Introduction

Managers of local, state, and national parks are under considerable pressure to maximize the number of park visitors, as high visitation rates stimulate local economies and provide justification for sustained funding. However, parks are also charged with protecting the natural resources that attract visitors in the first place, and these two goals are sometimes in conflict (Goodwin 1996, Hammitt and Cole 1998). Visitors can have a number of negative effects on park ecosystems, and an important aspect of park management is the identification and assessment of visitor impacts (Sun and Walsh 1998). Some sorts of visitor-related problems, such as trail erosion, traffic congestion, and waste management, are easily identified (e.g., Marion and Farrell 2002). Other effects of visitation are more subtle, less appreciated, and too often go unaddressed. In particular, the effects of park visitors on ecosystem-level processes such as nutrient cycling are not visually apparent and thus are often not considered. However, perturbations to ecosystem level functions can trigger large-scale disruptions to the ecological functions of a protected area. In order to manage and mitigate the effects of visitation on park resources, managers must learn to identify and evaluate all visitor impacts, not just the visually obvious ones. Here we report on a common tourism practice that may degrade water quality, with the goal of increasing the awareness of such problems so they can be more readily identified and corrected.

The Linesville spillway of Pymatuning State Park, Pennsylvania, USA, is one of the most visited tourist attractions in the region, averaging more than 450,000 visitors each year (Linda Armstrong, Pymatuning State Park, personal communication). Water from the uppermost basin of Pymatuning Reservoir, Sanctuary Lake, passes over the spillway into the Middle Basin of the lake (Figure 1). Since shortly after the reservoir was created in 1934, large numbers of fish (mostly common carp, Cyprinus carpio L.) and waterfowl have congregated at the spillway where they are fed bread and other foods by park visitors. Casual observations suggest that quite large quantities of food are dumped into the lake each day, with the typical visiting family carrying along several loaves of bread to feed the fish. These foods, like all organic material, contain phosphorus (P), commonly a limiting nutrient in lakes. Much of the phosphorus is bread is in the form of calcium phosphates. A portion of the phosphorus in bread is bound up in organic forms, but microbial activity results in the metabolism of organic material and remineralizes P, making it available to support plant growth.

Fig. 1
figure 1

Map of Pymatuning Reservoir, showing Linesville spillway, Sanctuary Lake, Middle Lake, Lower Lake, the lower portion of Linesville Creek, and the boundaries of Pymatuning State Park. Linesville spillway is the outlet from Sanctuary Lake, and drains into Middle Lake

Full size image

Excessive phosphorus loadings can result in blooms of nuisance or toxic algae, low water clarity, hypoxia, fish kills, loss of biological diversity, and increased likelihood of invasion by exotic species (Carpenter and others 1998, Smith 1998). Phosphorus enrichment is a leading cause of surface water impairment in the United States (U.S. EPA 1996), because it limits the use of water for drinking, recreation, agriculture, and industry. Because of the detrimental effects of phosphorus on aquatic ecosystems, point source discharges of P into surface waters are limited and regulated by state and federal laws, including the Federal Clean Water Act of 1972.

Study System

Pymatuning Reservoir, Pennsylvania’s largest inland lake (surface area = 6645 ha), consists of three distinct basins: Sanctuary Lake, Middle Lake, and Lower Lake (Figure 1). Pymatuning Reservoir was formed in 1934 when the headwaters of the Shenango River were impounded by a dam built at Jamestown, PA. At the same time, another dam was built to separate Sanctuary and Middle Lakes, and water drains over the Linesville spillway from Sanctuary Lake into Middle Lake and thence into Lower Lake before exiting the reservoir via the Shenango River.

Based on limnological parameters measured in the summer of 2002 (Table 1), as well as previous investigations of Pymatuning Reservoir (Tryon and Jackson 1952, Hartman and Graffius 1960, U.S. EPA 1975) and traditional limnological criteria (Carlson 1977, Wetzel 2001), Sanctuary Lake and Middle Lake are both hypereutrophic, but Sanctuary Lake is even more productive than is Middle Lake. Water column P concentrations are very high and water clarity is low in both lakes. We focused our investigations on phosphorus because Pymatuning Reservoir has a N:P molar ratio >16:1 in late spring and early summer, and thus is likely P limited during this time (U.S. EPA 1975, A. Turner unpublished data).

Table 1 Limnological parameters for Sanctuary Lake and Middle Lake basins of Pymatuning Reservoir
Full size table

There are a number of nutrient sources to Pymatuning Reservoir. The basins are shallow and unstratified, and thus sediment resuspension and other forms of internal loading are likely important. The watershed is largely agricultural, and thus watershed loading is likely significant. Both the Linesville sewage treatment plant and the Linesville fish hatchery discharge effluent into Sanctuary Lake. A large number of resident Canada geese (Branta canadensis Linnaeus) reside on the lake and make a significant contribution to the overall P budget (U.S. EPA 1975).

We hypothesized that the “breadthrowers” may constitute a significant nutrient vector to the upper portion of Pymatuning Reservoir. To our knowledge, however, there are no existing studies of recreational fish feeding that would inform an assessment of the environmental impact of this practice (but see Wolos and others 1992 and Arlinghaus and Mehner 2003 for an analysis of phosphorus loadings associated with carp angling). Therefore, we estimated phosphorus loadings attributable to breadthrowers at the Linesville spillway. For reference, we measured phosphorus exports from Linesville Creek, the largest stream discharging into Sanctuary Lake. Linesville Creek has a watershed area of 25.2 km2 (34% of the overall Sanctuary Lake watershed), and is predominantly agricultural.

Methods

Spillway phosphorus loadings and Linesville Creek phosphorus exports were estimated each day from June 30 through August 3, 2002. Estimates of Linesville Creek P exports were based on daily measurements of stream discharge and collections of water samples for analysis of phosphorus concentrations. Because P concentrations of stream water tended to increase during high flows, we estimated P exports over the course of the study as the sum of the daily product of discharge and stream water P concentration.

Our study site was approximately 500 m upstream from the mouth of Linesville Creek and immediately upstream of the Linesville sewage treatment plant (Figure 1). Except during high flow periods, stream discharge was measured at a semipermanent station with smooth flow through a confined reach. We estimated discharge from measurements of stream width along with depth and velocity measurements at six stations placed 30 cm apart across the width of the stream. Total phosphorus concentration of stream water samples was measured by subjecting the samples to a persulfate oxidative digestion followed by colorimetric analysis (Perkin-Elmer Lambda 20 spectrophotometer) using the ascorbic acid–molybdate blue method (Strickland and Parsons 1972, APHA 1998).

We estimated phosphorus loadings associated with spillway breadthrowers by measuring the amount of bread thrown per visitor, the number of spillway visitors, and the phosphorus content of bread:

$$ ({\rm g\,P/day})\,=\,({\rm visitors/day})\,*\,({\rm g\,bread/visitor})\,*\,({\rm g\,P/g\,bread}) $$
(1)

Phosphorus content of bread is well documented, and we used published sources including the USDA Food and Nutrition Information Center (http://www.nal.usda.gov/fnic/). Phosphorus is a constituent of the organic compounds in plant seed tissue from which bread is made, but it is also added as a salt preservative to commercially prepared breads (e.g., dicalcium phosphate, diammonium phosphate). Different sorts of bread vary in P content, and our mean P value is weighted by the observed proportions thrown at the spillway.

Determination of mass of bread thrown per visitor was based on direct observations of randomly selected spillway visitors (n = 346 visitors) conducted in July 2002. Food items other than bread were frequently fed to fish, and these were noted, but this material was not included in our calculations.

We used repeated counts of cars in the spillway parking lot to estimate daily visitation because censusing cars was more efficient than censusing individual visitors. Using a simple steady-state model, hourly car counts (C i ) were converted into estimates of visitation by taking into account the average residence time of visitors (r), and average number of passengers per car (p):

$$ {\rm Daily\,visitation}\,=\,\sum (C_i/r)\,p $$
(2)

The number of vehicles in the spillway parking lot was estimated from hourly censuses (174 observations over 35 days for an average of 5.0 observations/day, missing hourly values fit with linear interpolation between counts). The average duration of visit (r = 0.47 hr, n = 47 observations) and the average number of visitors per vehicle (p = 3.1 passengers / car; n = 346 cars observed) were measured through direct observation.

We conducted a Monte Carlo simulation in order to estimate the error associated with our model estimate of P loading. For each model parameter (P content of bread, bread thrown per visitor, daily visitation), we generated a set of input values by randomly sampling a normal distribution with mean and standard deviation drawn from the distribution of the sample means. We lack replicate observations for the P value of bread, so we assumed a standard deviation equal to 10% of the mean. We then conducted 50,000 model runs, and then calculated the standard deviation of the resulting estimated mean P loads.

Results

Spillway P Loading

Spillway visitation averaged 2747 people per day over the course of the study (Table 2). Our observations revealed that spillway visitors fed a variety of items to fish during the study period, including bread, donuts, bagels, canned corn, popcorn, corn chips, hot dogs, birthday cakes, and dog food. Considering only the bread portion of the feeding activity, each visitor threw an average of 1084 g of bread into the lake (Table 2); 83% of this total was white bread, with wheat, rye, and corn bread making up the remainder. Because the phosphorus content of bread depends on bread type, we used an average P content weighted by the proportions of different bread types thrown, yielding a mean of 1.08 mg/g (Table 2). The product of these model parameters is an average phosphorus loading estimate of 3.23 kg·day−1 and estimated loads over 5 weeks totaling 113.1 kg. The standard deviation of daily P loading estimates yielded by repeated model runs was 563 g, 17% of the estimated mean.

Table 2 Key parameters used to estimate daily P loading at Linesville spillway
Full size table

Daily loads varied from a low of 1.49 kg on July 3 to a high of 7.93 kg on July 6 (Figure 2). Temporal trends showed a clear tendency for feeding activity to peak on weekends (Figure 2), with P loading on weekends and holidays averaging 5.15 kg·day−1, but just 2.23 kg·day−1 on weekdays.

Fig. 2
figure 2

Daily phosphorus loading (g P day−1) to Pymatuning Reservoir attributable to tourists feeding bread to fish at the Linesville spillway in summer 2002. Daily phosphorus exports from Linesville Creek watershed are shown for reference

Full size image

Linesville Creek

Median daily discharge of Linesville Creek was 3828 m3·day−1, and phosphorus content of streamwater averaged 72 μg/L. However, discharge increased more than 10-fold after rainfall events, and stream P concentration was positively correlated with discharge, so most of the P exports occurred after heavy rainfall. In fact, 83% of the Linesville Creek P exports measured over 5 weeks occurred in 1 day after very heavy rainfall (July 28 and 29). A weather-reporting station near Linesville reported 3.7 cm of rain over the 48-hr period. On July 29, discharge peaked at 235,525 m3·day−1, and P concentrations rose to 276 μg·L−1, for a calculated export of 65 kg over a 24-hr period. Phosphorus exports from the Linesville Creek watershed averaged 2235 g·day−1, and exports over 5 weeks totaled 78.2 kg. Given a watershed area of 25.2 km2 , the estimated phosphorus export coefficient for the watershed is 32.4 mg·m−2·year−1, a value quite typical of agricultural drainages. Thus, even though our estimate of P exports was heavily influenced by a storm event, the resulting estimate is a robust value for reference purposes.

Discussion

Our study shows that fish feeding by park visitors can potentially result in substantial environmental degradation. Averaged across time, breadthrowers contributed 1.45-fold more P to Pymatuning Reservoir than did the Linesville Creek watershed. P loading attributable to breadthrowers exceeded that of the Linesville Creek watershed on 33 of the 35 days of study, with only a heavy rainfall event triggering watershed exports that exceeded spillway contributions. If the areal export rates measured for the Linesville Creek watershed are extrapolated to the entire Sanctuary Lake watershed, spillway contributions of P contribute an additional 48% above the total non-point source watershed loads of P to the lake during the course of our study. We note, however, that we lack a comprehensive nutrient budget for Sanctuary Lake, making evaluation of the biological importance of the additional P at the scale of the entire lake difficult. The lake is quite eutrophic, because there are several large point sources of P to the lake, including a fish hatchery and sewage treatment outflow. In addition, the sediments of the lake become anoxic in summer, creating conditions favorable for significant internal P loading. Thus, even though spillway contributions of P are large relative to the measured watershed inputs, they may well be dwarfed by other sources of P. Lacking a P budget for Sanctuary Lake, the overall effect of fish feeding on water quality cannot be known with any certainty.

There are several potential errors that stem from the use of a relatively simple model of P loading. We have estimated gross P input, but some portion of the bread thrown to the fish will be ingested, absorbed, and retained by carp, resulting in a lower net P input. Our observations suggest that ingestion efficiency is relatively high, but retention efficiency varies widely depending on food quality, temperature, and feeding rate (Lall 1991, Kibria and others 1998, Jahan and others 2003, Niesar and others 2004), making a precise estimate of retention difficult. Furthermore, a portion of absorbed P is lost through excretion via the kidneys and gills (Jahan and others 2001). Niesar and others (2004) found that for carp fed a range of diets, P retention ranged from 3.6% to 32.1%, with the lowest retention efficiencies associated with low P content food types (see also Jahan and others 2001). The P content of bread is lower than any of the foods tested by Jahan and others (2001) or Niesar and others (2004), suggesting that most (>90%) P added to Pymatuning Reservoir becomes biologically available within a short time. Phosphorus retained by the carp is immobilized for the lifetime of the fish, and P fixed in the skeleton is slow to remineralize even after the death of the fish (Kitchell and others 1975, 1979). Nevertheless, a portion of the P sequestered by fish will eventually be remineralized and add to overall lake nutrient loads unless the fish is somehow removed from the lake. Although there is a limited sport fishery targeting carp in Pymatuning Reservoir, most anglers practice catch and release, so angling likely does not constitute a significant loss of P from the system. In sum, because the P retention efficiency of carp feeding on bread is quite low, short-term net input of P is not substantially less than gross P input, and a portion of the assimilated P will become biologically available over longer time scales.

We chose to focus on bread and ignore other sorts of food additions to the lake because we had relatively few observations of addition rate for these other foods and thus our estimates of the P loading associated with other foods would have had low precision because of small sample size. The additional foods added to the lake, however, contain significant amounts of phosphorus, and their omission would contribute to an underestimate of the true amount of P added at the spillway.

Is the Phosphorus Added Biologically Important?

In the absence of a P budget, it is difficult to resolve the extent to which spillway-derived P has an effect on water quality in Sanctuary Lake. In a more general sense it is possible, however, to determine whether anthropogenic inputs of the magnitude measured here have the potential to degrade water quality in a lake of this size. The eutrophication models of Vollenweider (Vollenweider and Dillon 1974, Vollenweider 1976) can be used to assess the effect of increased P loading on water quality. Following these models, a critical loading, defined as the P loading rate beyond which an oligotrophic lake will become mesotrophic, can be calculated with a knowledge of mean depth and mean hydraulic retention time (e.g., the loading plot of Vollenweider 1975). Using such data for Sanctuary Lake (Table 1) yields a critical loading of 160 mg P·m−2·yr−1. Assuming 450,000 visitors per annum, the spillway contributes 530 kg P each year to Sanctuary Lake. Standardized by the area of Sanctuary Lake (8.1 × 106 m2), spillway P loading is 65 mg·m−2·yr−1, which by itself is 41% the critical loading rate. Thus, there is clearly the potential for fish feeding to have significant effects on water quality in lakes the size of Sanctuary Lake.

Perhaps the most useful way to view the issue is to express spillway P loading (or any other point-source effluent) in terms of “watershed equivalents.” A watershed equivalent is the watershed area that would export an amount of material equivalent to the observed point source discharge. Using the data of the 1975 EPA study (U.S. EPA 1975), the mean export coefficient for the entire Pymatuning watershed was estimated to be 38.9 kg P·km−2·yr−1. This figure agrees with our own study of Linesville Creek, where we found that P exports extrapolated to a full year would be 32.4 kg P·km−2·yr−1 (assuming that export rates did not change seasonally), and with published studies of phosphorus export coefficients from mixed farmland/forest watersheds (ca. 35 kg P km−2 yr−1; Dillon and Kirchner 1975, Wetzel 2001). Using 35 kg P km−2 yr−1 as a working value, the spillway contribution would be equivalent to that of 15.1 km2 of Pymatuning Reservoir watershed, or roughly the same as the P loading of a typical small stream to the lake.

Excessive P loading can switch lakes over to nitrogen limitation, and there is some evidence that Pymatuning Reservoir is reaching this threshold. N:P molar ratios in Sanctuary Lake and Middle Lake are >16:1 in late spring and early summer, but fall to <10:1 in late summer (U.S. EPA 1975, A. Turner personal observation). Portions of Middle Lake suffer from profuse growth of N-fixing blue–green algae (Cyanobacteria) in the late summer. Seasonal shifts from P to N limitation are evidence that the lake is becoming P saturated because of excessive loadings of phosphorus.

We have measured oxygen concentrations within carp feeding aggregations in order to assess whether carp activity results in localized oxygen depletion and hypoxia. Midday oxygen concentrations fell from a background level of 9–10 mg·L−1 to less than 2.0 mg·L−1 where carp densities exceeded 1.0 fish·m−2 (A. Turner and S. Montgomery, unpublished data). Carp density within feeding aggregations at the Linesville spillway often exceeds 6 fish·m−2. Carp require protein and other essential nutrients in their diet, so consumption of low-protein food such as bread may stunt growth or cause other health problems for the fish (Niesar and others 2004, Arlinghaus and Niesar 2005). Wild fish, however, may be able to meet their nutritional requirements by supplementing their diet with natural foods.

The feeding activity at the Linesville spillway also attracts a sizable flock of waterfowl, mostly resident giant Canada geese (B. canadensis). Because Canada geese feed largely in terrestrial habitats but take refuge on water, they may constitute a significant nutrient vector into lakes (Manny and others 1994, Post and others 1998, Kitchell and others 1999). Ferris and Jones (2003) evaluated the role of resident Canada geese in transporting phosphorus into Sanctuary Lake. Based on direct observations of goose population size, activity patterns, and defecation habits along with published accounts of goose manure phosphorus content, they estimated P loading attributable to geese in Sanctuary Lake of 320–1200 g P·day−1, similar to the daily P exports from Linesville Creek. The EPA study of 1973 (U.S. EPA 1975) also cited the resident Canada geese as a water quality concern.

A Solution to the Problem?

Bread sales at the Linesville spillway, handled by a private concessionaire, were replaced with catfish food pellets in summer 2003. Although this shift has somewhat reduced the unsightly and unseemly habit of park visitors throwing human food into a public waterway, it has not addressed the root problem of water quality degradation. The spillway concessionaire estimates food sales of more than 500 kg on a busy day (personal communication). Phosphorus content of catfish food is 0.8%, 8.5-fold higher than that of white bread. Thus, current P loads from fish food alone may exceed 4000 g day−1 (a large portion of the tourists continue to supply their own bread). More generally, it seems unlikely that the switch to the P-enriched fish food has been accompanied by an 8.5-fold reduction in amount of food fed to fish, so the overall result of this conversion has likely been to elevate the already problematic P loading to Pymatuning Reservoir.

The effect of fish feeding on water quality can be moderated in some cases by regulating the types of food fed to the fish. For example, Arlinghaus and Niesar (2005) show that various foods typically fed to carp differ substantially from each other in P content and retention and suggest that the most ecologically sound foods are those with the lowest P content (see also Jahan and others 2001, Niesar and others 2004). In the case of Pymatuning Reservoir, however, the bread predominantly fed to fish has a P content that is very low relative to alternate foods (Table 2). Thus, P enrichment of the reservoir is driven by the quantity of food added to the lake, not the quality, and any effort to moderate the impact of fish feeding on water quality at Pymatuning State Park should focus on limiting the amount of food fed to fish.

Conclusion

In the 1930s when tourists began feeding fish at the Linesville spillway, animals were commonly fed within parks, and such activities were popular tourist attractions. However, biologists and park managers have come to recognize that it is inappropriate for humans to interact with wildlife by feeding them, because this poses a number of dangers to both wildlife and humans. Well-known problems associated with feeding wildlife and the resulting aggregation of animal populations at feeding stations include aggression toward humans, malnutrition, dependency, transmission of disease, and environmental degradation (Orams 2002). For these reasons, managers of public lands generally discourage or prohibit the feeding of wild animals, and decades of education have sought to inform the public of the detrimental effects of such practices (Mallick and Driessen 2003). Curiously, however, the practices of feeding fish continue in some instances, as does the feeding of waterfowl, perhaps under the presumption that the problems associated with feeding other wildlife species do not apply to these taxa.

Our study demonstrates that fish-feeding activity by park visitors has the potential for local degradation of the aquatic ecosystem. This sort of visitor impact is not visually obvious, yet has large-scale and persistent ecosystem level effects. Lakes, streams, and rivers are especially vulnerable to disruptions of nutrient cycles, because they are preferred recreational sites, but are sensitive to even small perturbations of biogeochemical cycles. Park managers need to consider such disruptions to ecosystem level processes when evaluating visitor impacts.

Outright prohibition of feeding fish by park visitors would likely be unpopular and unpalatable to park managers. The 450,000 people that visit the Linesville spillway each year are a powerful testament, however, to the innate curiosity of humans regarding fish and aquatic ecosystems. The presence of so many curious park visitors presents a valuable opportunity to implement educational efforts. Exhibits or other educational efforts could be devised with the goal of educating visitors about the various threats to water quality. The most difficult step in implementing any sort of public education program is generating an interested audience, and the Linesville spillway attraction provides a large captive audience each day. By developing educational resources at the site, park managers would be taking best advantage of the lure of the carp.