Ecosystem recovery in restored headwater streams: the role of enhanced leaf retention

Authors

  • Timo Muotka,

    Corresponding author
    1. Department of Biological and Environmental Science, University of Jyväskylä, PO Box 35, 40351 Jyväskylä, Finland; and Oulanka Biological Station, 93999 Kuusamo, Finland
      Timo Muotka, Department of Biological and Environmental Science, University of Jyväskylä, PO Box 35, 40351 Jyväskylä, Finland (e-mail tmuotka@dodo.jyu.fi).
    Search for more papers by this author
  • Pekka Laasonen

    1. Department of Biological and Environmental Science, University of Jyväskylä, PO Box 35, 40351 Jyväskylä, Finland; and Oulanka Biological Station, 93999 Kuusamo, Finland
    Search for more papers by this author

Timo Muotka, Department of Biological and Environmental Science, University of Jyväskylä, PO Box 35, 40351 Jyväskylä, Finland (e-mail tmuotka@dodo.jyu.fi).

Summary

  • 1There is controversy over how the success of ecological restoration should be measured. Traditionally, emphasis has been placed on species diversity and other community attributes, whereas the restoration of ecosystem processes has received less attention. Here, we combine replicated field experiments and a field survey to provide an ecosystem-level measure of stream restoration success.
  • 2Numerous headwater streams in Finland, and in many other parts of the world, have been channelized for timber transport, resulting in channels with simplified structure and flow. Recently, programmes have been launched to restore these streams to their pre-channelization condition. While the efficacy of restoration in improving fish habitat has been tested, little is known about effects on other stream biota or on the retention of leaf litter, despite its importance in trophic dynamics of forested headwater streams. Using a before-after-control-intervention (BACI) designed experiment with multiple reference and experimental streams, we examined restoration-induced changes in retention efficiency by conducting leaf-release experiments before (1993) and after (1996) restoration.
  • 3Substrate heterogeneity increased, but moss cover decreased dramatically following restoration. Retention efficiency in restored streams was higher than in channelized, but lower than in natural, streams. Algae-feeding scrapers were the only macroinvertebrate group whose density increased significantly after restoration.
  • 4Aquatic mosses were a key retentive feature in both channelized and natural streams, but their importance to retention was strikingly reduced by restoration. During restoration work, mosses are detached from large areas of the stream bed, exposing bare stone surfaces for colonization by periphytic algae.
  • 5A more effective restoration technique would involve the use of moss transplants, or the addition of large woody debris, to increase retentiveness and thus enhance the availability of organic material to benthic consumers. This case study on rivers illustrates how restoration projects benefit from an ecosystem perspective and from measures of ecosystem processes in assessing restoration success.

Introduction

During the last few decades, ecosystem restoration has attained a central role in natural resource management, and this trend has been paralleled by a dramatic expansion of restoration ecology as a scientific discipline (Young 2000). Although many practitioners still view restoration as a ‘poor second to the preservation of original habitats’ (Young 2000), its potential as a management tool, or even as a means of preventing the continual loss of biodiversity (Wilson 1992), has been widely acknowledged. Unfortunately, restoration decisions are often based on intuition rather than rigorous science. For example, although it is generally recognized that each restoration project should have a clearly defined goal, there is still considerable controversy over how to assess restoration success. Should it be evaluated through the measurement of species richness, diversity or other community attributes (= structural restoration endpoints), or should we aim to restore ecosystem processes (= functional endpoints) (Palmer, Ambrose & Poff 1997)? Assessing restoration success is not easily amenable to hypothesis testing using controlled experiments. At the scales involved, there are problems with balanced experimental design, lack of controls, and the time–course of experimentation (Michener 1997). Nevertheless, restoration provides an opportunity to monitor ecosystem-level, human-controlled changes at relatively large spatial or temporal scales, and therefore its potential contribution to testing general ecological theory is obvious (Bell, Fonseca & Motten 1997; Palmer, Ambrose & Poff 1997).

Factors threatening the biodiversity of headwater streams are numerous, but habitat loss and degradation probably rank highest (Allan & Flecker 1993). Channelization is one of the major causes of habitat degradation in headwater streams. Streams have been channelized for diverse purposes, but consequences for stream habitat structure and ecosystem functioning are the same: channelized streams have lost much of their natural heterogeneity, being characterized by simplified flow patterns, poorly retentive channels and weakening of the aquatic–terrestrial linkage (Petersen et al. 1987).

The biotic communities of headwater streams in temperate forested areas are heavily dependent on allochthonous detritus, which enters the stream mainly in the form of autumn-shed leaves. Once retained onto the stream bottom, leaves enter a processing sequence whereby they are integrated into detrital food webs. Many detritivorous invertebrates have their major growth period in winter, coinciding with the peak availability of well-conditioned leaf detritus (the ‘shredder response model’ of Cummins et al. 1989). However, the amount of benthic organic matter is not regulated solely by litter input, but also, and perhaps even more importantly, by the capacity of a stream to retain terrestrial inputs (Cummins et al. 1989). It has frequently been shown that debris dams dramatically alter the retention characteristics of a stream, thereby regulating the abundances of benthic organisms, especially detritivorous invertebrates (Bilby & Likens 1980; Angermeier & Karr 1984; Smock, Metzler & Gladden 1989; Trotter 1990; Ehrman & Lamberti 1992; Wallace, Webster & Meyer 1995a; Wallace et al. 1999). While most studies addressing the role of bed retentivity to stream ecosystem dynamics have been based on addition or removal of large woody debris or macrophytes (Koetsier & McArthur 2000), Dobson & Hildrew (1992) observed enhanced detritus availability on stream beds following the addition of small leaf traps. Dobson et al. (1995) suggested that such retention devices could be used as a management tool to increase invertebrate production of headwater streams. Although such small, non-woody, structures clearly enhance bed retentiveness, their potential use in the restoration of channelized headwater streams has remained largely unexplored.

In Finland, as also in vast areas of north-western Russia (Jutila 1992) and forested parts of northern USA and Canada (Sedell, Leone & Duval 1991), numerous streams have been dredged to facilitate timber floating. Water transport of timber ceased in Finland in the late 1970s, and extensive restoration programmes were launched thereafter to rehabilitate these streams to their original pre-channelization condition. Most of these streams are unregulated and, as most boreal rivers have restricted floodplains (Petersen, Gislason & Vought 1995), the return of a natural flow regime (Poff et al. 1997; Robertson, Bacon & Heagney 2001) is not an issue in their restoration. Instead, restoration mainly aims at providing better habitat for important game fish such as brown trout Salmo trutta L. It is usually conducted using a bulldozer, and the most commonly used restoration measures include installation of boulders, flow deflectors, cobble ridges and other enhancement structures. Furthermore, cobble-to-pebble sized stones are used to create nursery habitats for juvenile trout (Yrjänä 1998). While the efficacy of these measures for fish habitat improvement has been tested using hydraulic modelling (Huusko & Yrjänä 1997), very little is known about their long-term effects on other stream biota and ecosystem processes. Obviously, increased habitat heterogeneity should enhance the retentive capacity of a stream, which in turn should lead to increased densities of benthic organisms, especially those directly dependent on benthic detritus.

Our aim in this study was to examine restoration-induced changes in the retentive characteristics of boreal headwater streams. Leaf litter and its retention are critical for the energy flow and trophic dynamics of forest streams, providing an ecosystem-level perspective for our study. Specifically, we addressed the following questions. (i) Is the retentive efficiency of a stream enhanced by installing restoration structures? (ii) Are there any adverse effects of restoration related to changes in stream habitat structure? (iii) Is the effect of varying discharge on retention efficiency similar in channelized, restored and natural streams? (iv) Do benthic invertebrates respond to habitat restoration as predicted based on changes in retention efficiency? We believe that our results are not specific to boreal streams channelized for timber floating but, due to the potentially far-reaching effects of enhanced leaf retention on stream ecosystem functioning, they should be applicable to a much wider range of channelized streams or, more generally, streams with reduced substrate heterogeneity.

Methods

STUDY SITES AND STREAM HABITAT CHARACTERISTICS

The study was conducted in eight headwater streams in north-eastern Finland (65°35′ to 67°31′ N; 27°58′ to 29°31′ E). All drain forested lowland areas, and their riparian zones are dominated by deciduous trees, especially birch Betulapubescens Ehrh., alder Populustremula L., European aspen Alnusincana L. and willows Salix spp. They are second- to third-order streams with circumneutral, oligotrophic and often slightly humic water. Four of the streams (Kutinjoki, Kosterjoki, Loukusajoki and Poika-Loukusa) were heavily dredged in the 1950s to facilitate log transport. These streams were reconstructed by the Ostrabothnian Environment Centre in 1993 to restore their original heterogeneous bed structure. We selected these streams randomly from a large number of channelized headwater streams in northern Finland for which a restoration scheme was available by 1992. As references, we used four approximately similar-sized natural streams (Aventojoki, Kalliojoki, Merenoja and Putaanoja) from the same area, also selected randomly from a large pool of unmodified streams. The main difference between the two sets of streams was that the natural streams had undisturbed, highly heterogeneous, stream beds. A feature common to nearly all headwater streams in our study area was that their riparian zones had been under heavy forestry practices (e.g. timber harvesting and forest drainage), especially during the 1950s–1970s. The riparian zones of our study streams have now remained intact for about 30 years, which, however, is far too short an interval for streamside forests to mature. Thus, input of large woody debris has been minimal and the streams, including the ones with unaltered channels, contained very few debris dams. The term ‘natural’ is used here to refer to the stream channel, not the surrounding terrestrial landscape. Obviously, in a truly pristine condition, boreal forest streams should contain abundant debris dams. This is indicated by the fact that streams draining across the Finnish–Russian border typically harbour large woody debris (LWD) densities almost 20 times higher on the Russian side where the impact of forestry on riparian zones has been negligible (P. Liljaniemi, personal communication; Vuori, Luotonen & Liljaniemi 1999).

Restoration measures were similar in all four streams, consisting of placement of boulder dams and flow deflectors, and digging excavations. Boulder dams were constructed by setting boulders side by side across the river, and the inside of the structure was partly filled with cobbles and pebbles. Deflectors of approximately half the width of the channel were created to increase diversity in water velocities and to deepen the channel. Deflectors are known to be effective in guiding the current in places where water velocity exceeds 60 cm s−1 (Brookes 1988). The spawning habitats for salmonid fish were improved by using sorted gravel to create spawning grounds in suitable stream areas with relatively swift currents. Addition of woody debris is rarely used as a habitat improvement technique in Finland (Yrjänä 1998).

Because the construction of restoration structures was restricted by site-specific variation in the availability of suitable material along or close to stream edges, the resulting bed structure was not exactly the same among the restored streams. Therefore, to assess the within-group similarity of treatment and reference streams, and to determine whether they differed from streams in other groups, we measured several habitat characteristics within 50-m sections at each stream. We positioned transects perpendicular to the flow at 3-m intervals, and for each transect we recorded depth, water velocity, moss cover, substrate size and presence/absence of wood in 0·5-m intervals. To assess whether restoration enhanced bed heterogeneity, we quantified bed roughness (k) at each site using a contour-plotting device modified from Young (1993). The device was 50 cm long, consisting of a continuous row of measuring rods (diameter 0·8 cm). Measurements were made in 1·5-m longitudinal transects, each transect consisting of three successive 50-cm sections. To obtain a measurement, the device was pressed firmly against the bottom and distance from a horizontal support was measured for each rod. The standard deviation for the length of the rod below the support was calculated for each transect (Statzner, Gore & Resh 1988). Eight 1·5-m transects, distributed evenly across the study section, were established at each site. Distances from permanent reference marks on stream banks were used to ascertain that measurements before and after restoration were made at exactly the same positions. Relative bed roughness, k/D (roughness/depth), was used as an indicator of substrate heterogeneity (Gordon, McMahon & Finlayson 1992). Finally, stream flow was monitored in each stream through permanent gauging stations located at the same or a nearby stream.

LEAF-RELEASE EXPERIMENTS

To examine any changes in the retention capacity of a stream after restoration, we performed a set of artificial leaf releases. Experiments before and after restoration were conducted 3 years apart, 1–2 weeks before and 3 years after restoration (in September to early October 1993 and 1996, respectively). Leaf retention is known to be controlled by stream discharge, and even a slight variation in discharge may cause drastic changes in retention (Snaddon, Stewart & Davies 1992). Using stream flow data, we therefore conducted the releases at closely corresponding discharges in all restored streams on both years (Table 1). By contrast, discharge in two of the natural streams (Aventojoki and Putaanoja) was highly variable in September–October 1996, and therefore the experiment could be repeated in only two of the natural streams (Merenoja and Kalliojoki) in 1996 at discharges approximating those during releases in 1993. Due to inadequate replication, only data for the two streams where the experiment was repeated on both occasions were included in statistical analysis, but data for the other two streams are also reported for comparative purposes.

Table 1.  Habitat characteristics (mean and range; n = 4 for each stream type) of the study streams. Relative bed roughness is calculated as the ratio of bed roughness/water depth (k/D; Gordon, McMahon & Finlayson 1992). Stone size was measured as the largest stone diameter
 Stream type
 ChannelizedRestoredNatural
Discharge (m3 s−1)0·93 (0·64–1·11)0·90 (0·55–1·11)0·63 (0·35–1·10)
Stream width (m)6·0 (5·5–7·0)7·0 (6·0–8·0)5·5 (5·0–6·5)
Depth (cm)30 (26–32)33 (28–36)24 (20–30)
Current velocity (cm s−1)33 (25–37)31 (27–36)27 (22–31)
Stream gradient (%)0·64 (0·56–0·71)0·64 (0·56–0·71)0·74 (0·63–0·95)
Stone size (cm)16·7 (11·5–22·3)18·5 (12·3–31·8)29·8 (20·9–38·8)
Bed roughness (cm)5·8 (5·5–7·0)8·0 (6·7–9·5)9·1 (8·4–11·4)
Relative roughness0·20 (0·18–0·22)0·26 (0·23–0·28)0·38 (0·30–0·45)
Moss cover (%)35 (16–44)9 (5–19)47 (31–58)

A 50-m long study section was established at a relatively uniform riffle section of each stream. We first surveyed the number of suitable riffles in each stream (three to five per stream) and then randomly selected the site to be used for the experiments. We used red and yellow plastic strips (8 × 4 cm) as artificial leaves in our experiments. In preliminary trials, we used spray-painted leaves of birch and alder, but were unable to relocate most of the leaves retained by the substratum. We thus decided to use artificial leaves, because they are easier to find, have been used successfully in previous experiments, and are known to behave much like freshly fallen natural leaves entering a stream (Speaker et al. 1988). In each experiment, 2000 plastic strips were scattered across the width of the channel during a 5-min period at the upstream end of the study section. The downstream end of the section was blocked with a wire screen (mesh size 7 mm). Three hours after the release, we counted the number of leaves that had travelled through the study section and collected on the screen, and searched the entire reach for leaves that had been retained within the section. Speaker, Moore & Gregory (1984) and Petersen & Petersen (1991) have reported that the number of leaves in transport stabilizes within 2–3 h of release.

For each leaf found, we recorded the distance travelled (nearest 1·0 m). We also recorded the retaining object (boulder, cobble, gravel, coarse woody debris, moss, other aquatic vegetation, backwater, stream edge) for each leaf. The streams were nearly devoid of bole wood, and most of the woody material contributing to leaf retention was small twigs < 2 cm in diameter.

The effect of discharge on leaf retention was measured in one randomly selected stream for each stream group. Stream Kalliojoki represented a natural stream, Kosterjoki a channelized, and Kutinjoki a restored stream. In each of these streams, we conducted a leaf-release experiment with 2000 artificial leaves at four or five discharges, using the methods described above. The discharges for conducting the experiments were selected to represent the range of flows for each stream. In this experiment, only the number of leaves that collected on the downstream screen after 3 h was recorded.

BENTHIC SAMPLING

Macroinvertebrate samples were collected by kick-sampling (net frame 25 × 25 cm, mesh size 0·25 mm), and four timed samples were taken from each site. To enable comparisons of benthic densities among the stream types, we standardized our sampling effort as much as possible. The distance kicked along the stream was exactly 1·0 m, and the person taking the sample kicked the substratum vigorously for 60 s. All benthic collections were made by the same person (PL), and a field assistant ascertained that the 1-m line was not crossed. Samples were taken 1–2 weeks before, and again 3 years after, restoration, in both treatment and reference streams. The post-restoration samples were collected at the same time of the year as the pre-samples 3 years earlier (September–early October). Samples were preserved in 70% alcohol, and invertebrates were later sorted in the laboratory. Animals were identified to species or genus, and they were assigned to functional feeding groups according to Malmqvist & Brönmark (1985) and Merritt & Cummins (1996).

Our study design was a partial before-after-control-intervention (BACI)-type design (Underwood 1994): we had replicates for ‘control’ sites (here, the natural reference streams) and ‘impacted’ sites (the restored streams); however, our ‘before’ and ‘after’ samples were unreplicated through time. Therefore, because we lacked a true temporal trajectory of changes in benthic communities before and after restoration, it must be remembered that any differences between the before (1993) and after (1996) samples in the natural streams could potentially be due to other changes that happened to be coincident with the restoration works.

STATISTICAL ANALYSES

Leaf transport distances were fitted to the negative exponential model of Ld = L0e−kd, where L0 is the number of leaves released into the reach and Ld is the number of leaves in transport at distance d from the release point. The slope, −k, is the instantaneous leaf-retention rate and 1/k is the average distance travelled by a leaf in the stream before its retention (Speaker, Moore & Gregory 1984).

The aim of restoration is usually either to restore a system to its near-natural pre-disturbance state, or simply to improve its current condition. Thus, the key questions when testing for restoration-induced changes are: (i) was a response variable positively affected by restoration, and (ii) was the status of an undisturbed near-pristine system achieved? Therefore, we were mainly interested in two comparisons: differences between (i) channelized and restored streams; and between (ii) restored and natural streams, whereas comparison between channelized and natural streams was deemed less interesting. Our basic approach was to perform planned, a priori, comparisons using t-tests (Winer 1971; Day & Quinn 1989) for each response variable to test for the predetermined patterns of differences among our ‘treatment’ (stream) groups. These comparisons are non-orthogonal (restored streams are included in both comparisons) but, as argued by Winer (1971), ‘the meaningfulness of a comparison is more important than its orthogonality’. The response variables examined were: key habitat variables (substrate heterogeneity, moss cover), retention efficiency, retention coefficient and densities (log-transformed) of functional feeding groups of benthic macroinvertebrates. All statistical tests were performed using SPSS 7.5 for Windows (SPSS Inc. 1997).

Results

HABITAT CHARACTERISTICS

Stream discharge during the leaf-release experiments was closely similar in streams before and after restoration. The four natural streams, however, had slightly lower discharges (Table 1). This was unavoidable because practically all larger streams in our study area have been channelized for log floating, and only the smallest streams (generally less than 5 m wide) have been left undisturbed. Nevertheless, size differences among our streams were not too pronounced: the average discharge before and after restoration was 0·93 (1993) and 0·90 m3 s−1 (1996), respectively, while it was 0·63 m3 s−1 (1993) in the natural streams.

The most distinct differences in habitat structure between the stream types were related to substrate heterogeneity and moss cover (Table 1). Restored streams had significantly higher substrate complexity than channelized streams (planned comparison test, t = 2·46, P = 0·036, d.f. = 7), and did not differ from natural streams (t = 1·33, P = 0·217, d.f. = 7). Moss cover in restored streams was significantly lower than in both channelized (t = 3·67, P = 0·005, d.f. = 7) and natural streams (t = 3·92, P < 0·001, d.f. = 7). Overall, natural streams were characterized by relatively large heterogeneous substrates with extensive moss cover, whereas the channelized streams had fairly homogeneous substrates composed mainly of cobbles and small boulders. Restoration clearly enhanced substrate heterogeneity, approaching the level of natural streams. Moss cover in the channelized streams was, however, drastically reduced following restoration (Table 1). The two variables most relevant to leaf retention (moss cover, bed roughness) showed no temporal trends between the sampling occasions (P > 0·20 for both variables).

LEAF-RETENTION EXPERIMENTS

Restoration enhanced the retentive capacity of a stream to leaf litter inputs, although the pre- vs. post-restoration difference in retention efficiency was not quite significant (planned comparison test t = 2·06, P = 0·069, d.f. = 7). Natural streams retained leaves more efficiently than the restored streams (t = 3·87, P = 0·040, d.f. = 7) (Fig. 1a), but they showed considerable variation in their retention efficiency (Fig. 1b). Calculated across all sites, number of leaves retained was not significantly correlated with stream discharge (rs = 0·32, P = 0·310, n = 12) but was significantly correlated with substratum heterogeneity (rs = 0·98, P < 0·001, n = 12).

Figure 1.

Retention efficiency (percentage of leaves retained out of 2000; mean ±1 SE) of the three stream types (a), and of each study stream separately (b). Streams before and after restoration: Kutinjoki (KU), Loukusajoki (LO), Poika-Loukusa (PO), Kosterjoki (KO); natural streams: Kalliojoki (KA), Merenoja (ME), Putaanoja (PU), Aventojoki (AV). Sample size is four for each stream type, except for natural streams in 1996 where n = 2.

Leaf-retention curves conformed generally well to a negative exponential model, all coefficients of determination (r2) being > 0·90 (Fig. 2). Nevertheless, differences in retention patterns among the stream types were distinct. Retention rates were significantly higher in the natural (mean k ± 1 SE 0·0203 ± 0·0061, n = 2) than in the restored streams (0·0075 ± 0·0025, n = 4; planned comparison test t = 2·63, P = 0·027, d.f. = 7), and also in streams after restoration compared with before (0·0020 ± 0·00031, n = 4) restoration (t = 3·09, P = 0·013, d.f. = 7). In the channelized streams, leaves were retained at relatively uniform rates throughout the study reach, whereas in the natural streams and in some, but not all, restored streams, retentive structures were patchily distributed (Fig. 2). The average distance travelled by a leaf was almost an order of a magnitude higher in the channelized than in the natural streams (mean distance of 528·5 vs. 65·2 m, respectively), but intermediate (mean: 179·0 m) in the restored streams.

Figure 2.

Relationship between leaf retention (percentage of leaves in transport) and distance travelled from the release point in each study stream. The regression line represents fit to the negative exponential model. (a–d) streams before (BE, black dots) and after (AF, open dots) restoration; (e–h) natural streams. Due to extensive overlap, only data for 1993 have been presented for the two natural streams (e, f) where the release experiment was repeated in both years.

Cobbles and mosses were the most retentive structures in the channelized streams, whereas the role of mosses was negligible in the restored streams where the majority of leaves was retained by cobbles (Fig. 3). In the natural streams, mosses were the most important retentive feature, followed by boulders. The only significant difference between the stream types was found for mosses, which were less important in the restored than in the channelized (planned comparison test on arcsin-transformed data t = 2·74, P = 0·023, d.f. = 7) or natural streams (t = 2·32, P = 0·045, d.f. = 7). Woody debris, stream margins and aquatic vegetation (other than mosses) retained 15–20% of the leaves, with negligible differences between the stream types.

Figure 3.

The proportion of artificial leaves retained (mean percentage ±1 SE) by various retentive structures in each stream type (n = 4 streams per group).

EFFECTS OF DISCHARGE ON RETENTION EFFICIENCY

Linear regression models provided good fit to data on leaf-retention efficiencies at various discharges in all three streams studied (Fig. 4). Slopes of regression lines varied significantly among the streams (F2,8 = 8·35, P = 0·011). Retention efficiency decreased with increasing discharge more dramatically in stream Kalliojoki (a natural stream) than in the other two streams (Tukey’s test for multiple comparisons P < 0·05), whereas slope for the restored stream (Kutinjoki) did not differ significantly from that of the channelized stream (Kosterjoki; Tukey’s test P > 0·05).

Figure 4.

Relationship between stream discharge and retention efficiency in a natural (Kalliojoki), channelized (Kosterjoki) and restored (Kutinjoki) stream.

BENTHIC MACROINVERTEBRATES

The only feeding group whose densities increased significantly after restoration was algae-scraping invertebrates, with densities almost three times higher in the restored than in the channelized streams (Fig. 5; planned comparison test t = 2·40, P = 0·048, d.f. = 7). Densities of detritivores (shredders, filterers and collector-gatherers) changed little during the 3-year period between the pre- and post-restoration samples. Densities in natural streams did not differ between the years (paired sample t-tests on log-transformed data, P > 0·30 for all groups). All feeding groups had a tendency towards maximum density in the natural streams, but due to large among-stream variation differences between the restored and the natural streams were significant only for shredders (t = 2·89, P = 0·037, d.f. = 7) and scrapers (t = 3·41, P = 0·019, d.f. = 7) and nearly so for predators (t = 2·23, P = 0·061, d.f. = 7).

Figure 5.

Densities of macroinvertebrate feeding groups in the pre-(1993) and post-(1996) restoration samples, and in the natural streams in the same years. Values are means (±1 SE) of four streams per stream type.

Discussion

The in-stream enhancement structures used for habitat restoration in our study streams caused a general increase in the capacity to retain leaf litter. Nevertheless, after 3 years, restored streams did not approach the retentive efficiency of natural streams. Compared to most previous studies, our channelized streams showed extremely poor retentivity, which was clearly related to their highly simplified bed structure. Our figure of 8% of leaves retained was much lower than the 40% reported by Petersen & Petersen (1991) for channelized agricultural streams in southern Sweden. However, our natural streams were also less retentive than theirs, and the difference between these two studies is mainly explained by the fact that our streams were larger, with higher discharges, than the headwater streams studied by Petersen & Petersen (1991).

Leaf transport curves conformed well to the negative exponential loss model, although a linear model produced nearly as good a fit for all channelized and most restored streams. This means that retentive structures were much less patchily distributed in the channelized and the restored streams than in streams with unmodified bed structure. A highly patchy distribution of retention sites is typical of streams with debris dams, which contribute markedly to the distribution of organic matter, resulting in extreme patchiness of community and ecosystem functions in these streams (Smock, Metzler & Gladden 1989). Although debris dams occurred rarely in our study sites, our natural streams were characterized by the presence of a few exceptionally retentive patches (Fig. 2). This suggests that, in the absence of large woody debris, other bed structures assume a key role in the retention process of unmodified headwater streams. The only striking difference in the relative importance of retention structures among our stream types was the reduced importance of mosses in recently restored streams, compared with channelized and natural streams. It may indeed be that mosses are the key retentive feature in many headwater streams lacking woody debris, although differently sized stones are also important. Overall, our results suggest that structures used for stream enhancement in Finland do not effectively mimic the physical complexity of naturally retentive stream channels.

It is well known that stream discharge greatly influences retention efficiency: the higher the discharge, the less retentive a stream is to allochthonous inputs (Speaker et al. 1988; Jones & Smock 1991; Snaddon, Stewart & Davies 1992). According to our study, channelized, restored and natural streams differ in how tightly leaf retention is linked to discharge. The channelized stream retained leaves ineffectively at all discharges, whereas the natural stream was highly retentive at base flows, but much less so at higher discharges. A corresponding observation was made by Webster et al. (1987) in a set of laboratory trials where discharge and substrate were directly manipulated. They found that smooth surfaces with little structural complexity were ineffective in trapping seston on all discharges, whereas increased discharge greatly reduced retention on more complex (artificial turf, gravel) substrates. It must be noted, however, that the effect of discharge on retention is closely linked to water depth, because the probability of a leaf coming into contact with the substrate decreases with increasing depth (Webster et al. 1994). Our natural streams were generally shallower than the other streams, and therefore we cannot exclude water depth as a partial explanation for the observed differences in leaf-retention rates.

Because most CPOM (Coarse Particulate Organic Matter) export occurs during major storms (Wallace et al. 1995b), the timing and extent of spates may be critical determinants of organic matter storage in streams (Jones & Smock 1991; Maridet et al. 1995). The flow regime of boreal streams typically exhibits an annual peak during spring, with a secondary peak in late autumn, during or immediately after leaf fall (Haapala & Muotka 1998). Thus, the retention efficiency of channelized and natural (or restored) streams may differ least during the period of major leaf input, resulting in less distinct differences in resource availability for benthic consumers than could be expected based on differences in retention potential (Laasonen, Muotka & Kivijärvi 1998). The lack of debris dams may further reduce differences in organic matter storage among streams with differing bed retentivity, because stones are effective in leaf retention only at low discharges, whereas the role of debris dams increases with rising discharge (Smock, Metzler & Gladden 1989; Jones & Smock 1991; Raikow, Grubbs & Cummins 1995). It must be stressed, however, that this part of our study was unreplicated, and therefore any generalizations must await further experimentation.

The dramatic loss of mosses during restoration works may have far-reaching effects on benthic communities and ecosystem processes of these headwater streams. A recurring theme in stream ecology has been a consistent relationship between retention efficiency and abundance of detritivorous invertebrates (Rounick & Winterbourn 1983; Smock, Metzler & Gladden 1989; Prochazka, Stewart & Davies 1991; Dobson & Hildrew 1992; Wallace, Webster & Meyer 1995a). Therefore, considering the increased bed complexity and a corresponding change in retention capacity, one would have expected an increase in the importance of detritivores following stream habitat restoration. This, however, did not happen, and the only significant change was an increase in the abundance of algae-scraping invertebrates. We believe that this result is connected to the removal of aquatic mosses during restoration works. Channelized woodland streams characteristically support abundant moss growth (Laasonen, Muotka & Kivijärvi 1998), probably because timber floating mainly ceased in the 1960s, and mosses have had enough time to recolonize these streams. Mosses clearly are key retentive features in headwater streams devoid (or nearly so) of macrophytes and debris dams. However, during restoration works mosses are detached from large areas of the stream bed, and bare stone surfaces are exposed for colonization by periphytic algae. Thus, the resource base for benthic consumers shifts from terrestrially derived detritus to autochthonous production by benthic algae, with a concomitant increase in scraper abundance. It is probable that, as mosses recover, retention efficiency of these streams will increase, and benthic communities will gradually shift to detritivore-dominated assemblages dependent on allochthonous organic material. The effect of mosses on stream ecosystem dynamics goes far beyond their role in particle retention, however. For example, they afford optimal ‘nursery’ habitats for the early instars of many benthic invertebrates (Suren & Winterbourn 1992). By altering near-bed flow regimes (Nikora et al. 1998), mosses provide benthic animals with hydraulic refugia where environmental conditions remain essentially unchanged during high-flow events (Lancaster & Hildrew 1993). Unfortunately, little is known about the colonization and growth rates of aquatic bryophytes, but it can be safely assumed that the full recovery of large canopy-forming species (e.g. Fontinalis spp. and Hygrohypnum spp.) most effective in particle retention will take many years, if not decades (Muotka et al., in press). This suggests that if the recovery of benthic communities and ecosystem processes to pre-channelization state is to be enhanced, current restoration practices must be modified. A more successful rehabilitation scheme might be achieved by relatively simple measures, however, for example by using moss transplants or, better still, leaving relatively large areas of the stream bed untouched to serve as colonization centres for mosses after restoration.

Given the above reasoning, one might ask why, then, are the densities of scrapers even higher in natural streams, although these generally support much higher densities of mosses than restored streams? This is probably also connected to the prominent role of mosses in organic matter retention in natural forest streams. Mosses retain not only leaf litter, but also fine suspended particles, thus providing rewarding feeding arenas for deposit-feeding invertebrates. Many of the invertebrates categorized as scrapers are secondarily collector-gatherers (e.g. Baetis and Heptagenia mayflies), feeding on fine organic material deposited on the stream bed (Huhta, Muotka & Tikkanen 1995; Merritt & Cummins 1996). Woodland stream ecosystems are largely fuelled by inputs of organic matter from the terrestrial environment, and it may well be that these invertebrates attain a different functional role (i.e. collectors instead of scrapers) in natural forest streams with abundant moss cover.

If the desired goal of restoration is to increase system productivity by enhancing a stream’s retention capacity (improvement of the present, degraded status of the system), this can be achieved by a slight modification of current restoration practices. However, if the more demanding goal of mimicking the pre-disturbance near-natural status is to be achieved, more radical measures will be needed. It may be that such fundamental changes cannot be achieved without the use of large woody debris to enhance the retention efficiency and habitat diversity of previously debris-free channels. No other natural in-stream structure contributes as importantly to retention efficiency as do debris dams, especially during high discharges. As the effects of detritus manipulation are known to propagate up in lotic food webs (Wallace et al. 1997), even single-interest restorations aimed at improved salmonid fisheries may ultimately benefit from restoration measures that enhance the availability of organic material to benthic consumers.

In conclusion, our study shows that in-stream restoration does enhance stream bed complexity, but this comes with a cost: the use of heavy machinery during restoration works caused a drastic reduction of moss biomass. Because aquatic mosses are clearly a key retentive feature in boreal headwater streams, the retentive capacity of a stream following restoration does not increase as much as could be expected based on enhanced substrate heterogeneity. The retentive potential of a stream will probably improve as mosses recover, but because mosses are slow-growing plants this may take years or even decades. Therefore, from an ecosystem perspective, better results might be achieved by using softer technology, for example by adding moss transplants or large woody debris to stream channels, to enhance the transference of terrestrially derived organic matter to detritus-based food webs. More generally, the ecosystem-level perspective adopted in this study might serve as a basis for an effective assessment of stream restoration. The use of replicated field experiments with multiple reference sites could aid in placing restored sites along a recovery gradient from channelized (or otherwise degraded) to near-pristine streams. In this respect, our method resembles the ‘start site–target site’ approach of Mitchell et al. (1999) for assessing restoration of heathland vegetation. An important aspect of our approach is that it combines well-targeted field experiments (leaf releases) with a field survey of stream habitats and macroinvertebrate communities, to provide a combined measure of restoration success. Indeed, as suggested by Manel, Buckton & Ormerod (2000), such a combination of large-scale surveys and site-specific process-orientated studies may provide the most powerful tool available for ecologists to assess large-scale anthropogenic changes on ecosystems.

Acknowledgements

We thank M. Erkinaro and P. Hilkuri for help in the field and Oulanka Biological Station for logistical support. We also acknowledge the two anonymous referees for their constructive comments on an earlier draft of the manuscript. This study is part of the Finnish Biodiversity Research Program (FIBRE), and it was financed by the Academy of Finland (grant numbers 35586 and 39134 to T. Muotka) and the Olvi Foundation (to P. Laasonen).

Ancillary