-
statistical significance of GIW hydrological connectivity to downstream tion from terrestrial land surfaces (soils, water bodies, and vegetative
water and its cumulative hydrologic influence over seasonal to multi- surfaces), transpiration from plants, and sublimation. Three methods
decadal time scales, in comparison with simulation results under are available to calculate potential evapotranspiration (PET): the Har-
NoWetLU. We tested: (1) if the simulation results between the two sce- greaves, the Priestly-Taylor, and the Penman-Montheith methods. Ac-
narios were statistically different, using matched pairs difference tests tual evapotranspiration (AET) is then estimated by adjusting PET
(both parametric t-test and nonparametric Wilcoxon matched-pairs considering water uptake by plants from soils, soil depth, and soil
signed-ranks test); and (2) if simulated hydrologic fluxes showed statis- water content. Water uptake by plants is calculated based on biomass
tically significant change patterns from seasonal to inter-annual time production and depth of root development in soil. Biomass production
scales, with a linear regression method and nonparametric Mann- is quantified by leaf area index (LAI; the relative area covered by leaves
Kendall trend test. in a HRU), and root depth is estimated as a fraction of total biomass
The following subsections are organized to provide a brief overview partitioned to roots according to plant type. SWAT estimates LAI as a
of SWAT (Section 2.2.1) and SWAT-WET (Section 2.2.2), and to explain function of Potential Heat Units (PHU) on a daily basis. Further details
input data, model set-up (Section 2.2.3), and calibration (Section 2.2.4) on SWAT are available from Neitsch et al. (2011).
of SWAT-WET to simulate the study watershed.

2.2.1. Model description on SWAT 2.2.2. SWAT-WET: Improved wetland representation


SWAT is a semi-distributed, process-based, continuous watershed Significant efforts have been made to enhance process representa-
model. It was developed to predict long-term impacts of agricultural tion of wetlands in SWAT and to develop methods to set-up more spa-
land management practices on water resources, sediment, and other tially explicit wetland parameters using remotely sensed and other
water quality variables at multiple spatial scales (Neitsch et al., 2011; geospatial data (Lee et al., 2017, 2018b; Yang et al., 2016; Evenson
Arnold et al., 1998). Basic hydrologic simulation (i.e., calculation of all et al., 2015, 2016, 2018; Liu et al., 2008). Based on previous studies,
vertical water and other solute fluxes) is performed at the hydrologic SWAT-WET was improved to: (1) constrain inflow into a wetland
response unit (HRU) level. HRUs are defined by unique combinations based on its relative spatial positioning within a subwatershed; and
of soil, land use, and slope, the basic landscape parameters that deter- (2) allow for bi-directional water exchange between riparian wetlands
mine rainfall-runoff characteristics. Simulated water and other solute (referred to as SCWs, surface water connected wetlands, hereafter)
fluxes at the HRU level (e.g., Eq. (1)) are then aggregated at the and a stream segment at the subwatershed scale. These changes directly
subwatershed (as a weighted sum by the HRU area) and routed to the affected the water budget of individual wetlands (e.g., GIW HRUs in
watershed outlet following the stream network. Subwatersheds are de- Fig. 2b) and other landscape elements and improved the representation
lineated as the contributing area for individual stream segments. SWAT of hydrologic connectivity between wetlands and other landscape
simulates water and solute transport mechanisms via surface water elements.
runoff, lateral flow (i.e., interflow from unsaturated soils), and ground- SWAT-WET further subdivides a subwatershed into three drainage
water from HRUs to the streams. Based on the water mass balance equa- zones (Fig. 2a), i.e., SCW drainage, GIW-drainage, and direct (i.e.
tion, soil moisture storage at the HRU level (Fig. 2b) is given as: stream) drainage zones, based on the spatial distribution of wetlands
relative to flow paths to a stream segment. Only those water fluxes (in-
cluding surface runoff (SR), lateral flow (LW), and groundwater flow
X
t
SW t ¼ SW 0 þ Pi −ET i −Q i −Di ; ð1Þ (GW)) generated from HRUs in a SCW drainage zone enter into SCWs
i¼1 (Fig. 2c). SCWs are simulated at the subwatershed scale, while GIWs
are simulated at the HRU level. The wetland (w) water budget, Vw (in
where SWt is soil water content above the wilting point at the end of day m3 H2O), is estimated as:
t, SW0 is initial soil water content, Pi is daily precipitation, ETi is daily
evapotranspiration, Qi is daily surface and sub-surface runoff flow, and
V w ¼ V w;stored þ V w;pcp −V w;evap þ V w;flowin −V w;seep −δGIW VGIW
flowout
Di is daily drainage into deep aquifer from the soil profile. All hydrologic
components are estimated as mm H2O normalized by the HRU area.  δSCW VSCWrcfflow ; ð2Þ
1562 I.-Y. Yeo et al. / Science of the Total Environment 653 (2019) 1557–1570

where Vw, stored is previously stored water in the wetland, Vw, pcp is pre- 2015). Routing excess surface water runoff and subsurface flows from
cipitation falling onto wetland surface area, Vw, evap is ET from open sur- GIWs directly to the subwatershed stream reach without interception
face and wetland vegetation, Vw, flowin is inflow to wetland (including SR, from intermediate GIW storage reservoirs, potentially overestimated
LW, and GW), Vw, seep is water loss from the wetland bottom by seepage, the quantity and rate of overland and subsurface flows entering the
VGIW SCW
flowout is outflow occurring as spillage from a GIW, and Vrcfflow is the vol-
stream network without time lags, and therefore underestimated the
ume of surface and sub-surface water exchange between a SCW and the hydrologic impacts of GIWs.
stream reach segment determined by relative water levels between the
two systems moving towards an equilibrium. δGIW and δSCW (∈{0, 1}) in-
2.2.3. Input data and model set up for SWAT-WET
dicate the membership of w, so that that VGIW
flowout occurs only for GIWs
SWAT-WET input data consist of various geospatial and climate data
(i.e., δGIW = 1 only if w = GIW; otherwise, 0) and VSCW rcfflow for SCWs (Table 1). A high resolution (1-m) lidar-based Digital Elevation Model
only (i.e., δSCW = 1 only if w = SCW; otherwise, 0), respectively. (DEM), land use map, and SSURGO were used to characterize the
SWAT-WET allows water inflow to a GIW to be directly associated study site and to delineate the stream network, subwatersheds, wet-
with its spatial positioning and other locational characteristics. For ex- land drainage zones, and HRUs. Daily precipitation and temperature
ample, SWAT-WET delineates the upstream catchment (i.e., those up- were obtained from three weather stations located near the GBW, and
gradient contributing HRUs) that directly drains to a GIW HRU and al- operated by the NOAA National Climate Data Center (NCDC)
lows their surface and subsurface (including soil and groundwater) (Chestertown: USC00181750, Royal Oak: USC00187806, and Greens-
water inflow draining to a GIW HRU (Evenson et al., 2015) (Fig. 3). boro: US1MDCL0009; Fig. 1a). Daily solar radiation, relative humidity,
Not only precipitation input to a GIW, but also partitioning of vertical and wind speed were not monitored, so they were generated using
fluxes occurring within a GIW is determined by site-specific character- SWAT's built-in weather generator (Neitsch et al., 2011). A daily
istics. For example, the amount of water loss by seepage (Vw,seep) is de- streamflow dataset collected at the outlet of the GBW was obtained
termined by dominant soil characteristics (i.e., the saturated hydraulic from the US Geological Survey (USGS) gauge station (#01491000;
conductivity) in the uppermost soil horizon below the GIW (Neitsch Fig. 1a). A commonly available drought index (i.e., Palmer Drought Se-
et al., 2011). AET (Vw,evap) from a GIW was estimated by adjusting PET verity Index, PDSI) was used to characterize weather conditions
with the typical evaporation compensation factor for open water sur- (e.g., drought, normal, and wet) based on historical climate records. It
faces (i.e., 0.5 for the potholes/depressions) and seasonal variation of was downloaded from the NOAA NCDC for the climate division (MD Re-
leaf area index (LAI) relative to open water surface area (i.e., the sea- gion 5) within which the study watershed is situated (NOAA NCDC,
sonal fractional vegetation cover for transpiration estimation) as an in- 2015). A summary of input data used for this study is presented in
dependent HRU (Neitsch et al., 2011). Outflow from a GIW-HRU occurs Table 1.
as seepage to soils and groundwater (Vw,seep), or spillage (VGIW
flowout ) to the
The study watershed was divided into subwatersheds and further
stream segment (i.e., direct drainage zone) or to a SCW (i.e., SCW drain- into HRUs following Winchell et al. (2010). HRUs were delineated first
age zone) depending on spatial position (Fig. 3). by applying a threshold (%) of 0, 0, and 0 to land use, soil type, and
Note that representation of wetland processes was improved within slope (i.e., no overlapping characteristics amongst HRUs), respectively.
the semi-distributed structure of SWAT. SWAT-WET does not represent To simplify spatial representation of wetlands and their computation re-
spillage from wetlands to the downstream landscape along the flow quirement, NWI polygons were first aggregated based on their proxim-
path. Hence, it cannot simulate “spill-and-fill” or “spill-and-merge” ity to each other (with b1 m threshold). Then, HRUs were further
inter-GIW hydrologic relationships (Lane et al., 2018; Evenson et al., processed to identify individual GIWs and their up-gradient catchment
2015, 2018). This process has been observed elsewhere (e.g., prairie areas as unique HRUs following Evenson et al. (2015) (Fig. 3). In addi-
potholes and California vernal pools), but it is less likely to occur tion, SCW NWI polygons that have their boundary spread over multiple
amongst wetlands in this study area (Lane et al., 2018; Evenson et al., subwatersheds were separated into multiple entities to conform to
2015, 2018). Given climatic and topographic characteristics, wetlands subwatershed delineation (Lee et al., 2017). In total, ~3500 GIW and
in our study site would not frequently exhibit overland flow between ~90 SCW NWI polygons were processed, and the study watershed was
GIWs, and between GIWs and the reach network (Evenson et al., divided into 80 subwatersheds and ~23,340 HRUs (Fig. 1).

Fig. 3. The spatial delineation of GIW-HRUs and contributing catchment HRUs illustrated in a hypothetical subwatershed [adapted from Evenson et al., 2015].
I.-Y. Yeo et al. / Science of the Total Environment 653 (2019) 1557–1570 1563

Table 1
List of input data for SWAT-WET.

Data Source Description Year

DEM MD-DNR lidar-based 1 m resolution 2006


[http://dnrweb.dnr.state.md.us/gis/data/lidar/].
Land use USDA-NASS Cropland Data Layer (CDL) 2008–2012
[https://www.nass.usda.
gov/Research_and_Science/Cropland/Release/index.php]
MRLC National Land Cover Database (NLCD) 2006
[https://www.mrlc.gov/]
USDA-FSA-APFO National Agricultural Imagery Program digital Orthophoto quad imagery 1998
[Sexton et al., 2010]
US census bureau TIGER road map 2010
[https://www.census.gov/geo/maps-data/data/tiger.html]
Soils USDA-NRCS Soil Survey Geographical Database (SSURGO) 2012
[https://www.nrcs.usda.gov/wps/portal/nrcs/main/soils/survey/]
Climate NCDC Daily precipitation and temperature 1983–2015
[NOAA NCDC, 2015]
Streamflow USGS Daily streamflow 1983–2015
[https://waterdata.usgs.gov/nwis/rt]
Wetland NWI National Wetlands Inventory
[https://www.fws.gov/wetlands/]
Inundation Huang et al. (2014) Annual time series inundation maps developed from Landsat images and 1985–2010
maps Jin et al. (2017) LiDAR intensity data
(SWF Maps)

Note: the CDL shows the spatial distribution of croplands and placement of crop rotations. The digital Orthophoto quad imagery, NWI and NLCD were used together to identify croplands,
wetlands, other land use types (e.g., forest, urban, etc.). MD-DNR: Maryland Department of Natural Resources, USDA-NASS: USDA-National Agricultural Statistics Service, MRLC: Multi-
Resolution Land Characteristics Consortium, USDA-FSA-APFO: USDA-Farm Service Agency-Aerial Photography Field Office, TIGER: Topologically Integrated Geographic Encoding and
Referencing, and USDA-NRCS: USDA-Natural Resources Conservation Service.

Then, the post-processed NWI map, lidar DEM, and SWF maps were volume-surface area relationship are presented by Lee et al. (2017,
further analyzed using the ArcGIS Cartography Tool (ArcGIS v 10.2), to 2018b) and Liu et al. (2008). Table 2 shows the list of wetland parame-
describe wetland geometry and to compute water storage capacity per ters used for SWAT-WET.
NWI polygon. We used the GIS-based method by Lane and D'Amico Note that we defined SCWs rather specifically to support the concep-
(2010) to estimate maximum storage capacity. Inundated surface area tual representation of riparian wetlands in SWAT-WET; i.e., SCWs were
within a wetland, estimated from the SWF maps produced under differ- identified as those wetlands that are likely to be connected directly to a
ent weather conditions, was used to set up key parameters that control stream segment through ‘bi-directional’ surface water flow exchange
net flow moving through a wetland (Lee et al., 2017) and to estimate in- (Lee et al., 2017; Yang et al., 2016; Liu et al., 2008). These wetlands
undated surface area from the water volume calculation as shown in were identified using an ‘intersect’ operation in ArcGIS 10 (i.e., based
Eq. (2). Further details on wetland geometry parameters and a wetland on spatial adjacency) that employed the aggregated NWI data and the

Table 2
List of calibrated parameters.

Parameter Description (unit) Range Value

CN⁎ SCS runoff curve number −50–50% −12%


ESCO⁎ Soil evaporation compensation factor 0–1 0.01
OV_N Manning's “n” value for land cover to estimate overland flow 0.01–3 2
SLSUBBSN Average slope length (m) −50–50% 24%
EPCO⁎ Plant uptake compensation factor 0–1 0.72
CANMX# Maximum canopy storage (mm H2O) 0–100 0.27
GW_DELAY⁎ Groundwater delay (days) 0–500 3.67
ALPHA_BF⁎ Baseflow alpha factor (1 days−1) 0–1 0.01
GWQMN⁎ Threshold depth of water in the shallow aquifer required for return flow to occur (mm H2O) 0–5000 4.06
GW_REVAP⁎ Groundwater “revap” coefficient 0.02–0.2 0.15
REVAPMN⁎ Threshold depth of water in the shallow aquifer for “revap” to occur (mm H2O) 0–500 164
SOL_AWC⁎ Available water capacity of the soil layer (mm H2O·mm soil−1) −50–50% 20%
SOL_K⁎ Saturated hydraulic conductivity (mm·hr−1) −50–50% −24%
SOL_Z⁎ Depth from soil surface to bottom of layer (mm) −50–50% −22%
CH_N2† Manning's “n” value for the main channel 0.01–0.3 0.01
CH_N1† Manning's “n” value for the tributary channels 0.01–30 0.12
SURLAG⁎ Surface runoff lag coefficient 0.5–24 1.88
SCW_NSA The surface area of a SCW at the normal water level (ha) – 0.01–36.3
SCW_MXSA The surface area of a SCW at the maximum water level (ha) – 0.23–79.6
SCW_NVOL The volume of a SCW at the normal water level (104·m3) – 0.01–6.4
SCW_MXVOL The volume of a SCW at the maximum water level (104·m3) – 0.02–98.5
SCW_K^ Hydraulic conductivity of the SCW wetland bottom (mm·hr−1) – 0.1
SCW_EVCOEFF^ Evaporation coefficient for SCW – 0.6
GIW_MXVOL The volume of a GIW at the maximum water level (104·m3) – 1.1–742.8
GIW_K Hydraulic conductivity of a GIW wetland bottom (mm·hr−1) – 72–2160.0
GIW_EVCOEFF^ Evaporation coefficient for GIW – 0.5

Note: The ranges of parameters with superscripts (*, #, †, and ^) are derived from Gitau and Chaubey (2010), Wu and Xu (2006), and Yen et al. (2014), respectively. Wetland parameters
(shaded) were derived from the SWF maps and a high resolution LiDAR. Refer to Lee et al. (2017, 2018b) for further details on the data preparation and processing for wetland parameters
in the study site.
1564 I.-Y. Yeo et al. / Science of the Total Environment 653 (2019) 1557–1570

stream map. All other wetlands were operationally defined as GIWs. It is PPU). The 95 PPU was calculated using all simulation outputs obtained
noteworthy to mention that the identification of GIWs and SCWs at a during the calibration procedure with different combinations of param-
landscape scale is subject to the unique character of geospatial input eter values in the allowable range.
data (e.g., NWI, stream maps, and DEM) and the methods used for the SWAT-WET was calibrated to daily stream flows collected at the wa-
spatial analysis (e.g., buffer distance and flow routing) (Lang et al., tershed outlet over the period of 2001–2005, and validated against
2012). Due in large part to the targeted mapping unit employed by those of 2006–2010, after a 2-year warm-up period (1999–2000).
NWI data creators, use of NWI data as part of this study will lead to a Then SWAT-WET was set to simulate hydrology retrospectively, over
higher level of uncertainty than use of detailed field data (Lang et al., the period of 1985–2015 (after 2-year spin-up period of 1983–1984),
2012). However, the use of field data is not practical and at the time of in order to assess the long-term cumulative impacts of GIWs for the
data creation the NWI geospatial dataset is the most accurate, as well study watershed. Table 2 provides the list of calibrated parameters
as spatially and categorically detailed wetland dataset available for the and allowable ranges reported from previous work. Note that the two-
US. Thus, NWI is the most suitable dataset for parameterizing hydro- step calibration procedure used in this study was necessary to reduce
logic models, especially those that take into account the spatial relation- the computational complexity and high resource demand imposed by
ship between wetlands and streams, since it provides more spatially the use of wetland modules with daily simulation, as they involve ex-
explicit wetland boundaries and positions, as well as information on hy- tremely large numbers of HRUs and associated parameters. This calibra-
drologic regimes. Furthermore, it should be noted that the stream map tion scheme (i.e., setting the initial parameter values from the previous
used in this modeling study was delineated to support SWAT modeling calibration obtained with the unmodified SWAT and then allowing it to
and landscape-level (e.g., HRU and subwatershed) analysis within its be changed with SWAT-WET) may lead to the local optima, given the
semi-distributed structure. Though the SWAT derived stream map was highly nonlinear nature of the hydrologic system and its representation
similar to the USGS NHD High Resolution Data Product (which is now in SWAT (Arnold et al., 2012; Yeo and Guldmann, 2010). More thorough
provided with NWI Version 2 geospatial dataset), it underestimated evaluation with different initial conditions may produce different cali-
stream and ditch presence and length (Lang et al., 2012). Hence the ‘in- bration results.
tersect’ operation based on this stream network overestimated GIWs.
Regardless, these NWI polygons are likely to be headwater wetlands 3. Results and discussion
that contribute to first order streams via a one-way surface water con-
nection (McDonough et al., 2014; Lang et al., 2012). The hydrologic con- 3.1. Model calibration and validation
tribution from headwater wetlands is captured appropriately within
SWAT-WET, as their outflow is directly routed to streams. Given process Fig. 4 and Table 3 show that calibrated SWAT-WET reproduced
representation of wetlands within SWAT-WET, it is most unlikely that streamflow reasonably well. Both daily and monthly simulation results
this delineation results in significant change in simulation outputs. met the ‘satisfactory’ performance criteria according to the hydrologic
model performance rating described by Records et al. (2014) and
2.2.4. SWAT-WET calibration Moriasi et al. (2007), i.e., NSE N 0.2 and P-bias ≤ ±25% for daily simula-
Several SWAT modeling studies have already been conducted on the tion (Records et al., 2014), or 0.5 b NSE ≤ 0.65, 0.6 b RSR ≤ 0.7, and ± 15%
study area to simulate stream flows and other water quality variables ≤ P-bias b ±25% for monthly simulation (Moriasi et al., 2007). Note that
(Lee et al., 2016, 2017, 2018a, 2018b; Sharifi et al., 2016; Yeo et al., model performance criteria are generally relaxed for daily simulations,
2014; Sexton et al., 2010). Previous studies reported a list of sensitive as they capture much more detailed high frequency processes and
parameters, their allowable ranges, and recommended (optimized) pa- hence naturally produce less accurate simulation results, compared to
rameter values for the study site that exceeded the performance criteria longer time scale (e.g., monthly or yearly) modeling outputs (Sharifi
as outlined by Moriasi et al. (2007) and Records et al. (2014). Capitaliz- et al., 2016; Records et al., 2014; Arnold et al., 2012).
ing on previous SWAT studies conducted on the study site, we cali- Overall, SWAT-WET performed better during the calibration period,
brated SWAT-WET with additional wetland parameters. A number of but a large bias was observed during the validation period. The different
parameter values uniformly distributed within the allowable ranges climate and streamflow conditions between the two split sample evalu-
identified from the previous studies were generated and evaluated fol- ation periods contributed to the model bias. The validation period
lowing the SWAT model technical guidelines (Arnold et al., 2012). We showed more variable weather conditions with extreme storms and
selected those parameter values that provided best stream flow simula- droughts, leading to more variable streamflow responses compared to
tion results while meeting the performance criteria by Moriasi et al. the calibration period. Those parameters calibrated to less variable
(2007) and Records et al. (2014). Three different model performance weather conditions may not reproduce highly variable flow conditions.
statistics, Nash-Sutcliffe efficiency coefficient (NSE), root mean squared For example, the coefficient of variation (CV, the ratio of standard devi-
error (RMSE)-standard deviation (STDEVobs) ratio (RSR), and percent ation to mean) of precipitation and streamflow was 2.6 and 1.5, respec-
bias (P-bias), were considered to evaluate simulated stream flows tively during the calibration, and 2.8 and 1.8, respectively during the
against observed ones, as: validation period. The modeled discharge pattern did not capture peak
" P # streamflow after large storm events while it overestimated streamflow
n
ðO −Si Þ2 during wetter weather conditions. This inaccurate simulation of peak
NSE ¼ 1− Pn i¼1 i 2
; ð3Þ flow or streamflow during wet seasons can be potentially attributed to
i¼1 ðOi − O Þ
the inherent limitation of SWAT and the quality of input climate data.
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi The SCS-CN method used in SWAT has a limited ability to accurately
Pn 2
RMSE i¼1 ðOi −Si Þ predict storm effects because its surface runoff calculation does not con-
RSR ¼ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P
ffi; ð4Þ
STDEV obs n 2 sider duration and intensity of precipitation (Neitsch et al., 2011; Arnold
i¼1 ðOi − O Þ et al., 1998). In addition, the semi-distributed structure of SWAT, which
Pn  simplifies routing within subwatersheds, could amplify model bias dur-
i¼1 ðOi −Si Þ  100 ing wetter conditions. SWAT routes excess surface water runoffs and
P−bias ¼ Pn ; ð5Þ
i¼1 Oi subsurface flows from HRUs directly to the subwatershed stream
reach, without time lags and interception from intermediate landscape
where Oi are observed data, Si are simulated data, Ō is observed mean units along down gradient flowpaths (Neitsch et al., 2011). Lastly, this
value, and n is number of observations. The prediction uncertainty of model error could be due to climate input data, which were acquired
the model was assessed using the 95% prediction uncertainty (95 from two weather stations located either ~15 or ~35 km away from
I.-Y. Yeo et al. / Science of the Total Environment 653 (2019) 1557–1570 1565

Fig. 4. Observed and simulated daily streamflow during calibration (2001–2005) and validation (2006–2010) periods.

the study watershed. Observed precipitation from the somewhat dis- lowered AETa and SWa, but increased WYa, compared to NoWetLU.
tant stations showed some degree of inconsistency with streamflow ob- This difference in water budget partioning, led to clearly different
served from the study watershed outlet. This likely indicates that the streamflow generation mechanisms and flow regime characteristics be-
common assumption of spatially homogeneous climate input data did tween the two scenarios (p-value b 0.01; Fig. 5c & d). WetLU produced
not sufficiently characterize the spatial distribution of precipitation higher but less variable annual mean WYa, as it was largely contributed
events and failed to capture localized storm effects (Sexton et al., 2010). from GWa (Fig. 5c). Overall, ~75% of WYa was generated from GWa,
while SRa and SWa (shown as lateral flow generated from saturated
3.2. The overall mean impact of GIWs soils, LWa) only contributed to 21% and 4% of WYa, respectively. Mean-
while, NoWetLU showed reduced but more variable WYa, as it was more
Fig. 5 shows the 30–year average annual watershed water budget for influenced by SRa (36%) and LWa (7%) (Fig. 5.d). SWAT-WET showed re-
the study watershed (a-b) and major sources of water (i.e., groundwater duced water stored in the unsaturated zone (i.e., less soil water (SWa),
(GWa), lateral flow from soils (LWa), and surface water runoff (SRa)) that shown by reduced lateral flow to the stream (LWa)), but increased
generated streamflow at the watershed outlet (c-d), under the two sce- groundwater influence (indicating higher groundwater table with in-
narios. Note that the subscript ‘a’ indicates annual simulation results. creasing water stored in the saturated zone) under WetLU. The results
The partitioning of the mean annual precipitaiton (Pa) into AETa, soil indicated an overall stronger connection between groundwater and
water storage (SWa), and water yield (WYa, streamflow normalized by streamflow under WetLU over long run, in comparison to NoWetLU.
the watershed area) under the two scenarios (WetLU vs. NoWetLU)
was statistically different (p-value b 0.01).
Overall, WetLU produced less variable annual water fluxes for all 3.3. The inter-annual time scale impacts of GIWs
components. When compared to forest land cover (NoWetLU), GIWs
could potentially reduce AET as reduced biomass (LAIw, estimated Fig. 6a showed that the study site experienced increasing precipita-
based on vegetation cover fraction relative to flooded surface water tion at a rate of 10.3 [±4.5] mm/year over 1985–2015 (Note that the
area within a wetland) decreased transpiration by plant water uptake numeric value in brackets [] is the standard error of the estimate, indi-
(see Section 2.2). As expected, simulation results showed that WetLU cating the uncertainty of estimated parameter value). While this in-
creasing trend was consistent with that of the Northeastern region of
Table 3 the US (Section 2.1; Kunkel et al., 2013), the local trend (Fig. 6b) was
Model performance measures computed for streamflow. much higher than the positive anomalies reported for the US Northeast
region. This large positive trend was likely due to the specific character-
Period NSE RSR P-bias Rating
istics of local weather conditions during the simulation period
a b a b a b
Calibration 0.31 0.59 0.83 0.64 9.50 9.70 Satisfactory (1985–2015) relative to longer-term (1895–2011), regional climate
(2001–2005)
a b a a b conditions. For example, wetter, more variable regional climate changes
Validation 0.38 0.57 0.79 0.65 −24.00 −23.90 Satisfactory
(2006–2010) b were reported since 1970s, compared to the long-term historic patterns.
In addition, this local trend analysis was based on mean annual precip-
Note: Model performance statistics were calculated based on the daily (a) and monthly
mean (b) simulation outputs. Model performances were rated based on the criteria ap- itation of 1985–2015. During this period, local precipitation showed a
plied to the monthly simulation (Moriasi et al., 2007) as follows: * Satisfactory (0.5 cyclic pattern with frequency of 2–8 years, most likely due to the El
b NSE ≤ 0.65, 0.6 b RSR ≤ 0.7, and ± 15 ≤ P-bias b ± 25), ** Good (0.65 b NSE ≤ 0.75, 0.5 Niño–Southern Oscillation (ENSO) effect. It happened to be drier in
b RSR ≤ 0.6, and ±10 ≤ P-bias b ±15), and *** Very Good (0.75 b NSE ≤ 1.0, 0.0 b RSR ≤ the beginning (1985), but became much wetter towards the end
0.5, P-bias b ±10). Use of more relaxed criteria was recommended for daily simulation
as model prediction naturally becomes less accurate at a finer time step. Improved perfor-
(2010s) of the simulation period, causing a large artifact in the overall
mance statistics over the calibration period was in part due to the difference in the climate trend fitted over this time period (Fig. 6b). Local temperature showed
conditions between the two split sample evaluation periods. See Fig. 4. a slight increase in minimum temperature (0.03 [±0.01] °C per year;
1566 I.-Y. Yeo et al. / Science of the Total Environment 653 (2019) 1557–1570

Watershed Water Budget, 1985-2018

Contribution to Streamflow, 1985-2018

Fig. 5. The 30-year average impact of GIWs on the watershed water budget (a, b) and the streamflow generation process (c, d).

p-value b 0.06) and SWAT-WET predicted clearly upward PETa (at a rate (~1171 mm). However, these two scenarios produced clearly different
of 7.1 [±3.6] mm/yr; p-value b 0.06). seasonal partitioning of all water fluxes (p-value b 0.01). Overall,
Both scenarios showed similar soil water change patterns mirroring WetLU showed less variable seasonal fluxes (as shown by a lower coef-
this increasing precipitation trend (0.9 [±0.3] mm/yr–1.1 [±0.4] mm/ ficient of variation (CV) in Fig. 7). It showed lower AETm, slightly higher
yr; p-value b 0.01), but resulted in different streamflow responses. WYm, and more subtle seasonal change in soil water (as shown by soil
WetLU showed streamflow (shown as WYa) increasing at a similar moisture anomaly, SMAm, calculated by taking the monthly mean soil
rate as precipitation (10.9 [± 3.1] mm/yr; p-value b 0.001), but moisture difference from the 30-year annual mean) throughout sea-
NoWetLU suggested a higher increasing trend for WYa at a rate of 12.6 sons. Both scenarios showed that AET peaked in July (~91 mm under
[±3.5] mm/yr (p-value b 0.001) (Fig. 6c & d). The change trend in WetLU and ~113 mm under NoWetLU) when solar radiation and tem-
flow regime variables (GWa, LWa, and SRa) that was responsible for perature reached their highest levels. Hence, overall monthly watershed
streamflow generation offered an explanation for this (Fig. 6e & f). water budget (when estimated as Pm-AETm-WYm) would hit its lowest
Both scenarios showed similar change rates for GWa over time (6.2 [± in July. However, SWm continued to be depleted until August, showing
2.1] mm/yr for NoWetLU and 6.6 [±2.2] mm/yr for WetLU; p-value b its persistent, “memory”, to the dry state, and reached its lowest with
0.001). However, NoWetLU showed much higher increasing SRa contri- a one month time lag from the highest AETm. WYm showed even slower
bution to streamflow generation at a rate of ~ 6.1 [±2.0] mm/yr (with p- response than SWm, reaching its lowest level (~38 mm under WetLU
value b 0.01), which explained higher increasing WYa under this sce- and ~31 mm under NoWetLU) in September with ~2-month time lag
nario. Under WetLU, the increasing trend for SRa was kept at 4.1 [± from peak AETm, and then started increasing rapidly with reduced AET
1.3] mm/yr (p-value b 0.01). This difference in SRa contribution to and seasonal fall storms (Fig. 7b).
WYa, led to different flow regime characteristics under the two scenar- Gradually changing seasonal groundwater flow (GWm) offered in-
ios, and the increasing influence of SRa on WYa became more noticeable sights into the time delay and offset in seasonal streamflow pattern
under wetter conditions. This higher SRa offset reduced AETa under (WYm). As shown in Fig. 7b, GWm plays an important role in streamflow
NoWetLU (occurred at a rate of 2 [±0.8] mm/yr; p-value b 0.01), generation for this humid low-lying coastal watershed. Monthly varia-
while WetLU did not show any statistically significant change trend tion in GWm was low compared to other hydrologic variables (see the
for AETa. CV values in Fig. 7), and it remained the primary source of streamflow
The inter-annual trend analysis highlighted the role of densely dis- throughout all seasons. Mean seasonal GWm contribution to WYm was
tributed GIWs in exerting strong hydrological influence on downstream ~35 mm (i.e., 74% of WYm, with the standard deviation of 3.5 mm)
water by regulating surface water runoff, while maintaining groundwa- under WetLU, though it decreased to ~24 mm (i.e., 57% of WYm, with
ter recharge and AET under the changing climate condition for the study the standard deviation of 5.9 mm) under NoWetLU. GWm showed de-
watershed. Note that the linear trend was analyzed using both paramet- layed and subtle response to seasonal climate variations with
ric and non-parametric methods, and reported p-values from these two 2–3 month time lag from peak AETm. With seasonal fall storms and re-
methods were very similar for all water flux terms discussed above. duced AET, groundwater slowly recharged while increasing seasonal
surface water runoff (SRm) increased streamflow rapidly. Hence, WYm
3.4. Seasonal variations of hydrological fluxes started increasing in October before GWm.
Overall, WetLU showed consistently higher but less seasonally
Fig. 7 showed a seasonal variation in water fluxes (i.e., Pm, WYm, variable GWm. Note that both scenarios showed similar WYm during
AETm, and SWm) with monthly mean simulation outputs over winter and spring (Fig. 7b), but WetLU maintained much higher sea-
1985–2015 (note the subscript ‘m’ indicates monthly output). It sonal GWm, which showed subtle changes throughout year. For exam-
showed that the study site received relatively constant monthly precip- ple, NoWetLU showed much faster GWm dropping over the growing
itation (Pm) throughout the simulation period, with the mean and seasons (i.e., April–October) at a rate of 2.3 mm/month (with relative
standard deviation of 97.6 mm and 11.9 mm, respectively. Overall, cu- change of 50%), whereas WetLU showed GWm decreasing at a rate of
mulative PETm (~1200 mm) was slightly higher than cumulative Pm 1.4 mm/month (with relative change of 27%) over the same period.
I.-Y. Yeo et al. / Science of the Total Environment 653 (2019) 1557–1570 1567

Fig. 6. Inter-annual change trend in annual mean precipitation (a, b), the watershed water budget (c, d) and the stream flow generation (e, f) with the presence [W] and absence [N] of GIWs.

This highlights the hydrological influence of GIWs on downstream wa- by the seasonal change pattern of GWm. The analysis of seasonal
ters through the groundwater connection. In part through the hydro- water fluxes showed the overall effectiveness of GIWs on controlling
logic continuum created by GIWs, surface water and groundwater surface water runoff and buffering dynamics of shallow groundwater
systems in the watershed became tightly coupled, and GIWs helped to and stream base flow throughout the year.
improved terrestrial water storage by recharging groundwater during
wet seasons. This, in turn buffered the watershed water budget and 3.5. Implication from the multi-temporal analysis
shallow groundwater dynamics during the dry season. In addition, a
large monthly difference in surface water runoff (SRm) between the This comparative analysis allowed us to infer watershed-level hy-
two scenarios especially during leaf-off wet season, emphasized effec- drological connectivity of GIWs with downstream waters from simu-
tiveness of GIWs on controlling and storing surface water runoff, lated hydrological functions. The long term (30+ year) average, inter-
which eventually were released into soils and groundwater as shown annual trend, and seasonal variability of hydrological fluxes suggest
1568 I.-Y. Yeo et al. / Science of the Total Environment 653 (2019) 1557–1570

Fig. 7. Seasonal variation in the watershed water budget (a) and the flow regime (b) under WetLU [W] and NoWetLU [N] scenarios.

very strong, temporally persistent hydrological connection between that demonstrated the importance of GIWs in stabilizing downstream
surface water and groundwater systems with the presence of GIWs flow fluctuations in response to large storm effects. Note that these
over multiple time scales for the study watershed. At the seasonal long-term effects of GIWs on modulating flow regimes, may not directly
time scale, GIWs were effective in increasing terrestrial water storage apply for other regions. This is because GIW effects on watershed hydro-
during wet seasons when precipitation exceeded actual evapotranspira- logic functioning largely depend on the dominant type of hydrologic
tion. Similarly, previous field-based studies conducted in the study re- connectivity between wetlands and surrounding landscapes and local
gion (Denver et al., 2014; Phillips and Shedlock, 1993) reported the climate conditions. For example, depressional wetlands in the Prairie
development of transient mounds of shallow groundwater near or Pothole Region (PPR) are primarily hydrologically connected via over-
under selected seasonal GIWs, indicating strong groundwater coupling land flow (i.e., surface water connection) and have been shown to be ef-
with GIWs during wet (winter and spring) seasons. In addition, recent fective at storing precipitation and snowmelt (Evenson et al., 2018).
field monitoring demonstrated that the study area's poorly drained While GIWs in the PPR attenuated downstream peak flow, they were
soils overlaid with silty and sandy top soils with high hydraulic conduc- less effective at maintaining baseflow (Evenson et al., 2018).
tivity (Section 2.1) could allow the formation of shallow perched Overall, this simulation study provided additional explanation and
groundwater (Lee et al., 2018c). Shallow perched groundwater could support for previous studies that highlight landscape-level hydrological
flow into GIWs, and sustain the period of inundation for GIWs while re- connectivity and functioning of GIWs in the study region. Hydrological
ducing water loss due to seepage (Phillips and Shedlock, 1993; Lee et al., connection between selected GIWs and downstream waters via
2018c). When evapotranspiration started exceeding precipitation, groundwater at the seasonal time scale (during the wet period) has
GIWs helped to maintain soil and groundwater storage through seepage been demonstrated through field data (Denver et al., 2014; Phillips
and reduced AET (i.e., less water loss to the atmosphere). The inundated and Shedlock, 1993) and modeling (Golden et al., 2016). It was sug-
land surface increased evaporation but decreased transpiration at a gested that the long-term annual hydrological regime of seasonal for-
greater rate than evaporation loss, compared to forested land cover. ested GIWs in the temperate region responded to seasonal weather
Overall, GIWs in the study region helped to maintain higher terres- patterns (precipitation and evapotranspiration) (Brooks, 2006). Simi-
trial water storage throughout all seasons, in comparison to NoWetLU larly, our analysis in Part 1 (Yeo et al., 2018) of this two part manuscript
(with absence of GIWs), which resulted in higher stream flow discharge series illustrated the statistically meaningful relationship between inun-
over multiple time scales. However, streamflow from the GIW domi- dation patterns and the stream flow change trend at the seasonal time
nated watershed (i.e., WetLU) was more stable seasonally and inter- scale. Furthermore, it suggested the correlation between the long-
annually, in spite of an increasing precipitation pattern with more fre- term inter-annual water regime of GIWs and seasonal dynamics of in-
quent seasonal storms. This was because streamflow in this region undation, by showing consistent inundation patterns between SWF
was primarily contributed from groundwater, which showed persis- maps and NWI hydrologic modifiers. This simulation study provides a
tence to changing weather and climate conditions. In particular, logical explanation to relate inter-annual water regime to seasonal
WetLU showed higher seasonal groundwater flow with more subtle change dynamics and supports the geospatial modeling approach pro-
fluctuations, highlighting the role of GIW's in maintaining base flow. posed in the Part 1 (i.e., mapping hydrologic connectivity between wet-
Note that the study watershed has very dense GIW distribution, which lands and downstream waters using multi-temporal remote sensing
covers ~25% of the watershed area. Despite their small size, this large such as SWF maps, albeit at a coarse temporal resolution and with lim-
coverage of headwater GIWs relative to watershed area collectively pro- ited availability throughout season).
vided a very high water retention threshold (Evenson et al., 2018; Lane The next logical step is to quantify the spatial and temporal variabil-
et al., 2018; Lee et al., 2018b; Fossey et al., 2016). In particular, the inter- ity of connectivity and isolation between wetlands and downstream
annual change trend of surface water runoff contribution to streamflow waters and other landscape components at multiple spatial scales. Ex-
underscored the importance of the water holding capacity of GIWs in plicit representation of the spatial and temporal patterns of connectivity
flood attenuation and maintaining base flow through redistribution of between streams and wetlands at the individual level (e.g., wetland-
surface runoff, under changing (wetter) weather conditions. This retro- wetland, wetland-stream) could be impractical for GIW dominated wa-
spective multi-decadal analysis substantiated the potential role of GIWs tersheds similar to this study site. However, quantifying connectivity at
in modulating the long-term hydrologic effects of changing weather/cli- a finer landscape scale than a watershed (e.g., the subwatershed) in a
mate conditions in this study region, building upon previous studies ‘quasi-explicit’ manner (Golden et al., 2017), would be informative for
I.-Y. Yeo et al. / Science of the Total Environment 653 (2019) 1557–1570 1569

the development of management policies pertinent to wetland protec- et al., 2018). Part 1 capitalized on a time series of remotely sensed sur-
tion and restoration (Creed et al., 2017). For such modeling, characteri- face water inundation information (i.e., SWF maps) to inform local hy-
zation of relevant spatial resolution and dynamic inundation extent drological conditions of a groundwater dominated, wetland dense
would need careful consideration as inundation within wetlands and landscape and to map landscape-scale hydrological connectivity of
river networks changes across the landscape at different time scales. GIWs with downstream waters. This two-part study (Parts 1 & 2) pro-
Surface water dynamics characterized by multi-temporal remote sens- vides crucial information to infer watershed-scale hydrological connec-
ing could be helpful in supporting this effort (Vanderhoof et al., 2016, tivity of GIWs in groundwater dominated systems based on convergent
2017; Huang et al., 2014). evidence from field-based studies, remote sensing, and modeling, and
To provide reliable quantitative information on hydrologic connec- thus highlights the importance of protecting GIWs for improved hydro-
tivity (with details on the frequency, magnitude, duration, timing, and logical and ecological watershed functioning.
rate of change considering different flowpath types) at a finer spatial
resolution, semi-distributed simulation models, such as SWAT-WET,
need further enhancement. For example, SWAT (similar to other Acknowledgements
models) needs to improve process representation of GIWs considering
their interaction with contributing areas and down gradient ecosystems This research project was funded by the NASA's Land Cover and Land
along flow paths, so that more complete hydrologic behaviours of the Use Change (LCLUC) Program (contract No: NNX12AG21G) and by
wetland complex (including the fill-and-merge or fill-and-spill) and in- USDA NRCS Conservation Effects Assessment Project (CEAP) Water-
duced impacts can be considered (Lane et al., 2018; Evenson et al., 2016, sheds and CEAP Wetlands components. We are thankful to Dr. Yongbo
2018). In addition, improved representation of groundwater process Liu from University of Guelph, Canada for providing the riparian wet-
would be needed, so that groundwater change dynamics, considering land module (RWM) and Dr. Heather Golden from U.S. Environmental
their interaction with wetlands and redistribution of water flows (or Protection Agency (EPA) for providing crucial insights and suggestions.
discharge to surface stream water or soils) with time lags, could be ex- We greatly appreciate valuable comments and suggestions from the
plained more explicitly (Ameli et al., 2017; McLaughlin et al., 2014). reviewers.
Lastly, more detailed field-based data are needed to provide insights The findings and conclusions in this article are those of the author
into wetland-groundwater processes and other internal landscape pro- (s) and do not necessarily represent the views of the U.S. Fish and Wild-
cesses (e.g., inundation of GIWs) (Lee et al., 2017; McDonnell and life Service (FWS) or the U.S. Department of Agriculture-Agricultural
Beven, 2014), as well as to provide crucial information on land surface Research Service.
properties (e.g., soil depths and properties, such as the hydraulic con-
ductivities) that affect soil water storage, the rate of seepage, and References
hence the hydroperiod of GIWs. With increasing availability and capa- Alexander, R.B., Boyer, E.W., Smith, R.A., Schwarz, G.E., Moore, R.B., 2007. The role of head-
bility of remote sensing, we expect to have more frequent, finer resolu- water streams in downstream water quality. J. Am. Water Resour. Assoc. 43, 41–59.
tion surface inundation or soil moisture data products in the near future. Alexander, L.C., Fritz, K.M., Schofield, K.A., Autrey, B.C., DeMeester, J.E., Golden, H.E.,
Goodrich, D.C., Kepner, W.G., Lane, C.R., LeDuc, S.D., Leibowitz, S.G., McManus, M.G.,
This information would be extremely valuable for calibrating land sur- Pollard, A.I., Raanan, H., Ridley, C.E., Vanderhoof, M.K., Wigington Jr., P.J., 2018. Fea-
face parameters (near surface soil properties such as hydraulic conduc- tured collection introduction: connectivity of streams and wetlands to downstream
tivity) (Salvucci and Entekhabi, 2011) and for evaluating emergent waters. J. Am. Water Resour. Assoc. 54 (2), 287–297.
Ali, G., Oswald, C., Spence, C., Wellen, C., 2018. The T-TEL method for assessing water, sed-
properties of hydrological fluxes and associated landscape processes at iment, and chemical connectivity. Water Resour. Res. 54, 634–662. https://doi.org/
multiple spatial scales. 10.1002/2017WR020707.
Ameli, A.A., Creed, I.F., 2017. Quantifying hydrologic connectivity of wetlands to surface
water systems. Hydrol. Earth Syst. Sci. 21, 1791–1808. https://doi.org/10.5194/hess-
4. Conclusion 21-1791-2017.
Arnold, J.G., Srinivasan, R., Muttiah, R.S., Williams, J.R., 1998. Large area hydrologic model-
This study elucidates how GIWs in aggregate could influence ing and assessment part I: model development1. J. Am. Water Resour. Assoc. 34,
73–89. https://doi.org/10.1111/j.1752-1688.1998.tb05961.x.
partitioning of precipitation into evapotranspiration and terrestrial Arnold, J.G., Moriasi, D.N., Gassman, P.W., Abbaspour, K.C., White, M.J., Srinivasan, R.,
water storage, and how this further affects water transport mechanisms Santhi, C., Harmel, R.D., Van Griensven, A., Van Liew, M.W., Kannan, N., 2012.
and lateral routing processes that generate streamflow over multiple SWAT: model use, calibration, and validation. Trans. Am. Soc. Agric. Biol. Eng. 55
(4), 1491–1508.
time scales across the landscapes. Based on a retrospective modeling
Brooks, B., 2006. A review of basin morphology and pool hydrology of isolated ponded
analysis conducted in an agricultural watershed with densely distrib- wetlands: implications for seasonal forest pools of the northeastern United States.
uted GIWs, this study demonstrated GIWs in aggregate could improve Wetl. Ecol. Manag. 13, 334–335. https://doi.org/10.1007/s11273-004-7526-5.
hydrologic functioning at a watershed scale. GIWs in aggregate were Calhoun, A.J.K., Mushet, D.M., Bell, K.P., Boix, D., Fitzsimons, J.A., Isselin-Nondedeug, F.,
2018. Temporary wetlands: challenges and solutions to conserving a ‘disappearing’
found to significantly influence the overall water budget over the long ecosystem. Biol. Conserv. 211, 3–11.
run. Seasonal variations in water fluxes showed the overall effectiveness Cohen, M.J., Creed, I.F., Alexander, L.C., Basu, N.B., Calhoun, A.J., Craft, C., D'Amico, E.,
of GIWs in attenuating surface water runoff and buffering dynamics of DeKeyser, E., Fowler, L., Golden, H.E., Jawitz, J.W., Kalla, P., Kirkman, L.K., Lane, C.R.,
Lang, M., Leibowitz, S.G., Lewis, D.B., Marton, J., McLaughlin, D.L., Mushet, D.M.,
shallow groundwater and stream base flow throughout the year. Raanan-Kiperwas, H., Rains, M.C., Smith, L., Walls, S.C., 2016. Do geographically isolated
Through the hydrological continuum generated via GIWs in this wetlands influence landscape functions? Proc. Natl. Acad. Sci. 113 (8), 1978–1986.
groundwater dominated watershed, surface water and groundwater Creed, I.F., Lane, C.R., Serran, J.N., Alexander, L.C., Basu, N.B., Calhoun, A.J.K., Christensen,
J.R., Cohen, M.J., Craft, C., D'Amico, E., DeKeyser, E., Fowler, L., Golden, H.E., Jawitz,
systems become more tightly coupled during wet seasons. Hence, the J.W., Kalla, P., Kirkman, L.K., Lang, M., Leibowitz, S.G., Lewis, D.B., Marton, J.,
study watershed demonstrated enhanced terrestrial water storing ca- McLaughlin, D.L., Raanan-Kiperwas, H., Rains, M.C., Rains, K.C., Smith, L., 2017. En-
pacity during wet seasons, which in turn affected the long-term annual hancing protection for vulnerable waters. Nat. Geosci. 10, 809–815.
Denver, J.M., Ator, S.W., Lang, M.W., Fisher, T.R., Gustafson, A.B., Fox, R., Clune, J.W.,
hydrological regime of GIWs and watershed streamflow, creating a McCarty, G.W., 2014. Nitrate fate and transport through current and former
feedback loop. Furthermore, GIWs provide important hydrological ben- depressional wetlands in an agricultural landscape, Choptank Watershed, Maryland,
efits by maintaining base flow and attenuating flood effects during wet- United States. J. Soil Water Conserv. 69 (1), 1–16.
Evenson, G.R., Golden, H.E., Lane, C.R., D'Amico, E., 2015. Geographically isolated wetlands
ter periods with more seasonal variability, by controlling the timing,
and watershed hydrology: a modified model analysis. J. Hydrol. 529, 240–256.
amount, and dominant flow paths to the river network across the land- Evenson, G.R., Golden, H.E., Lane, C.R., D'Amico, E., 2016. An improved representation of
scape, under changing weather conditions. geographically isolated wetlands in a watershed-scale hydrologic model. Hydrol. Pro-
Furthermore, findings on watershed-scale wetland hydrologic func- cess. 30 (22), 4168–4184.
Evenson, G.R., Golden, H.E., Lane, C.R., McLaughlin, D.L., D'Amico, E., 2018. Depressional
tion from this study (Part 2), support the basis for the geospatial model- wetlands affect watershed hydrological, biogeochemical, and ecological functions.
ing approach presented in Part 1 of this two part manuscript series (Yeo Ecol. Appl. 28 (4), 953–966. https://doi.org/10.1002/eap.1701.
1570 I.-Y. Yeo et al. / Science of the Total Environment 653 (2019) 1557–1570

Fossey, M., Rousseau, A.N., Savary, S., 2016. Assessment of the impact of spatio-temporal Liu, Y.B., Yang, W.H., Wang, X.X., 2008. Development of a SWAT extension module to sim-
attributes of wetlands on stream flows using a hydrological modelling framework: a ulate riparian wetland hydrologic processes at a watershed scale. Hydrol. Process. 22
theoretical case study of a watershed under temperate climatic conditions. Hydrol. (16), 2901–2915.
Process. 30 (11), 1768–1781. McDonnell, J.J., Beven, K., 2014. Debates-the future of hydrological sciences: a (common)
Freeman, M.C., Pringle, C.M., Jackson, C.R., 2007. Hydrologic connectivity and the contri- path forward? A call to action aimed at understanding velocities, celerities, and resi-
bution of stream headwaters to ecological integrity at regional scales. J. Am. Water dence time distributions of the headwater hydrograph. Water Resour. Res. 50,
Resour. Assoc. 43, 5–14. 5342–5350. https://doi.org/10.1002/2013WR015141.
Gitau, M.W., Chaubey, I., 2010. Regionalization of SWAT model parameters for use in McDonough, O.T., Lang, M.W., Hosen, J., Palmer, M., 2014. Surface hydrologic connectivity
ungauged watersheds. Water 2 (4), 849–871. between Delmarva Bay wetlands and nearby streams along a gradient of agricultural
Golden, H.E., Lane, C.R., Amatya, D.M., Bandilla, K.W., Kiperwas, H.R., Knightes, C.D., alteration. Wetlands 35. https://doi.org/10.1007/s13157-014-0591-5.
Ssegane, H., 2014. Hydrologic connectivity between geographically isolated wetlands McLaughlin, D.L., Kaplan, D.A., Cohen, M.J., 2014. A significant nexus: geographically iso-
and surface water systems: a review of select modeling methods. Environ. Model lated wetlands influence landscape hydrology. Water Resour. Res. 50, 7153–7166.
Softw. 53, 190–206. Moriasi, D.N., Arnold, A.G., Van Liew, M.W., Bingner, R.L., Harmel, R.D., Veith, T.L., 2007.
Golden, H.E., Sander, H.A., Lane, C.R., Zhao, C., Price, K., D'Amico, E., Christensen, J.R., 2016. Model evaluation guidelines for systematic quantification of accuracy in watershed
Relative effects of geographically isolated wetlands on streamflow: a watershed-scale simulations. Trans. ASABE 50 (3), 885–900.
analysis. Ecohydrology 9, 21–38. National Oceanic and Atmospheric Administration (NOAA) National Climate Data Center,
Golden, H.E., Creed, I.F., Ali, G., Basu, N.B., Neff, B., Rains, M.C., McLaughlin, D.L., Alexander, 2015. Available from: http://www.ncdc.noaa.gov/oa/climate/research/prelim/
L.C., Ameli, A.A., Christensen, J.R., Evenson, G.R., Jones, C.N., Lane, C.R., Lang, M., 2017. drought/palmer.html.
Scientific tools for integrating geographically isolated wetlands into land manage- Neitsch, S.L., Arnold, J.G., Kiniry, J.R., Williams, J.R., 2011. Soil and Water Assessment Tool
ment decisions. Front. Ecol. Environ. 15, 319–327. https://doi.org/10.1002/fee.1504. Theoretical Documentation Version 2009. Texas Water Resources Institute Technical
Huang, C., Peng, Y., Lang, M.W., Yeo, I.-Y., McCarty, G.W., 2014. Wetland inundation map- Report Texas A&M University System, College Station, Texas, p. 618.
ping and change monitoring using Landsat and airborne LiDAR data. Remote Sens. Phillips, P.J., Shedlock, R.J., 1993. Hydrology and chemistry of groundwater and seasonal
Environ. 141, 231–242. ponds in the Atlantic Coastal Plain in Delaware, USA. J. Hydrol. 141, 157–178.
Jin, H., Huang, C., Lang, M.W., Yeo, I.-Y., Stehman, S.V., 2017. Monitoring of wetland inun- Records, R.M., Arabi, M., Fassnacht, S.R., Duffy, W.G., Ahmadi, M., Hegewisch, K.C., 2014.
dation dynamics in the Delmarva Peninsula using Landsat time-series imagery from Climate change and wetland loss impacts on a western river's water quality. Hydrol.
1985 to 2011. Remote Sens. Environ. 190, 26–41. Earth Syst. Sci. 18 (11), 4509–4527.
Kunkel, K.E., Stevens, L.E., Stevens, S.E., Sun, L., Janssen, E., Wuebbles, D., Rennells, J., Salvucci, G.D., Entekhabi, D., 2011. An alternate and robust approach to calibration for the
DeGaetano, A., Dobson, J.G., 2013. Technical Report NESDIS 142-1: Regional Climate estimation of land surface model parameters based on remotely sensed observations.
Trends and Scenarios for the US National Climate Assessment- Part 1. Climate of Geophys. Res. Lett. 38, L16404. https://doi.org/10.1029/2011GL048366.
the Northeast U.S., Silver Spring, MD. USA. Sexton, A.M., Sadeghi, A.M., Zhang, X., Srinivasan, R., Shirmohammadi, A., 2010. Using
Lane, C.R., D'Amico, E., 2010. Calculating the ecosystem service of water storage in isolated NEXRAD and rain gauge precipitation data for hydrologic calibration of SWAT in a
wetlands using LiDAR in North Central Florida, USA. Wetlands 30 (5), 967–977. Northeastern watershed. Trans. ASABE 53 (5), 1501–1510.
Lane, C.R., D'Amico, E., 2016. Identification of putative geographically isolated wetlands of Sharifi, A., Lang, M.W., McCarty, G.W., Sadeghi, A.M., Lee, S., Yen, H., Yeo, I.-Y., 2016. Im-
the conterminous United States. J. Am. Water Resour. Assoc. 52, 705–722. proving model prediction reliability through enhanced representation of wetland
Lane, C.R., Leibowitz, S.G., Autrey, B.C., LeDuc, S.D., Alexander, L.C., 2018. Hydrological, soil processes and constrained model auto calibration: a paired watershed study.
hysical, and chemical functions and connectivity of non-floodplain wetlands to J. Hydrol. 541, 1088–1103.
downstream waters: a review. J. Am. Water Res. Assoc. 54 (2), 346–371. https:// Vanderhoof, M.K., Alexander, L.C., Todd, M.J., 2016. Temporal and spatial patterns of wet-
doi.org/10.1111/1752-1688.12633. land extent influence variability of surface water connectivity in the Prairie Pothole
Lang, M., McDonough, O., McCarty, G., Oesterling, R., Wilen, B., 2012. Enhanced detection Region, United States. Landsc. Ecol. 31, 805–824.
of wetland-stream connectivity using LiDAR. Wetlands 32, 461–473. Vanderhoof, M.K., Distler, H.E., Lang, M.W., Alexander, L.C., 2017. The influence of data
Lee, S., Yeo, I.-Y., Sadeghi, A.M., McCarty, G.W., Hively, W.D., Lang, M.W., 2016. Impacts of characteristics on detecting wetland/stream surface-water connections in the
watershed characteristics and crop rotations on winter cover crop nitrate uptake ca- Delmarva Peninsula, Maryland and Delaware. Wetl. Ecol. Manag. https://doi.org/
pacity within agricultural watersheds in the Chesapeake Bay Region. PLoS One 11 (6), 10.1007/s11273-017-9554-y.
e0157637. Winchell, M., Srinivasan, R., Di Luzio, M., Arnold, J., 2010. ArcSWAT Interface for
Lee, S., Yeo, I.-Y., Lang, L.W., McCarty, G.W., Sadeghi, A.M., Sharifi, A., Jin, H., Liu, Y., 2017. SWAT2009 User's Guide. Texas Agricultural Experiment Station and U.S. Department
Improving the catchment scale wetland modeling using remotely sensed data. Envi- of Agriculture, Temple, TX, p. 415.
ron. Model Softw. https://doi.org/10.1016/j.envsoft.2017.11.001 (In press). Wu, K., Xu, Y.J., 2006. Evaluation of the applicability of the swat model for coastal water-
Lee, S., Yeo, I.-Y., Sadeghi, A.M., McCarty, G.W., Hively, W.D., Lang, M.W., Sharifi, A., 2018a. sheds in Southeastern Louisiana. J. Am. Water Resour. Assoc. 42 (5), 1260.
Comparative analyses of hydrological responses of two adjacent watersheds to cli- Yang, W., Liu, Y., Ou, C., Gabor, S., 2016. Examining water quality effects of riparian wet-
mate variability and change using the SWAT model. Hydrol. Earth Syst. Sci. 22, land loss and restoration scenarios in a Southern Ontario watershed. J. Environ.
689–708. https://doi.org/10.5194/hess-22-689-2018. Manag. 174, 26–34.
Lee, S., Yeo, I.-Y., Lang, L.W., Sadeghi, A.M., McCarty, G.W., Moglen, G., Evenson, G., 2018b. Yen, H., Jeong, J., Feng, Q., Deb, D., 2014. Assessment of input uncertainty in SWAT using
Assessing the cumulative impacts of wetlands on the watershed hydrology using the latent variables. Water Resour. Manag. 29 (4), 1137–1153.
SWAT model coupled with improved wetland modules. J. Environ. Manag. 223, Yeo, I.-Y., Guldmann, J.-M., 2010. Global spatial optimization with hydrological systems
37–48. simulation: application to land-use allocation and peak runoff minimization. Hydrol.
Lee, S., McCarty, G.W., Moglen, G., Lang, L.W., Sadeghi, A.M., Green, T.R., Yeo, I.-Y., Earth Syst. Sci. 14 (2), 325–338. https://doi.org/10.5194/hess-14-325-2010.
Rabenhorst, R.C., 2018c. Effects of subsurface soil characteristics on wetland- Yeo, I.-Y., Lee, S., Sadeghi, A.M., Beeson, P.C., Hively, W.D., McCarty, G.W., Lang, M.W.,
groundwater interaction in the Coastal Plain of the Chesapeake Bay watershed. 2014. Assessing winter cover crop nutrient uptake efficiency using a water quality
Hydrol. Process. https://doi.org/10.1002/hyp.13326 (In press). simulation model. Hydrol. Earth Syst. Sci. 18 (12), 5239–5253.
Leibowitz, S.G., Wigington Jr., P.J., Schofield, K.A., Alexander, L.C., Vanderhoof, M.K., Yeo, I.-Y., Lang, M.W., Lee, S., McCarty, G.W., Sadeghi, A.M., Yetemen, O., Huang, C., 2018.
Golden, H.E., 2018. Connectivity of streams and wetlands to downstream waters: Mapping landscape-level hydrological connectivity of headwater wetlands to down-
an integrated systems framework. J. Am. Water Resour. Assoc. 54 (2), 298–322. stream waters: A geospatial modeling approach - Part 1. Sci. Total Environ. https://
https://doi.org/10.1111/1752-1688.12631. doi.org/10.1016/j.scitotenv.2018.11.238 (In press).

You might also like