Environmental Management

, Volume 50, Issue 3, pp 462–477

Dealing With Uncertainty When Assessing Fish Passage Through Culvert Road Crossings

Authors

  • Gregory B. Anderson
    • Odum School of EcologyUniversity of Georgia
    • Department of Fish and Wildlife ConservationVirginia Polytechnic Institute and State University
    • United States Geological Survey, Patuxent Wildlife Research CenterUniversity of Georgia
  • Byron J. Freeman
    • Odum School of Ecology and Georgia Museum of Natural HistoryUniversity of Georgia
  • Carrie A. Straight
    • Odum School of EcologyUniversity of Georgia
  • Megan M. Hagler
    • Odum School of EcologyUniversity of Georgia
  • James T. Peterson
    • United States Geological Survey, Georgia Cooperative Fish and Wildlife Research UnitUniversity of Georgia
    • USGS Oregon Cooperative Fish and Wildlife Research Unit
Article

DOI: 10.1007/s00267-012-9886-6

Cite this article as:
Anderson, G.B., Freeman, M.C., Freeman, B.J. et al. Environmental Management (2012) 50: 462. doi:10.1007/s00267-012-9886-6

Abstract

Assessing the passage of aquatic organisms through culvert road crossings has become increasingly common in efforts to restore stream habitat. Several federal and state agencies and local stakeholders have adopted assessment approaches based on literature-derived criteria for culvert impassability. However, criteria differ and are typically specific to larger-bodied fishes. In an analysis to prioritize culverts for remediation to benefit imperiled, small-bodied fishes in the Upper Coosa River system in the southeastern United States, we assessed the sensitivity of prioritization to the use of differing but plausible criteria for culvert impassability. Using measurements at 256 road crossings, we assessed culvert impassability using four alternative criteria sets represented in Bayesian belief networks. Two criteria sets scored culverts as either passable or impassable based on alternative thresholds of culvert characteristics (outlet elevation, baseflow water velocity). Two additional criteria sets incorporated uncertainty concerning ability of small-bodied fishes to pass through culverts and estimated a probability of culvert impassability. To prioritize culverts for remediation, we combined estimated culvert impassability with culvert position in the stream network relative to other barriers to compute prospective gain in connected stream habitat for the target fish species. Although four culverts ranked highly for remediation regardless of which criteria were used to assess impassability, other culverts differed widely in priority depending on criteria. Our results emphasize the value of explicitly incorporating uncertainty into criteria underlying remediation decisions. Comparing outcomes among alternative, plausible criteria may also help to identify research most needed to narrow management uncertainty.

Keywords

CulvertFish passageStream habitatImperiled fishesBayesian belief network

Introduction

Fragmentation of stream systems by dams and other human-made structures is a major cause of declining aquatic biodiversity worldwide (Strayer 2006; Dudgeon and others 2006; Helfman 2007). The effects of dams as barriers to migrations and dispersal of a variety of species (plants, mussels, fishes) are well documented (Ward and Stanford 1979; Petts 1984; Poff and Hart 2002; Freeman and others 2003), but managers are increasingly focused on the smaller and more pervasive barriers created by road crossings on streams, especially where crossings are constructed with culverts. These structures frequently constrict the stream through a hardened structure such as a metal or concrete pipe that can impede movements by fishes and other aquatic or semi-aquatic organisms (Derksen 1980; Belford and Gould 1989; Utzinger and others 1998; Warren and Pardew 1998; Toepfer and others 1999; Schaefer and others 2003; Gibson and others 2005; Benton and others 2008; Norman and others 2009). Consequently, federal and state resource-management agencies have developed programs to identify and remediate culverts that are barriers to stream biota movements. Examples include the U.S. Fish and Wildlife Service’s National Fish Passage Program and the U.S. Forest Service’s National Inventory and Assessment Procedure (Clarkin and others 2005), the latter of which focuses specifically on road crossings as barriers in streams. State fish and wildlife agencies, including those in Alaska, California, Maine, Massachusetts, Oregon, Washington, and Vermont, have similarly developed guidelines for recognizing and avoiding construction of culverts likely to impede fish movement (Mirati 1999; Taylor and Love 2003; Reback and others 2004; Milone & MacBroom, Inc 2009; WDFW 2009; ADF&G 2010).

Deciding where to invest restoration resources on culvert remediation (i.e., removal or replacement with an alternative structure) poses a distinct problem from designing road crossings that will not impede passage by stream biota. Current thought on design dictates placing culverts so that the streambed maintains natural channel characteristics as it passes under the road, with similar depths, bed sediments and water velocities to those upstream and downstream of the crossing (Bates and others 2003; Norman and others 2006; River and Stream Continuity Partnership 2006). However, a program to reduce fragmentation of habitat and populations by replacing existing poorly designed culverts involves the initial questions of (1) which culverts are barriers to organism movement, and (2) which structures would, if remediated, most benefit aquatic biota. Addressing these questions provides a foundation for considering other benefits (e.g., reducing crossing maintenance and increasing safety), along with the costs of remediation and additional considerations (e.g., public and partner support).

Identifying whether a culvert is a barrier can be difficult because aquatic organisms vary widely in their ability to traverse steep grades or move against high velocity currents. Additionally, organisms may require passage only at particular times, for example to reach spawning, rearing or overwintering habitats (Fausch and Young 1995) or only periodically to facilitate re-colonization (Meffe and Sheldon 1990; Detenbeck and others 1992; Roghair and Dolloff 2005; Albanese and others 2009) and genetic exchange (Gagen and Rajput 2002). Thus, a culvert that is a barrier to movement at extreme low flows might provide passage at higher flows adequate to maintain populations. These considerations have led to differing criteria for identifying culverts as barriers, even for fishes with well-studied leaping and swimming abilities. For example, standards for designating a culvert as impassable by adult salmonids have included a drop from the culvert outlet to the water surface of 0.24 m (WDFW 2009), 0.3 m (NMFS Southwest Region 2001; Milone & MacBroom, Inc. 2009), and 0.6 m (Taylor and Love 2003; Coffman 2005). For fishes with lesser leaping or swimming abilities, thresholds for what constitutes barriers are even less clear. Most standards recognize uncertainty in identifying culverts as impassable, and several recognize that passage might not be a binary state (i.e., passable vs. impassable) (Coffman 2005; Diebel and others 2010; WDFW 2009; Kemp and O’Hanley 2010).

In typical landscapes with multiple road crossings on streams, culverts may be prioritized for remediation to maximize expected benefit to aquatic organisms, perhaps while minimizing costs or meeting other objectives. Often, culvert remediation decisions are based on scoring and ranking schemes that are independent of spatial arrangement of the culverts or other barriers (O’Hanley and Tomberlin 2005; Kemp and O’Hanley 2010). However, the remediation of a barrier within close proximity of another barrier may do little to improve stream connectivity. By considering the spatial context of culverts and other barriers (e.g., impoundments), the amount of habitat potentially reconnected by remediation could be incorporated in prioritization scores. For example, Cote and others (2009) propose the Dendritic Connectivity Index (DCI) as a measure of how multiple barriers decrease the probability that an organism is able to move between two randomly selected points in a stream network. Simulating DCI with each barrier in a watershed eliminated in turn can identify barrier removal that would maximize network connectivity, thus accounting for the spatial context of each barrier (Cote and others 2009; Bourne and others 2011). Diebel and others (2010) provide a similar measure of culvert effects on watershed-wide stream connectivity, in this case summed across habitat types (e.g., stream orders) and weighted by habitat quality and distance between stream segments. Measures of network connectivity also can be used in optimization models to rank barriers for removal with the goal of maximizing watershed-wide stream connectivity given economic constraints (O’Hanley and Tomberlin 2005; Diebel and others 2010; Kemp and O’Hanley 2010).

Our objective has been to develop an approach for prioritizing culverts for removal or replacement to benefit small-bodied species of imperiled stream fishes, recognizing the interrelated challenges posed by uncertainty in assessing culverts as barriers, and the influence of the spatial arrangement of culverts and other barriers on potential benefits. Managing for persistence of small-bodied imperiled stream fishes may differ from maximizing basin-wide connectivity for anadromous or migratory (potomodromous) fishes (O’Hanley and Tomberlin 2005; Cote and others 2009; Bourne and others 2011). Many imperiled stream fishes occur in small, often disjunct areas within a basin, where local habitat fragmentation may reduce gene flow, effective population size (Allendorf and Luikart 2007) and recolonization potential (Albanese and others 2009), and increase the potential for inbreeding, genetic drift, and loss of ecological and genetic diversity. In these cases, the availability of unfragmented stream habitat in the areas occupied by each species of concern may be more relevant to species conservation than basin-wide connectivity (which could include connectivity to areas beyond the dispersal distances of target species). Therefore, our approach has involved estimating potential gain in connected habitat for targeted species of conservation concern from remediating individual culverts, accounting for spatial context relative to upstream and downstream barriers. We also explicitly examine how the choice among alternative, plausible criteria for evaluating fish passage at a culvert may influence relative rankings of culverts for remediation in an actual river system. Our analysis illustrates a general approach for prioritizing culvert remediation to benefit aquatic species, the potential importance of explicitly recognizing uncertainty in how culverts form barriers, and areas where research could narrow that uncertainty.

Methods

Study Area

The study area encompassed the Conasauga, Coosawattee, and Etowah river systems, having a combined area of 8,941 km2 in the Upper Coosa River system, northern Georgia and southeastern Tennessee (Fig. 1). This area was selected for its high species diversity (97 native extant fish species), conservation concerns (six federally protected and eight state protected species; Table 1), and high level of projected urban and suburban development (Freeman and others 2005; Wenger and others 2010). Hundreds of small impoundments and farm ponds were known to exist within these systems, in addition to three main-stem impoundments: Allatoona Reservoir (49 km2) on the Etowah River, and Carters Reservoir (13 km2) and a reregulation pool (3.5 km2) on the Coosawattee River. An estimated 9,954 road-stream crossings existed in the study area: 3,701 in the Conasauga River system, 2,530 in the Coosawattee River system, and 3,723 in the Etowah River system; however, it was not known how many of these crossings are culverts.
https://static-content.springer.com/image/art%3A10.1007%2Fs00267-012-9886-6/MediaObjects/267_2012_9886_Fig1_HTML.gif
Fig. 1

Study area comprising the Conasauga, Coosawattee, and Etowah river systems in the Upper Coosa River system, northwest Georgia and southeast Tennessee

Table 1

Fish species used to prioritize culvert remediation within priority sub-systems of the Upper Coosa River system

Family common name

Species

Status

Cona

Coos

Etow

Maximum total length (cm)

Minimum watershed size (km2)

Cyprinidae

 Blue shiner

Cyprinella caerulea

FT

+

+

 

9

10

 Etowah chub

Hybopsis sp. cf. H. winchelli

   

+

8.4

0.8

 Coosa chub

Macrhybopsis sp. cf. M. aestivalis

SE

+

+

+

7.6

42

Percidae

 Holiday darter

Etheostoma sp. cf. E. brevirostrum

SE

+

+

+

6.4

4

 Coldwater darter

Etheostoma ditrema

SE

+

  

5.4

0.1

 Etowah darter

Etheostoma etowahae

FE

  

+

6.5

2

 Cherokee darter

Etheostoma scotti

FT

  

+

7.1

0.5

 Trispot darter

Etheostoma trisella

SE

+

+

 

5.9

0.3

 Goldline darter

Percina aurolineata

FT

 

+

 

9

25

 Bridled darter

Percina kusha

SE

+

+

+

7.8

11

Conservation status (Status) refers to federal and state listings: FT federally threatened, FE federally endangered, ST state threatened, and SE state endangered; two FE species (Percina jenkinsi and P. antesella) and two SE species (P. lenticula and Noturus sp. cf. N. munitus) are not listed because minimum watershed sizes for known occurrence (68, 154, 387 and 240 km2, respectively) exceed those of streams typically crossed using culverts. Pluses (+) indicate which priority sub-systems, Conasauga (Cona), Coosawattee (Coos), Etowah (Etow), are within each species distributional range. Maximum lengths are from Page and Burr (2011). Minimum watershed sizes are based on known occurrences

Site Selection for Culvert Assessment

We created a stream drainage network based on a 30-m resolution digital elevation model and the 1:24,000 National Hydrography Dataset (NHD) for the project area. Streams were created using ArcHydro 9 Toolbox with ArcGIS 9.2 (Esri, Redlands, California) and were set to originate at 120 cells (i.e., 0.1 km2 drainages). Using the 1997 1:24,000 Georgia Department of Transportation (GDOT) road coverage, we recorded the intersection of all streams and road crossings. We also mapped locations of impoundments using National Inventory of Dams, NHD, and dams identified in aerial photography of the study area (Appendix). To narrow the pool of sites for field inspections, we omitted all road crossings likely to be bridges based on drainage area over 52 km2 (based on GDOT policy). Additionally, all crossings within 500 m downstream of an impoundment were eliminated from further consideration; dams were considered impassable by fishes, and thus habitat gained by remediating a near-by downstream culvert would be minor. Similarly, because impoundments did not provide suitable habitat for the target fish species, road crossings upstream of an impoundment were eliminated if: (1) the crossing was within 250 meters of the impoundment border (exclusive to impoundment outlines documented within the NHD); (2) the crossing was within 250 meters of any plotted impoundment; or (3) the drainage size above the impoundment was less than 1 km2 (i.e., considered too small to support populations of target fish species).

We calculated the smallest stream size likely to be occupied for each target species based on minimum drainage areas upstream of known occurrences (Table 1). We eliminated from consideration all stream crossings where the upstream drainage area was too small for target species, excepting the coldwater darter, Etheostoma ditrema, which primarily inhabits springs and spring runs, but also occurs in the mainstem Conasauga River. Thus, for sub-basins (12-digit USGS hydrologic units, or HUC12s) containing the coldwater darter, we included crossings between springs identified on USGS topographical maps and the nearest connected stream that had a drainage greater than 52 km2, and any road crossings upstream or downstream of a known occurrence of the coldwater darter.

The included stream crossings were prioritized for field surveys based on conservation priority in the Etowah and Conasauga River systems (Wenger and others 2009; Wenger and others 2010), location relative to impoundments, and drainage area if located above an impoundment. The highest priority areas in the Etowah and Conasauga systems were those that support federally-protected fishes. Stream crossings within these areas were surveyed first; lowest priority sites for field assessment were those nearest to dams and those located in relatively small drainages upstream from an impoundment. A total of 404 road crossings within the priority sub-basins of the Upper Coosa River system were visited between June 7, 2006 and March 12, 2008. Of the road crossings visited, 256 culverts were assessed for potential to impede passage by small-bodied fishes. The sites not assessed: lacked access or there was no road (n = 45); had a bridge, bottomless culvert, ford, or no crossing (n = 76); or were dry, intermittent or impounded streams (n = 23). An additional four sites were omitted from analysis due to incomplete data. Most sampled sites were in the Etowah River system (n = 195), with 48 sites assessed in the Conasauga River system and 13 within the Coosawattee River system.

Culvert Measurements

Surveys were conducted from June 2006 to March 2008 by a team of three or four people, usually at baseflow conditions. Characteristics recorded at each crossing included type (freespan bridge, bottomless culvert, pipe culvert, or box culvert), culvert material (smooth metal, corrugated metal, or concrete), number of culvert openings, and occurrence of a concrete “apron” beneath the culvert. All fords, freespan bridges, and bottomless culverts encountered were assumed passable and omitted from further analysis. For all culverts, the length and diameter (or width) of each opening were measured from inside the culvert, along with the wetted width, the width of deposited sediment (if any), and the distance from the top of the pipe or box to the water surface and to the streambed sediments. If either end of the culvert was perched above the stream, the distances from the pipe or box lip (or apron lip, if applicable) to the water surface and to the streambed sediments were measured. At each outlet, water depth and velocity were measured in the center main flow from the pipe or box using a Marsh-McBirney Inc. FLO-MATETM Model 2000 portable flow meter with a top-setting wading rod at 60 % stream depth.

Stream metrics were also collected upstream, downstream and adjacent to the culvert. Average stream width was measured and the dominant bed sediment type noted upstream and downstream of the culvert. Width and depth of any scour pools adjacent to the culvert were measured, along with maximum stream width. We noted apparent bank erosion or occurrence of sediment deposits near the culvert.

Estimating Culvert Effects on Passage

Field measurements taken at culverts were used to estimate the probability that each culvert impeded passage by small-bodied fishes, using literature-derived criteria and hypotheses. We assumed all species of concern in our study system would be equally impeded by any given culvert, and thus estimated probability of passage for a generalized small-bodied (i.e., having maximum total length typically ≤10 cm; Table 1) fish species.

We assessed passage at culverts using four criteria sets that represented differing hypotheses about how culverts impede fish passage (Table 2; Fig. 2). We used Bayesian belief networks (BBNs) to model probability of passage based on each set of hypothesized relations. BBNs are increasingly applied in natural resource management, as described in recent reviews including Peterson and Evans (2003); Marcot and others (2006) and Stewart-Koster and others (2010). In brief, a BBN models the relations between key drivers (in our case, culvert characteristics) and response variables (e.g., fish passability at a culvert), using conditional probabilities to link drivers to responses. Each causal and response variable is described as a series of discrete states that define possible conditions (e.g., probabilities that water velocity in a culvert is low, moderate, or high) or outcomes (e.g., probability that a culvert is passable or impassable), so that uncertainty in characteristics of model components is explicit (Marcot and others 2001). Two of our modeled criteria sets used culvert attributes that were usually observable during a survey and linked these to certain outcomes, whereas two other sets included estimated culvert parameters and linked these to probabilities of fish passage. The first two criteria sets, based on literature values, were used to define “Wide” and “Narrow” standards in an earlier culvert assessment conducted in a portion of the Upper Coosa system (Millington 2004). The Wide standards specified a wider range of conditions for culvert passability by small fishes, requiring either a baseflow velocity greater than 0.4 m/s (Warren and Pardew 1998) or a drop from the culvert to the water surface greater than 0.15 m (ODFW Oregon Department of Fish and Wildlife (ODFW) 2010) (Table1; Fig. 2a), for a culvert to be considered impassable. When more than one culvert opening was present, the opening with the lowest velocity or smallest drop was used. Under the narrow standards (i.e., narrow with respect to conditions allowing passage; Fig. 2b), any culvert with a baseflow velocity greater than 0.25 m/s (Toepfer and others 1999) or having any drop from the culvert outlet to the water surface at baseflow (QDPIF 2004) was considered impassable. Using these two criteria sets, each culvert was assessed as either “passable” or “impassable” based on field measurements of outlet drop to the water surface and water velocity in the culvert at baseflow. The exceptions were at sites where water level in the culvert was too shallow to permit velocity measurements (n = 66), in which case we used field measurements from Millington (2004) to estimate probabilities that baseflow velocity was moderate (0.25–0.40 m/s) or fast (>0.40 m/s). Millington visited 70 randomly selected stream crossings (52 culverts) in the upper Coosa River system during 2003 and 2004, also during baseflow conditions but during a wetter period than the current study, so that water levels in culverts were sufficient for velocity measurements at all sites. For the 66 culverts in the present study where occurrence of moderate or fast velocities were estimated rather than measured, the binary criteria resulted in passage probabilities that could fall between zero and one.
Table 2

Alternative criteria sets used to assess the probability that culverts impede passage (“mostly impassable”) by small-bodied fishes, listing the probabilities of impassability assigned to different culvert conditions under the two-level and three-level criteria

Criteria Set

Outlet drop (m)

Baseflow velocity (m/s)

High-flow velocity

Mostly impassable probability

Criteria narrative: culvert has some probability of blocking movement by small bodied fishes if:

Wide

>0.15

Any

1.00

High drop from culvert to water surface

 

Any

>0.40

1.00

OR

     

Fast velocity at base-flow

Narrow

>0.00

Any

1.00

Any drop from culvert to water surface

 

Any

>0.25

1.00

OR

     

Moderate velocity at base flow

Two-level

>0.15

Any

Not-fast

0.50

High drop to water surface, especially with evidence of Fast high-flow velocity

 

>0.15

Any

Fast

1.00

Three-level

0.00

0.25-0.40

Fast

0.20

Moderate or Fast velocity at base flow with evidence of Fast high-flow velocity

 

0.00

>0.40

Fast

0.50

OR

 

0.01-0.15

Any

Fast

0.75

Any drop to water surface and evidence of Fast high-flow velocity

 

>0.15

Any

Not-fast

0.50

OR

 

>0.15

Any

Fast

1.00

High drop to water surface, especially with evidence of Fast velocity at high flows

Outlet drop is the distance from the culvert outlet to water surface at base-flow. For combinations of outlet drop, base-flow velocity and high-flow velocity other than those listed for each criteria set, the probability that culverts are impassable is set to zero

https://static-content.springer.com/image/art%3A10.1007%2Fs00267-012-9886-6/MediaObjects/267_2012_9886_Fig2_HTML.gif
Fig. 2

Alternative hypotheses used to assess small-bodied fish passage through culvert road crossings: a “Wide”, and b “Narrow” criteria, from Millington (2004); c Two-level and d Three-level criteria, which incorporate effects of velocity at high flow. Ovals enclosed with broken lines represent unobserved states

Two additional criteria sets (Table 2) were based on the hypothesis that culverts perched above the water surface may allow fish passage during high flows when the drop from the outlet is eliminated, provided that water velocities through the culvert at those higher flows are not too fast for small-bodied fishes to swim against. These criteria sets estimated probability of passage given measured outlet drop to the water surface and water velocity in the culvert at baseflow (as above), and the additional probability that high flows sufficient to submerge the outlet also do not entail fast water velocities. The Two-Level criteria (Fig. 2c) used two binary variables to determine passability (i.e., slow or fast velocity; outlet > or <15 cm above the water surface; Table 2). The three-level criteria (Fig. 2d) used three-factor categorical variables (i.e., slow, moderate, or fast velocity; outlet 0 cm, 1–15 cm, or >15 cm above the water surface; Table 2). Using either the two- or three- level criteria, the BBN calculated conditional probabilities for three possible states at a culvert: “always passable”, “partially passable”, and “mostly impassable”. The middle category was included to represent uncertainty in relations within the network (Marcot and others 2006), however only the “mostly impassable” probability (henceforth “impassability”, which could range from 0 to 1) was used to prioritize culverts for replacement. Because we did not measure water velocity at high flows, the BBN models also included a component that related fast velocity at high flows to three field measurements (Fig. 2c, d): (1) occurrence of scour pools downstream from the culvert, or of mid-stream sediment deposits upstream from the inlet (Bates and others 2003; RSCP 2006; WDFW 2009), (2) deposited sediment within the culvert (Bates and others 2003; WDFW 2009); and (3) the velocity at baseflow (i.e., a culvert with fast baseflow velocity was assumed to have fast high-flow velocity; Table 3).
Table 3

Conditional probabilities for estimating the presence of fast velocity in culverts during high flows, based on observed velocity at baseflow and on occurrence of extensive scour or sediment deposition, or of sediment in the culvert

Velocity conditions

Associated probabilities

 

High flow velocity

Base flow velocity:

Fast

Not-fast

Slow

0.50

0.50

Moderate

0.50

0.50

Fast

1.00

0

 

Extensive scour or sediment deposition

High flow velocity:

True

False

Fast

0.90

0.10

Not-fast

0.30

0.70

 

Sediment in culvert

High flow velocity:

True

False

Fast

0.20

0.80

Not-fast

0.50

0.50

Extensive scour was defined as any scour pool with depth or width twice or more that of the natural stream channel (RSCP 2006). Extensive sediment deposition was defined as increased mid-channel bar deposition adjacent to inlet (WDFW 2009)

We used the software Netica (Norsys Software Corp., Vancouver, British Columbia) to build the BBNs and to evaluate the probability of fish passage at each culvert using hypothesized conditional probabilities. The conditional probabilities are central to a BBN (Marcot and others 2006), specifying outcome probabilities in each node (model component) for all possible combinations of the factors immediately linked to that node. For our two- and three-level criteria, the probabilities of passage conditional on culvert out-flow drop, baseflow velocity or high-flow velocity were not known, nor were the probabilities of downstream scour or upstream deposition, or occurrence of sediment in the culvert, conditional on fast velocity during high-flow conditions. Therefore, we assigned conditional probabilities (Tables 2, 3) using plausible, judgment-based values, and then evaluated sensitivity of model outcomes to changes in those assigned probabilities. Specifically, we defined the combination of no drop from the culvert outlet and evidence of fast velocities at high flow as “possibly a barrier, but unlikely”, with probability of impassability defined as 0.20. Probability a culvert was impassable given either a large outlet drop or evidence of fast velocities at any flow was set to 0.50 (Table 2). We assigned a 0.75 probability of impassability for culverts with intermediate outlet drop in combination with evidence of fast, high-flow velocities (Table 2). One level up in the BBN (Fig. 2c, d), we assigned a 0.90 probability that fast high-flow velocity resulted in scour downstream or deposited sediment upstream from the culvert, and a 0.30 probability of these conditions in the absence of fast velocity at high flow (Table 3). We assumed fast high-flow velocity was 80 % likely to remove sediment from the culvert, whereas sediment was equally likely to occur or not (i.e., 50 %) in the absence of fast velocity at high flow (Table 3).

To evaluate the sensitivity of projected culvert impassability to conditional probability values used in the BBN for the two- and three-level criteria, we conducted a one-way sensitivity analysis (van der Gaag and Coupé 2000; Coupé and van der Gaag 2002). Specifically, we increased and decreased the conditional probabilities underlying conditions of (1) fast velocities at high flow, (2) extensive scour or sediment deposition, (3) occurrence of sediment in the culvert, and (4) the probability of culvert impassability, by increments of 0.10, up to a total change of plus and minus 0.50. The full dataset was re-analyzed to estimate mean culvert impassability for each incremental change in the conditional probability.

Culvert Prioritization

A prioritization score was calculated for each culvert in the database as the total amount of habitat potentially gained by removing or remediating the culvert, summed over all species of concern. We assumed that any stream segment draining the minimum watershed size for a given species could be occupied by that species, provided there was at least one record of occurrence within the encompassing HUC12 (average drainage area = 54 km2). To estimate the potential habitat gain at a given culvert, we first totaled the amount of connected habitat upstream and downstream of the road crossing for each species of concern. Connected habitat was calculated as the summed lengths of potentially occupied stream segments, weighted by culvert-specific passage probability (i.e., upstream passage probability = 1 − probability [impassable]; downstream passage probability assumed = 1) for all culverts in the stream network that limited access to other segments (Fig. 3). Connectivity was treated as bidirectional, measuring the potential for fish to move back and forth between segments separated by culverts (Cote and others 2009; Diebel and others 2010). Thus, fish upstream of a culvert were disconnected from downstream segments to the extent that the culvert impeded return passage upstream. We chose the lesser of connected habitat upstream or downstream as the amount of habitat that would be added to existing if a given culvert were remediated (avoiding inflating apparent habitat gain in cases where a small downstream segment was disconnected from a more extensive upstream segment). Potential habitat gain at that culvert was then estimated as the total connected habitat (i.e., upstream or downstream, whichever was less), minus that same amount weighted by passage probability at the culvert (Fig. 3). Thus, if a culvert was scored as ‘passable’ under binary criteria, there would be no projected habitat gain by replacing it. Otherwise, projected habitat gain increased as the estimated impassability increased at the culvert being considered. Prioritization scores were calculated using each of the four criteria sets for assessing culvert impassability to examine the sensitivity of the net outcome, prioritization, to the choice of passage criteria.
https://static-content.springer.com/image/art%3A10.1007%2Fs00267-012-9886-6/MediaObjects/267_2012_9886_Fig3_HTML.gif
Fig. 3

Computation of potential habitat gain, showing a hypothetical stream system with two culverts (A and B) that differ in probability of being mostly impassable for small-bodied fishes

Results

Culvert Characteristics

The majority (60.9 %) of sampled road crossings were pipe culverts (n = 156). The mean watershed size upstream of pipe culverts was 3.28 km2, with a maximum of 27.5 km2. Ninety-nine (38.7 %) of the sampled road crossings were box culverts, with a mean watershed size of 9.1 km2 and a maximum of 51 km2. On average, pipe culverts were shorter with smaller openings than box culverts (Table 4). Most sampled pipe culverts had a single opening, although five pipe culverts had 4–6 openings. Box culverts modally had two openings, and a maximum of four. Current velocity at baseflow through pipe culverts averaged higher than through boxes (Table 4), with a maximum of 2.71 m/s observed at a pipe. Pipe culverts averaged a lower mean outlet drop (culvert lip to water surface) than box culverts (Table 4), but the highest measured outlet drop (0.91 m) was at a pipe. In total, 69 (44 %) pipe culverts and 48 (48 %) box culverts had outlets elevated above the water surface; inlets were elevated above the water surface at 11 pipe culverts and two box culverts.
Table 4

Summary statistics for sampled culverts in the Upper Coosa River system

 

Pipe

 

Box

 

Min

Mean (SD)

Max

Min

Mean (SD)

Max

No. barrels

1

1.70 (0.93)

6

1

2.27 (1.00)

4

Length (m)

5.13

14.32 (6.51)

40.62

9.13

24.14 (28.47)

152

Wetted width (m)

0.19

0.83 (0.50)

3.22

0.94

2.23 (0.57)

3.17

Sediment width (m)

0.00

0.50 (0.57)

2.75

0.00

1.33 (1.11)

3.27

Top to water surface (m)

0.34

1.34 (0.6)

3.31

0.93

2.15 (0.55)

3.63

Top to bed sediments (m)

0.43

1.47 (0.69)

3.58

1.05

2.1 (0.59)

3.63

Lip to bed sediments (m)

Inlet

0.00

0.04 (0.09)

0.70

0.00

0.13 (0.16)

0.93

Outlet

0.00

0.16 (0.23)

1.54

0.02

0.46 (0.33)

1.45

Lip to water surface (m)

Inlet

0.00

0.00 (0.02)

0.25

0.00

0.00 (0.04)

0.39

Outlet

0.00

0.07 (0.14)

0.91

0.00

0.17 (0.37)

0.75

Velocity (m/s)

-0.18

0.38 (0.39)

2.71

-0.06

0.19 (0.16)

0.58

Depth (m)

0.01

0.11 (0.14)

0.67

0.02

0.10 (0.10)

0.64

Assessment and Prioritization

Depending on the criteria set, between 90 and 202 of the 256 assessed culverts were estimated to be passable (i.e., probability[impassable] = 0) by small-bodied fishes (Table 5). Conversely, the number of culverts considered as barriers also depended on criteria. Applying the binary (i.e., wide or narrow) criteria resulted in three to almost six times more culverts assessed as essentially impassable (i.e., probability[impassable] = 1.0 or > 0.9) than obtained using the two-level and three-level criteria (Table 5). The two-level criteria estimated more culverts to be completely passable than did any of the other criteria (Table 5). Otherwise, impassability ranged from 0.53 to 1.00 under two-level criteria, compared to a broader range of estimates between 0 and 1.00 using the three-level criteria (Table 5). Only nine culverts were assessed as completely impassable by all four criteria sets; all were pipe culverts with outlets elevated above the water surface by 0.17–0.8 m at baseflow, and with baseflow velocities ranging from 0.4–2.7 m/s.
Table 5

Counts of road crossings tabulated by level of impassability by small-bodied fishes, maximum priority score (weighted habitat gain in km), and counts of crossings with priority scores of at least 20, 10 and 5 km, as assessed using each of four criteria sets

 

Criteria set

Wide

Narrow

Two level

Three level

Probability (impassable) = 0

124

90

202

90

Probability (impassable):

>0.00 & <0.30

32

0

0

54

>0.30 & <0.60

0

11

14

45

>0.60 & <0.90

0

0

13

40

Probability (impassable) = 1.0 or >0.9

100

155

27

27

Maximum priority score

77.22

77.22

57.52

57.43

Number of crossings, priority score (km):

≥20

5

6

2

4

≥10

9

12

4

7

≥5

21

26

9

15

All counts are out of 256 total crossings assessed. Use of prior distributions to estimate baseflow velocity (as slow, moderate or fast) at sites missing velocity data resulted in estimated probabilities of impassability >0 and <1 at 32 and 11 culverts for the binary wide and narrow criteria, respectively

Small differences were observed in the scores assigned by culvert type (i.e., pipe or box) within each criteria set. Under the Wide criteria, 40 % of pipe culverts and 41 % of box culverts were considered impassable, compared to 62 % of pipe culverts and 63 % of box culverts under the narrow criteria. For both criteria sets, 46 culverts were deemed impassable because of high baseflow velocity. Forty-four culverts were deemed impassable due to outlet drop (i.e., >15 cm) under the wide criteria, compared to 72 under the narrow criteria (i.e., outlet drop > 0 cm). Ten and eight culverts were assessed as impassable under the wide and narrow standards, respectively, due to both velocity and outlet drop. Most pipe culverts assessed as impassable exceeded the velocity criterion, whereas outlet drop was the primary reason box culverts were deemed impassable. Using the two-level standards, probabilities of “mostly impassable” for pipe culverts and box culverts averaged 0.14 and 0.21, respectively, compared to 0.33 and 0.29 for the three-level standards.

The one-way sensitivity analysis assessed the effects of changing the conditional probabilities used in the BBN by as much as ±0.5 on the mean probability that culverts in the data set were impassable, estimated using the two- and three-level criteria sets. Changing the conditional occurrences of high flow velocity, scour or upstream sediment deposition, and occurrence of bed sediments within the culvert had relatively small effects on the mean probability of culvert impassability. The greatest effect was for the state of fast high-flow velocity, given slow or moderate baseflow (p[Fast HF|Slow/Mod BF]; Fig. 4); increasing the conditional probability from 0.5 to 1.0 changed average probability of culvert impassability from 0.31 to 0.44 (three-level criteria) or 0.16 to 0.21 (two-level criteria; Fig 4).
https://static-content.springer.com/image/art%3A10.1007%2Fs00267-012-9886-6/MediaObjects/267_2012_9886_Fig4_HTML.gif
Fig. 4

Results from one-way sensitivity analysis for effects of changing conditional probabilities used in BBNs on mean probability of impassability (averaged over 256 culverts assessed for the case study; horizontal axis). Sensitivity is estimated using two-level criteria (circles onleft side of chart) and three-level criteria (circles on right side of chart). The change in mean probability of impassability with each incremental change (in steps of 0.10) in conditional probabilities is indicated by bar width, from ± 0.10 (widest) to ± 0.50 (narrowest). For example, the top-most set of figures shows net effects on mean probability of impassability of reducing the [probability of fast velocity at high flow conditional on occurrence of fast velocity at baseflow] from 1.0 to 0.5 in increments of 0.1. Abbreviations are: HF (velocity at high flow), BF (velocity at baseflow), Scour or Dep (bed scour at outlet or sediment deposition at inlet of culvert), High OD (outlet drop to water surface >15 cm), Elev OD (outlet drop >0, <15 cm), No OD (outlet drop = 0 cm), p(Impass) (probability culvert is mostly impassable using two-level [2L] or three-level [3L] criteria)

Average culvert impassability was most sensitive to changes in conditional probabilities of impassibility in the absence of a high outlet-drop and absence of fast velocity at high-flow and baseflow (No/Elev. OD, Not-Fast HF, Slow or Slow/Mod BF; Fig. 4). Incrementally increasing the probability that the culvert was impassable for these conditions from 0 in the current analysis, to as high as 0.5 increased mean culvert impassability from 0.16 to > 0.4 or from 0.31 to 0.47 for two- and three-level criteria, respectively (Fig. 4). Although this represented a more than doubling of the mean probability of impassability under the two-level criteria, the change in conditional probability (from 0 to 0.5) in this case amounted to changing the criteria set (i.e., by removing the requirement for a high outlet drop for a culvert to be considered a barrier to passage; Table 2). Net effects of all other assessed changes in conditional probabilities were substantially smaller (Fig. 4).

Culvert priority scores and rankings for replacement were sensitive to choice of assessment criteria. The highest priority scores based on the two- and three-level criteria were about 25 % lower than those based on the binary criteria (Table 5). This difference occurred because potential habitat gain was weighted by estimated impassability at a culvert, which was usually < 1 using the non-binary criteria. In contrast, impassability was normally 0 or 1 (i.e., except when baseflow velocity conditions were estimated) under the wide and narrow criteria, so that existing habitat was either completely fragmented or completely connected. The two-level criteria set the strictest criteria for impassability, requiring a drop at the culvert outlet of at least 0.15 m to the water surface at baseflow with evidence of fast velocity during high-flow events for a culvert to be considered completely impassable (Table 2). The four culverts ranked highest for replacement under the two-level standard were also highly ranked by the other criteria (Table 6). However, three culverts that had no replacement value under the two-level criteria were ranked as high priorities (i.e., within the top 5) for replacement under the other criteria (Table 6; Fig. 5). Remediating these culverts potentially could provide as much or more habitat, for as many or more target species, as high-priority culverts under the two-level standard. Overall, priority rankings under the wide, narrow and three-level criteria shared seven to eight out of the ten top-ranked structures for remediation. However, none of the nine culverts assessed as completely impassable by all four criteria reconnected sufficient habitat to rank in the top ten of priorities for remediation.
Table 6

Culverts ranking among the top five in priority for remediation under at least one criteria set

 

Culvert description

Priority rank (probability [mostly impassable], priority score)

 

Name

Type (# of outlets)

Outlet drop (m)

Base flow velocity (m/s)

Probability [fast high-flow velocity]

Two level

Three level

Wide standards

Narrow standards

No. species affected (# listed)

PA-06

Box (4)

0.75

*

0.87

1 (0.94, 57.5)

1 (0.94, 57.4)

2 (1.0, 60.8)

2 (1.0, 60.1)

3 (2)

CONC-129

Pipe (4)

0.17

0.26

0.19

2 (0.59, 20.4)

3 (0.59, 20.3)

4 (1.0, 30.4)

4 (1.0, 30.4)

2 (0)

C-47

Box (4)

0.45

0.15

0.83

3 (0.91, 14.1)

6 (0.91, 14.1)

6 (1.0, 15.4)

7 (1.0, 15.4)

1 (1)

D-46

Box (2)

0.56

0.20

0.83

4 (0.91, 11.1)

7 (0.91, 10.8)

7 (1.0, 11.9)

9 (1.0, 11.9)

2 (2)

D-32

Pipe (2)

0.35

1.09

1.00

5 (1.00, 7.3)

9 (1.00, 7.3)

15 (1.0, 7.3)

17 (1.0, 7.3)

3 (2)

CONC-155

Box (3)

0.05

0.44

1.00

51 (0.00, 0)

2 (0.75, 40.4)

1 (1.0, 77.2)

1 (1.0, 77.2)

4 (1)

B-05

Pipe (4)

0.09

0.40

1.00

51 (0.00, 0)

4 (0.75, 20.0)

5 (1.0, 24.8)

5 (1.0, 24.8)

3 (2)

CONC-079

Box (3)

0.00

0.43

1.00

51 (0.00, 0)

5 (0.50, 19.6)

3 (1.0, 37.2)

3 (1.0, 34.8)

4 (1)

Priority rank is the ranked order among all culverts assessed for small-bodied fish passage based on the length of stream habitat potentially reconnected by remediation and summed for special concern species affected (priority score), which is weighted by the probability that the culvert is mostly impassable. The number of special concern species that could gain habitat from culvert remediation is shown for each structure, with the number of those species that are listed as federally protected under the Endangered Species Act (in parentheses)

* Base flow velocity was not measured because stream flow ran under the concrete base of the culvert

https://static-content.springer.com/image/art%3A10.1007%2Fs00267-012-9886-6/MediaObjects/267_2012_9886_Fig5_HTML.gif
Fig. 5

Probability of impassability (shade gradient) and priority score (symbolsize) for culverts assessed in the Upper Coosa River system (n = 256), comparing results using two-level and three-level criteria

Discussion

This study compared assessment outcomes using alternative, plausible criteria to estimate the ability of small-bodied fishes to pass through culvert road crossings, for a dataset of 256 measured culverts in a river system containing multiple species of conservation concern. We found that use of alternative criteria sets could influence culvert rankings for remediation. Using previously developed binary criteria (Millington 2004) as many as 155 culverts (61 % of the total) were identified as probable barriers to small-bodied fishes, because of high baseflow velocity or elevation of the culvert outlet above the water surface at baseflow. Using models that explicitly incorporated uncertainty in effects of baseflow velocity and outlet drop on fish passage (i.e., our two-and three-level criteria), 27 culverts (11 %) were estimated as >90 % likely to be impassable. Out of the 256 culverts assessed, only nine culverts were identified as complete barriers (i.e., probability of impassability = 1) by all four criteria sets used, but relatively little habitat would be reconnected if these structures were to be replaced; thus, they did not rank highly in replacement prioritization. Based on the priority strategy used in this study, up to 77 km of stream habitat summed for priority species could be reconnected through replacement of a single culvert (CONC-155) in the study area. However, assessing which culverts would provide the most benefit (or even any benefit) depended on criteria used to assess impassability.

The four criteria sets used to assess culverts as barriers in this study reflect alternative ideas about how culverts may block passage by small-bodied fishes. Field studies of fish movements through culverts support the assertion that culverts perched above the bed sediments, or with strong velocity at baseflow or during high flow events may impede movement by small-bodied fishes (e.g., Toepfer and others 1999; Warren and Pardew 1998; Coffman 2005; Benton and others 2008; Norman and others 2009). However, even culverts with outlets elevated above the water surface at low flows may not form complete barriers to upstream movement by small fishes. For example, Norman and others (2009) report upstream movement by small-bodied minnows (Cyprinidae) through culverts with outlet drops at baseflows, but only during intervals including higher flows that eliminated the drops. Coffman (2005) similarly reports infrequent movements by minnows and also sculpin (Cottidae) through culverts that had been classified as impassable on the basis of outlet elevation or slope (and hence velocity), possibly facilitated by higher flows. Thus, binary classification of culverts as either passable or impassable is likely an over-simplification for many species, because passage may be possible during occasional periods when flow levels provide sufficient depth to submerge an elevated outlet without creating velocities that are too great for small fishes to swim against.

The two-level and three-level criteria evaluated in this study allowed some possibility of movement through culverts that would have been assessed as impassable under the binary (i.e., wide and narrow) criteria. Constructing the BBNs to assess impassability using the two- and three-level criteria required that we assign conditional probabilities relating observations of scour or sediment deposition to occurrence of high velocities at high flows, and fish passage to conditions of outlet drop and current velocities. However, estimated impassability on average was not greatly altered by changing those assigned conditional probabilities except when the changes amounted to changing from one criteria set to another. In contrast, our analysis was sensitive to the choice between the two-level and three-level criteria, in that different culverts ranked highly for remediation under the two criteria sets (Fig. 5). These criteria primarily differed in whether a culvert with an outlet drop < 15 cm either could (three-level) or could not (two-level) form a passage barrier. If moderate or high velocities were believed to impede fish passage to some extent even in the absence of an outlet drop, then it would be more appropriate to apply criteria like the three-level set. Although the three-level criteria provided similar estimates of weighted potential habitat gain as the two-level criteria, some structures with outlet drops < 15 cm ranked highly for remediation under the former criteria, but had no remediation value under the latter.

A primary lesson from this effort is the importance of making priorities on the basis of transparent and explicit criteria and management objectives. Selection of thresholds of culvert elevation or current velocity to assess a culvert as likely impassable clearly affects outcomes of the prioritization process; however, there is considerable uncertainty about how culverts impede passage by aquatic organisms. The ability of fish to pass through road crossings is generally assumed to depend on species-specific swimming and leaping abilities, and the timing of the movement relative to flow conditions (Coffman 2005; Bourne and others 2011). Although any plausible criteria may place high priority on remediating structures that provide little possibility of passage even to strong swimmers and that fragment large areas of habitat (Bourne and others 2011), some criteria may fail to identify opportunities to reconnect substantive habitat areas for smaller species. For the case study presented here, the species of special concern span adult sizes of about 5 to 10 cm. Criteria that allow some probability that an embedded culvert with moderate velocities may impede passage (such as the three-level criteria) may be more appropriate for smaller or benthic species, whereas requiring an outlet drop to consider a culvert a barrier could be more appropriate in reconnecting habitat for larger, stronger swimmers (Coffman 2005). Thus, the choice of criteria for assessing culvert impassability may reasonably differ depending on management objectives, e.g., increasing overall network connectivity (Cote and others 2009; Diebel and others 2010) or reconnecting habitat for specific species (as in the Upper Coosa River example). The swimming ability of non-game and small-bodied fishes has received less attention than game fishes, and estimates may vary depending on the experimental design of the study (Coffman 2005). Validating criteria for impassability for stream fishes and other aquatic organisms will be essential and may require, e.g., long-term assessments of movements (Norman and others 2009) or analyses of genetic variation among putatively fragmented populations (Neville and others 2006) in relation to stream culverts across a range of outlet drops and velocities to test hypotheses about what structures substantially fragment populations (Kemp and O’Hanley 2010). In the interim, managers who need information on actions most likely to benefit imperiled species will necessarily use models based on uncertain relations. For that reason, it is useful from the outset to understand the sensitivity of prioritization schemes to underlying assumptions.

The approach illustrated in this study was designed to allow managers to incorporate new information pertaining to model assumptions as it becomes available. For example, we have assumed that all streams above a minimum threshold size within an occupied HUC12 have the same probability of being occupied. Although watershed size has been shown to account for much of the variance within species distributions, other parameters including land use characteristics (Wenger and others 2008), downstream link (Osborne and Wiley 1992), and stream slope (Walters and others 2003) can be equally, if not more predictive. Several of the species of concern from this study have been shown to have distributions that are influenced by land use characteristics (e.g., impervious area, forest cover; Wenger and others 2008); the prioritization strategy used here could incorporate these characteristics, e.g., by using land use or modeled habitat quality to weight potentially reconnected stream kilometers according to habitat value. Additionally, our prioritization strategy assumes that culverts that block access to larger areas for these species have not been over-looked; our sampling approach aimed to minimize this possibility, but visiting each of the road crossings (n > 9900, including bridges) in the project area was beyond the scope of this project. Obviously, new information on species ranges, occurrence data for newly considered species, or data on additional culverts could be incorporated directly into the spatial database to modify prioritizations.

Our prioritization strategy did not account for differential costs among culverts for replacement or remediation, or other potential benefits (e.g., improving public safety or lowering maintenance costs by replacing problematic structures). Similarly, for illustration, we have weighted all species of concern equally in calculating culvert priority ranks, whereas resource managers may wish to weight species differently, depending on conservation objectives or degree of imperilment. Where spread of non-native species are of concern, maintaining culverts that block passage could become a priority (Fausch and others 2009). Other considerations including costs of remediation (O’Hanley and Tomberlin 2005), or age of the structure could be directly incorporated into prioritizations. Finally, whereas we have only ranked culverts by remediation benefit, the approach of estimating habitat gain by incorporating culvert position relative to upstream and downstream barriers could be used to optimize culvert remediation (O’Hanley and Tomberlin 2005; Kemp and O’Hanley 2010), incorporating decisions regarding species priorities and variation in resource availability or opportunities afforded by partners.

The prioritization scheme developed here for culvert remediation may provide a useful starting point for incorporating habitat connectivity for particular species into decisions for allocating limited road maintenance and conservation dollars. Our case study suggests that research on the extent to which small drops and high water velocity actually fragment populations (e.g., using genetic or movement studies) could narrow uncertainty that affects management choices for culvert remediation. Until such information becomes available, assessments that explicitly incorporate uncertainty in how culvert characteristics relate to impassability, as in the case of the BBNs illustrated here, could help managers more realistically estimate potential habitat gain by accounting for potential occasional passage at less restrictive barriers while also identifying severe barriers. Managers weighing other considerations, such as stakeholder support and project costs, or needs to improve habitat for particular critically imperiled species, may wish to compare degree of uncertainty in culvert impassability among restoration options and for differing criteria sets. Given the conservation emphasis on restoring connectivity for stream organisms and the accompanying uncertainty in identifying culverts as impassable, prioritization approaches that explicitly examine effects of uncertainty on expected outcomes may support better-informed management choices in addition to identifying research that could have the greatest effect in narrowing management uncertainty.

Acknowledgments

Many thanks to those who helped conduct field sampling: Christina Baker, Bill Bouthiller, Jeffery Garnett, Jason Hunt, Rachel Katz, Jason Lang, Sam Miles, Amanda Neese, James Norman, Nicole Pontzer, and Randy Singer. We also thank Brett Albanese and Katie Owens for supplying us with an impoundment layer for the Coosawattee River system. Brenda Rashleigh, William Fisher, Joe Anderson, Frank Dirrigl, Jr. and an anonymous referee provided insightful comments and helpful suggestions on an earlier version of the manuscript. Funding for this research was provided by a grant from the U.S. Fish and Wildlife Service. Use of trade, product, or firm names does not imply endorsement by the U.S. Government. The Georgia Cooperative Fish and Wildlife Research Unit is sponsored by the U.S. Geological Survey, the U.S. Fish and Wildlife Service, the Georgia Department of Natural Resources, the University of Georgia, and the Wildlife Management Institute.

Copyright information

© Springer Science+Business Media, LLC (outside the USA) 2012