Responses of riparian plants to flooding in free-flowing and regulated boreal rivers: an experimental study

Authors

  • M. E. Johansson,

    Corresponding author
    1. Landscape Ecology Group, Department of Ecology and Environmental Science, Umeå University, SE-901 87 Umeå, Sweden; and
    Search for more papers by this author
  • C. Nilsson

    1. Landscape Ecology Group, Department of Ecology and Environmental Science, Umeå University, SE-901 87 Umeå, Sweden; and
    2. Department of Natural and Environmental Sciences, Mid Sweden University, SE-851 70 Sundsvall, Sweden
    Search for more papers by this author

Dr M. E. Johansson, Landscape Ecology Group, Department of Ecology and Environmental Science, Umeå University, SE-901 87 Umeå, Sweden (fax + 46 90 7867860; e-mail mats.johansson@e.g.umu.se).

Summary

  • 1The long history of river regulation has resulted in extensively changed ecosystem structures and processes in rivers and their associated environments. This fact, together with changing climatic and hydrological conditions, has increased the need to recover the natural functions of rivers. To develop guidelines for river restoration, comparative ecological experiments at contrasting water-level regimes are needed. We compared growth and survival of transplanted individuals of four riparian plant species (Betula pubescens, Carex acuta, Filipendula ulmaria and Leontodon autumnalis) over 2 years on four free-flowing and four regulated riverbank sites in northern Sweden. The species were chosen as representatives of dominating life-forms and species traits on different elevations of the riverbanks.
  • 2In Betula and Filipendula, mean proportional growth rates were significantly higher at free-flowing sites than at regulated sites, whereas no consistent differences between free-flowing and regulated sites were found in Carex and Leontodon. Differences among species were generally in accordance with natural distribution patterns along riverbank elevation gradients and with experimental evidence on flooding tolerance, although plants of all species survived and even showed positive growth rates on elevations below their natural range of occurrence.
  • 3Partial least squares regression was used to relate plant performance (growth and survival) to duration, frequency and timing of flooding at the different sites. Flood duration and frequency typically reduced performance in all species and during all time periods, although to various degrees. Flood events early in the experiment determined the outcome to a high degree at all sites. Variables indicating a regulated regime were mostly negatively related to plant performance, whereas free-flowing regime variables were positively related to plant performance.
  • 4We used two of the regression models generated from our data with an acceptably high predictive power to simulate a hypothetical re-regulation scenario in run-of-river impoundments. With an overall reduction in flooding duration and frequency of 50–75%, plant performance of Filipendula at low riverbank elevations showed predicted increases of about 20–30%, levelling off to zero at the highest elevations. Reductions in summer floods represented about one-third to half of this increase.
  • 5We conclude that for a range of species individual plant performance is clearly reduced on banks of impoundments and storage reservoirs due to changes in the water-level regime. Furthermore, our model simulation suggests that rather substantial reductions of flood duration and frequency are needed to improve plant performance on riverbanks upstream from dams in impounded rivers. River restoration principles should, however, be based on a combination of experimental data on plant performance of individual species and observed long-term changes in plant communities of regulated rivers. Consequently, successful re-regulation schemes in boreal rivers should include both reductions of summer and winter floods as well as re-introduced spring floods.

Introduction

An estimated two-thirds of the fresh water flowing to the oceans is obstructed by more than 840 000 dams (Petts 1984; McCully 1996). In the United States, Canada, Europe and republics of the former USSR, 85 of the 139 largest rivers, representing 77% of the total water discharge, are strongly or moderately regulated and fragmented by dams (Dynesius & Nilsson 1994). In Latin America, Africa and South-East Asia, impact conditions are similar (Revenga et al. 2000). These hydrological alterations have changed ecosystem structures and processes in running waters and associated environments all over the world. For example, riparian plant communities in regulated rivers often have a lower richness and density of species and reduced plant cover compared with free-flowing rivers (Nilsson et al. 1991a; Nilsson, Jansson & Zinko 1997; Jansson et al. 2000). Plant dispersal is constrained by dams, the flora is fragmented, and species composition altered (Friedman et al. 1998; Johnson 1998; Andersson, Nilsson & Johansson 2000; Jansson, Nilsson & Renöfält 2000; Merritt & Cooper 2000). These ecological changes are strongly associated with changes in flow and water-level regimes (e.g. Jansson et al. 2000). During recent years, the importance of natural flooding regimes for maintaining river ecosystems has become widely appreciated (Poff et al. 1997; Puckridge et al. 1998; Tockner et al. 1999; Robertson, Bacon & Heagney 2001) and artificial flooding has been used as an important means for river rehabilitation (Barinaga 1996; Wuetrich 1996). These activities increase the demands for more detailed guidelines on how to rehabilitate rivers, including the identification of bottlenecks for riparian plant recruitment, i.e. what plant life stages in different species are most affected by changes in flow variation. It is known, for example, that required water levels for successful establishment may be quite different from optimum conditions for subsequent survival and growth (e.g. Mahoney & Rood 1998) and that the response to water levels may vary among species. However, very few comparative studies between free-flowing and regulated rivers have been made to determine experimentally the influence of contrasting water-level regimes on plant performance.

To study how duration, frequency and timing of flooding influence plant performance in different water-level regimes, we grew plants of four representative but contrasting riparian plant species on riverbanks of free-flowing and regulated rivers in northern Sweden and recorded their growth and survival during 2 years. We hypothesized that plants would have lower rates of growth and survival at regulated sites than at free-flowing sites due to longer, more frequent and ill-timed flooding events.

Methods

study sites

To reflect different hydrological regimes of free-flowing and regulated rivers in northern Sweden, eight riverbank sites within an area of 300 × 100 km were chosen for study. The sites were situated in: (i) two free-flowing small rivers with a mean annual discharge < 50 m3 s−1 and with a pronounced but rather short spring flood (hereafter denoted FS1 and FS2); (ii) a large free-flowing river with a significantly greater discharge and longer spring flood than the smaller rivers (FL1 and FL2); (iii) two run-of-river impoundments with an extremely variable water level throughout the year (RI1 and RI2); and (iv) two storage reservoirs with a flood during the latter part of the summer (SR1 and SR2; Figs 1 and 2). The regulated sites were all situated within the Ume River system (15–20° E, 64–66° N), where the main channel has been transformed into a chain of consecutive reservoirs and impoundments with virtually no unimpounded reaches. They were also all within the damming area of the nearest downstream dam, which means that influences of water-level on plant performance can be regarded as upstream effects.

Figure 1.

Map of the study area in Northern Sweden indicating the location of study sites in two small free-flowing rivers (FS1 and FS2), a large free-flowing river (FL1 and FL2), and two run-of-river impoundments (RI1 and RI2) and two storage reservoirs (SR1 and SR2) in a regulated river.

Figure 2.

Hydrologic data from (a) free-flowing and (b) regulated study sites. Left columns show water-level variation above lowest recorded values during the study period. Greyed areas define the sections used for the experiment and are also equal to the naturally vegetated portions of the riverbanks. Right columns show variation in cumulative duration (solid line) and frequency (dashed line) of flooding on the experimental sections of the gradients (greyed areas in left column). Relative riverbank elevation is defined as actual elevation divided by total used riverbank elevation and expressed as the percentage of ‘elevation’ below maximum water level. Note the changes in y-axis scales. All variables describe the experimental period, June 1993 to September 1994, and are derived from data supplied by the Swedish Meteorological and Hydrological Institute, Båkab Energi AB, and Nordkraft Service.

To accurately estimate the water-level variation during the study period, the criteria for choosing a site were that it should be located close to a daily recording water-level (stage) gauge, and that no tributaries larger than second order streams should enter the stretch between the study site and the water-level gauge. Finally, to reduce any influence of environmental factors other than water-level variation, the riverbanks should be gradually sloping and only moderately exposed to waves and currents.

At all sites, upland vegetation consisted of mixed coniferous forest. The soils were composed of fine sediments and sand except at the storage reservoirs where stone and gravel dominated a 1–2 m broad, wave-eroded, zone slightly below the maximum damming level. The riverbank vegetation was generally dense and continuous at the free-flowing sites but sparse and patchy at the regulated sites. Further environmental characteristics of the sites are summarized in Table 1.

Table 1.  Environmental characteristics of the study sites located in two small free-flowing rivers, the Sävar River (FS1) and the Öre River (FS2), a large free-flowing river, the Vindel River (FL1 and FL2), two run-of-river impoundments in the Ume River (RI1 and RI2), and two storage reservoirs in the upper reaches of the Ume River (SR1 and SR2)
  Regulated
 Free-flowingRun-of-river impoundmentsStorage reservoirs
 SmallLarge
 FS1FS2FL1FL2RI1RI2SR1SR2
  • *

    Data from the Swedish Meteorological and Hydrological Institute.

  • Negative value due to recharge of water to the system from a hydroelectric pump station.

  • Total height and width of the riverbank was defined as the range between average highest and average lowest water level over a 10-year period preceding the experiment.

  • §

    The part of the riverbank used in the experiment, which was equal to the part of the riverbank that had a continuous vegetation cover of non-aquatic species.

Annual discharge at study site (m3 s−1)*
Mean  7 31176123221207   28 117
Mean minimum  2  6 31 15 23 21  −27   0·3
Mean maximum 31237909806473447   92 374
Distance to water-level gauge (km)  0·7  0·5  4·5  5·8  7  7   24  26
Length of growing season (d)133143133123138128  113 123
Height above sea level (m)176 55160342176252  412 395
Riverbank elevation (m)
Total  1·41  2·46  4·02  2·98  0·98  0·98   14·16  20·05
Used§  1·15  1·80  2·45  1·41  0·49  0·50    1·90   1·95
Riverbank width (m)
Total 17 49 29 65  9 26> 100> 10
Used§ 11 48 28 34  4  7   14  21
AspectNENENENESWNENWSW

study species

The vegetation on riverbank elevation gradients in boreal rivers is organized in a characteristic zonation of life forms, species traits and species groups. Typically, the zonation proceeds from a riparian forest at the top of the riverbank to a Salix-dominated shrub vegetation, then to a forb- and grass-dominated zone at mid-elevation, and a graminoid belt dominated by Carex spp. towards the average summer low-water level (Nilsson 1999). For this study, we selected four plant species that were representative of dominating life-forms and species traits within each of these different sections of the gradient. They were: (i) Betula pubescens Ehrh., a deciduous tree that often dominates the riparian forest; (ii) Filipendula ulmaria (L.) Maxim., a tall herb commonly found at mid-elevations; (iii) Leontodon autumnalis L., a low rosette species occurring mostly on disturbed patches around mid-elevations; and (iv) Carex acuta L., a tall graminoid forming dense stands on the lowermost parts of the riverbank. These species are all common in the riverbank flora of the studied rivers. Mean species frequency of occurrence among previously studied reaches of these rivers ranges from 0·37 for Leontodon to 0·97 for Betula (determined across 25 200-m sections in each river and expressed on a 0–1 scale; Nilsson et al. 1991a,b). All species are perennial but differ in time to reproductive maturity. In Betula, maturation normally exceeds 10 years, in Filipendula 5 years, and in Carex 2 years, whereas Leontodon normally flowers during the first year. The responses are therefore representative for juvenile stages of all species except Leontodon.

experimental procedure

The plants used in the experiment were raised from seeds originating from a riverbank site at the Ume River about 10 km from the mouth. Seeds were harvested from 10 individuals of Betula and approximately 100 individuals of each of the other species. The seeds were mixed and sown in trays filled with potting soil in late March 1993. In mid-May, individual seedlings of each species were transferred to 0·5-L black plastic pots filled with a soil mixture consisting of 20% topsoil, 30% glaciofluvial silt and 50% sand. The plants were grown in a glasshouse with ambient light conditions, temperature ranges of 12–20 °C by day, and 8–12 °C by night, daily watering, and no fertilizing. One week before transplantation to the experimental sites, 30 individuals of each species were measured for a number of morphological traits, harvested, dried and weighed. The dry masses were then regressed on the different measured traits and the best trait for each species was chosen for the estimation of initial dry mass of the transplants. For Betula, initial biomass was estimated by the equation dry mass = 0·0017 (plant height) – 0·010 (F = 42·15, P= 0·0003, r2 = 0·61); for Carex, dry mass = 0·035 (number of leaves) – 0·11 (F = 65·69, P < 0·0001, r2 = 0·70); for Filipendula, dry mass = 0·0043 (rosette diameter) – 0·17 (F = 69·19, P < 0·0001, r2 = 0·71); and for Leontodon, log dry mass = 1·31 (log rosette diameter) – 3·30 (F = 93·60, P < 0·0001, r2 = 0·78). These traits were measured on all plants on the day before transplantation into the field.

The water-level range and the vegetation-covered part of the riverbank varied greatly among sites (Fig. 2 and Table 1). To make the comparison reasonable, the experiment was made only within the range of the riverbank gradient that had a more or less continuous vegetation cover of non-aquatic riparian plant species, i.e. within the range of potential survival and growth of transplants. In the free-flowing rivers, the upper limit of the experimental gradient equalled the level of the average spring flood maximum, 1982–91. At the regulated sites, the upper limit was equal to the legislated maximum damming level. Along a 40-m stretch of the riverbank, 40 elevations, randomly pre-selected within the determined elevational range, were located to the closest centimetre with a levelling instrument. At each of these locations, single transplants of all species were planted at least 20 cm apart and with the substrate level of the pot at ground level. Tree and shrub branches immediately overtopping the planting sites were removed, whereas ground floor vegetation was left intact. Due to extended high water levels during the first summer, the transplantation of the lowermost eight groups of plants at site FL1 had to be postponed by 1 month. During this time, the plants were temporarily placed on adjacent non-flooded ground. This delay was corrected for in the calculations of water-level variables (see below).

To control for differences in plant performance among sites due to factors other than water-level variation, 20 groups of pots were also planted on non-flooded ground immediately above the riverbank. One-quarter of the experiment (10 of 40 flooded groups and 5 of 20 terrestrial groups) was harvested at the end of the first growing season. The remaining plants were harvested at the end of the second growing season. The harvested plants were partitioned into above- and below-ground biomass, dried at 60 °C for 72 h, and weighed. Survival of all plants was recorded in autumn 1993, spring 1994 and autumn 1994. A plant was recorded as dead when no living shoots (apical or lateral) were found.

data analysis

The response variable, plant performance, was expressed as the growth rate, r, defined as

r = [ln(M1/M0)]/d(eqn 1)

where M0 is the initial plant biomass and M1 the harvested plant mass. The time period in days between the start and the end of the experiment, d, only included days within the growing seasons (see below). As mortality was substantial in some species and at some sites, dead plants were also included in the analyses. If a plant was found dead in autumn 1993 or spring 1994, it was assigned the minimum recorded r-value of the living plants harvested at that site in 1993, and, similarly, if it was recorded dead at the end of the experiment, it received the minimum value of the plants harvested in 1994. If the minimum values were positive, the r-value was set to zero. The measure r is thus incorporating both growth and mortality and is therefore referred to as plant performance. Alternatively, growth and mortality could have been analysed separately, but mortality was so unequal between species and sites that this would have restricted the analysis to only growth or only mortality for certain species and sites. Potential shortcomings with this procedure are discussed below. In analyses, where data from different sites were pooled (see below), we used proportional growth rate (rp) to decrease variability among sites due to site-intrinsic factors not related to water-level variation. This was defined as

rp = (r − rt)/rt(eqn 2)

where r is as defined above, and rt is the mean growth rate of the plants grown on non-flooded ground immediately above the riverbank.

The predictor variables described actual water-level variation, riverbank elevation and type of water-level regime. For each plant location on the riverbanks, the duration and frequency of flooding were calculated from the water-level data. For a given riverbank elevation, the duration was expressed as the cumulative number of flooded days and the frequency as the cumulative number of flood events. The effects of the timing of flood events were evaluated by calculating flood duration and frequency for five non-overlapping time periods during the experiment delimited by climatic data from stations as near as possible to the study sites. Summer year 1 was defined as the period from the date of planting to the end of the first growing season; autumn year 1, from end of summer year 1 to the establishment of a permanent ice-cover; winter, from end of autumn year 1 to the start of spring flood; spring year 2, from end of winter to the date with a daily mean air temperature > 10 °C; and summer year 2, from end of spring year 2 to the date of harvest, which equalled the end of the second growing season. The calculations were based on daily water-level recordings, except for the run-of-river impoundments, where hourly recordings were used because of the large within-day variation. We also added a standardized measure of riverbank elevation as a predictor variable to examine if plant growth was influenced by factors that were not directly related to the water-level variables but still varied with riverbank elevation. This relative riverbank elevation was defined as the actual elevation of a plant on the riverbank divided by the total used riverbank elevation (see Table 1) and expressed as the proportional ‘elevation’ below the defined maximum water level. In models fitted for the pooled data set (see below), we included dummy variables that indicated type of water-level regime, i.e. small free-flowing, large free-flowing, run-of-river impoundment, and storage reservoir, respectively. In this way, residual variation, including possible regulation effects that were not included in the measured water-level variables, could be evaluated. We first included both linear and quadratic variables in the models but quadratic variables did not improve fit and were not ranked among the most influential variables in any model. Therefore, only linear variables were included in the final models.

Several of the predictor variables were highly correlated. For example, the correlation between the first and second summer-flood duration ranged from 66% in one of the storage reservoirs to 99% in the run-of-river impoundments (not shown). Even when data from all sites were pooled, correlations between water-level variables were in many cases very strong. Strong collinearity among independent variables in a ‘traditional’ least-square multiple regression leads to unreliable estimates of the regression parameters and to models that are difficult to interpret. To solve this, one can either choose to omit variables that are almost completely expressed in others, with the apparent loss of ability to evaluate all individual variables, or one can use alternative statistical methods that allow collinearity. We chose to analyse the data with partial least square (PLS) regression (Geladi & Kowalski 1986; Martens & Naes 1989), which combines regression and ordination and which is increasingly used in analyses of biological data (e.g. Zhang, Malmqvist & Englund 1998). The method circumvents the problem with highly correlated predictor variables by projecting the measured x-variables onto a few ‘latent’ variables or components, rather than by selecting among variables as in stepwise multiple regression. These components are then used as independent variables in a regression. The number of extracted components and the statistical significance in the resulting models are determined by cross-validation through jack-knifing (Wold 1978). Three measures of validity of the models are generated: (i) R2X, the proportion of the variance in the matrix of predictor variables that is used in the models; (ii) R2Y, the proportion of variance in the response variable that is explained by the model (corresponds to the multiple correlation coefficient, R2); and (iii) Q2, the proportion of the variance in the response variable that can be predicted by the model. The relative influence of each predictor variable in the model can be expressed as the variable importance in the projection, VIP, which is the sum of the variable influence over all model dimensions divided by the total explained variation by the model (Fridén, Koivula & Wold 1994). Numerically, the squared sum of all VIPs is equal to the number of predictor variables in the model. This measure enables a direct comparison of the degree of influence of each variable in a model and those with VIP > 1 are considered the most relevant to explain variation in the response variable. Prior to the PLS-analysis, all variables were standardized to unit variance (i.e.  = 0 and SD = 1) by subtracting the mean and multiplying by the inverse of the standard deviation. This procedure is recommended when no information is available on the relative importance of the different variables. To decrease deviations from normal distribution in the data, all predictor variables were also (log10 + 1)-transformed. After analyses, we examined the data for outliers and their influence was evaluated by removing them to see if the results were affected. In the results, predictor variables are ranked in order of importance as indicated by their VIP-values and their influence (size and sign) is expressed as the centred and standardized regression coefficient. A model was considered significant when Q2 > 0·097 (Fridén, Koivula & Wold 1994).

The data were analysed in two steps. First, the data from all sites were pooled to determine the overall influence of duration and frequency of flooding on each species. This was done separately for plants harvested in autumn 1993, which were influenced by water-level fluctuation during the first growing season only, and for plants harvested in the second year, which represented the cumulative impacts from both 1993 and 1994. This analysis also evaluated effects of a specific water-level regime, including possible regulation effects, that could not be ascribed to water-level variation expressed in the measured variables. Secondly, to determine site-specific effects, we analysed the second-year data set for each species and site, separately. First-year sample sizes were too small for this purpose.

The PLS regressions were made with the software SIMCA-S for Windows, version 5·1 (Fridén, Koivula & Wold 1994), and all other analyses were made with SPSS for Windows, version 9·0 (SPSS Inc. 1999).

Results

hydrological variation

The 1993 summer was rainy, resulting in a high mean water level and recurrent floods in the free-flowing rivers. In 1994, precipitation was moderate and water levels gradually receded throughout the summer in the free-flowing rivers. In the Vindel River, for example, mean summer discharge (May–September) was almost twice as large in 1993 (484 m3 s−1) as in 1994 (263 m3 s−1). Long-term mean summer discharge (1961–90) was 234 m3 s−1 (SMHI 1995). The storage reservoirs were filled later than normal in 1994 and in SR2 less than half of the storage capacity was used. This resulted in such a low water level in SR2 during 1994 that none of the plants in the experiment were flooded (Fig. 2). Plant performance in SR2 could therefore only be analysed against water-level variation in 1993 in the site-level analysis. The run-of-river impoundments had a more or less regular daily water-level fluctuation throughout the study period.

Flood duration showed an almost linear decrease with increasing riverbank elevation at all sites, regardless of water-level regime (Fig. 2). However, the average cumulative duration over the entire study period differed considerably (27–59 days at free-flowing sites and 73–264 days at regulated sites).

At the large free-flowing sites and the impounded sites, flood frequency was highest at intermediate elevations. In the small free-flowing rivers, frequency was highest at lower elevations, whereas in the reservoir sites, frequency was highest at higher elevations (Fig. 2). There were relatively few flood events at the free-flowing and reservoir sites (range: 0–9), whereas flood frequency was extremely high in the impoundments (maximum: 220 and 396).

plant performance

Both mortality and growth rates differed considerably between species and sites (Fig. 3). In Betula, about 30 days of flooding greatly reduced survival and growth. At the regulated sites, most plants died during the first summer, whereas at the free-flowing sites, cumulative mortality gradually increased (FS2 excepted). In Carex, little influence of flood duration could be noticed. Filipendula and Leontodon were also able to survive relatively long periods of submersion, but growth and survival clearly decreased for cumulative flooding over 200 days. Reproduction only occurred in Leontodon and flowering individuals appeared at all sites but were generally few, ranging from 5% at the reservoirs to 20% at FS1.

Figure 3.

Growth rates in relation to cumulative duration of flooding in Betula pubescens, Carex acuta, Filipendula ulmaria and Leontodon autumnalis transplants grown for 15 months at two sites in small free-flowing rivers (FS1 and FS2), two sites in a large free-flowing river (FL1 and FL2), two run-of-river impoundments (RI1 and RI2), and two storage reservoirs (SR1 and SR2). Filled circles represent plants surviving the entire experiment. Open circles are plants that died during the experiment and for which growth values were assigned as described in Methods. Note differences in y-axis scales.

In all species, there was a significant between-site variation in the average proportional growth rate of flooded plants (P < 0·001, one-way anova; Table 2). In Betula and Filipendula, proportional growth rates were significantly higher at three of the four free-flowing sites than at any of the regulated sites (P < 0·05; Table 2). In Carex and Leontodon, no consistent significant differences between regulated and free-flowing sites were found (Table 2).

Table 2.  Proportional growth rate, rp, of transplants grown for 15 months at the eight study sites. Values are given ± 1 SE. Values with different superscripts are significantly different (P < 0·05) according to the Tukey's HSD multiple comparison test
  Regulated
 Free-flowingRun-of-river impoundmentsStorage reservoirs
 SmallLarge
 FS1FS2FL1FL2RI1RI2SR1SR2
Betula pubescens−0·11a−0·95ab−2·00b−0·45ab−5·82cd−1·39ab−4·20c−6·46d
±0·26±0·63±0·43±0·09±1·09±0·12±0·84±1·19
Carex acuta 0·01a−0·38b−0·04b−0·06b−0·16b−0·04b−0·34b−0·33b
±0·04±0·22±0·05±0·04±0·08±0·05±0·10±0·06
Filipendula ulmaria 0·12a−0·02ab−0·24ab−0·05ab−0·80c−0·20ab−0·57bc−0·67c
±0·05±0·10±0·05±0·05±0·16±0·09±0·09±0·12
Leontodon autumnalis−0·16a−1·27ab−1·04ab 0·04a−0·67ab−1·22ab−1·23ab−0·56ab
±0·06±1·08±0·28±0·09±0·24±0·29±0·18±0·11

The variation in root : shoot ratios did not show any consistent trends along the flooding gradients for any of the species and will consequently not be treated further.

plant performance vs. water-level variation

In this section, we summarize the results from the PLS-regression. For significant models (Q2 > 0·097), we account for the most influential variables as indicated by their VIP-values (see Methods). We first describe the analysis of the pooled data sets for each species, and then the site-level analysis for each species based on the 2-year data set.

Although variation was large within the pooled data sets, significant models were obtained for all species in both the 1-year (Fig. 4a) and the 2-year (Fig. 4b) data sets, except for Carex in the 1-year data set. Flood duration and frequency typically influenced growth negatively in all species and during all time periods, although to various degrees of importance. The duration of the first summer flood was generally the most influential water-level variable. The dummy variable indicating reservoirs was negatively related to plant performance for all species, both in the 1-year and the 2-year data sets, and was ranked among the three most important variables for Betula, both after 1 and 2 years, and for Carex and Filipendula after 2 years. The impoundment variable was positively related to plant performance in Leontodon in the 2-year data set and negatively related in all other models but was in no case ranked among the most important variables. Variables indicating free-flowing regimes were positively related to plant performance in all species, and the small free-flowing regime was notably important for Betula, both after 1 and 2 years, and for Carex and Filipendula after 2 years. Riverbank elevation was always positively related to plant performance but was ranked as highly influential only in Leontodon in the 2-year data set.

Figure 4.

Influence of regulation type and water-level variation (duration, frequency, and timing of flooding) on the performance of four riparian plant species. Results are illustrated for (a) the 1993 data sets (n = 80) and (b) the 1994 data sets (n = 240). Note differences in x-axis scales. The models were generated with partial least squares regression analysis. The seven most influential variables are included and black bars indicate variables with VIP-values > 1. Three measures of the validity of the models are given: R2X, the proportion of the variance in the matrix of predictor variables that is used in the models; R2Y, the proportion of variance in the response variable that is explained by the model (corresponds to the multiple correlation coefficient, R2); and Q2, the proportion of the variance in the response variable that can be predicted by the model. Models in the 1993 data set were fit on the first component only, and models in the 1994 data set on the first two components. A model was considered significant when Q2 > 0·097.

The site-level analysis identified several differences between the species and between the water-level regimes. In Betula (Table 4a), negative impacts of flood duration and frequency during the early phase of the experiment were most important in explaining variation in performance at the small free-flowing sites as well as in one of the storage reservoirs, SR1. At the large free-flowing sites, performance was largely explained by a positive influence of riverbank elevation, together with negative effects of second-year flooding variables. In both impoundments, negative impacts from most variables were indicated and winter flood frequency was ranked as the most important.

For Carex (Table 3b), no clear relationships were found between water-level variation and plant performance. Only at one of the storage reservoirs, a weakly significant model indicated the importance of second-summer flooding (negative effects) and riverbank elevation (positive effect).

Table 3.  Influence of water-level variation on performance of (a) Betula pubescens, (b) Carex acuta, (c) Filipendula ulmaria and (d) Leontodon autumnalis transplants grown for 15 months at the eight study sites. Values are standardized regression coefficients in models generated with partial least squares regression analysis. The analyses are based on the 1994 data sets (n = 30). The most influential variables (VIP > 1) are printed in bold face. Three measures of the validity of the models are given: R2X, the proportion of the variance in the matrix of predictor variables that is used in the models; R2Y, the proportion of variance in the response variable that is explained by the model (corresponds to the multiple correlation coefficient, R2); and Q2, the proportion of the variance in the response variable that can be predicted by the model. All models were fit on the first component only, and a model was considered significant when Q2 > 0·097
Variable    Regulated
Free-flowingRun-of-river impoundmentsStorage reservoirs
SmallLarge
FS1FS2FL1FL2RI1RI2SR1SR2
(a) Betula pubescens
Flood duration
Summer 1993 −0·15 −0·23  0·08 −0·03 −0·06 −0·070·13 −0·10
Autumn 1993 –  0·04 – – −0·06 −0·08−0·14 –0·05
Winter 1993–94 – – – – −0·07 −0·08 –
Spring 1994 −0·09 −0·14 −0·17 −0·16 –0·06 −0·07 –
Summer 1994 −0·12  0·00 −0·17 −0·14 –0·06 −0·06−0·09 –
Flood frequency
Summer 1993 −0·17 −0·25  0·07 −0·02 −0·05 −0·08 0·08  0·09
Autumn 1993 –  0·04 – – −0·05 −0·08−0·12 –0·07
Winter 1993–94 – – – – −0·07 −0·09 –
Spring 1994 −0·08 −0·18 −0·14 −0·16 –0·06 −0·07 –
Summer 1994 −0·13 –0·02 −0·15 –0·08 −0·06 −0·07−0·09 –
Riverbank elevation  0·13  0·11  0·15  0·17  0·04  0·04 0·10  0·06
R2X  0·72  0·70  0·67  0·59  0·83  0·81 0·73  0·73
R2Y  0·57  0·63  0·60  0·37  0·35  0·52 0·40  0·10
Q2  0·54  0·62  0·57  0·31  0·14  0·39 0·19<0·097
(b) Carex acuta
Flood duration
Summer 1993  0·10  0·04  0·04  0·09 −0·05 −0·040·07  0·04
Autumn 1993 – −0·07 – – −0·05 –0·04−0·09  0·04
Winter 1993–94 – – – – −0·05 −0·04 –
Spring 1994  0·03 −0·03 −0·05  0·04 −0·05 –0·04 –
Summer 1994  0·12 –0·05 −0·08 −0·06 −0·04 −0·04−0·12 –
Flood frequency
Summer 1993  0·04 −0·04  0·03  0·07 –0·01 −0·04 0·04 −0·02
Autumn 1993 – −0·07 – – −0·01 −0·03−0·03  0·09
Winter 1993–94 – – – – −0·03 −0·02 –
Spring 1994  0·02 −0·02 −0·06  0·08 –0·03 −0·05 –
Summer 1994  0·11 –0·04 −0·10 –0·02 −0·01 −0·040·12 –
Riverbank elevation −0·10  0·06  0·03  0·03  0·06  0·05 0·10  0·01
R2X  0·71  0·69  0·66  0·35  0·80  0·82 0·71  0·72
R2Y  0·22  0·10  0·11  0·04  0·12  0·14 0·26  0·03
Q2< 0·097< 0·097< 0·097< 0·097< 0·097< 0·097 0·18< 0·097
(c) Filipendula ulmaria
Flood duration
Summer 1993  0·03 −0·07  0·05  0·06 −0·08 −0·08−0·10 –0·11
Autumn 1993 – −0·11 – – −0·08 –0·060·10 –0·12
Winter 1993–94 – – – – −0·08 –0·06 –
Spring 1994  0·06 –0·06 −0·10  0·06 −0·08 −0·08 –
Summer 1994  0·04 −0·11 −0·11 –0·02 −0·08 −0·08−0·10 –
Flood frequency
Summer 1993 −0·02 −0·05  0·04  0·06 –0·06 −0·06 0·04 −0·14
Autumn 1993 – −0·11 – – −0·02 −0·01 0·05  0·13
Winter 1993–94 – – – −0·06 −0·01 –
Spring 1994  0·08 –0·05 −0·08  0·09 –0·06 −0·07 –
Summer 1994  0·05 −0·10 –0·07  0·06 –0·03 −0·06−0·09 –
Riverbank elevation −0·04  0·11  0·07 −0·02  0·08  0·09 0·13 –0·11
R2X  0·67  0·70  0·67  0·34  0·82  0·79 0·72  0·74
R2Y  0·05  0·46  0·20  0·04  0·51  0·45 0·29  0·27
Q2< 0·097  0·31  0·12< 0·097  0·48  0·44 0·23  0·11
(d) Leontodon autumnalis
Flood duration
Summer 1993 −0·13 −0·12  0·10 –0·03 −0·08 −0·060·04  0·11
Autumn 1993 – −0·07 – – −0·08 –0·05−0·08  0·10
Winter 1993–94 – – – – −0·08 –0·05 –
Spring 1994  0·08 −0·09 −0·09  0·12 −0·08 −0·06 –
Summer 1994 −0·14 –0·09 −0·10  0·06 −0·07 −0·07−0·13 –
Flood frequency
Summer 1993 −0·08 −0·12  0·09 –0·04 −0·03 −0·04 0·04 −0·07
Autumn 1993 – −0·08 – – −0·01 −0·01−0·03  0·10
Winter 1993–94 – – – – −0·05 −0·01 –
Spring 1994  0·07 −0·08 −0·08  0·13 –0·05 −0·05 –
Summer 1994 −0·13 –0·10 −0·03  0·08 −0·01 −0·050·12 –
Riverbank elevation  0·05 0·11  0·10 –0·03  0·10  0·08 0·11 –0·07
R2X  0·56 0·70  0·67  0·64  0·80  0·79 0·70  0·74
R2Y  0·20 0·54  0·22  0·18  0·36  0·24 0·22  0·15
Q2  0·18 0·48  0·15< 0·097  0·30  0·18 0·14< 0·097

Also in Filipendula (Table 3c), plant performance was generally weakly related to the examined variables. At the impounded sites, however, positive effects of riverbank elevation and negative effects of flood duration during most time periods were clearly important.

In Leontodon (Table 3d), models for the small free-flowing sites indicated the importance of negative impacts of summer flooding in both years. In the impoundments, positive effects of riverbank elevation and negative effects of flood duration during the entire experiment were important. At large free-flowing sites and storage reservoirs weak or non-significant models were obtained.

Discussion

The analysis of the pooled data sets demonstrated that the explained fraction of the variation in plant performance was largely related to water-level variation (Fig. 4). The general implication of this result is that reduced duration and frequency of flooding would improve plant performance, irrespective of water-level regime. This prediction is obviously only valid over shorter time-spans and not in the long term, as the studied species are all common along riverbanks of free-flowing rivers. Short-term negative effects of flooding on plant performance are outweighed by long-term positive effects, such as reduced interspecific competition (e.g. Malanson 1993).

The models also indicated that plant performance in storage reservoirs was poorer than could be explained by water-level variation only. This can probably largely be attributed to direct (on plants) and indirect (on substrate) wave disturbance. Wave and ice erosion can strongly limit shoreline vegetation (Keddy 1983), especially in impoundments and reservoirs (Nilsson & Keddy 1988; Hellsten & Riihimäki 1996). Contrary to the free-flowing rivers, the regulated sites are flooded also during the winter, when the river is ice-covered. This increases the total annual flood duration and the risk for ice-erosion of the shoreline area. This is especially pronounced in the impoundments, with their high flood frequency, indicated by our results on Betula in the impounded sites, where winter flood frequency was the principal limiting factor (Table 3a). However, we used relatively sheltered study sites in the regulated rivers, and our models therefore probably do not fully account for the effects of wave and ice erosion usually occurring in regulated rivers.

Another general observation was that plants of all species survived and even showed positive growth rates on elevations below their natural range of occurrence (Wassén 1966; Nilsson 1983; M. Johansson & C. Nilsson, personal observations). This was especially pronounced at the regulated sites but also was apparent at the free-flowing sites and may have several explanations. First, this pattern could arise if water levels were lower than normal during the study period, but this was only the case in the storage reservoirs during the second year (SMHI 1995; Fig. 2). Alternatively, and more likely, it could imply that it takes more than 15 months of influence from flooding before a successfully established individual is eventually eliminated. The observed species distributions are the results of long-term water-level variation and need not coincide with plant performance registered during 2 years. Furthermore, it may illustrate that plant colonization and establishment are the stages most sensitive to flooding (Rood & Mahoney 1990; Krahulec & Lepš 1994; Lenssen, Menting & Blom 1998). In this study, plants were 11–12 weeks old when transplanted to the field sites, implying that the plants may have passed the most critical life stage and therefore could survive below their natural distribution ranges. However, the results clearly showed that flood events early in the experiment were very influential. Among the evaluated water-level variables, first summer flood duration was ranked as the most important in all models generated from the pooled data sets, although the high water levels in the first growing season probably reinforced the negative influence of this variable. Finally, survival and growth outside natural distribution ranges might be an effect of interspecific competition being reduced because plants were grown in pots.

The timing of floods is critical for successful plant performance and most species tend to have a lower flooding tolerance during growth periods (e.g. Siebel & Blom 1998). Our results support this view by the fact that duration and frequency of summer flooding were the most influential variables in all models generated from the pooled data sets and in 65% of the significant models in the site-level analysis.

The riverbank represents an obvious environmental gradient as flood duration decreases almost linearly with increasing elevation. Nevertheless, there are other factors varying along this gradient in patterns that may reinforce or reduce the effects of flooding. These factors include light and nutrient availability, soil texture, and biotic properties of the soil (Menges & Waller 1983; Nilsson & Wilson 1991; Lenssen et al. 1999; Miller 2000). The riverbank elevation variable in our models may indicate, when significant, the importance of such factors (Table 3). Alternatively, the water-level variables included in the models may not have fully accounted for all direct effects of flooding. Regardless, plant performance was generally very variable along the elevational gradients and the models were only able to account for about half this variation at the best, suggesting that environmental heterogeneity along the studied gradients was high.

Our measure of plant performance incorporated both responses in growth and mortality because we included dead plants in the analyses and assigned those with minimum values of recorded relative growth rates. This procedure may be questionable if mortality patterns along the riverbank elevation gradient differ notably from growth patterns, i.e. if flooding affects plant survival differently than it affects plant growth or if other disturbance factors, e.g. herbivores or pathogens, act selectively on certain riverbank elevations. We found no indication of such complicating patterns in our results. On the contrary, at elevations with a high incidence of mortality, growth was also low (Fig. 3), which supported our treatment of the data.

There are also certain site- or species-specific phenomena that have to be considered when evaluating the results. First, the strong negative influence of first-summer flood frequency on Betula in the small free-flowing rivers is probably a threshold effect. Only one or two flood events occurred during the first summer and almost all the plants experiencing these floods either died or had growth rates close to zero (Fig. 3). Secondly, floods deposited considerable amounts of silt at the study site FS2, which buried some groups of plants at lower elevations. This made plants perform less well than if flooded only and explains the relatively low average plant performance at FS2 compared with other sites (Fig. 3). Thirdly, at SR2, no flooding occurred in the second year. Plant performance could therefore only be related to water-level variation in 1993 and the resulting models have low validity (Table 3).

The observed differences in plant performance among species were generally in accordance with observed natural distribution patterns along riverbank elevation gradients (Wassén 1966; Nilsson 1983; M. Johansson & C. Nilsson, personal observations) and with experimental evidence on flooding tolerance (see below). Betula pubescens was least tolerant to flooding and is also well known to show reduced growth and survival when exposed to moderate degrees of flooding (e.g. Rinne 1990; Frye & Grosse 1992). Carex acuta, on the other hand, showed little or no response to flooding and is reported to tolerate extended periods of flooding and be able to grow in constantly anoxic soil, provided leaves are at least temporarily emergent (Van den Brink et al. 1995). Filipendula ulmaria is generally regarded as a flood-tolerant species, and can endure prolonged anoxic conditions (Barclay & Crawford 1982). It does not, however, produce aerenchyma when flooded (Smirnoff & Crawford 1983) and endures oxygen deprivation in a quiescent state (Braendle & Crawford 1987; Studer-Ehrensberger, Studer & Crawford 1993). In agreement with these findings, we observed that Filipendula survived long periods of submersion almost without growth. Leontodon autumnalis behaved similarly, but responses to flooding have, to our knowledge, not been experimentally studied in this species. However, it is generally regarded as a ruderal typically found on disturbed ground. It was the only species for which the model on the pooled second-year data set did not include a significant negative influence by reservoirs but did include a positive effect from impoundments (Fig. 4). Also, proportional growth rates of Leontodon did not differ significantly between free-flowing and regulated sites (Table 2). This indicates that it may have a relative advantage at regulated sites due to its ruderal habit and low stature, compared with free-flowing sites, where it is probably suppressed by a denser and taller riverbank vegetation.

management implications: a simple simulation example

There are numerous examples of models predicting vegetation changes following modifications of hydrological regimes in regulated rivers (Rood & Mahoney 1990; Auble, Friedman & Scott 1994; Richter et al. 1997; Hill, Keddy & Wisheu 1998; Friedman & Auble 1999). These models are usually generated from long-term data on hydrological variation and vegetation dynamics. Our study is an attempt to downscale re-regulation effects to individual plants. Individual plant performance is admittedly too simplistic to be used as an indicator of ecosystem health, but our approach provides an opportunity to precisely evaluate effects of different water-level variables on the target species and relate the response to species traits. As our statistical models showed, however, the predictive precision at the level of individual plants was rather low. Environmental variation within sites always produced a rather high fraction of unexplained variation. Nevertheless, to provide an indication of the direction and magnitude of predicted change after a hypothetical re-regulation, we selected the two models on plant performance with the highest predictive power among all models from the impounded sites and used them to simulate such a scenario. The chosen models were those of Filipendula at the impounded sites with Q2-values of 0·48 and 0·44, respectively (Table 3c).

First, we tested the effects of reducing overall duration and frequency of flooding on the impounded sites during a 15-month period to a maximum of 200 days and 100 events, respectively, by multiplying input values of the water-level variables to the models with a factor of 0·5 or 0·25. Secondly, we simulated the same reduction of duration and frequency of summer floods only. With an overall flood reduction, performance of Filipendula at low riverbank elevations showed increases of about 20–30%, levelling off to zero at the highest elevations (Fig. 5). Reductions in summer floods represent about one-third of the increase in RI1 and half of the increase in RI2.

Figure 5.

Predicted changes in performance of Filipendula ulmaria in the run-of-river impoundments (RI1, upper graph, and RI2, lower graph) following simulated reductions in flood duration and frequency. The predictions are restricted to the conditions and time-span modelled by the data in this study.

This simulation indicates that rather substantial reductions of flood duration and frequency are needed to improve plant performance on riverbanks of impounded boreal rivers. It should be emphasized that this prediction is formally restricted to young established plants of Filipendula and to the environmental setting and time-span for which the models were generated. Obviously, the response to flood reduction will vary between species depending on their flood tolerance, e.g. the least flood tolerant species should have the highest potential to increase its performance. The combined result of individual species’ responses and competitive interactions between species (Keddy, Twolan-Strutt & Wisheu 1994), will determine the succession of the riparian plant community following flood reduction.

As often stated, however, water levels (and flows) also need to be high enough to maintain the ecology of a river channel in the long term (e.g. Poff et al. 1997). Flood reduction, used in isolation, may lead to an encroachment of the riparian forest onto lower, herb and grass dominated parts of the riverbank. In practice therefore successful re-regulation schemes should include both flood reduction (summer and winter) and re-introduced spring flood (e.g. Nilsson 1996). Moreover, in areas where rivers pass through arid regions, high flows during the establishment phase may be essential for recruitment of riparian plants (Mahoney & Rood 1998). A reduced flow for increased growth and survival and a periodically increased flow to promote establishment or to prevent forest encroachment are different aspects of the same management concept that is sometimes called environmental flows (Petts 1996; Gippel & Stewardson 1998). It is, however, complicated to set realistic values of water-level variation and flow for a re-regulation, as they are highly constrained by economical, practical and legislative limitations (Nilsson 1996). There is also an important time component in resolving these constraints, especially for the storage reservoirs. As time goes by, erosion of the riparian zone will continue, further reducing the plant cover (Nilsson, Jansson & Zinko 1997). Restored water levels and flows would thus need to be accompanied by costly reapplication of lost sediments in the riparian zone.

Acknowledgements

We thank M. Danvind, D. Evander, K. Hägglund, M. Karlsson, P. Karlsson, U. Larsson, A. Rydh, M. Svedmark, S. Xiong and U. Zinko for practical help at various stages of this project. Stewart Rood and anonymous referees provided helpful comments on the manuscript. Water-level data were obtained from the Swedish Meteorological and Hydrological Institute, Båkab Energi AB, and Nordkraft Service. Financial support was provided by the Swedish Environmental Protection Agency, the Swedish National Board for Industrial and Technical Development (NUTEK), and Vattenfall AB.

Ancillary