- Top of page
- Supplementary material
- Supporting Information
Regulation and fragmentation of river–floodplain systems by dykes and dams, to control flooding, facilitate navigation and make use of hydroelectric power, have altered the flow patterns of nearly all of the major river–floodplain systems in Europe and throughout the northern Hemisphere (Petts 1984, 1989; Dynesius & Nilsson 1994; Nilsson & Berggren 2000; Tockner & Stanford 2002). River regulation measures have strong impacts on riparian vegetation. River damming severely alters species richness and composition, because of the change in hydrological regimes and disruption of the longitudinal flow (Auble, Friedman & Scott 1994; Nilsson & Jansson 1995; Nilsson, Jansson & Zinko 1997; Jansson, Nilsson & Renöfält 2000; Merritt & Cooper 2000). The loss and degradation of natural floodplains caused by dykes (also called levees) are considerable. These structures are usually several metres high and run parallel to the course of a river, preventing flooding of the adjoining countryside as well as restricting the river to its main course. The renewal of fluvial landforms, an important process in natural floodplains, is constrained. Nearly all large rivers and their floodplains in central Europe and the USA are affected by dykes, with an estimated 40 000 km of dykes in the USA alone (Johnston Associates 1989). In central Europe, the floodplain landscapes of the Upper and Lower Rhine River, the Lower Danube River and the Middle Elbe River, for example, have lost more than three-quarters of their natural inundation area as a result of dykes (Dister 1991; Schneider 1991; Dahl & Flade 1994). The ecological impacts of dykes have largely been ignored, although recent studies have demonstrated that separating floodplains from their river channels results in a profound change in floodplain structure, process and species composition (Marston et al. 1995; Girel, Garguet-Duport & Pautou 1997; Trémolières et al. 1998; Deiller, Walter & Trémolières 2001; Gergel, Dixon & Turner 2002; Leyer 2004).
The main effect of a dam is to disrupt longitudinal river flow, whereas dykes disrupt the hydrological connectivity between river and floodplain. Both structures alter natural water level fluctuation patterns: the seasonal flow variability and peak flows are considerably reduced. Therefore, river regulation measures are invariably accompanied by declining water level fluctuations. The alternating wet–dry cycle of periodically flooded areas is of great importance for riparian and lake vegetation (Crawford 1996; Davis et al. 1996; Gowing & Spoor 1998; Hill, Keddy & Wisheu 1998). Few studies, however, have analysed quantitative data on water level fluctuations in relation to riparian vegetation patterns, and there is an almost complete lack of knowledge on specific plant species’ responses to reduced water level fluctuations. A pioneering study by Johansson & Nilsson (2002) compared the growth and survival of transplanted individuals of four riparian plant species on free-flowing and regulated riverbanks in Sweden. They found a clear reduction in the performance of plants growing on banks of regulated rivers because of changes in the water level regime.
This study investigated the response of floodplain species to reduced water level fluctuations on a gradient from highly fluctuating to stable water tables. Highly fluctuating regimes are typical of natural floodplains, being characterized by a considerable flooding depth in the high water season and a dry period with a strong decrease in groundwater in the low water season. Stable conditions are typical of floodplains that have been isolated from the river flow by dykes. Hydrological parameters, including average groundwater level, flooding duration and flooding depth, as well as disturbance and dispersal processes, will influence the response patterns of riparian vegetation. The aim of this study was to predict shifts in floodplain vegetation in response to human-caused alterations in floodplain landscapes and to provide information for an assessment of river regulation and floodplain restoration measures.
- Top of page
- Supplementary material
- Supporting Information
Hydrological variables differed between the recent and older floodplains (Table 1). The recent floodplain varied widely in average groundwater level and had a lower median value than the older floodplain (U-test, P < 0·0001). Furthermore, the recent floodplain was characterized by significantly higher fluctuating water tables (U-test, P < 0·0001) and a greater number of water levels higher than 50 cm above the soil surface (U-test, P < 0·0001). There was no difference in the flooding duration of the floodplain types.
Table 1. Comparison of hydrological variables between the floodplain types ‘recent’ and ‘older’ calculated over the time period 1997 and 1998
|Hydrological variable||Average groundwater level (cm)||Size of water level fluctuations (cm)||Duration of flooding (days year−1)||Water levels > 50 cm (days year−1) |
|Minimum||−298||−149||50·7||16·4|| 0|| 0|| 0|| 0|
|25 percentile value||−200|| −87||60·8||29·8|| 6·5|| 0|| 2|| 0|
|Median||−134|| −63||70·4||38·3|| 17|| 30|| 9|| 0|
|75 percentile value|| −75|| −39||77·1||40·0|| 49|| 71|| 17|| 0|
|Maximum|| 15|| −4||78·5||64·7||204||187||114||21|
Within the 182 plots, 234 plant species were identified. Species richness did not exceed 49 species 12-m−2 plot, with a median value of 22. The frequency in occurrence of the 30 species that were selected for analysis ranged between 20% and 68%, with a median value of 60%.
Sigmoid or unimodal models of logistic regression were developed for each species in relation to the average groundwater level (AGL). Most of the species tested were strongly related to this parameter (Table 2). Only Rumex acetosa showed no significant relation; however, in combination with the water level fluctuation gradient (WLF), multiple logistic regression detected the independent importance of AGL, even for Rumex acetosa. For all species, the explained deviance varied widely and accounted, on average (median), for about 20% of the total deviance. The explained deviance significantly increased if the water level fluctuation parameter (in its linear form) was also included in the model. Only five species (Carex praecox, Glechoma hederacea, Vicia cracca, Poa trivialis and Rumex thyrsiflorus) were not affected by the water level fluctuation gradient, even in those cases where it was only added in the regression model. The occurrences of these species seemed to be independent of water level fluctuation changes, and were therefore not considered in further analysis. The explained deviance of the combination AGL + WLF (with or without an interaction term consisting of both parameters) for all other species ranged from 14% to 59% of the total deviance and had a median value of about 33%.
Table 2. Results of logistic regression analyses for species in relation to AGL and the combination with WLF (AGL + WLF). AUC values and classification rates based on popt and pfair are displayed as criteria for the predictive power of the models. ***P < 0·001; **P < 0·01; *P < 0·05
| ||Training data set||Testing data set|
|Total deviance||AGL explained deviance (%)||AGL + WLF explained deviance (%)||AUC (CI)||Correct (%, popt)||Correct (%, pfair)|
|Achillea millefolium||102·9||27·0***||49·2***||0·91 (0·83, 0·96)***||91·2||80·2***|
|Agrostis capillaris||116·8||39·4***||48·6**||0·94 (0·89, 0·98)***||86·8||85·7***|
|Agrostis stolonifera||112·3||16·8***||20·0*||0·73 (0·64, 0·83)**||73·6||72·4***|
|Alopecurus geniculatus|| 93·3||32·8***||54·0***||0·90 (0·85, 0·96)***||83·5||82·4***|
|Alopecurus pratensis||108·9||21·0***||31·4*||0·85 (0·76, 0·93)***||80·2||79·1***|
|Cardamine pratensis||105·0||13·3***||23·1**||0·79 (0·68, 0·88)***||75·8||72·5***|
|Carex acuta||105·0||33·2***||36·7*||0·85 (0·75, 0·92)***||80·2||79·1**|
|Carex vulpina|| 87·7||25·5***||33·1**||0·84 (0·72, 0·95)***||84·6||78·0***|
|Cerastium holosteoides||118·0||10·1**||14·0*||0·84 (0·76, 0·91)***||78·0||73·6***|
|Cnidium dubium||100·7||10·3**||17·5*||0·84 (0·74, 0·93)***||84·6||82·4***|
|Deschampsia cespitosa||102·9||16·2***||32·9***||0·85 (0·76, 0·92)***||85·7||79·1***|
|Elymus repens||126·1||33·1***||38·2*||0·92 (0·85, 0·97)***||86·9||86·9***|
|Festuca pratensis||107·0|| 5·6*||19·4***||0·83 (0·72, 0·92)***||82·4||78·0***|
|Galium verum agg.|| 93·2||39·3***||49·5**||0·93 (0·86, 0·97)***||87·9||86·8***|
|Holcus lanatus||102·9||19·7***||35·6***||0·81 (0·65, 0·91)***||90·1||70·3**|
|Persicaria amphibia||105·0||17·8***||26·3**||0·85 (0·77, 0·93)***||82·4||72·5***|
|Phalaris arundinacea||121·3||20·8***||25·3*||0·79 (0·69, 0·88)***||75·8||75·8***|
|Plantago intermedia|| 78·1|| 6·2*||15·0**||0·78 (0·67, 0·89)***||84·6||71·4**|
|Poa palustris||116·8||15·7**||29·1***||0·85 (0·75, 0·90)***||76·9||74·7***|
|Poa pratensis agg.||125·3||46·4***||59·0***||0·88 (0·79, 0·95)***||84·6||84·6***|
|Potentilla anserina||112·3||29·8***||33·2*||0·89 (0·86, 0·91)***||84·6||83·5***|
|Ranunculus repens||122·2||20·5***||45·8***||0·82 (0·72, 0·90)***||79·1||73·6***|
|Rorippa sylvestris|| 92·3||11·3**||35·8***||0·85 (0·75, 0·92)***||81·3||78·0***|
|Rumex acetosa|| 95·9||– NS||29·5***||0·89 (0·87, 0·95)***||81·1||78·9***|
|Stellaria palustris||116·8||23·9***||31·6*||0·88 (0·80, 0·95)***||83·5||77·4***|
Results of the ROC technique (AUC values), as well as the number of correct classifications based on two cut-off values (popt and pfair), were generated both for the training and testing data to discriminate and validate the models. On the whole, the AUC values of the training data were slightly better because they were optimally adjusted. This was also true for the classification results. In further analyses and discussion, only the results of the testing data set were considered, which represented a more realistic estimate of prediction accuracy.
The models consisting of AGL (sigmoid or unimodal), WLF (sigmoid) and partly an interaction term yielded highly significant results (Table 2). The AUC values (≥ 0·73) reflected the high predictive power of each model. The number of correct classifications supported these results.
Two groups of species’ response patterns along a gradient of water level fluctuations could be distinguished. The first group was characterized by the fact that the species’ optima towards AGW changed along a gradient of WLF (Fig. 2). Most species belonging to this group had a preference for a lower AGL where the water tables fluctuated widely, in comparison with the optima under stable water tables. While the first group had similar occurrence probabilities over the whole range of water level fluctuations, the second group responded to reduced fluctuations with an increase or decrease in the probability of occurrence (Fig. 3). Some species from the second group had a tendency towards different AGL optima (Deschampsia cespitosa, Stellaria palustris and Carex vulpina). In the strict sense, these species belonged to both groups of response patterns.
Figure 2. Species’ probability of occurrence in relation to the size of WLF and the AGL. The latter correspond to an elevation gradient. First group of species’ response patterns: change of optimum AGL along a gradient of fluctuations.
Download figure to PowerPoint
Figure 3. Species’ probability of occurrence in relation to the size of WLF and the AGL. Second group of species’ response patterns: declining or increasing probability of occurrence along the gradient of reduced WLF.
Download figure to PowerPoint
To examine species’ responses to the fluctuation gradient at a finer scale, the data set was partitioned into two subsets: the recent and the older floodplains. As both floodplain types were well differentiated with regard to the fluctuation gradient (Table 1), the two groups of species’ response patterns should also be reflected at this scale.
The majority of the species was significantly related to the AGL gradient (see the Appendix). Species’ response curves strongly differed between the recent and the older floodplains. Species belonging to the first group preferred different AGL values in the recent floodplain compared with the older floodplain (e.g. Poa palustris and Galium verum agg.; Fig. 4a,b). Species from the second group either had a quite high probability of occurrence within the older floodplain, while they rarely occurred in the recent floodplain (e.g. Deschampsia cespitosa and Rumex acetosa; Fig. 4c) or vice versa (e.g. Rorippa sylvestris and Plantago intermedia). Some species were not significantly related to AGL in the recent floodplain because of a complete or almost complete absence (e.g. Cardamine pratensis, Holcus lanatus and Potentilla anserina). These species belonged to the second group as well.
Figure 4. (a–c) Examples of species’ response curves in relation to AGL for the recent (dotted line) and older (solid line) floodplains.
Download figure to PowerPoint
The results confirmed the patterns that were detected by analysing the whole data set. Generally, the results of the contingency tables, which were used to assess whether the occurrence of each species was independent of the floodplain type, conformed to the described patterns as well. Eight species revealed significant associations with the older floodplain (Cardamine pratensis, Carex acuta, Deschampsia cespitosa, Festuca pratensis, Holcus lanatus, Potentilla anserina, Rumex acetosa and Stellaria palustris) and three with the recent floodplain (Elymus repens, Plantago intermedia and Rorippa sylvestris), while the remainder did not exhibit a relationship with a particular floodplain type at all. Species with a declining probability of occurrence along a gradient of declining water level fluctuations were significantly more frequent in the recent floodplain and vice versa. Only in the case of Alopecurus geniculatus and Carex vulpina did the G-test not recognize a significant preference for floodplain type, while the response patterns detected by logistic regression displayed the opposite (Fig. 3). Carex acuta represented a converse case.
The optimum (u) of the AGL gradient was determined for each species’ response curve of the recent and older floodplains. Species without a model for each floodplain type were not considered. The relation between the difference (urecent − uolder) and the optimum of the recent floodplain, urecent, showed the direction of species’ responses (Fig. 5).
Figure 5. Differences in species’ optima between recent and older floodplains in relation to the species’ optima of the recent floodplain.
Download figure to PowerPoint
Species preferring low average groundwater levels in the recent floodplain (indicating high elevation habitats) selected habitats with much higher average groundwater levels in the older floodplain (indicating a preference for lower elevation habitats). These differences became smaller along a gradient of declining elevations. Where average groundwater levels were very high, around zero (very low elevation habitats of the recent floodplain), species showed the reverse behaviour. They tended to prefer lower groundwater levels, i.e. higher elevation habitats in the older floodplain. The relation between the optimum difference (urecent − uolder) and the gradient of urecent was strongly linear (F = 79·59 on 1 and 16 d.f., P < 0·0001).