The Need for Hydrologic Connectivity in Lidar Elevation Surfaces

This section explains the importance of achieving hydrologic connectivity in lidar DEMs. It also describes hydrologic connectivity issues that have been addressed in the scientific literature, and the lack of reference points to validate hydro-enforcement results in drainage structure locations.

As the need for coastal mapping, monitoring, and change detection increases in response to inundation hazards that impact vulnerable coastal zones (Brock and Purkis, 2009; Burkett and Davidson, 2012; Buxton et al., 2013; Gesch, 2009; Gesch, Gutierezz, and Gill, 2009; Stoker et al., 2009; Turnipseed et al., 2007), hydrologic connectivity in lidar DEMs has become essential for storm surge and sea-level rise hydrodynamic modeling (ARCADIS, 2011; Gesch, 2009; Gesch, 2013; Li et al., 2009; MacDonald, 2012; NOAA, 2010; Poulter, Goodall, and Halpin, 2008; Poulter and Halpin, 2008; Sheets, Brenner, and Gilmer, 2012; Westerink et al., 2008; Zhang et al., 2011). Although lidar has become a commonly used technology for the collection of highly accurate elevation information (Buxton et al., 2013; Poppenga et al., 2010; Schmid, Hadley, and Wijekoon, 2011; Stoker, Harding, and Parrish, 2008; Webster and Forbes, 2006) that is used for hydrologic applications (Brock and Sallenger, 2001; Medeiros, 2012; Medeiros et al., 2011; Poppenga et al., 2010; 2012; 2013), according to Barber and Shortridge (2005), a high-resolution, high-accuracy elevation dataset does not necessarily produce a highly reliable model of overland surface flow. Lidar DEMs capture elevated features, such as road fill overlaying culverts, that impact any standard surface hydrology model by impeding the representation of overland surface flow (Figure 1A) (Barber and Shortridge, 2005; Duke et al., 2003; Maune et al., 2007; Poppenga et al., 2013; Poppenga et al., 2014a,b). These elevated road features were not as problematic in the past with lower resolution (~30-m) DEMs derived from topographic maps because the level of topographic detail was much lower than lidar DEMs.

Figure 1.

Topographic digital elevation model (DEM) representation of road surface elevation overlying a culvert (A). Hydrologic DEM representation showing hydrologic-enforcement at the culvert location (B).

In situ, water is carried through culverts; however, because they are located beneath elevated road features, culverts cannot be represented in a lidar (topographic) DEM. Unless hydro-corrected, or hydro-enforced, these road features, so finely detailed in available lidar DEMs, create hydrologic connectivity issues (Poppenga et al., 2010; 2012; 2013; Webster and Stiff, 2008). Without modifications to the elevation surface, water flow would be functionally dammed by the raised topography, creating artificial pooling of depressions on the upstream side (Heidemann, 2012a; 2012b; 2014; Maune, 2007; Poppenga et al., 2010, 2014a). Therefore, a hydro-enforced lidar DEM representing hydrologically connected modeled surface flow is needed for hydrodynamic modeling (Figure 1B) (Heidemann, 2012a; Poppenga et al., 2014a,b).

Some hydrologic connectivity has been addressed with visual identification of culverts and bridges because (lidar) DEMs did not account for these features (Webster and Forbes, 2006; Webster and Stiff, 2008). However, these techniques do not consider the time consuming visual identification and manual hydro-enforcement. Other methods superimpose ancillary line networks into DEMs, such as stream burning (Saunders, 2000) or culvert burning (Stiff, Hopkinson, and Webster, 2008), or adaptive drainage enforcement (Kenny and Matthews, 2005; Soille, Vogt, and Colombo, 2003; Turcotte et al., 2001; Zhang and Huang, 2009). These approaches may result in alignment issues, especially in high-resolution (lidar) DEMs (Chow, 2010), and may alter the integrity of an elevation surface, especially in stream network locations (Poppenga et al., 2013).

Superimposing line networks does not address the extraction of fine-scale (2–5 m) stream channels (Cho et al., 2011), which are not available as vectors (line networks). Additionally, raised surfaces with underlying drainage pipes are not always located along roads (Chow, 2010). Although Cho, Kampa, and Slatton (2007) demonstrated the feasibility of fine-scale stream extraction, they had difficulties connecting stream segments. Thus, a simple vector superimposition does not solve the hydrologic connectivity issue.

Alternatively, a method that can produce raster (grid) and vector (line) hydrologic connectivity is a digital hydro-enforcement approach as presented by Poppenga et al. (2010; 2012; 2013; 2014a,b). This approach uses residuals calculated between unfilled and filled lidar elevation surfaces to define artificially pooled depressions that form and are in need of hydro-enforcement. This method creates a hydrologically connected lidar DEM that models the flow of water across the land surface and consists of a suite of hydrologic derivatives that are consistently integrated with the high-resolution elevation data (Poppenga et al., 2009). To determine the feasibility of this digital hydro-enforcement method, two different types of quantitative assessments were conducted. The goal of the first assessment was to use culvert and bridge datasets, collected by public agencies, to validate the hydro-enforcement. However, because of the lack of comprehensive culvert/bridge reference point datasets that can be used to conduct hydro-enforcement assessments (Chow, 2010), it was necessary to conduct a second assessment of image-interpreted culverts and bridges.

Hydrologic-Enforcement Quantitative Assessments

This section describes the paucity of reference points to validate hydro-enforcement, and the challenges that arise when conducting assessments on hydro-enforced drainage structure locations.

In the scientific literature, hydrologic uncertainty assessments have been presented, however, these studies pertain to rainfall, river discharge, or water quality (Heistermann and Kneis, 2011; McMillan, Krueger, and Freer, 2012; Seibert, 2001). A Hydrologic Benchmark Network was established by Leopold (1962), but at most this network contains 58 drainage basins in 39 states and was developed to provide long-term measurements of streamflow and water quality in areas virtually free of human activities (Mast and Turk, 1999; Murdoch et al., 2005). Therefore, the challenge of hydro-enforcement validation originates from the lack of hydrologic benchmarks that can be referenced as ground truth (Chow, 2010). This problem is compounded by the paucity of hydro-enforcement accuracy assessments, which lag behind the ever-increasing demand for lidar-derived hydrologic information.

Hydro-enforcement validation is problematic because drainage structure reference points are not readily available or even as accurate as control points that are available for DEM accuracy assessments. Considering the lack of drainage structure reference points that contain as high of a degree of accuracy, frequency, and availability as the control points that are used for elevation assessments (Evans et al., 2014; Gesch, Oimoen, and Evans., 2014; National Geodetic Survey, 2012; Samsung et al., 2013; Thatcher et al., 2014), conducting a hydro-enforcement assessment requires a creative approach.

Hydrologic studies are often conducted over relatively large regions using high-resolution lidar-based DEMs. Collecting survey-grade reference points for thousands of culverts and bridges in an attempt to conduct a hydro-enforcement assessment is not feasible during a project timeframe. Because an all-encompassing nationwide drainage structures database was not available, georeferenced coordinates for culverts and bridges were obtained from various public agencies. Combining these disparate field surveys into one dataset would seem to be a viable solution for obtaining an independent reference point dataset. However, collectively, these surveyed data are a medley of drainage structure types, geographic locations, and collection dates obtained using various mapping techniques with varying levels of positional accuracy.

Therefore, the challenge of hydro-enforcement assessments lies not only in obtaining the actual field survey datasets from public agencies but also in the lack of a comprehensive drainage structure dataset. Ancillary field survey collections will never be 100% of what is needed to assess hydro-enforcement because of the staggering amount of depressions that are created when hydro-conditioning (filling) a lidar DEM. Although many of the depressions are caused by noise and artifacts in the detailed lidar DEM, field-surveyed data cannot contain coordinates for every location that becomes impounded by elevated features other than road fill in the lidar DEM. Therefore, in this article, an approach is presented that encompasses two types of hydro-enforcement assessments in select watersheds in the Delaware River Basin.

Study Areas

An assessment was conducted of hydro-enforcement in Tinicum Creek, Lockatong Creek, and Jacobs Creek in the Delaware River Basin (Figure 2). These watersheds are part of the Newark Basin in Pennsylvania and New Jersey and consist of rural landscapes. The watersheds were selected because they are designated by the U.S. Department of the Interior as named tributaries included in the National Wild and Scenic Rivers System tributaries of the Lower Delaware River that drain into the coastal Delaware Bay (Delaware River Basin Commission, 2004; National Park Service, 2012). The landscape of each study site is both variable and representative of coastal watershed regions.

Figure 2.

Lidar-derived watersheds in the Delaware River Basin study area (A). Tinicum Creek watershed, Bucks County, Pennsylvania (B). Lockatong Creek watershed, Hunterdon County, New Jersey (C). Jacobs Creek watershed, Mercer County, New Jersey (D).

The first study area is located in the Tinicum Creek watershed in Bucks County, Pennsylvania (Figure 2B). Tinicum Creek surface waters flow generally from north to south through its upper two branches, Rapp and Beaver Creek, and flow northeastward to the Delaware River confluence. The watershed drains an elevation-derived area of 62.60 km² with a minimum elevation value of 29.25 m and a maximum elevation value of 256.57 m. The landscape consists of steep vertical cliffs exposed along the streambed and forests interspersed with cultivated crops in the higher elevations.

The second study area is located in the Lockatong Creek watershed, a tributary of the Delaware River in Hunterdon County, New Jersey (Figure 2C). Lockatong Creek surface waters flow generally from north to south through an elevation-derived area of 62.33 km². The minimum elevation value in this watershed is 20.32 m and the maximum elevation value is 211.31 m. The upper part of the watershed landscape consists of cultivated crops and pasture interspersed with deciduous forests, and the lower part of the watershed consists mainly of deciduous forests with some cultivated crops and pasture.

The third study area is located in the Jacobs Creek watershed in Mercer County, New Jersey (Figure 2D). Jacobs Creek surface waters flow generally from north to south through an elevation-derived area of 34.28 km². The minimum elevation value in this watershed is 7.90 m and the maximum elevation value is 139.05 m. This watershed contains state and nationally designated historic districts and bridges that were impacted by flooding associated with Hurricane Irene (County of Mercer, 2011; Cusido, 2014). The landscape consists of both urban and commercial development with agriculture interspersed among forested areas.

Lidar Data Sources

Effective hydrologic modeling using lidar DEMs requires a single hydro-enforced surface model for the study area. Unfortunately, the extents of topographic data acquisitions are typically based on political boundaries such as county or state boundaries rather than watershed boundaries, and the most commonly delivered data are the traditional topographic DEMs (Poppenga et al., 2014a). Because the topographic DEMs were not hydrologically-corrected for hydrologic modeling, it was necessary to merge multiple topographic DEMs from disparate sources into a single surface prior to hydro-enforcement in the study areas.

For a part of the Delaware River watershed in New Jersey and Pennsylvania, three disparate airborne lidar datasets collected between 2006 and 2009 were obtained from the USGS. Because the lidar datasets were collected in various projections and at different time periods, a final mosaicked DEM was processed into one projection as Universal Transverse Mercator (UTM) Zone 18N, North American Datum of 1983 (NAD83), and North American Vertical Datum of 1988 (NAVD88) (Figure 3). The resulting 1-m resolution, 23-gigabyte mosaicked lidar DEM was used as the foundation layer for hydro-enforcement (Poppenga et al, 2014a). The metadata regarding each lidar dataset is provided in Table 1.

Figure 3.

Map of disparate airborne lidar DEM datasets in New Jersey and Pennsylvania mosaicked as a high-resolution topographic DEM used to generate a hydrologically-corrected DEM.

Table 1.

Lidar source data and digital elevation model (DEM) metadata that were used to generate the mosaicked DEM in the Delaware River Basin.

Field-Surveyed Culvert and Bridge Data Sources

Field-surveyed culvert and bridge datasets were obtained from public agencies in Pennsylvania and New Jersey (Table 2). Obtaining these datasets was a time-consuming process that consisted of contacting each public agency by email or phone to request various field survey data, and at times requiring signed agreements. Most drainage structure datasets obtained from the public agencies were in geographic information system (GIS) format; however, a few datasets were only available in spreadsheet or text format, thus necessitating a conversion to georeferenced coordinates. Table 3 lists the numerous data types and sources, sizes, maintenance, and ownership of the field-surveyed data that were obtained. All disparate culvert and bridge data were combined into one georeferenced dataset for the assessment analysis.

Table 2.

Surveyed culvert and bridge coordinate data obtained from public agencies.

Table 3.

Surveyed culvert and bridge source data and ownership in the Delaware River Basin study areas.

Field-surveyed metadata did not accompany the drainage structure datasets received from public agencies. Therefore, positional accuracy of the drainage structure locations were undetermined.

Digital Orthophoto Quadrangles Sources

Additional ancillary data used in this hydro-enforcement assessment included DOQs acquired by various vendors that were available from the USGS. Bucks County, Pennsylvania, DOQs were acquired by Aero Metric, Inc. (2010) on flight dates spanning from March 2010 to April 2010. Hunterdon County, New Jersey, DOQs were acquired by Photo Science, Inc. (2012a) on flight dates spanning from March 2012 to April 2012. Mercer County, New Jersey, DOQs were acquired by Photo Science, Inc. (2012b) on flight dates spanning from March 2012 to April 2012. The data consisted of 0.3-m pixel resolution (~1-ft), natural color orthoimages (Table 4).

Table 4.

Digital orthophoto quadrangles (DOQs) source metadata for the study areas in the Delaware River Basin.

In addition to merging multiple, disparate lidar DEMs into a single surface, it was also beneficial to merge the DOQs into one orthoimagery surface as well. This allowed seamless movement throughout a GIS environment during the hydro-enforcement assessments.

Hydrologic-Enforcement Methods

Summary information for the hydro-enforcement methods is given herein; additional details regarding the methodology can be found in Poppenga et al (2010; 2012; 2013; 2014a). A combination of filling depressions in the DEM (hydro-conditioning) and draining techniques (hydro-enforcement) (Poppenga et al., 2010; 2012) was used to define road fill, railroad grade, and other elevated features that caused artificially pooled depressions to form upstream of drainage structures (culverts or bridges). The elevations of these features were digitally modified to achieve hydrologic connectivity in lidar DEMs.

In the Delaware River Basin study areas, the mosaicked bare-earth lidar DEM was hydro-conditioned, or filled, to facilitate cell-to-cell routing (Figure 4). A difference grid (referred to hereafter as DiffGrid) was then generated by calculating the residuals between the bare-earth (unfilled) lidar DEM and the hydro-conditioned (filled) lidar DEM using Equation 1 (Poppenga et al., 2010):

Figure 4.

An overview of the hydrologic-conditioning and hydrologic-enforcement methods applied to the mosaicked lidar DEM study areas to obtain hydrologic connectivity. Detailed information regarding this methodology is provided in Poppenga et al. (2010; 2012).

Because the DiffGrid can contain millions of un-drained depressions, it was not plausible to manually inspect all of them; only a subset of DiffGrid depressions was suitable for draining. To address the large number of un-drained depressions, criteria were set to exclude small depressions that were not likely to impact surface changes by draining them. DiffGrid summary statistics including area, minimum, maximum, and sum (volume), were calculated for each of the remaining depressions. Note that the volume parameter is a sum of the DiffGrid and represents the volume filled for each depression. Empirical testing was then conducted using various DiffGrid area and sum (volume) parameters to define un-drained depression subsets that prevented surface flow connectivity. These subsets were overlain on DOQs to visually analyze if the un-drained depressions needed draining and to determine which parameters provided the most plausible DiffGrid subset for hydro-enforcement. The empirical testing was based upon the hypothesis that depressions in the study areas within the following parameters were a proxy for candidate depressions that should be considered for draining. The set of candidate depressions that should be considered for draining, D_c, based on empirically-determined thresholds for planimetric area and volume, were identified:

Using the lidar DEM, the following hydro-enforcement methods were then executed for each location where depressions needed draining. A least cost path analysis was computed by calculating Euclidian distance between the lowest elevation value (source pixel) inside a depression location and the next lowest elevation value (target pixel) downstream of the raised elevation feature (Poppenga et al., 2010; 2012; 2013). The calculated distance between the source and target pixels was used to extract a 1-pixel-wide transect DEM from the lidar DEM. Because the transect DEM represented raised elevation values that artificially impede representative overland surface flow, the transect DEM values were lowered to equal that of the target pixel. The modified elevation values were mosaicked into a copy of the lidar DEM to preserve the topographic DEM while creating a hydrologically-corrected DEM (Figure 4). The hydro-enforcement process thereby creates monotonically decreasing elevation values in the hydrologically-corrected lidar DEM to achieve hydrologic connectivity. Note that the interactive part of the hydro-enforcement methodology is the selection of depressions that need draining. All other hydro-enforcement processes are automated.

Quantitative Assessments

This section describes the methods used to compare spatial correspondence between semi-automated hydro-enforcement and ancillary vector drainage structure data and DOQs.

Corresponding Field-Surveyed Drainage Structure Coordinates and Hydro-Enforced Transects

This quantitative assessment is a distance-based error metric defined between the lidar-derived hydro-enforced transect and a corresponding ancillary drainage structure. Because of the absence of hydrologic benchmarks that would be viable for hydro-enforcement assessments (Chow, 2010), field-surveyed culverts and bridges vector data that were obtained from numerous public agencies were used as reference points to assess the hydro-enforced transects (Tables 2 and 3). Because the field-surveyed datasets were a plethora of spatial coordinates collected on various dates by various entities with differing equipment, some hydro-enforced transects were near more than one set of field-surveyed coordinates. To eliminate redundant coordinates for any one hydro-enforced transect, the field-surveyed point data were compared with ancillary DOQs. If there were more than one set of spatial coordinates for any specific hydro-enforced transect, the coordinate that was nearest to the culvert or bridge identified in the DOQ was selected, and any duplicate coordinates near that location were eliminated.

A proximity analysis was conducted by defining corresponding relationships between each ancillary field-surveyed coordinates and their nearest hydro-enforced transect. The offset distance between each set of field-surveyed coordinates that corresponded with each hydro enforced transect was measured. These measurements were used to assess the hydro-enforcement techniques used in the Delaware River Basin study areas.

Corresponding Aerial Photograph Interpreted Drainage Structures and Hydro-Enforced Transects

In lidar DEMs, depressions do not always form near roads, where field-surveyed drainage structures are typically collected. Therefore, a merged dataset of disparate field-surveyed coordinates was not comprehensive enough to correspond with all DiffGrid depressions. A different approach was developed to define drainage structure reference points by interpreting DOQ aerial photographs in a GIS environment. This interpretation was conducted exclusively with DOQs and no other elevation or ancillary data.

The DOQs were interpreted to create points for culverts and bridges that were near each depression that was selected to hydro-enforce. Each interpreted point represented the center most point between the inlet and outlet of a drainage structure. Culvert locations were identified by having a visible inlet and outlet with vegetation often surrounding the road and culvert openings. Bridges were identified along larger streams and had rectangular features adjacent to the road sometimes with shadows below the bridge deck. Bridge decks often had contrasting brightness compared to the adjacent roads. At times, it was not possible to classify the drainage structure as a culvert or bridge, but an interpreted point was designated nevertheless because there were clearly defined stream channels on both sides of the impoundment. The drainage structure points interpreted from DOQs were used to define offset distances with hydro-enforced transects. The purpose of the assessment between DOQ-interpreted drainage structures and hydro-enforced transects was to identify additional reference points for validating hydro-enforcement.

First Assessment Using Field-Surveyed Drainage Structure Coordinates

The number of ancillary field-surveyed drainage structures that corresponded with hydro-enforced transects was limited in all three study areas. For example, in Table 5, Column 2, although there were 33 artificially pooled depressions that needed hydro-enforcement in the Tinicum Creek study area, only five field-surveyed drainage structures corresponded to hydro-enforced transects (Column 3). This limited corresponding trend continued in the Lockatong Creek watershed where only 13 field-surveyed drainage structures corresponded with 39 hydro-enforced transects, and in the Jacobs Creek watershed where only 10 field-surveyed drainage structures corresponded to 49 hydro-enforced transects (Table 5). However, where there was a match between the predicted hydro-enforced transects and the observed field-surveyed data, the weighted mean offset distance was 3.91 m with a maximum distance of 28.61 m (Table 6). The 95^th percentile ranking of the offset distance for Tinicum Creek was 3.53 m; Lockatong Creek was 18.13 m; and Jacobs Creek was 6.17 m (Table 6). Although these summary statistics imply successful hydro-enforcement where correspondence existed, the sample size of field-surveyed data was not sufficient in all three watersheds to derive a valid quantitative assessment (Table 5). An analysis was needed to determine the causes of the limited corresponding trend.

Table 5.

Comparison of hydro-enforced transects with ancillary field-surveyed drainage structures from public agencies.

Table 6.

Summary statistics for offset distances between field-surveyed drainage structures and hydro-enforced transects.

One cause of limited correspondence between hydro-enforced transects and field-surveyed data was removal (by the lidar data vendors) of elevation data representing bridge decks. In topographic DEM processing, bridge deck elevations are typically removed to obtain the bare-earth topographic surface.

Thus, a DiffGrid depression will not form in bridge deck locations where the lidar DEM has essentially been hydrologically-corrected. Although this may be an ideal situation for deriving monotonic surface flow, bridge deck removal had a negative effect in this hydro-enforcement assessment. Comparing a field-surveyed bridge coordinate with a DiffGrid depression that did not exist resulted in a limited correspondence. To compound this issue, the bridge removal process was not consistent in all study areas. Therefore, field-surveyed data were excluded in locations where bridge decks were removed from the lidar DEM. This issue did not apply to underground culvert drainage structures.

Temporal differences also caused limited correspondence between hydro-enforced transects and field-surveyed drainage structures. For example, newly constructed culverts or bridges that were not included in the field-surveyed datasets but appear in lidar DEMs will not correspond with hydro-enforced transects. Additionally, the plethora of collection methods and equipment accuracies may have contributed to the limited corresponding trend.

Combining disparate field-surveyed drainage structures into one dataset revealed that some watersheds contained more bridge coordinates than culvert coordinates. This issue surfaced in the Tinicum Creek watershed (Figure 2B) where only five field-surveyed drainage structures corresponded with the hydro-enforced transects (Table 5). Because most bridge decks in this watershed were already removed from the lidar DEM, the field-surveyed bridge coordinates were also removed from this assessment. The remaining few culvert coordinates did not correspond with the numerous DiffGrid depressions in the lidar DEM. The result was a limited correspondence in this watershed.

The parameters defined in the Methods section were used to select depressions that needed hydro-enforcing in the study areas. The purpose of the parameters was to reduce the unreasonably high number of DiffGrid depressions to a manageable size so that hydro-enforcement processes could be executed. If the parameters were set too low, the results contained depressions (noise) that did not need hydro-enforcement. If the parameters were set too high, small depressions in need of hydro-enforcement were excluded from the selection. This resulted in errors because some smaller depressions were near culvert locations that actually corresponded to a field-surveyed drainage structure. This scenario is virtually impossible to predict until after hydro-enforcement results are reviewed.

Because of the extreme detail in the mosaicked lidar DEM, the number of DiffGrid depressions defined by the criteria that needed hydro-enforcement greatly exceeded the number of ancillary field-surveyed drainage structures. This was because DiffGrid depressions form not only near elevated road surfaces where culverts and bridges are field-surveyed but also near other features, such as dams, weirs, private drives, and bike path structures where public agencies may not have collected coordinates for drainage structures. Because of the complex nature of lidar DEMs, there will always be depressions in need of hydro-enforcement in locations other than elevated road features. Some DiffGrid depressions were located in the middle of land sections or in remote areas and did not correspond to any field-surveyed data. Therefore, because bridge and culvert coordinates are typically surveyed on roads, dependence upon the disparate field-surveyed datasets for hydro-enforcement validation resulted in an incomplete assessment.

Second Assessment Using Aerial Photograph Interpreted Drainage Structures

Due to the limited sample size between the predicted hydro-enforced transects and the observed data field-surveyed drainage structures (Table 5), an additional survey of aerial photograph-interpreted culverts and bridges from DOQs was developed to define drainage structures that were impounded in locations other than elevated road surfaces. This assessment, conducted exclusively with DOQs, was feasible for the project timeframe rather than conducting field surveys for the numerous depressions in the study areas.

The DOQ-interpreted assessment resulted in a greater number of corresponding drainage structures (Table 7, Column 3) than the field-surveyed data assessment (Table 5, Column 3). For example, in Tinicum Creek, there was a match between 26 air photograph-interpreted drainage structures and hydro-enforced transects (Table 7, Column 3), whereas there were only 5 matches with the field-surveyed data (Table 5, Column 3). A few drainage structures could not be confidently interpreted in the orthoimagery because of leaf-on conditions, especially in heavily vegetated areas surrounding the drainage structure. Additionally, the size of smaller drainage structures made it challenging to discern an inlet or outlet in the DOQs.

Table 7.

Comparison of hydro-enforced transects with air photograph-interpreted drainage structures from orthoimagery.

Although the DOQ-interpreted drainage structure assessment resulted in more matches with the hydro-enforced transects (Table 7, Column 3), the weighted mean offset distance was slightly higher (6.71 m) (Table 8) than the field-surveyed weighted mean offset distance (3.91 m) (Table 6). The maximum offset distance also increased for the interpreted data to 70.607 m (Table 8), while the field-surveyed maximum offset distance was 28.614 m (Table 6).

Table 8.

Summary statistics for offset distances between air photograph-interpreted drainage structures and hydro-enforced transects.

Because of these increases, the spread of offset distances, as illustrated in the box plots shown in Figure 5, was examined. The bottom of each plot is the 25^th percentile and the top is the 75^th percentile. Extreme offset values exceeding 1.5 times the interquartile range (25–75) are plotted as blue crosshairs. The median values are shown as horizontal red lines. The values along the vertical axis are the offset distances in meters. The characteristics for Lockatong Creek and Jacobs Creek display extreme offset distances while Tinicum Creek has good agreement. Further examination of the air photograph-interpreted data showed that the extreme offset distances were due to detention basin depressions that were impounded by features other than elevated road surfaces. Ironically, detention basin coordinates were usually not available in the field-surveyed data. However, because the detention basin points were interpreted in the orthoimagery, they caused extreme offset distances in this second hydro-enforcement assessment.

Figure 5.

Box-and-whisker plot of the distance between air photograph-interpreted drainage structures and hydro-enforced transects in the three Delaware River study area watersheds.

There are a variety of detention basin outlet structure types. For the most part, those identified were discharge stand pipe or culvert points in the orthoimagery that corresponded with DiffGrid depressions that were selected to drain. At times, it was not clear in the orthoimagery if the detention basin structure drained overland or sub-surface. The impounding feature (berm) and the stand pipe or culvert was visible, yet the downstream drainage was not evident in the imagery. For sub-surface drainage, if a detention basin discharge point was designated, the result was an extreme offset distance. This was because hydro-enforcement, initiated at the lowest point in the detection basin (usually an outlet point), generated a lengthy transect when searching for a nearby channel to drain downstream. In some of the detention basins the lowest elevation was not located near a stand pipe or culvert; perhaps this was because detention basins are designed to impound water for a short duration, or perhaps this was due to bare-earth processing errors in heavily vegetated areas. Regardless of the cause, the result was also an extreme offset distance between the interpreted points and the hydro-enforced transects. Therefore, detention basin depressions and their corresponding interpreted points needed to be removed from this second assessment to avoid degrading the lidar DEM with long hydro-enforced transects.

Summary statistics and percentile rankings for remaining DOQ-interpreted points exclusive of detention basin outlet structure points were then re-calculated. The weighted mean offset distance improved from 6.71 m (Table 8) to 4.94 m (Table 9). The results in Table 9 show an improved corresponding agreement between interpreted points and hydro-enforced transects, particularly in the Lockatong Creek watershed (Figure 6). The cumulative percent graphs, shown in Figure 7, illustrate that 95% of the offset distances were less than 7.04 m in Tinicum Creek, 21.69 m in Lockatong Creek, and 10.44 m in Jacobs Creek (Table 9).

Figure 6.

Box-and-whisker plot of the distance between air photograph-interpreted drainage structures and hydro-enforced transects excluding detention basin depressions in the three Delaware River study area watersheds. ND = no detention basins.

Figure 7.

Cumulative percentage graphs (excluding detention basins) of the distance between air photograph-interpreted drainage structures and hydro-enforced transects in the three Delaware River study area watersheds.

Table 9.

Summary statistics for offset distance between air photograph-interpreted data excluding detention basins and hydro-enforced transects.

In general, the variability of hydro-enforcement assessment results in the Lockatong Creek and Jacobs Creek watersheds was most likely caused by bare-earth processing, or in other words the failure to remove lidar points classified as vegetation. Additionally, in vegetated areas, as well as in some channels, elevation errors appeared in the DEM, thereby causing extreme offset distances in the hydro-enforcement assessments. Perhaps these errors occurred because of insufficient lidar points in specific areas during DEM generation and artifacts from natural neighbor resampling.