1. Evaluating the effectiveness of stream restoration is often challenging because of the lack of pre-treatment data, narrow focus on physicochemical measures and insufficient post-restoration monitoring. Even when these fundamental elements are present, quantifying restoration success is difficult because of the challenges associated with distinguishing treatment effects from seasonal variation, episodic events and long-term climatic changes.
2. We report results of one of the most comprehensive and continuous records of physical, chemical and biological data available to assess restoration success for a stream ecosystem in North America. Over a 17 year period we measured seasonal and annual changes in metal concentrations, physicochemical characteristics, macroinvertebrate communities, and brown trout Salmo trutta populations in the Arkansas River, a metal-contaminated stream in Colorado, USA.
3. Although we observed significant improvements in water quality after treatment, the effectiveness of restoration varied temporally, spatially and among biological response variables. The fastest recovery was observed at stations where restoration eliminated point sources of metal contamination. Recovery of macroinvertebrates was significantly delayed at some stations because of residual sediment contamination and because extreme seasonal and episodic variation in metal concentrations prevented recolonization by sensitive species.
4.Synthesis and applications. Because recovery trajectories after the removal of a stressor are often complex or nonlinear, long-term studies are necessary to assess restoration success within the context of episodic events and changes in regional climate. The observed variation in recovery among chemical and biological endpoints highlights the importance of developing objective criteria to assess restoration success. Although the rapid response of macroinvertebrates to reduced metal concentrations is encouraging, we have previously demonstrated that benthic communities from the Arkansas River remained susceptible to other novel anthropogenic stressors. We suggest that the resistance or resilience of benthic macroinvertebrate communities to novel stressors may be effective indicators of restoration success that can account for the non-additive (e.g. synergistic) nature of compound perturbations.
The likelihood that degraded ecosystems can recover following the removal of a stressor and the length of time required for systems to return to pre-disturbance conditions remain critical research questions in applied ecology (Millennium Ecosystem Assessment 2005). If recovery is gradual and dependent on the magnitude of a stressor or if a system has shifted to an alternative stable state, recovery may require long periods of time (Scheffer et al. 2001). Furthermore, because recovery trajectories are often complex or nonlinear (Clements, Vieira & Sonderegger 2010), some communities may continue to display a legacy of impacts long after stressor removal (Matthews, Landis & Matthews 1996; Zhang, Richardson & Pinto 2009). In a meta-analysis of 240 independent investigations across a wide range of ecosystems and disturbance types, Jones & Schmitz (2009) concluded that the duration of many studies was insufficient to document recovery.
Long-term monitoring programmes that evaluate recovery of degraded ecosystems following the removal of a stressor are relatively rare. Failure to implement effective pre- and post-restoration monitoring remains a serious impediment to our ability to assess effectiveness of restoration programmes (Bernhardt et al. 2005; National Research Council 2007). Field assessments of how communities respond to restoration will further our knowledge on the resilience of natural systems. In addition, such studies can also be designed as ‘natural experiments’ (sensuDiamond 1986) to evaluate the ecological benefits of specific restoration activities and to quantify stressor-response relationships and recovery trajectories after an anthropogenic perturbation (Wiens & Parker 1995). Understanding these relationships is necessary to determine restoration effectiveness and to identify which biological responses are the best indicators of ecological recovery. Currently, there is little consensus among resource managers on these issues (McClurg et al. 2007). Furthermore, there have been few attempts to integrate basic ecological principles of disturbance and succession theory into studies of restoration (Lake, Bond & Reich 2007).
The recovery of communities from anthropogenic disturbance is influenced by numerous factors, including the specific nature of the perturbation (e.g. physical vs. chemical stressor; Rapport, Regier & Hutchinson 1985), the amount of natural variation that a community experiences (Kiffney & Clements 1996a), the presence of other stressors (Paine, Tegner & Johnson 1998), availability of colonists (Niemi et al. 1990), life-history characteristics of dominant species (Gunderson 2000) and functional redundancy (Bellwood et al. 2004). Quantifying recovery in natural systems is often difficult (Van Nes & Scheffer 2007), in part because of the limited number of long-term studies conducted after natural or anthropogenic disturbance. Furthermore, apparent recovery of some measures, such as abundance or species richness, may occur despite persistent alterations in community composition or loss of ecological resilience (Berumen & Pratchett 2006).
Evaluating the effectiveness of stream restoration programmes following improvements in water quality requires long-term monitoring of physicochemical characteristics and contaminant concentrations, and then relating these changes to a suite of ecologically relevant metrics. In Rocky Mountain streams where historical mining operations have resulted in widespread contamination by heavy metals and other stressors, long-term monitoring has revealed considerable variation in recovery rates among biological measures (Clements 2004). Delayed recovery of aquatic communities has been attributed to ‘biological resistance’, which occurs when tolerant species become established and impede recolonization by sensitive species (Frost et al. 2006). Because reductions in aqueous metal contamination typically occur well before improvements in sediment quality (Burton 1992), it is not surprising that recovery trajectories vary among biological measures. For example, organisms associated with the water column are likely to recover more rapidly than organisms closely associated with highly contaminated sediment. Furthermore, without relating ecosystem recovery to specific restoration activities, variation in responses observed among physical, chemical and biological measures cannot be fully explained.
A long-term (17 year) research programme conducted in the Arkansas River near Leadville, CO (USA), provided a unique opportunity to evaluate the effectiveness of restoration activities designed to mitigate historical mining impacts. We measured water quality, metal concentrations in sediments and biological (fish and macroinvertebrate) responses at stations upstream and downstream from two sources of metal contamination. Data were collected before and after several restoration events, including installation of water treatment facilities, removal of mine tailings and revegetation of riparian areas. The specific objectives of this research were to (i) document the effectiveness of restoration activities in the upper Arkansas River basin; (ii) compare responses of macroinvertebrates and brown trout Salmo trutta Linnaeus populations to post-restoration improvements in water quality; and (iii) quantify the relative importance of local (e.g. physicochemical) and regional (e.g. hydrology) factors that influenced rates of recovery.
Materials and methods
Study site and description of restoration activities
Our long-term sampling programme focused on the upper Arkansas River, located c. 110 km south-west of Denver, CO (Fig. 1). Mining activities in the watershed throughout the late 19th and early 20th century focused primarily on deposits in California Gulch (CG), a tributary to the Arkansas River. CG and much of the surrounding watershed was designated a U.S. EPA Superfund Site in 1983. Installation of a water treatment facility on the Leadville Mine Drainage Tunnel (LMDT), which discharged metal-contaminated water into the East Fork of the Arkansas River, was completed in 1992 and designed to improve water quality in the upper portion of the watershed (U.S. EPA 2007). Restoration activities in the lower portion of the watershed included construction of a water treatment facility for the Yak Tunnel (completed in 1994), removal of c. 150 000 m3 of tailings from CG and revegetation of riparian areas (both completed in 1999).
Sampling stations were located upstream (EF5 and AR1) and downstream (AR3 and AR5) from CG, the major source of contamination in the system (Fig. 1). Although stations EF5 and AR1 were located downstream from LMDT, elimination of this source of metals in 1992 allowed us to compare biological responses with those at highly contaminated stations below CG (Clements 2004). We also compared benthic communities to those sampled in 62 reference streams from Colorado. To characterize the large-scale spatial distribution of metals, we also report concentrations in sediments collected throughout the upper 110 km of the watershed.
The upper Arkansas River displays a typical snowmelt-dominated hydrograph, with peak discharge occurring in early to mid-June. Streambed substrate consists primarily of medium to large cobble, with underlying pebble and coarse sand. At c. 3000 m elevation, riparian vegetation is mostly willow (Salix spp.), whereas upland vegetation consists primarily of sagebrush and grasses. Historically, fish communities were dominated by salmonids, including the native greenback cutthroat trout Oncorhynchus clarki stomias Cope. Native fishes were extirpated in the early 1900s (Behnke 2002) and introduced brown trout now comprise over 90% of the fish community.
Conductivity, pH and water temperature were measured at each station in spring and autumn from 1989 to 2006. Stream depth and current velocity were measured immediately adjacent to where benthic samples (see below) were collected. Water samples (0·5 L) were collected in the field and immediately placed on ice for determination of water hardness and alkalinity, two factors that determine metal toxicity. Hardness and alkalinity were measured in the laboratory using standard titration procedures. Long-term (1968–2007) stream discharge data were obtained from a U.S. Geological Survey (USGS) gauge at station AR1.
Water samples collected for determination of dissolved Cd, Cu and Zn were filtered (0·45 μm) and acidified with nitric acid to a pH of <2. Depending on concentrations, metals were analysed using either flame or furnace atomic adsorption spectrophotometry. Routine QA/QC procedures included analysis of blanks and spiked water samples. To examine spatial and temporal variation in sediment contamination, 3 kg of fine sediment was collected from 13 stations (Fig. 1) along a 110 km reach over three time periods: 1993–1994 (two sampling occasions), 2001 and 2005–2006 (two sampling occasions). At each site, sediment was collected from 5 to 10 locations within a 30 m stream reach to obtain a representative sample. Sediment was sieved in the field to remove particles >2 mm, stored in a plastic container and transported to the USGS laboratories in Denver, CO, for processing. After air-drying at ambient temperature, the sample was mixed, sieved to <177 μm and sub-sampled for geochemical analysis. Samples were then ground to <120 μm and digested using a mixed-acid procedure consisting of HCl, HNO3, HClO4 and HF acids, with a final dilution factor of 1 : 100. Samples were analysed for total metals using inductively coupled plasma atomic emission spectroscopy.
Macroinvertebrate and brown trout sampling
A total of 700 quantitative macroinvertebrate samples were collected using a 0·1-m2 Hess sampler in spring and autumn at all four stations from 1989 to 2006. Five replicate samples were collected from each station. Samples were washed through a 350-μm sieve and organisms were preserved in 70% ethanol in the field. Macroinvertebrate samples were sorted in the laboratory and all organisms, except chironomids, were identified to genus or species. Chironomids were identified to subfamily. Benthic samples were sorted completely from 1989 to 2004, but sub-sampled using a standard 300-count protocol thereafter (Moulton et al. 2000). In this study, we report results for total macroinvertebrate abundance and species richness of mayflies (Ephemeroptera). We included macroinvertebrate abundance because of its potential importance to brown trout populations. Mayfly richness was included because our previous studies have shown this metric to be especially sensitive to metals (Clements et al. 2000). To place these findings in a broader context and to objectively define reference conditions, results were compared to a large database of reference streams (n = 62) sampled in Colorado from 2003 to 2007 using these same techniques (Schmidt 2007).
Fish surveys were conducted via bank electroshocking in August 1991, 1994, 1996, 1997, 1999, and annually from 2001 to 2006. Density estimates were based on the two pass removal method (Seber & LeCren 1967). Fish were identified and individual lengths (mm) and weights (g) were measured in the field. Estimates of brown trout density and biomass, based on trout greater than 1 year of age (1+), were reported for each station. In most cases, 1+ or older were classified as fish ≥10 cm in length, based on length–frequency distributions.
Because the Arkansas River is contaminated by a mixture of metals (Cd, Cu, Zn), we used an additive measure of toxicity to express metal concentrations relative to the U.S. EPA criteria maximum concentration (CMC). The CMC is the highest concentration of a metal to which an aquatic community can be exposed briefly without resulting in an unacceptable effect (U.S. EPA 2002). We define the cumulative criterion unit (CCU) as the ratio of the measured dissolved-metal concentration to the hardness-adjusted CMC and summed for each metal:
where mi is the measured concentration of the ith metal and ci is the hardness-adjusted CMC for the ith metal. Criterion values were adjusted for water hardness, which influences the toxicity of metals to aquatic organisms (Penttinen, Kostamo & Kukkonen 1998). If metal concentrations were below the analytical limits of detection, we used half of the detection limit to calculate the CCU. Assuming that responses to metals were additive, a CCU ≤1·0 represents a metal concentration that should be protective of aquatic organisms (Clements et al. 2000).
To quantify responses of macroinvertebrates and fish to restoration treatments, we examined long-term changes in total macroinvertebrate abundance, species richness of mayflies, and density and biomass of brown trout. We examined differences in community composition based on changes in relative abundance of metal-sensitive [heptageniid mayflies, Paraleptophlebia sp. (Ephemeroptera), Heterlimnius corpulentus (Coleoptera), Pericoma sp. (Diptera)] and metal-tolerant [hydropsychid and brachycentrid caddisflies (Trichoptera), orthoclad chironomids (Diptera), oligochaetes] groups. Metal-sensitive and metal-tolerant groups were identified based on the results of spatially extensive surveys in Colorado (Clements et al. 2000) and mesocosm experiments conducted with benthic communities (Clements 2004). All statistical analyses were conducted using SAS (version 9·0, 2003; SAS Institute Inc., Cary, NC). A Generalized Linear Model procedure (PROC GLM) was used to examine variation in metals and biological metrics among stations (EF5, AR1, AR3 and AR5), seasons (autumn, spring) and restoration treatments (before and after 1999, the year all major restoration activities were completed). To determine if effects of restoration varied among stations, we also tested for a significant station × restoration treatment interaction term for each variable.
To assess the influence of metals relative to sampling date, station and other physicochemical characteristics, we conducted linear regression analyses (PROC REG) using Akaike Information Criterion (AIC) on a suite of candidate models (Burnham & Anderson 1998). For each biological measure we tested five a priori models: (i) metals only; (ii) metals, station and sampling date; (iii) metals, station, date and stream discharge; (iv) metals, station, date and all physicochemical variables except discharge; and (v) all of the above predictors. Stream discharge was included as a separate predictor in these analyses because previous research showed that metal contamination in the Arkansas River and other Colorado streams increased during periods of high run-off (Clements 1994; Brooks, McKnight & Clements 2007). Other physicochemical variables included conductivity, pH, water temperature, stream depth, current velocity and alkalinity. All variables were transformed to meet assumptions of normality and homogeneity of variance and models were checked for independence, outliers and collinearity using diagnostic tools in SAS (version 9·0, 2003). Likely candidate models were defined as those with a ΔAIC value <2·0 (Burnham & Anderson 1998).
Although there were some differences among sites, routine water quality and physicochemical characteristics showed few obvious longitudinal trends (Table 1). The relatively high alkalinity, hardness and conductivity at station EF5 were most probably a result of the LMDT water treatment facility. The increased conductivity observed between stations AR1 and AR3 resulted from elevated SO4 concentrations associated with the contaminant load from CG. Mean annual stream discharge at station AR1 was highly variable over the 17 year period, ranging from 0·83 to 3·41 m3 s−1, with the highest values observed in the mid-1990s (1995–1997) and the lowest value observed during a severe drought in 2002 (Fig. 2).
Table 1. Mean (SD) and range of physicochemical characteristics measured at stations located upstream (EF5, AR1) and downstream (AR3, AR5) from the California Gulch Superfund site, 1989–2006. N = 35 for each station
Alkalinity (mg L−1)
72·7 (9·8) 48–86
54·8 (16·3) 27–96
55·5 (16·1) 27–85
53·2 (11·3) 29–70
Hardness (mg L−1)
113·4 (23·5) 66–166
77·9 (20·4) 36–112
101·4 (25·5) 52–164
84·9 (15·2) 52–120
7·7 (0·6) 5·7–8·6
7·6 (0·4) 6·4–8·3
7·4 (0·5) 6·4–8·4
7·5 (0·6) 4·8–8·7
Conductivity (μS cm−1)
211·3 (54·8) 108–365
141·3 (39·0) 70–237
187·1 (66·5) 80–360
154·0 (38·2) 65–230
7·6 (2·9) 2–15
6·7 (2·6) 2–13
6·5 (2·6) 1–12
5·9 (2·7) 1–13
Current velocity (m s−1)
0·7 (0·3) 0·2–2·0
0·7 (0·2) 0·3–1·6
0·9 (0·6) 0·2–1·7
0·8 (0·3) 0·3–1·8
29·6 (8·9) 17–55
37·5 (9·8) 25–70
30·4 (11·5) 15–61
38·0 (10·5) 19–58
Concentrations of metals in sediment decreased with distance downstream from CG, reflecting deposition in the upstream reaches and dilution by uncontaminated sediment further downstream (Fig. 3). Metal concentrations in sediment also varied among years; however, except for the most contaminated stations immediately downstream from CG, concentrations did not decrease consistently over time. For example, concentrations of Cd were generally higher in 2001 compared to other years. Despite a gradual reduction in the level of sediment contamination downstream, Cd and Zn concentrations remained above the probable effects concentrations throughout the upper Arkansas River basin.
Spatiotemporal patterns of metal contamination in the water reflected seasonal and long-term variation in stream discharge and responses to restoration treatments (Fig. 4). Metal concentrations at stations EF5 and AR1 decreased immediately after completion of the LMDT treatment facility in 1992. Metals remained well below toxic concentrations (CCU = 1·0) at station EF5 and fluctuated seasonally between 0·5 and 2·0 at station AR1. In contrast, metal concentrations at station AR3 remained elevated (>20 CCU) until spring 1998 and then decreased (CCU < 12) after completion of major restoration activities in CG and the Arkansas River floodplain. CCU values decreased downstream at station AR5, but remained elevated above upstream levels. Metal concentrations were routinely elevated during spring snowmelt at all Arkansas River stations, particularly in high-flow years (e.g. 1996), but decreased during years of low stream discharge (e.g. 2002). Across all stations and dates, CCU was significantly lower (P <0·0001) after restoration (Table 2). The station × restoration treatment interaction term was also highly significant (P =0·0031), indicating that improvements in water quality varied among stations.
Table 2. Results of Generalized Linear Model analyses (PROC GLM) showing effects of station (EF5, AR1, AR3, AR5), season (spring, autumn) and restoration treatment (before, after 2000) on metals (as CCU), macroinvertebrates and brown trout in the Arkansas River. Seasonal variation in brown trout was not assessed because fish were sampled only in summer. Table shows mean values and results of Ryan’s Q multiple range multiple tests (means with the same letter were not significantly different)
Station × restoration
na, data not collected.
F3,131 =48·6 (<0·0001)
F1,131 = 37·8 (<0·0001)
F1,131 = 20·8 (<0·0001)
F3,131 = 4·9 (<0·0031)
F8,131 =27·4 (<0·0001)
F3,131 =10·1 (<0·0001)
F1,131 = 7·4 (0·0076)
F1,131 = 81·3 (<0·0001)
F3,131 = 9·1 (<0·0001)
F8,131 = 18·3 (<0·0001)
F3,131 =34·5 (<0·0001)
F1,131 = 4·7 (0·0322)
F1,131 = 79·1 (<0·0001)
F3,131 = 6·7 (<0·0006)
F8,131 = 25·8 (<0·0001)
Trout biomass (kg ha−1)
F3,34 =18·8 (<0·0001)
F1,34 = 67·4 (<0·0001)
F3,34 = 18·4 (<0·0001)
F7,34 = 22·2 (<0·0001)
Trout density (no. ha−1)
F3,34 = 17·7 (<0·0001)
F1,34 = 42·4 (<0·0001)
F3,34 = 8·9 (<0·0002)
F7,34 = 17·8 (<0·0001)
Macroinvertebrate responses and brown trout
Macroinvertebrate communities in the Arkansas River varied significantly among stations and responded to temporal (seasonal and annual) changes in water quality (Figs 5–7). Across all stations, total macroinvertebrate abundance and mayfly richness were significantly (P =0·0001) greater after restoration (Table 2). Responses to restoration also varied among stations, as indicated by a highly significant (P <0·0001) station × restoration interaction term for both metrics. Macroinvertebrate abundance (P =0·0076) and mayfly richness (P =0·0322) were also significantly greater in autumn than in spring.
Temporal patterns of macroinvertebrate communities varied between upstream and downstream stations. Total macroinvertebrate abundance at the two upstream stations increased immediately after completion of the LMDT water treatment facility and approached or exceeded values at reference stations for the duration of the study (Fig. 5). Macroinvertebrate communities at the most contaminated station (AR3) displayed extreme seasonal variation prior to 1997, as most organisms were eliminated during periods of high metal concentrations in spring. However, macroinvertebrate abundance at station AR3 responded to the completion of restoration treatments in 1999 and increased for the duration of the study. In contrast to these patterns, macroinvertebrate communities at station AR5 showed little response to restoration. Abundance increased steadily until 1995 then decreased abruptly in spring 1996, coinciding with a period of high stream discharge and a greater than twofold increase in metal concentrations (Fig. 4).
Species richness of mayflies responded to long-term improvements in water quality at stations EF5, AR1 and AR3 (Table 2; Fig. 6) and exceeded reference values after completion of restoration treatments. In contrast to the relatively abrupt response to restoration at station AR3, mayfly richness at station AR5 was either similar to or less than reference values.
Although total macroinvertebrate abundance at the downstream stations was consistently greater after restoration, community composition, based on relative abundance of metal-sensitive and metal-tolerant taxa, was markedly different between upstream and downstream stations (Fig. 7). Between 2000 and 2006, metal-sensitive taxa on average accounted for c. 30% of total macroinvertebrate community at the two upstream stations. The relative abundance of these groups was much lower at the downstream stations where communities were dominated by metal-tolerant taxa, which accounted for 47–66% of the total macroinvertebrate community.
Density and biomass of brown trout populations varied among stations and responded significantly (P <0·0001) to restoration treatments (Table 2, Fig. 8). The station × restoration treatment interaction term was also highly significant (P <0·0002) for both measures, indicating that the effects of restoration varied among stations. Density and biomass at the two upstream stations were variable but generally increased over time. Brown trout populations at the downstream stations showed little change throughout the 1990s, but recovered quickly at station AR3 in 1999. Improvements in brown trout populations at station AR5 over this period were more gradual and density remained low compared with other stations.
Abiotic factors influencing biological recovery
AIC model selection was used to determine the relative importance of metal concentration, station, sampling date, stream discharge and other physicochemical characteristics for each of the biological measures we examined (Table 3). AIC values, corrected for small sample size, were calculated and ΔAIC values were used to rank candidate models. Models for all response variables were highly significant (P <0·0001), and showed a negative relationship with metal concentration and a positive relationship with time since restoration. Metal concentration alone was a relatively weak predictor of all biological variables, explaining only 21–25% and 34–38% of the observed variation in macroinvertebrate and brown trout metrics, respectively. AIC analysis showed no support for these simple models. Total macroinvertebrate abundance was best described by a model that included all predictor variables (R2 = 0·70); however, there was strong support for a much simpler model that only included CCU, station and date (ΔAIC = 1·2; R2 = 0·68). The best models identified by AIC for mayfly richness (R2 = 0·66), brown trout biomass (R2 = 0·51) and brown trout density (R2 = 0·54) included CCU, station, date and stream discharge. However, as with the macroinvertebrate metrics, there was strong support for simple models that only included CCU, station, and date (ΔAIC = 0·1–1·1).
Table 3. Results of AIC model selection for benthic macroinvertebrate and brown trout metrics. The table shows adjusted R2 values, number of parameters (k) and AICc values (corrected for small sample size) for each model. AIC weight refers to the relative support for each model. ΔAIC is the difference between each model and the best model. All models with ΔAIC values ≤2·0 (shown in bold) were considered likely models
AIC, Akaike Information Criterion; PCHEM, conductivity, pH, water temperature, stream depth, current velocity and alkalinity; CMS, stream discharge.
2. CCU, station, date
3. CCU, station, date, CMS
4. CCU, station, date, PCHEM
5. CCU, station, date, PCHEM, CMS
2. CCU, station, date
3. CCU, station, date, CMS
4. CCU, station, date, PCHEM
5. CCU, station, date, PCHEM, CMS
Brown trout density
2. CCU, station, date
3. CCU, station, date, CMS
4. CCU, station, date, PCHEM
5. CCU, station, date, PCHEM, CMS
Brown trout biomass
2. CCU, station, date
3. CCU, station, date, CMS
4. CCU, station, date, PCHEM
5. CCU, station, date, PCHEM, CMS
Biological monitoring programmes for aquatic and terrestrial ecosystems are generally designed to assess status and trends and to identify specific stressors that adversely impact communities. A less common objective is to quantify ecological responses to improvements following the elimination of a stressor. We believe that monitoring programmes intended to assess the restoration effectiveness in damaged ecosystems should be conducted similar to manipulative experiments (Benayas et al. 2009), thereby strengthening our ability to establish causal relationships between restoration treatments and recovery.
Although restoration efforts may significantly improve biodiversity and ecosystem services, these systems often remain impaired relative to reference systems (Benayas et al. 2009). Restoration in the upper Arkansas River resulted in significant improvements in water quality, abundance and richness of macroinvertebrate communities and brown trout populations, but the effectiveness of restoration varied temporally, spatially and among response variables. Faster recovery was observed at the two upstream stations where restoration focused exclusively on improving water quality. CCU levels at both stations dropped sharply and macroinvertebrate communities responded immediately after completion of the LMDT water treatment facility. CCU values were generally below 2·0, a level previously shown to have little impact on benthic communities in Rocky Mountain streams (Clements et al. 2000). The rapid response of macroinvertebrates to improvements in water quality demonstrates the high resilience of these communities to chemical stressors, a finding that has been reported in other systems (Watanabe, Harada & Komai 2000; Adams, Ryon & Smith 2005). These results are also consistent with the findings of Jones & Schmitz (2009) who reported relatively fast recovery for freshwater benthic communities (10 years) relative to other community types. We speculate that the proximity of upstream and lateral (e.g. from unpolluted tributaries) sources of organisms that quickly re-colonized following treatment of LMDT accounted for the rapid recovery at upstream stations. These results also highlight important differences in how benthic communities respond to physical and chemical stressors. Unlike responses to physical disturbances, which often involve habitat alterations, responses to elimination of chemical stressors may occur quite rapidly (Niemi et al. 1990).
The density and biomass of brown trout at the two upstream stations were variable, making it difficult to associate improvements in these populations directly with restoration activities. Although density and biomass at these stations increased from 1991 to 1994, coinciding with reduced metal concentrations and increased macroinvertebrate prey resources, both measures were lower on subsequent sampling occasions. These results suggest that fish populations were relatively insensitive to the moderate levels of metal contamination observed at the two upstream stations.
In contrast to responses at the upstream stations, recovery at the two downstream stations was less pronounced and significantly delayed. The slower recovery at these stations probably resulted from much higher and seasonally variable metal concentrations. CCU values after 1995 were generally less than 6·0 in autumn, but frequently exceeded levels considered to be highly toxic (10·0–20·0) in spring. Total macroinvertebrate abundance and mayfly richness at the most contaminated station (AR3) showed little indication of recovery until 2000, after which abundance steadily increased. Similarly, brown trout abundance and biomass at station AR3 increased only after 1999, coinciding with the completion of major restoration activities and significant improvements in water quality. Slower recovery of all metrics at the furthest downstream site, AR5, was probably influenced by residual mine tailings in the floodplain that contributed to the higher metal loadings observed in the spring at this station.
Population responses of brown trout to metals in the Arkansas River may have been complicated by exposure to the parasite Myxobolus cerebralis, the primary cause of whirling disease in salmonids. Although infection rates varied annually and showed no differences among our four sites (G. Policky, personal communication), they were consistently higher compared to downstream stations located below the influence of metals (Nehring 2006). These results are consistent with the hypothesis that metal-exposed populations were more susceptible to whirling disease, and that whirling disease may have contributed to variation in population density via poor recruitment in some years.
Although metals are the major stressor to fish and macroinvertebrates in the Arkansas River, the results of AIC model selection showed essentially no support for models that included only metal concentration as a predictor. However, the most complex models that included all predictor variables were also not consistently supported by AIC. Variation in all biological measures could be explained by relatively simple models that included metal concentration, station and sampling date. These results suggest that local physicochemical characteristics (temperature, pH, conductivity, alkalinity, depth, current velocity) were relatively unimportant in explaining long-term biological responses in the Arkansas River.
Seasonal variation and episodic effects of metals
In addition to long-term changes in metal concentrations and associated biological responses, we observed considerable seasonal variation in metal contamination. We hypothesize this seasonal variation played an important role in determining the recovery of aquatic communities, particularly at the two downstream stations. Researchers investigating responses to acidification in streams have also reported significant seasonal variation in ecological effects (Durance & Ormerod 2007; Kowalik et al. 2007; McClurg et al. 2007). Despite overall improvement in water quality, these episodic events may prevent sensitive species from recolonizing, thereby delaying recovery (Kowalik et al. 2007). Metal concentrations in the Arkansas River were consistently greater in spring than in autumn, and this pattern continued after completion of restoration activities. We hypothesize that the high metal concentrations at station AR3 in spring eliminated sensitive taxa, thereby delaying recovery. Seasonal variation in metals resulted from increased stream discharge and other episodic events during spring run-off, as others have reported for Rocky Mountain streams (Brooks, McKnight & Clements 2007). Model simulations of metal export to the Arkansas River showed significant movement of contaminated materials from the CG floodplain during high discharge events (Velleux et al. 2006).
The timing of these seasonal and episodic events relative to the presence of sensitive life stages complicates assessments of biological effects (Kowalik et al. 2007). Because tolerance of aquatic insects to metals generally increases with developmental stage (e.g. instar) (Kiffney & Clements 1996b), which varies seasonally, sensitivity to metals will be likely to differ among sampling dates. For example, field and microcosm experiments conducted seasonally showed significantly greater effects in summer when insect populations were dominated by early instars (Clark & Clements 2006). Seasonal changes in physicochemical factors that determine metal toxicity may also influence biological assessments. Dissolved organic carbon (DOC), an important constituent in aquatic ecosystems that reduces metal bioavailability, is greater in Rocky Mountain streams during spring run-off (Hornberger, Bencala & McKnight 1994; Clements et al. 2008). Because peak metal concentrations coincided with seasonal increases in DOC, organisms probably received some protection from toxic effects during this critical period.
Importance of long-term monitoring to evaluate restoration effectiveness
Previous studies have identified serious deficiencies associated with aquatic restoration programmes (Bernhardt et al. 2005; McClurg et al. 2007; National Research Council 2007). The primary criticisms of these programmes were inadequate study designs (e.g. lack of pre-treatment data), the narrow focus on physical and chemical characteristics relative to ecological measures, and a lack of agreement regarding what constitutes restoration success. Our ability to quantify the responses of stream communities to restoration is also impeded by the limited number of long-term studies that monitor communities after completion of restoration activities. Indeed, over 50% of the studies examined by Jones & Schmitz (2009) that failed to report recovery were simply not conducted for a sufficient duration to allow recovery to occur. A long-term perspective is necessary to identify potential shifts to alternative stable states following exposure to or recovery from a disturbance event (Robinson & Uehlinger 2008). The lack of long-term monitoring data also complicates our ability to separate responses to restoration from those associated with other stressors, particularly global change. Because of the relatively short duration (e.g. <9 years) of most ‘long-term’ macroinvertebrate studies (Jackson & Füreder 2006), aquatic ecologists are often unable to characterize the influence of climatic cycles such as El Nino/La Nina or Northern Atlantic Oscillation events. The increase in stream discharge and associated spike in metal concentrations that we observed in the mid-1990s coincided with one of the most severe El Nino events in the 20th century. It is unlikely that we could have identified the ecological consequences of this episodic event without a long-term perspective. These findings highlight the importance of assessing restoration success within the broader context of long-term changes in regional climate.
Has the Arkansas River recovered?
The answer to the fundamental question that motivated our research, has the Arkansas River recovered as a result of restoration, is complex and varied among response variables. Although the return to previous undisturbed conditions is a common restoration goal, the lack of pre-treatment data and the possibility that some systems may permanently shift to an alternative stable state may preclude this measure of restoration success. Despite significant decreases in aqueous metal concentrations in the Arkansas River, metals in sediment remained elevated. These findings suggest that controlling point sources of contamination in water does not necessarily translate into lower concentrations in other compartments (Burton 1992). Differences in the rate of recovery between water and sediments may partially explain the variation among biological responses (Clements 2004). For example, we speculate that organisms exposed to contaminants primarily by aqueous pathways recover faster than those that are exposed to metals through sediments or diet (e.g. grazing mayflies).
In addition to the conventional measures of biological integrity reported in this study, we suggest that susceptibility to novel perturbations is an important indicator of restoration effectiveness (Van Nes & Scheffer 2007; Thrush et al. 2008). Although rarely quantified in restoration studies, community resistance and resilience are fundamental components of ecological integrity (Karr & Dudley 1981). Because of the long history of metal contamination, susceptibility of Arkansas River communities to other stressors differs from that in uncontaminated streams. We have previously reported results of mesocosm experiments showing that despite tolerance to metals, macroinvertebrate communities from contaminated sites in the Arkansas River were more sensitive to acidification (Courtney & Clements 2000), UV-B radiation (Kashian et al. 2007; Clements et al. 2008) and stonefly predation (Clements 1999) compared to communities from reference streams. Other researchers have speculated that the loss of species in acidified systems increases susceptibility to climate change, UV-B and invasive species (Vinebrooke et al. 2003). The community-conditioning hypothesis (Matthews, Landis & Matthews 1996) has been proposed to account for the persistence of toxicant effects on communities long after a contaminant has degraded, and may explain differential responses to novel stressors between reference and impacted communities. Regardless of the underlying mechanisms, these patterns are consistent with the hypothesis that multiple perturbations often result in ecological surprises (Paine, Tegner & Johnson 1998). We suggest the susceptibility of macroinvertebrate communities to novel stressors is a useful measure of recovery and could be used to evaluate restoration success.
We are especially grateful to dozens of CSU undergraduates who have assisted with field sampling and laboratory sorting of macroinvertebrates over the past 17 years. Without the dedicated support of these students this long-term project could not be completed. D. Rees, D. Wade, T. Cady, L. Courtney, M. Pearson, R. Thorpe and K. Mitchell assisted with the identification of macroinvertebrates and B. Kondratieff verified many of our more difficult specimens. Thanks to G. Policky, S. Brinkman and P. Davies from Colorado Division of Wildlife (CDOW) for conducting the brown trout surveys and to D. Fey (USGS) for assistance with sediment collection. Comments by J. Carter and D. Carlisle on an earlier draft of this manuscript are greatly appreciated. Funding for sediment sampling and analysis was provided by the USGS, U.S. EPA and the U.S. Fish and Wildlife Service. Water quality and macroinvertebrate studies were funded by the U.S. EPA, CDOW, USGS and the National Institute of Health Sciences.