Landscape metrics as predictors of hydrologic connectivity between Coastal Plain forested wetlands and streams

Abstract Geographically isolated wetlands, those entirely surrounded by uplands, provide numerous landscape‐scale ecological functions, many of which are dependent on the degree to which they are hydrologically connected to nearby waters. There is a growing need for field‐validated, landscape‐scale approaches for classifying wetlands on the basis of their expected degree of hydrologic connectivity with stream networks. This study quantified seasonal variability in surface hydrologic connectivity (SHC) patterns between forested Delmarva bay wetland complexes and perennial/intermittent streams at 23 sites over a full‐water year (2014–2015). Field data were used to develop metrics to predict SHC using hypothesized landscape drivers of connectivity duration and timing. Connection duration was most strongly related to the number and area of wetlands within wetland complexes as well as the channel width of the temporary stream connecting the wetland complex to a perennial/intermittent stream. Timing of SHC onset was related to the topographic wetness index and drainage density within the catchment. Stepwise regression modelling found that landscape metrics could be used to predict SHC duration as a function of wetland complex catchment area, wetland area, wetland number, and soil available water storage (adj‐R 2 = 0.74, p < .0001). Results may be applicable to assessments of forested depressional wetlands elsewhere in the U.S. Mid‐Atlantic and Southeastern Coastal Plain, where climate, landscapes, and hydrological inputs and losses are expected to be similar to the study area.

2002), ecological function (Pringle, 2003) and ecosystem services (Brauman, Daily, Duarte, & Mooney, 2007). Hydrologic connectivity can be highly dynamic, particularly between transitional landscape elements such as wetlands, which vary in size and degree of hydrologic connectivity depending on their landscape position as well as on net inflows and outflows from ground, surface, and atmospheric water (Bracken & Croke, 2007).
Attributed in large part to their dynamic nature, wetlands perform a number of important hydrologic, biogeochemical, and habitat/food web functions with local and regional effects Sharitz, 2003). Wetlands supply materials such as water and organic matter (source function), remove harmful materials such as excess nutrients and pathogens (sink function), and provide habitat or refugia for organisms such as fish and aquatic insects (Leibowitz, Wigington, Rains, & Downing, 2008;Rains et al., 2016). Among depressional wetlands, longer wetland hydroperiods and occasional surface water connections to permanent waters have been linked to higher species richness (Snodgrass, Bryan, Lide, & Smith, 1996) and higher net primary productivity (Cook & Hauer, 2007). Wetland area within a watershed has been shown to be significantly related to flood control (Mitsch & Gosselink, 2000) and reduced nitrate concentrations in groundwater and surface water (Phillips, Denver, Shedlock, & Hamilton, 1993). The type, magnitude, and scale of these functions depend considerably on the degree and mechanism of hydrologic connectivity between wetlands and other landscape elements Leibowitz, Mushet, & Newton, 2016;Marton et al., 2015).
Processes at the landscape scale such as watershed runoff response downstream of depressional wetlands are also affected by hydrologic connectivity. Runoff is influenced by the spatial relationship between runoff-generating areas, flow pathways, surface water storage, and the catchment outlet (Ali et al., 2015;Nippgen, McGlynn, Marshall, & Emanuel, 2011;Shaw, Pietroniro, & Martz, 2013). Thus, changes in land use that alter hydrologic connectivity (e.g., urbanization, agriculture, mining, and channelization) can result in dramatic changes in runoff and the transport of pollutants to downstream waters (Marton et al., 2015) due to loss of wetland water storage and filtration functions (Zedler, 2000). Agricultural drainage through ditching and tiling has led to the greatest loss of wetlands globally (Bartzen, Dufour, Clark, & Caswell, 2010;Blann, Anderson, Sands, & Vondracek, 2009) and altered hydrologic connectivity over sufficiently large scales as to extirpate some wetland species (Jenkins, Grissom, & Miller, 2003). Despite significant loss of wetlands and dramatic changes in hydrologic connectivity, recent use of high resolution imagery has shown that remaining wetlands and small streams in ditched agricultural landscapes may be far more numerous than previously thought (Lang, McDonough, McCarty, Oesterling, & Wilen, 2012), and the plugging of ditches to restore wetlands can result in even greater surface hydrologic connectivity (SHC; Jones et al., 2017; McDonough, Lang, Hosen, & Palmer, 2015). Interest in preserving and restoring wetlands as well as protecting downstream waters has spurred interest in understanding the dynamics of SHC. Researchers are asking when and where temporary flow paths between features of the landscape exist and how important these dynamic connections are in ecological, chemical, and hydrological contexts.
Here, we focus on the seasonal hydrodynamics of Delmarva bays in the eastern Delmarva Peninsula (Figure 1) with the goal of advancing our understanding of the relationship between landscape characteristics and the hydrologic connectivity between these wetlands and the FIGURE 1 Location of the study area within the upper portions (Upper Choptank, Tuckahoe Creek) of the Choptank River watershed and Corsica River watershed surrounding stream network. Delmarva bays are geomorphological features characterized by their depressional shape. They have previously been classified as "geographically isolated" wetlands (Tiner, 2003)-wetlands completely surrounded by uplands-as the majority lack clearly defined surface water inlets or outlets. Despite this classification, the surface water extent within individual bays varies seasonally and inter-annually with rainfall and antecedent conditions (Pyzoha, Callahan, Sun, Trettin, & Miwa, 2008). During wet periods, surface water levels frequently exceed storage capacity and individual basins frequently merge to form large wetland complexes (Sharitz & Gibbons, 1982) that outflow to stream networks (McDonough et al., 2015).
Additionally, the Delmarva Peninsula is characterized by hand-dug ditches created in the early to mid-1900s to drain wetlands for agriculture and mosquito control ( Figure 2). Many of these ditches now function as temporary streams (i.e., streams in which surface flow is absent for some portion of the year) connecting wetlands to the perennial stream network. Using a semi-automated stream mapping approach on the basis of light detection and ranging (LiDAR) digital elevation maps, Lang et al. (2012) reported that 53% of semi-natural palustrine wetlands in a Maryland Coastal Plain watershed were directly connected to streams and 60% were stream-connected within a 10-m stream buffer.
Although distance-based methods may provide a reasonable firstorder approximation of structural physical wetland-stream connectivity at regional or national scales, more accurate approaches are needed to predict functional SHC at catchment scales. Given that surface water connections are easier to assess than groundwater dynamics, the importance of hydrologic connections from wetlands to downstream waters is often based on estimates of SHC. Paired with a mechanistic understanding of the drivers of local hydrology, recent advancements in remote sensing and geographic information system (GIS)-based methods provide an opportunity to more accurately map the spatial extent of streams and wetlands and predict the degree of connectivity between water features across the landscape. These advancements could also provide an important tool for managers and regulators in need of estimates of hydrologic connectivity. The goal of this study was to develop an approach for predicting connectivity using field and GIS-derived landscape predictor metrics representing drivers of connectivity duration and timing.
The specific objectives of this study were to (a) quantify seasonal variability in SHC patterns between Delmarva bay wetlands and perennial streams from field observations; (b) develop predictive metrics from hypothesized landscape drivers of wetland-stream SHC; and (c) model cumulative SHC duration, seasonal connection onset dates, and seasonal connection offset dates as a function of landscape predictor metrics.

| Location and characteristics
This research was completed in two coastal plain watersheds in Maryland: the Choptank River watershed (U.S. Geological Survey [USGS] hydrologic unit code 02060005,~1,756 km 2 ) and the Corsica River watershed (USGS hydrologic unit code 020600020411,~102 km 2 ), a tributary to the Chester River ( Figure 1). Both watersheds are dominated by agricultural land use (60%) with some forest (33% for the Choptank; 25% for the Corsica; [McCarty et al., 2008]). Twenty-three forested wetland complexes across a 150 km 2 area within the Choptank (n = 21) and the Corsica River (n = 2) watersheds were selected for this study. Most wetland catchments in this study (n = 14) were selected because of the ability to monitor the confluence between temporary and perennial streams at road crossings; the remaining catchments (n = 9) were located on state (Maryland Department of Natural Resources, DNR), conserved (The Nature Conservancy), or private lands with explicit permission from landowners.
Hydrology in this region, represented by seasonal and daily stream discharge patterns, is controlled by rainfall, temperature, evapotranspiration, topography, and soil drainage properties (Fisher et al., 2010).
Approximately 50% of annual precipitation is lost to the atmosphere via evapotranspiration while the remainder recharges groundwater or enters streams via surface runoff (Leahy & Martin, 1993). From approximately April to August, evapotranspiration and streamflow discharge rates exceed rainfall, leading to net water loss and falling groundwater levels (Fisher et al., 2010). Surface water levels typically reach peak expression in early spring (March/April) when levels of evapotranspiration are still relatively low (Lang et al., 2012). For the present study, historical monthly rainfall totals  were calculated using PRISM climate mapping system data (http://www.prism. oregonstate.edu/ downloaded on 13 October 2015; Supplemental Online, S2).

| Catchment and stream delineations
Wetland catchments were defined as the total contributing area draining one or more forested Delmarva bay wetlands to a confluence with the perennial stream network. Resulting in part from human perturbations (e.g., ditching), most forested wetlands in the study area connect seasonally to the perennial stream network via surface flow ( Figure 3). Catchment outlets (confluence points of temporary and perennial streams) were first identified within ArcGIS (ESRI; Redlands, CA) using a 2-m digital elevation model (DEM) flow accumulation layer (D8 flow routing algorithm) to find contributing areas immediately upstream of the perennial stream network. A hand-edited, flow accumulation-based stream data set (Lang et al., 2012) was used to represent the perennial/intermittent stream network for the Choptank River watershed. The Lang et al. (2012) stream data set was designed to map streams that are fed by groundwater at least part of the year (perennial and intermittent) hydrology. Considering that the vast majority of streams identified by this method have surface water flow at least 9 months, we refer to these streams as "perennial." For the present study, the Lang et al. (2012) methods were applied to the Corsica River watershed because the two wetlands in it extended beyond the spatial coverage of Lang et al.'s existing data layer (Supplemental Online, S1). Field visits with a handheld global positioning system (GPS) unit (Trimble Geo 7× model) were then conducted to validate catchment outlet locations and assess the eligibility of each site for long-term monitoring. Sites were chosen based upon position within the same watershed (i.e., within the Choptank River or adjacent Corsica River watersheds), receipt of permission to work on the property, exhibition of characteristics indicative of Delmarva bays (e.g., elliptical shape, upland rim, alternating seasonal hydrology), and at least 50% forested cover.

| Landscape metrics
Winter's (Winter, 2001) hydrologic landscape conceptual framework informed our selection of landscape predictor metrics representing the hypothesized drivers of hydrologic connectivity, from field and GIS-derived landscape variables generated at the reach and catchment scales (Table 1). This framework describes hydrologic landscapes based on land-surface form, geology, and climate, which can be used to develop hypotheses of how the hydrologic system functions. This study used metrics characterizing land-surface form and geology of catchments because climate conditions are similar across the sites.
Metrics were classified into four groups: catchment, temporary stream, wetlands, and soils.
Landscape predictor metric development and spatial analyses were conducted using ArcGIS (version 10.1; ESRI, Redlands, CA), R (version 3.2.2), and the Geospatial Modelling Environment (R-Team, 2015;Beyer, 2012). GIS data layers were selected that (a) had spatial coverage across the Upper Choptank, Tuckahoe Creek, and eastern portion of the Corsica River watersheds (1,069 km 2 ) useful for watershed-wide SHC predictions; and (b) maximized our ability to detect FIGURE 3 Schematic of forested wetland catchments, defined as relatively small areas of predominantly forested (generally, >50% forested) land (a) comprised of one or more seasonally inundated Delmarva bays (b) that produce episodic surface outflow into nonperennial streams (c), connecting them to the perennial stream network (d). Catchment outlets were defined as the nonperennial/ perennial stream confluence (e) fine-scale variability between study catchments ranging in area from <1 to ≥70 ha.

| Catchment metrics
Terrain analysis of high resolution digital elevation data was used to delineate flow paths, watersheds, and flow networks (after [Tarboton & Ames, 2001] (Committee, 1988) National Standards for Spatial Data Accuracy for data at 1:2,400. Estimated horizontal positional accuracy of LiDAR point returns exceeds 50 cm. Bridges, roads, and other impediments to two-dimensional flow were eliminated, and then bare earth LiDAR point data were rasterized to create a 2-m resolution DEM using inverse weighted distance interpolation. Stream flow paths were then delineated from this corrected DEM using methods described below (Lang et al., 2012). Catchment terrain slope was calculated as the median slope (m/m) value within the catchment; slope direction was calculated using the Terrain Analysis Using Digital Elevation Models D-Infinity flow routing algorithm. Topographic wetness index (TWI) was calculated for each catchment cell as ln(a/tan β), where "a" is the upslope area per unit contour length and "tan β" is the D-Infinity slope (Beven & Kirkby, 1979). The median TWI value within the catchment was then used.  1998]), and the elevation at catchment outlet. Catchment shape was defined as the catchment's length/width ratio (Bent & Steeves, 2006 (Fritz et al., 2006). Field-based estimates of temporary stream length were collected to compare the accuracy of field and GIS-based delineations.

| Wetland metrics
Past studies have demonstrated that wetland type and characteristics (e. Wetland area was defined as the total wetland area (excluding farmed wetlands, [Cowardin et al., 1979], "Pf" classification) within each catchment. Number of wetlands was determined by the number of wetland polygons within each catchment. Mean wetland distance and minimum wetland distance were calculated by determining the Euclidean distance between each wetland polygon centroid and the catchment outlet. Wetland spill threshold relief was used to estimate the wetland surface water level needed to generate a surface hydrologic connection with the nearby perennial stream. It was calculated using the 2 m DEM as the difference in elevation between the highest point along the temporary stream and the lowest point within the wetland nearest to the catchment outlet ( Figure 4).
The Cowardin et al. (1979) wetland classification includes a water regime modifier code, which describes hydrologic conditions during the growing season. Water regime values for wetlands within the study catchments ranged from saturated (substrate is saturated to the surface but typically no surface water present) to permanently flooded (water permanently covers the land surface; Cowardin et al., 1979). A wetland hydrologic permanence score was generated for each catchment by recoding wetland water regime values to a numerical scale from 1 (saturated) to 6 (permanently flooded), calculating an area-weighted mean water regime value, normalized by total wetland area.

| Soil metrics
Soil-based metrics were generated using Soil Survey Geographic Database (SSURGO) soils data (version 2.2). SSURGO maps are created using manual photo interpretation at scales ranging from 1:12,000 to 1:63,630; minimum delineation size for Maryland surveys is approximately 0.6 ha. County-level soils data were from the U.S. Department of Agriculture's Geospatial Data Gateway (https://gdg.sc.egov.usda. gov/; downloaded 18 August 2015) that were then clipped to the study area. Soil hydrologic groups range from "A" to "D," with "A" soils having a very high infiltration rate (and hence a relatively low runoff potential) and "D" soils having a very low infiltration rate (and hence FIGURE 4 Wetland spill threshold relief was defined as the difference between the minimum elevation within the wetland (a) nearest to the catchment outlet (x), and the highest elevation along the nonperennial stream (b) a relatively high runoff potential; NRCS, 2009). In some areas, soils are assigned a dual hydrologic group status (e.g., "A/D") to indicate soil drainage properties in both "drained" (areas where seasonal high water table is kept at least 60 cm below the soil surface where it would be higher in a natural state) and "undrained" conditions, respectively. A catchment-wide infiltration score was calculated using SSURGO data representing both drained (Infil drained ) and undrained (Infil undrained ) conditions. Hydrologic group values were assigned a numerical scale from 1 to 4 (high to very low infiltration), then aggregated to generate one area-weighted mean catchment value.
Available water storage represented an estimate of the water volume that soil (0-150 cm depth) can store after having been wetted and free drainage has ceased; higher values are generally associated with low infiltration soil types (loams and clays). Annual minimum water

| Relationship between SHC and landscape metrics
Preliminary exploration of the individual relationships between SHC metrics and landscape predictor metrics was conducted using Pearson's product moment correlation tests. Landscape predictor metrics that deviated substantially from normality based on the Shapiro-Wilk normality test were transformed (natural log or inverse of the metric). Spearman rank-order correlation was used for heavily-skewed predictor metrics. Strong correlations were noted as those with correlation coefficients greater than or equal to 0.40.
A forward stepwise linear regression approach (alpha-to-enter ≤0.05) was also used to model SHC patterns (cumulative connection duration, connection onset date, and connection offset date) as a function of the metrics. To reduce the number of metrics included (Austin & Steyerberg, 2015), separate stepwise regressions were first run using predictors from each of the four groups (catchment, temporary stream, soils, and wetlands). Significant predictors from these final regression models were then combined into a single data set to run a full, integrated stepwise regression with predictors from all four landscape predictor groups ( Figure 5). Variance inflation cofactor values, which represent the degree to which variance of the estimated regression coefficients are inflated as compared to when the predictor metrics are not linearly related, were used to assess multi-collinearity in final models (O'Brien, 2007). Landscape predictor metrics with Variance inflation cofactor less than 10 were included in final models.

FIGURE 5
Stepwise regression procedure workflow Moran's I test was used to test for spatial autocorrelation in SHC metrics (cumulative connection duration, connection onset date, and connection offset date) across forested wetland catchments (Paradis, Claude, & Strimmer, 2004).

| Comparing models with field versus GIS-based metrics
To assess model improvement with the addition of field-derived metrics, two stepwise regressions were run for each SHC metric: (a) using field and GIS-derived landscape predictor metrics and ( is most likely to occur. Total 5-day antecedent rainfall was significantly greater on days when a connection event occurred compared to 5-day antecedent totals when a connection did not exist (t = 9.07, df = 22, p < .001, mean of differences = 5.40 mm).

| Observed wetland-stream connectivity patterns
Surface flow patterns in temporary streams connecting forested wetlands to nearby perennial streams varied between wetland catchments (n = 23; cumulative wetland-stream connectivity duration range 64-298 days; mean = 164.6 days). Between late-spring and late-fall, connections were short-term (hours in duration) following rainfall events ( Figures 6 and 7). Median seasonal connection onset and offset dates (first and last >24 hr connection) were December 9 and July 5, respectively ( Overall differences in temporary baseflow among months were significant (F 7, 74 = 2.56, p = .04); peak discharge occurred in early spring (March/April), during which surface water levels generally reach peak expression (Lang et al., 2012; Figure 8).
Three temporary stream metrics were strongly correlated with all three SHC metrics: temporary stream length, temporary stream BFW, and temporary stream cross-sectional area (Table 3). Longer, deeper channels were associated with more prolonged periods of surface flow that initiated earlier and remained longer through the water year.
Models built as a function of both field and GIS-derived predictor metrics explained the most variability in cumulative connection duration (Adj. R 2 = 0.80), followed by seasonal connection onset date (Adj. R 2 = 0.69) and seasonal connection offset date (Adj. R 2 = 0.53;

| Temporal variability in wetland-stream connectivity patterns
Delmarva bays are complex systems whose degree of landscape connectivity is a function of both local and regional hydrological processes.
Like other depressional wetlands, SHC between Delmarva bays and streams is a function of water balance within wetland catchments and landscape attributes including soils and perhaps size Leibowitz & Nadeau, 2003;Sharitz, 2003  year, more than half of wetland-stream surface connections turned "on" and "off" within a 3-week period (Figures 6 and 7). The spatiotemporal homogeneity of SHC onset and offset dates and absence of spatial autocorrelation across the study area suggests that a seasonal drop in evapotranspiration, followed by a regional rise in groundwater table, exert first-order controls over sustained outflow of surface water ponding within bays to temporary streams when the water table is at or above the surface (Lide et al., 1995).
The shortened (minutes to hours), duration of SHC events observed between late spring and late fall in 2015 reflects a seasonal shift in the drivers of SHC. These shortened SHC events coincided with the seasonal peaks in transpiration, during which Delmarva bays typically lack surface water (Fisher et al., 2010;, and generally represented ephemeral surface water levels in temporary streams following rain events. Over the study year, recent rainfall amounts, as indicated by 5-day antecedent rainfall totals, were significantly higher on days when a SHC occurred compared to non-SHC days, indicating that antecedent conditions and local fill-spill dynamics also influence wetland-stream connectivity in Delmarva bays.  The delay from SHC onset to measureable baseflow discharge in temporary streams suggests a mechanistic shift in the SHC driver from groundwater (i.e., ponded surface water level with the groundwater table in the surrounding soils and wetland) to surface water outflow (i.e., wetland spillage) from fall to winter. Although the median seasonal onset date occurred on December 9, baseflow discharge was not measureable (i.e., water depth ≤ 3 cm and/or no measureable water velocity in channel) in most catchments until late January (Supplemental Online, Table T1). Field observations confirm that this shift was aligned with bay ponding levels exceeding their relative spill thresholds and flowing into the adjacent temporary streams (Figure 4). This finding is consistent with the model of wetland connectivity described by (Winter & LaBaugh, 2003), in which surface outflow is described as a function of groundwater flow, spill elevation above normal wetland water level, and the timing of precipitation events.
The effects of seasonal shifts in surface and groundwater dynamics on hydrology at the landscape scale are evident in the seasonal increase in baseflow discharge in Tuckahoe Creek, a tributary to the Choptank River, during the winter and early spring (Figure 9). It is well-documented that depressional wetlands, in aggregate, have a substantial effect on watershed-scale water balances by increasing seasonally defined subsurface storage and groundwater flow (Evenson, Golden, Lane, & D'Amico, 2015). During the 2015 water year, peak temporary stream baseflow discharge below forested wetlands was observed in early spring, indicating that Delmarva bay wetlands contribute to mainstem river discharge via direct surface outflow at least seasonally. This finding is consistent with  which reports a seasonal effect on the relationship between wetland characteristics and streamflow in the Middle Atlantic Coastal Plain ecoregion (North Carolina, United States), including a significant relationship between wetland area and streamflow during the spring, when poorly drained wetland systems respond rapidly to precipitation events.
Regressions in the current study indicated that one or of both the wetland area and wetland number metrics was related to SHC duration and seasonal offset date (Table 3). Future studies are needed to quantify the partitioned (surface water vs. groundwater) and/or aggregate effect of wetland-stream connectivity on downstream waters (e.g., mean seasonal increase in mainstem river baseflow during connections). In addition to linking SHC patterns to downstream ecological processes, future studies should investigate the relationship between wetland-stream groundwater hydrologic connectivity and landscape characteristics (e.g., bay size and soil type;  Note. Models built using full data set (n = 23); full predictor metric names and descriptions in Table 1. BFW = bankfull depth; TWI = topographic wetness index.  [McLaughlin, Kaplan, & Cohen, 2014]) to better quantify the downstream effects of such connections.

| Landscape characteristics as predictors of wetland-stream connectivity patterns
Among the landscape predictor metrics, temporary stream length, BFW, and cross-sectional area were strongly correlated with all three SHC metrics, and BFW was a significant predictor in all final regression models. Several studies have reported these physical measurements to be significant predictors of stream flow duration, including BFW (Fritz, Winerick, & Kostich, 2013;Svec et al., 2005) and entrenchment ratio (flood prone width divided by BFW; [Fritz, Johnson, & Walters, 2008;Svec et al., 2005]), though Fritz et al. (2013) caution they may be weak predictors in high rainfall, low topographic relief regions with low erosive potential. Although correlations between metrics do not imply causation, two mechanisms may explain the relationships between SHC metrics and temporary stream channel geomorphology.
First, surface water presence in temporary stream channels may be an expression of the shallow regional depth to groundwater in our study region ( In general, larger, wetter catchments (greater number and area of wetlands, higher wetland hydrologic permanence score) were associated with greater cumulative SHC duration and later SHC offset dates.
These results agree with earlier findings that wetland area (McDonough et al., 2015) and total catchment area (Lampo, 2015) are positively related to headwater stream flow duration in flat, welldrained landscapes. Larger, wetter catchments were also associated with larger temporary stream channels, illustrating the potential effect of collinearity among landscape predictor metrics in accurately identifying drivers of observed SHC patterns (Supplemental Online Figure   F1). Future studies should consider paired-sensor approaches to discriminate between periods of ponding (i.e., groundwater-fed) and streamflow (i.e., wetland surface outflow) in temporary streams connecting wetlands and streams (Bhamjee et al., 2016), which may help better link landscape characteristics to hydrological patterns.

| Evaluating the accuracy of landscape predictorbased regression models
Stepwise regression has been applied in several other studies within Based on final models and non-significant differences in observed versus predicted correlation strengths between models that included GIS + Field metrics and models with only GIS, the addition of fieldderived predictor metrics did not significantly improve model performance (Table 4). In fact, the removal of field-derived predictor metrics from stepwise regression led to a final GIS-based model of cumulative SHC duration with more support, as indicated by a decrease in AICc value of 4.2 (Burnham & Anderson, 2002). These results suggest that among the variables used in this study, GIS + Field and GIS-based models performed comparably in their ability to explain variability in SHC patterns among forested Delmarva bay wetland catchments.
In their study linking landscape attributes to channel head locations, Julian et al. (2012) concluded that the occurrence of channel heads across Maryland's Coastal Plain was most likely driven by saturation overland flow given the sandy soils and close proximity of the water Linking hydrologic dynamics to landscape structure provides an important foundation for GIS-based analysis of environmental drivers of SHC. Our finding that the addition of field-derived data did not improve predictive models suggests that GIS-derived metrics alone may be adequate for predicting multiple aspects of SHC. Recent technological improvements in the quality and spatial resolution of remote sensing and GIS products provide increasing opportunities to accurately model hydrological patterns as a function of GIS-based variables.
High-resolution (≤5 m) satellite imagery is currently available upon request (Vanderhoof, Alexander, & Todd, 2016;Vanderhoof, Alexander, & Todd, 2017) and will become available at near-daily recurrence intervals in the near future (Tiner, Lang, & Kleman, 2015). The coverage of high-resolution data is increasing as well. In the Unites States, for example, future versions of the National Hydrography Data set will incorporate national, high-resolution elevation data (Viger, Rea, Simley, & Hanson, 2016) from the 3D Elevation Program (Sugarbaker et al., 2017). The 3D Elevation Program, a partnership led by the U.S. Geological Survey, is systematically collecting and processing enhanced elevation data for all U.S. states and territories. Upon completion, anticipated in 2023, products and source data for the entire United States will be freely available online (Sugarbaker et al., 2017). These and other efforts, for example, the Open Geospatial Consortium (OCG, 2017) to provide increased access to high-quality, high-resolution, attribute-rich geospatial data sets are establishing new standards of precision and accuracy for analyses of bare-earth and above-ground features in natural and human-altered landscapes.
Given the role of seasonal groundwater table dynamics as a driver of wetland-stream connectivity in the broader region (Lide et al., 1995), our results likely reflect broad SHC patterns in forested wetlands across the Delmarva bay landscape and may be directly applicable to depressional bays across the U.S. Mid-Atlantic and Southeastern Coastal Plain, where climate and landscape controls of hydrological inputs and losses are expected to be similar to our study area. The SHC patterns and their relationships with predictor metrics in our study area will be most directly applicable to shallow, seasonal wetland complexes that are both precipitation and groundwater driven. As wetlands fill with seasonal precipitation and regional groundwater table rises, inundated wetlands reach a saturation threshold at which they merge with co-located wetlands and spill into defined, temporary flowpaths. In our study system, the outlets of wetland catchments were situated within a relatively short distance (e.g., <400 m) from perennial stream networks. Our results highlight the importance of local and regional surface water-groundwater interactions, wetland density, wetland area, and characteristics of the connecting flowpath on the onset and duration of wetland outflow to stream networks.
Our results can also broadly inform research on watershed connectivity by comparison to studies in contrasting landscapes or studies conducted at different spatial or temporal scales. For example, Vanderhoof et al. (2016, Vanderhoof, Distler, et al., 2017 used Landsat imagery to quantify landscape-scale connectivity in the Prairie Pothole Region. They demonstrated that connectivity via merging of wetlands and wetlands with streams was positively related to total wetland area (also reported by Kahara, Mockler, Higgins, Chipps, & Johnson, 2009) but sensitive to the definition of the wet-dry threshold used to classify Landsat pixels or subpixels as "water" or "land" (Vanderhoof et al., 2016). Temporal and spatial scales of analysis such as those in Vanderhoof et al. (2016) are not achievable using static data sets or most field studies. However, as the authors acknowledge, the resolution of Landsat imagery (30 m) is biased against detection of connectivity via small features, such as those measured here. Smaller scale studies, in turn, may miss variation in drivers of connectivity that occur over longer timeframes or over larger spatial scales. Integration of data sets from different sources to improve accuracy and resolution and modelling over larger spatial and temporal scales is an emerging science (DeVries, Pratihast, Verbesselt, Kooistra, & Herold, 2016;. However, trade-offs between model fidelity and complexity, as well as scale, and interpretation of differences in the climate and landscape that drive dominant connectivity mechanisms (e.g., expansion and merging of surface waters vs. precipitation/groundwater driven fill-and-spill) are required to determine the appropriate application of a model (Golden et al., 2017), including the one presented here.
Lastly, this study can inform stream and wetland restoration efforts in the Delmarva Peninsula. The wetland complexes in this study are all seminatural and are hydrologically connected to perennial waterways via small, intermittent channels (including some historical ditches that are no longer maintained and are structurally indistinguishable from natural streams). Across the Peninsula, approximately 70% of Delmarva bays have been hydrologically altered for agriculture (Fenstermacher, Rabenhorst, Lang, McCarty, & Needelman, 2014), typically by ditch drainage, which increases transport of water and contaminants (e.g., nitrogen) to streams and the Chesapeake Bay (Denver et al., 2004). In a recent study of potential surface water stor- ORCID Jacob D. Hosen http://orcid.org/0000-0003-2559-0687 Margaret A. Palmer http://orcid.org/0000-0003-1468-7993