Synergistic and singular effects of river discharge and lunar illumination on dam passage of upstream migrant yellow-phase American eels

Abstract

Monitoring of dam passage can be useful for management and conservation assessments of American eel, particularly if passage counts can be examined over multiple years. During a 7-year study (2007–2013) of upstream migration of American eels within the lower Shenandoah River (Potomac River drainage), we counted and measured American eels at the Millville Dam eel pass, where annual study periods were determined by the timing of the eel pass installation during spring or summer and removal during fall. Daily American eel counts were analysed with negative binomial regression models, with and without a year (YR) effect, and with the following time-varying environmental covariates: river discharge of the Shenandoah River at Millville (RDM) and of the Potomac River at Point of Rocks, lunar illumination (LI), water temperature, and cloud cover. A total of 17 161 yellow-phase American eels used the pass during the seven annual periods, and length measurements were obtained from 9213 individuals (mean = 294 mm TL, s.e. = 0.49, range 183–594 mm). Data on passage counts of American eels supported an additive-effects model (YR + LI + RDM) where parameter estimates were positive for river discharge (β = 7.3, s.e. = 0.01) and negative for LI (β = −1.9, s.e. = 0.34). Interestingly, RDM and LI acted synergistically and singularly as correlates of upstream migration of American eels, but the highest daily counts and multiple-day passage events were associated with increased RDM. Annual installation of the eel pass during late spring or summer prevented an early spring assessment, a period with higher RDM relative to those values obtained during sampling periods. Because increases in river discharge are climatically controlled events, upstream migration events of American eels within the Potomac River drainage are likely linked to the influence of climate variability on flow regime.

Introduction

Fish passes provide upstream dam passage for migratory fish, and allow for monitoring the number of individuals moving upstream, as well as environmental variables associated with upstream migration (Schilt, 2007; Roscoe and Hinch, 2010; Noonan et al., 2012; Welsh and Liller, 2013). For migratory American eel, monitoring of eel passes can be useful for management and conservation assessments, particularly if passage counts can be examined over multiple years (Marcogliese and Casselman, 2009). Researchers have examined the association of upstream migration of yellow-phase (immature) American eels with several environmental variables, including river discharge, lunar illumination (LI), and water temperature (WT; Jessop, 2003; McGrath et al., 2003; Verdon et al., 2003; Schmidt et al., 2009; Welsh and Liller, 2013). Also, cloud cover (CC) may be associated with upstream migration, given the nocturnal-nature and avoidance of light by anguillid eels (Lowe, 1952; Hain, 1975; Cairns and Hooley, 2003; Schmidt et al., 2009; Welsh and Liller, 2013).

The annual initiation and cessation of American eel upstream migration is controlled, in part, by seasonal change in WT. Walsh et al. (1983) found that immature American eels become torpid at WTs below 10°C. WTs between 10 and 16°C are associated with upstream movements of American eel elvers during spring (Smith and Saunders, 1955; Sorensen and Bianchini, 1986; Jessop, 2003; Schmidt et al., 2009). Hammond and Welsh (2009) found that the initiation of upstream movements of yellow-phase American eels during spring corresponded to WTs of 15°C in the Shenandoah River, USA. Unpublished data from other river systems suggest that yellow-phase American eels may initiate upstream movements during spring at WTs of 17–20°C (K. McGrath, pers. comm.). McGrath et al. (2003) and Hammond and Welsh (2009) reported that fall movements of yellow-phase American eel were reduced as WTs declined to 10°C. Thus, the annual onset and end of yellow-phase American eel movements likely coincide with spring and fall WTs of ∼15–20 and 10°C, respectively.

Bianchini et al. (1982), as cited in Oliveira (1997), suggested that short-term home ranges of American eels are punctuated by long-range movements. The environmental cues or correlates of punctuated upstream migration are not fully understood and are open for debate. Schmidt et al. (2009) and Welsh and Liller (2013) reported that upstream migration of yellow-phase American eels were positively associated with river discharge and negatively associated with LI. These short-term studies (≤2 years), however, did not address temporal variation in passage counts and associated environmental covariates. From a 3-year study on the St Lawrence River, Verdon et al. (2003) did not find a direct correlation between passage counts of American eels with river discharge or LI. Further research is needed to increase our understanding of associations among environmental variables and upstream migration of yellow-phase American eels.

During a 7-year study (2007–2013), conducted at the Millville Dam eel pass on the lower Shenandoah River (Potomac River watershed), we examined the association between daily passage counts of migrant yellow-phase American eels and environmental variables of river discharge, LI, WT, and CC. The Millville Dam eel pass is installed and removed seasonally (Welsh and Liller, 2013), and the timing of pass operation relative to the range of WTs conducive to American eel movements should be considered for maximizing American eel passage opportunities. We addressed the following two questions: (i) Are river discharge, LI, WT, and CC associated with daily counts of upstream migrant American eels?, and (ii) What is the relative importance of tributary and mainstem river discharges to passage counts at an eel pass? We also examined the timing of annual installation and removal of the eel pass, and its potential influence on annual passage counts of American eels.

Methods

Study site

The Millville Dam eel pass, located at a hydroelectric dam on the lower Shenandoah River, is ∼285 km upstream from the mouth of the Potomac River estuary (Figures 1 and 2). The low head (5 m height) Millville Dam is ∼296 m in width. The veil of water spillover on this run-of-the-river dam is maintained at 2.54 cm during low river discharges, and increases during periods of higher river discharge. Millville Dam and the design of the eel pass were previously described by Welsh and Liller (2013), and monitoring involves daily passage counts and length measurements of American eels by physical or photogrammetric methods (Welsh and Aldinger, 2014). Millville Dam is not a complete barrier to upstream passage, as evidenced by the pre-pass presence of American eels within the upper Shenandoah River watershed (Jenkins and Burkhead, 1994; Smogor et al., 1995; Goodwin and Angermeier, 2003). Welsh and Liller (2013), however, demonstrated that the eel pass at Millville Dam, during its first 2 years of operation (2004–2005), was used by upstream migrants during spring, summer, and fall periods. Personnel of the hydroelectric power company removed the eel pass each year during fall to avoid pass damage caused by debris during high river discharges of late fall and winter. The eel pass is installed each spring or summer, after the period of expected high river discharges during March and April.

Figure 1.

Map of the Potomac River drainage and locations of the Millville Dam eel pass and associated gauging stations.

Figure 2.

Photograph of the Millville Dam eel pass on the lower Shenandoah River, Millville, WV, USA. This figure is available in black and white in print and in colour at ICES Journal of Marine Science online.

Data collection

Passage counts and total lengths (mm) of American eels were recorded and measured physically or by photogrammetry during 2007 through 2013. Digital photographs from an infrared triggered camera located on the eel pass were used to obtain photogrammetric measurements of American eel lengths as described by Welsh and Aldinger (2014). Each annual sampling period was determined by dates of removal and re-installation of the eel pass at the discretion of the hydroelectric power company. During each year, we obtained daily measures of five environmental variables: river discharge (m³ s⁻¹) of the Shenandoah River at Millville Dam (RDM) and of the Potomac River at Point of Rocks, Maryland (RDP), an LI index, WT at Millville Dam (°C), and a measure of % CC. Daily estimates of RDM were obtained from the US Geological Survey gauge located ∼400 m downstream (western river bank) of the Millville Dam tailwater and hydro-canal tail race (Figure 1). Additionally, we lagged RDM by 1 and 2 d (RDM1 and RDM2), because American eels longitudinally distributed downstream of the eel pass may experience delayed movement in response to increased river discharge. Also, daily estimates of river discharge (RDP) were obtained for the RDP, located 28.5 km downstream of Millville Dam and 19.5 km downstream of the confluence of Shenandoah and Potomac rivers (Figure 1). A daily lunar index of the illuminated percentage of the moon face, which ranges from 0 (new moon) to 1 (full moon), was calculated from an astronomical algorithm (Meeus, 1991). Daily measures of WT were logged at the Millville Dam eel pass (StowAway Tidbit Temp Logger, Onset Computer Corporation). CC was obtained from the Eastern West Virginia Regional Airport (Martinsburgh, WV, ∼22 km NW of Millville Dam), where METAR (Meteorological Aerodrome Report) data represented night-time CC as “oktas” (cloud coverage in percentages as eighths of the sky).

Data analyses

We analysed daily American eel passage counts by fitting negative binomial regression models with time-varying environmental covariates (Liang and Zeger, 1986; Cui and Feng, 2009). Specifically, generalized estimating equations (GEE) assuming a negative binomial distribution and a log link function (SAS 9.2, PROC GENMOD, SAS, 2008) and QIC model selection (Hilbe, 2011) were used following methods of Welsh and Liller (2013). Annual time-series data for the seven annual periods of American eel counts were analysed together with and without a year (YR) effect. A GEE approach requires clusters within a time-series, where daily counts are expected to be serially correlated within a cluster period and independent among cluster periods (Hilbe, 2011). Given the cyclical nature of lunar data, we were concerned that the alignment of clusters within the time-series may influence selection of lunar models, particularly if the cluster periods were not centred on new and full moon periods. To address this concern and to determine the appropriate cluster period, we used the global model structure and fit models with clustered time-series data ranging from 3 to 10 d clusters, where cluster period models were included with clusters centred and not centred on new and full moon phases. The best cluster model was selected with the Quasi-likelihood information criterion (QIC_u) of Pan (2001). To address serial correlation for each time-series, we selected a working correlation structure by fitting models with autoregressive AR(1), compound symmetry, and independent working correlation matrices (Hardin and Hilbe, 2003) and used the correlation information criterion (CIC) to select a working correlation structure (Hin and Wang, 2009; Hilbe, 2011).

Using GEE, the model-selected cluster period, and the model-selected working correlation structure, a set of 46 candidate models were fit to the time-series of daily American eel passage counts. The first candidate model fitted an intercept to the time-series data. The next set of 14 models included seven single covariates with and without a YR effect of river discharge (RDM, RDM1, RDM2, and RDP), LI, WT, and CC. The remaining candidate models included additive-effects models of the seven covariates with and without YR effect (Appendix). Using an information-theoretic approach, the best model (or suite of competing models) was selected with the Quasi-likelihood Information Criterion (QIC_u) of Pan (2001). We also estimated QIC distances among models (ΔQIC_u) and QIC_u model weights (w_i) following methods of Burnham and Anderson (2002). Models, which represented alternative hypotheses, were considered to be supported by the data if ΔQIC_u values were <2.0 (Burnham and Anderson, 2002).

To aid interpretation of model selection results, we plotted annual time-series of daily passage counts, RDM, RDP, and LI. We also calculated Pearson correlation coefficients to assess correlation between environmental variables (RDM, RDP, LI, WT, and CC). The mean values of RDM, RDP, LI, WT, and CC were calculated and compared for days without passage vs. the mean values of those variables when American eels passed through the Millville Dam eel pass. In addition to examining RDM as cubic metre per second, we also calculated the per cent increase in RDM as the ratio of the largest RDM during a multiple-day passage event to the lowest RDM value within a 5-d period before the passage event. The per cent increase in RDM was examined for multiple-day passage events where >100 American eels used the eel pass.

To examine the relative importance of tributary vs. mainstem river discharges to passage counts, we calculated a daily proportion of the two river discharge values (i.e. RDM/RDP) and plotted these proportions vs. daily passage counts of American eels. The plot of RDM/RDP vs. daily passage counts provided a visual assessment of whether passage counts were higher on days when the Shenandoah River contributed a larger proportion of the river discharge of the lower Potomac River. We were also interested in understanding the timing of annual installation and removal of the eel pass, and its potential influence on annual passage counts of American eels. Thus, we calculated and compared mean values of RDM and WT for the annual sampling periods of eel pass operation with those values for the period from which WTs reached and remained above 15°C until the date of eel pass installation.

Results

A total of 17 161 American eels used the eel pass during the 7-year period, and annual counts ranged from 852 to 5394 (Table 1). Total lengths (mm, TL) were measured for 9215 individuals, with mean ± s.e. TL of 294 ± 0.49 mm, and range 183–594 mm. Annual sample days (when the eel pass was operational) ranged from 113 to 188 d, and were determined by the annual schedule of the electrical power company's dates of installation and removal of the eel pass. Daily passage counts of American eels fluctuated widely and were clumped in time (Figure 3). The number of days with American eel passage ranged from 38 of 181 sample days during 2007 to 91 of 113 sample days during 2013 (Table 1).

Table 1.

Annual passage counts of upstream migrant yellow-phase American eels at the Millville Dam eel pass, Shenandoah River, including sampling period duration and summary statistics of five environmental variables (mean daily river discharge from the USGS gauging stations on the lower RDM, WV, USA, and on the Potomac River at Point of Rocks, MD, USA, mean daily WT of the RDM, and mean daily values of % CC near Millville Dam).

Year	Sample period	Days	Days with eel passage	Passage count	Shenandoah River discharge (m3 s−1)		Potomac River discharge (m3 s−1)		Lunar cycles	Water temperature °C		Cloud cover (%)
					Mean ± s.e.	Range	Mean ± s.e.	Range		Mean ± s.e.	Range	Mean ± s.e.
2007	10 May–6 November	181	38	852	28.1 ± 1.0	13.2–68.0	76.4 ± 3.0	29.2–222.0	6.2	23.7 ± 0.34	8.5–30.4	26.2 ± 2.1
2008	6 June–6 November	154	58	1616	24.8 ± 0.91	11.6–68.5	98.0 ± 5.3	44.7–532.4	5.2	23.0 ± 0.44	9.6–29.1	29.2 ± 2.3
2009	22 June–9 November	141	46	1311	21.6 ± 1.1	11.5–107.3	85.9 ± 4.6	39.1–376.6	4.8	21.4 ± 0.45	8.6–28.3	36.7 ± 2.6
2010	6 May–9 November	188	60	5394	26.4 ± 1.4	10.6–174.4	99.2 ± 6.8	28.1–668.3	6.4	22.7 ± 0.44	6.5–30.5	29.7 ± 2.3
2011	28 June–7 November	133	69	1255	34.0 ± 1.6	16.1–121.2	165.9 ± 12.6	46.4–971.3	4.5	23.0 ± 0.64	5.6–32.4	36.9 ± 2.8
2012	9 May–12 November	188	90	4263	39.8 ± 4.3	15.1–750.4	162.6 ± 18.2	43–2579.7	6.4	22.8 ± 0.48	3.8–32.0	34.3 ± 2.3
2013	1 July–21 October	113	91	2470	44.5 ± 4.1	18.6–390.8	119.0 ± 7.7	39.6–509.7	3.8	24.4 ± 0.28	12.4–29.5	34.5 ± 2.8

Figure 3.

Daily counts of American eels at the Millville Dam eel pass and associated environmental variables for seven annual sampling periods from 2007 to 2013. The mean daily river discharge values (m³ s⁻¹) are from the Millville gauging station, WV, USA. Daily values of LI range from 0 to 100%. Shaded area represents the period of eel pass operation.

Environmental variables fluctuated widely within and among the seven study periods. The mean river discharge during the seven study periods ranged from 21.6 to 44.5 m³ s⁻¹ for the RDM, WV, USA, and from 76.4 to 165.9 m³ s⁻¹ for the RDP (Table 1). The seven study periods included from 3.8 to 6.4 lunar cycles (Table 1). Daily WTs ranged from 3.8 to 32.4°C, with annual mean values ranging from 21.4 to 24.4°C (Table 1). Daily CC ranged from 0 to 100% with annual mean values from 26.2 to 36.9% (Table 1).

The GEE analyses, using CIC-selected AR(1) working correlation structure and QIC_u-selected 7-d cluster periods (aligned and centred with lunar cycles), supported a single additive-effects model of YR + RDM + LI (QIC_u weight = 1.0) for the 7-year dataseries. The QIC_u-selected model had a positive coefficient for RDM (β = 7.3, s.e. = 0.01) and a negative coefficient for LI (β =− 1.9, s.e. = 0.34). Thus, passage counts were associated with elevated RDM and low levels of LI. Models with variables, RDP, WT, CC, and lag effects on RDM were not supported by the data. All models and associated model selection statistics are listed in Appendix.

The QIC_u-selected model of YR + RDM + LI was corroborated by graphing annual time-series data and by descriptive statistics. Graphs of annual time-series data depicted passage events clumped in time and often in association with increased RDM or low LI (Figure 3). The QIC_u-selected model was also corroborated by descriptive statistics of the mean values for days without passage vs. mean values of those variables when American eels passed through the Millville Dam eel pass (Figure 4). For each annual sampling period, the mean values of RDM when American eels passed through the Millville Dam eel pass exceeded that of days without passage (Figure 4). Also, annual mean values of LI when American eels passed through the Millville Dam eel pass were lower than that of days without passage (Figure 4).

Figure 4.

Mean values of five environmental variables for days without passage vs. mean values of those variables when American eels passed through the Millville Dam eel pass. Error bars are 95% confidence intervals. Annual means of daily river discharge values (m³ s⁻¹) are from the lower Shenandoah River at Millville, WV, USA (a), and the Potomac River at Point of Rocks, MD, USA (b). Also included are annual means of daily values of % LI (c), mean daily WT of the RDM (d), and daily values of % CC near Millville Dam (e).

The QIC_u-selected model was also supported by examining LI and the per cent increase in RDM for multiple-day passage events where >100 American eels used the eel pass. A total of 28, 2–13 d passage events occurred with >100 individuals, and 23 of these events were associated with either an increase in RDM (10 events, RDM ≥ 1.4 times that of the period before the increased flow event), low LI (6 events, LI ≤ 0.17), or both increased RDM and low LI (7 events, RDM increase of ≥1.6 times that of the previous period, and LI ≤ 0.16; Table 2). Passage events with >250 American eels were associated with periods of increased RDM or periods of both increased RDM and low LI, whereas passage events associated with low LI without increased RDM were <250 individuals (Table 2).

Table 2.

Summary of associations of river discharge and LI with upstream migration of American eels for passage events when >100 individuals used the Millville Dam eel pass.

Dates of passage	Passage count	LI	% increase RDM
River discharge associated with passage events
2–8 June 2007	280	0.83	1.4
23 May–4 June 2010	2041	0.83	1.9
7–10 June 2010	453	0.29	2.4
16–18 July 2010	167	0.37	1.9
2–6 October 2010	1427	0.39	9.5
15–22 October 2011	148	0.85	4.3
24–30 July 2012	173	0.71	2.2
19–27 September 2012	1896	0.32	4.7
13–20 July 2013	1680	0.3	10.4
10–18 October 2013	269	0.53	5.7
LI associated with passage events
13–16 June 2010	108	0.04	0.86
8–13 July 2010	241	0.04	0.98
28 July–2 August 2011	102	0	1
24–27 May 2012	182	0.04	1.2
6–10 July 2013	116	0	1
27 August–4 September 2013	203	0.17	1.1
River discharge and LI associated with passage events
6–12 June 2008	832	0.16	1.6
29 August–3 September 2008	144	0	5.3
29 September–3 October 2008	410	0.04	3.7
18–26 August 2009	367	0.01	1.7
26–30 September 2011	261	0.01	2.7
17–22 May 2012	1563	0.07	4
18–23 August 2012	132	0.03	2.1
Ladder installation associated with passage events
21–22 May 2007	297	0.24	1.1
24–29 June 2009	801	0.16	0.7
6–15 May 2010	778	0.5	0.95
2–14 July 2011	291	0.72	1
Culvert and bridge construction associated with passage event
15–24 September 2011	381	0.87	0.98

The value of LI is given for the day during the passage event with the highest count of American eels. The per cent increase in river discharge of the Shenandoah River at Millville, WV, USA, was calculated as the ratio of the largest RDM during the passage event to the lowest RDM value within a 5-d period before the passage event.

A total of five passage events of >100 individuals, however, were not associated with either an increase in RDM or low LI. Four of these passage events occurred within 1–10 d after the date of eel pass installation during 2007, 2009, 2010, and 2011, which was interpreted as delayed passage of upstream migrants (Table 2, Figure 3). Passage events of >100 individuals also occurred within 1–15 d after eel pass installation during 2008, 2012, and 2013, but these events were also associated with increased RDM or low LI. Also, one passage event of >100 individuals occurred during a period of bridge construction, where a causeway spanning the Shenandoah River (∼250 m downstream of Millville Dam) may have delayed upstream passage (Table 2). This 10-d passage event occurred during a decreasing period of RDM (39.6–24.6 m³ s⁻¹), but was 5–14 d after a high flow event (121.2 m³ s⁻¹) where RDM increased 6.9 times that of the previous level.

Models with WT and CC were not supported by the data based on negative binomial regression analysis. This finding was corroborated by descriptive statistics of the mean values of covariates for days without passage vs. mean values of those variables when American eels passed through the Millville Dam eel pass (Figure 4). Most (90.0%) American eels used the eel pass during 2007–2013 when WT was between 18 and 28°C, but WT was between 18 and 28°C on 63.3% of the sampling days. The late spring or summer installation of the eel pass, however, did not allow for full evaluation of American eel migration at colder temperatures at the expected onset of upstream migration. For the 17 161 American eels, the mean ± s.e. of WT on the day of passage was 23.1 ± 0.03°C, with a range of 11.2–31.5°C. A total of 65.7% of individuals passed when daily CC values ranged from 0 to 25%, which is contrary to the expectation that American eels counts would be higher during periods of increased CC. However, CC was ≤25% on most (59.0%) of the sampling days during 2007–2013.

The Potomac River discharge (RDP) included flows from the Shenandoah River (RDM), and RDP and RDM were correlated (r = 0.82) as expected (all other variable correlations were ≤0.20). The magnitude of increase in RDP, however, was not always consistent with that of RDM, and passage count data did not support models with RDP over those with RDM. The 3 years with highest mean values of river discharge for the Shenandoah and Potomac rivers were 2011–2013 (34.0, 39.8, and 44.5 vs. 165.9, 162.6, and 119.0 m³ s⁻¹, respectively, Table 1). Based on annual mean values of river discharge, RDP ranged from 2.7 times that of RDM in 2007 and 2013 to 4.9 times that of RDM in 2011 (Table 1). Based on daily maximum values of river discharge, RDP ranged from 1.3 times that of RDM in 2013 to 8.0 times that of RDM in 2011 (Table 1). A plot of the proportion RDM/RDP vs. daily passage counts depicted that the 10 highest daily counts of American eels (range 373–716 individuals) occurred when RDP was 1.3–3.8 times that of RDM, where the Shenandoah River contributed from 26.3 to 76.7% of the Potomac River discharge (Figure 5).

Figure 5.

Plot of daily passage counts of American eels at the Millville Dam eel pass from 2007 to 2013 vs. the ratio of daily river discharge values of the Potomac River at Point of Rocks, MD, USA, to that of the Shenandoah River at Millville, WV, USA. For example, a daily discharge value of 200 m³ s⁻¹ for the Potomac River and 50 m³ s⁻¹ for the Shenandoah River would equal a ratio of 200/50 or 4, which we interpret as a 25% discharge contribution of the Shenandoah River to the discharge of the Potomac River.

The timing and duration of annual sample periods differed, where eel pass installation dates occurred in May or June, and removal dates ranged from October to November (Table 1, Figure 3). The timing of annual eel pass installation did not correspond closely with the timing of WTs reaching and remaining above 15°C, when American eels were expected to begin moving upstream (Figure 3). Eel pass removal during fall, however, was closely associated with the date that WT reached 10°C (Figure 3). Dates of eel pass installation for the seven annual periods ranged from 17 to 63 d after WTs reached and remained above 15°C (Table 3). The mean river discharge values of the Shenandoah River during these periods before eel pass installation were >2.0 times that of the study periods during 2007, 2008, 2009, and 2011 when annual passage counts were relatively low (852–1255 individuals), and <2.0 times that of the 2010, 2012, and 2013 study periods when the highest annual counts were observed (2470–5394 individuals; Tables 1 and 3).

Table 3.

Annual summary statistics of mean daily river discharge and mean daily WT for the period from which WTs reached and remained above 15°C until the date of eel pass installation.

Year	Sample period	Days	Shenandoah River discharge (m3 s−1)		Water temperature °C
			Mean ± s.e.	Range	Mean ± s.e.	Range
2007	22 April–9 May	17	100.2 ± 5.0	70.0–147.2	17.5 ± 0.27	15.9–19.8
2008	18 April–5 June	47	137.1 ± 12.3	43.0–430.4	18.1 ± 0.37	13.8–23.9
2009	25 April–21 June	56	128.6 ± 10.3	62.6–438.9	19.3 ± 0.33	14.0–23.6
2010	4 April–5 May	31	84.3 ± 6.4	48.4–189.7	16.9 ± 0.39	15.1–22.5
2011	24 April–27 June	63	158.6 ± 22.1	43.9–1138.3	21.4 ± 0.44	15.4–26.9
2012	14 April–8 May	24	52.8 ± 2.0	40.5–78.7	17.8 ± 0.49	15.1–22.6
2013	9 April–5 May	27	93.2 ± 4.9	52.7–143.0	17.2 ± 0.31	15.3–23.6

Discussion

This study provides several findings relevant to upstream migration and dam passage of yellow-phase American eels. First, the QIC-selected model and supporting descriptive statistics provided evidence that increased Shenandoah River discharge (RDM) and darker periods near a new moon (low LI) were synergistically or singularly associated with passage counts of upstream migrants. Second, data did not support river discharge of the Potomac River (RDP) over RDM, suggesting that RDM on the day of passage had a larger influence on migration and dam passage than that of the Potomac River. Also, 1 and 2 d lags of RDM were not supported by the data, indicating that upstream migrants did not have a delayed response to increased river discharge. Third, no model selection support was given for WT or CC as correlates of upstream migration. Finally, highest daily counts of American eel passage occurred when RDM was contributing from 26.3 to 76.7% of the Potomac River discharge, suggesting that the Shenandoah River is more likely to experience mass migration events of American eels when river discharge is high relative to that of the Potomac River. We recognize, however, that our model inferences and supporting descriptive statistics are based on observational covariate data, a condition which precludes strong inference.

Passage counts

Annual passage counts of American eels at the Millville Dam eel pass, which fluctuated from 852 to 5394, did not likely represent the total number of upstream migrants within the Shenandoah River. The eel pass was not in operation during all periods of expected upstream migration, and American eels may pass Millville Dam by other routes. Passage counts of American eels may also be associated with variables not measured during this study. For example, annual passage counts at Millville Dam may reflect variation in the annual recruitment of elvers to the Potomac River estuary (Moriarty, 1986). Interestingly, separate studies on ages of American eels that used the Millville Dam eel pass in 2005 (n = 74, range 214–550 mm TL) and 2008 (n = 42, range 244–467 mm TL) reported age ranges of 3–10 and 4–11 years, respectively (Hildebrand, 2005; Zimmerman and Welsh, 2012), which demonstrates a wide-range of age classes for upstream migrants.

River discharge and LI

This study found that elevated river discharge and low levels of LI were associated with upstream migration of American eels in the lower Shenandoah River, which is consistent with results of previous studies by Schmidt et al. (2009) and Welsh and Liller (2013). Although our study found an association of RDM and LI with upstream passage counts, multiple-day passage events with the highest counts of American eels (>250 individuals) were associated with periods of increased river discharge. It is unknown why American eels in the Shenandoah River move upstream during periods of increased river discharge, but Welsh and Liller (2013) speculated that increased water volume may provide American eels with less restricted travel, or the reduced ambient light owing to increased water turbidity during high river discharges may provide more favourable conditions for upstream migration. Migrant American eels may reduce upstream movements at some threshold of increased river discharge, but we documented upstream dam passage at the highest river discharge (i.e. 750.4 m³ s⁻¹) during this study. Anguillid eels avoid light and are crepuscular and nocturnal (Hammond and Welsh, 2009; Schmidt et al., 2009); hence, upstream movements associated with low LI are likely also associated with reduced ambient light.

Potomac River discharge

Routes of migratory fish are often influenced by the flow regime of tributaries within a river network (Thorstad et al., 2008; Nunn et al., 2010). We expected that river discharge of the Potomac River (measured at 19.5 km downstream of the Shenandoah River mouth) would influence the number of upstream migrants entering the Shenandoah River. Although increased discharge in the Potomac River likely influences upstream movements of American eels, our passage count data did not support river discharge in the Potomac River over that of the lower Shenandoah River. This suggests that river discharge of a tributary is likely more important to upstream movements within that tributary relative to river discharge in the mainstem downstream of the tributary mouth. Possibly, when the magnitude of increase in river discharge of the Potomac River upstream of the Shenandoah River mouth exceeds that of the Shenandoah River, then American eels may continue migration upstream within the Potomac River mainstem, as opposed to entering the Shenandoah River. However, we do not have data on the number of American eels moving upstream in the Potomac River, so we were not able to evaluate this hypothesis. We did note, however, that the ten highest daily passage counts of American eels occurred when the Shenandoah River contributed from 26.3 to 76.7% of the Potomac River discharge. To our knowledge, however, no studies have examined upstream migration of American eels relative to differential flow regimes of tributaries and the mainstem within a river network.

Water temperature

Our data did not support WT as an important contributor to upstream migration, a finding that results, in part, from the timing of the annual sample periods. Because the eel pass was installed annually in late spring or early summer, we were not able to document the WT associated with the onset of upstream migration, which likely occurred when WTs were within the range of 15–20°C (Hammond and Welsh, 2009; K. McGrath, pers. comm.). During fall, the number of upstream migrants decreased with lower WTs. Individuals were not observed moving upstream during fall after WTs were below 11.2°C, a finding consistent with previous studies by McGrath et al. (2003) and Hammond and Welsh (2009).

Most passage counts (90.0%) were observed within a wide range of WTs (18–28°C), which is consistent with a 19–28°C range reported by Welsh and Liller (2013). Further, Verdon and Desrochers (2003) found that the warmest seasonal temperatures (>20°C) were associated with upstream migration of American eels between the Beauharnois and Moses-Saunders Power Dams, St Lawrence River. McGrath et al. (2003) and Schmidt et al. (2009) reported peak counts for upstream migrants during summer at WTs of ∼20°C at the Moses-Saunders Power Dam of the St Lawrence River, and a small tributary (Saw Kill) of Hudson River, New York, respectively. However, McGrath et al. (2003) also documented peak counts in early October, and mid-October at WTs ∼15 and 13°C, respectively. Sorensen and Bianchini (1986) suggested that WT after reaching the threshold of 10–15°C may have little influence on American eel elver movements. Similarly, WT after reaching 15–20°C may have little influence on the upstream migration of yellow-phase American eels.

Cloud cover

Hain (1975) noted that reduced ambient light was associated with movement activity of American eels, an environmental condition that could be attributed to low LI, water turbidity during periods of increased river discharge, or increased cloud coverage. Although our data supported elevated RDM and low LI, we did not find a relationship between CC and upstream passage counts of American eels. A total of 65.6% of individuals (11 265 of 17 161 American eels) used the eel pass when CC was ≤25%. Also, daily values of CC and RDM were not correlated (r = 0.07), an expected outcome given that rain events within the Shenandoah River watershed, which are often associated with heavy CC, often occur one or more days before a realized increase in RDM.

Timing of annual eel pass installation

To protect the Millville Dam eel pass from damage or destruction associated with debris from periods of high river discharge, the pass was installed by personnel of the hydroelectric power company during spring or summer, and removed during fall. Consequently, the eel pass was not in operation during spring when WTs reached and exceeded 15°C, a condition when American eels were expected to initiate upstream migration. During the 7-year study, WTs exceeded 15°C for 17–63 d before annual eel pass installation, and the mean RDM and maximum values of RDM during these pre-installation periods were as high as 6.0 and 9.4 times that of those values during the periods of eel pass operation.

Given the association of eel passage counts with increased RDM, perhaps large numbers of American eels migrated upstream during high river discharges of the 17–63 d pre-installation periods. Support for this, in part, is provided by peak passage counts that occurred within 2 weeks following eel pass installation, as large passage counts of upstream migrants (possibly delayed by the dam) were recorded soon after eel pass installation. In addition to delayed passage, American eels may have found other routes past the dam during periods before the eel pass was operational, thus reducing the effectiveness of the eel pass as a monitoring method for upstream passage counts. We found that mean values of Shenandoah River discharge during the pre-installation periods when WT exceeded 15°C were >2.0 times that of the sampling periods with low annual passage counts in 2007–2009 and 2011, and <2.0 times that of the sampling periods with the highest annual counts in 2010, 2012, and 2014. One possible explanation for this difference is that American eels passed the dam during the relatively high river discharges of the 2007–2009 and 2011 pre-installation periods. Consequently, the effectiveness of the eel pass at passing American eels may be increased by an earlier installation date. The trade-off for an earlier eel pass installation date, however, is that debris damage could result in an extended non-operational period.

In summary, our 7-year study provided evidence that dam passage counts at an eel pass and pulses of upstream migrant yellow-phase American eels are associated with periods of increased river discharge or low LI. Given that passage events with the highest counts (>250 American eels) were associated with periods of increased river discharge, then annual and seasonal variability in river discharge likely influence the numbers of American eels migrating upstream within the Potomac River drainage. Because increases in river discharge are climatically controlled events, pulsed or punctuated migration events of American eels within the Potomac River drainage are likely linked to the influence of climate variability on flow regime of the tributary/mainstem network. The Millville Dam eel pass is typically not in operation during early spring, a period when river discharge is relatively high and WT exceeds the 15–20°C threshold for upstream migration. Consequently, the eel pass is not in operation during all periods when American eels are expected to migrate upstream, and annual passage counts may have been underestimated owing to passage during high river discharges of spring pre-installation periods. Our results suggest that an increase in passage effectiveness at Millville Dam may result from an earlier annual eel pass installation period, a management strategy that should also consider alternatives to reduce the risk of eel pass damage.

Acknowledgements

We thank S. Eyler, G. Harbaugh, C. Hilling, K. Sheehan, D. Smith, N. Taylor, and D. Wichterman for assistance with data collection. Funding was provided by FrontierEnergy Corporation and West Virginia Division of Natural Resources. This study was performed under the auspices of a West Virginia University IACUC protocol. The use of trade names or products does not constitute endorsement by the US Government.

Appendix

QIC_u-model selection statistics for 46 candidate models fit to a 2007–2013 time-series of daily American eel counts from the Millville Dam eel pass, Shenandoah River. Single-variable and additive-effects models included per cent LI, river discharge (m³ s⁻¹) from the lower Shenandoah River at Millville, West Virginia (RDM), and the RDP, WT of the RDM, and % CC near Millville Dam. We also fit lag models of RDM for 1 and 2 d. ΔQIC_u is the difference between a model and the model with the lowest QIC_u value, and the model weight (w_i) is the normalized value of the model based on its ΔQIC_u value.

Table A1.

Model	QICu	ΔQICu	wi
YR + RDM + LI	−166 890.9	0.0	1.0
YR + RDM + LI + CC	−162 031.8	4 859.1	0.0
YR + RDM	−157 247.6	9 643.3	0.0
YR + RDM + CC	−156 758.0	10 132.9	0.0
RDM + CC	−138 968.2	27 922.7	0.0
YR + RDM + LI + WT	−138 045.2	28 845.7	0.0
RDM	−137 349.4	29 541.5	0.0
YR + RDM + WT	−136 964.2	29 926.7	0.0
RDM + WT	−135 147.1	31 743.8	0.0
YR + RDM + LI + WT + CC	−134 953.8	31 937.1	0.0
RDM + LI	−132 143.1	34 747.8	0.0
RDM + LI + WT	−120 105.9	46 785.0	0.0
RDM + LI + WT + CC	−117 246.9	49 644.0	0.0
YR + RDP	−115 664.9	51 226.0	0.0
YR + RDP + CC	−114 516.3	52 374.6	0.0
RDP	−113 691.7	53 199.2	0.0
RDP + CC	−113 381.7	53 509.2	0.0
YR + RDP + LI	−108 200.0	58 690.9	0.0
YR + CC	−106 610.9	60 280.0	0.0
RDP + LI	−105 098.7	61 792.2	0.0
YR + LI + CC	−103 647.9	63 243.0	0.0
YR + LI	−101 276.7	65 614.2	0.0
YR + RDM(2) + CC	−98 791.3	68 099.6	0.0
YR + LI + WT	−97 622.8	69 268.1	0.0
CC	−92 241.4	74 649.5	0.0
YR + WT	−91 998.7	74 892.2	0.0
YR + RDM(1) + CC	−91 537.7	75 353.2	0.0
YR + RDM(2)	−89 527.3	77 363.6	0.0
Intercept	−85 766.7	81 124.2	0.0
WT	−85 678.0	81 212.9	0.0
YR + RDP + WT	−85 099.8	81 791.1	0.0
RDP + WT	−84 612.9	82 278.0	0.0
YR + RDM(1)	−84 485.3	82 405.6	0.0
LI + CC	−83 013.6	83 877.3	0.0
YR + RDP + LI + WT	−82 673.4	84 217.5	0.0
YR + RDM(2) + WT	−82 671.1	84 219.8	0.0
YR + RDP + LI + WT + CC	−82 052.5	84 838.4	0.0
LI	−80 293.3	86 597.6	0.0
LI + WT	−79 675.6	87 215.3	0.0
RDM(1) + CC	−77 593.6	89 297.3	0.0
RDP + LI + WT	−76 991.5	89 899.4	0.0
RDP + LI + WT + CC	−76 785.4	90 105.5	0.0
RDM(2) + CC	−74 454.2	92 436.7	0.0
RDM(1)	−71 955.8	94 935.1	0.0
YR + RDM(1) + WT	−71 085.2	95 805.7	0.0
RDM(2)	−61 215.8	105 675.1	0.0

Author notes

Handling editor: Caroline Durif