This article summarizes the Ph.D. thesis dissertation of the author, who was granted by the Ministry of Education and Science of Spain (FPU program). The field works were funded by the Departments of the Environment (Reserva Natural Galachos) and Science, Technology and University (Research group E-61 on Ecological Restoration) — Aragon Government — and Ministry of Science and Innovation of Spain — MICINN (CGL2008—05153-C02-01/BOS). The author especially thanks his thesis supervisors FA Comín and E Muller and his group colleagues A. Cabezas, B. Gallardo, M. García and M. González.

1Riparian vegetation, as one of the main biotic compartments in fluvial ecosystems, is naturally adapted to the hydrogeomorphic regime of rivers and therefore relies upon its dynamic character (Poff et al., 1997; Stanford et al., 2005; Whited et al., 2007). Unfortunately, the progress in civil engineering during the 18^th, 19^th and, especially, the 20^th century to improve navigation, reduce flooding risks, reclaim land and water for agriculture, industry and domestic uses has caused dramatic changes in the hydrogeomorphology of water courses, in turn leading to a worldwide and massive loss and malfunctioning of all associated ecosystems and their ecological functions (Tockner and Stanford, 2002; Hughes, 2003; Hughes et al., 2008).

2In response to these impacts, there is an increasing interest on the part of river administrations and the scientific community in sustainable management, and particularly in restoration of riparian vegetation (Richardson et al., 2007; Hughes, 2003; Hughes et al., 2008). However, the lack of a profound understanding on the ecology of these systems, and particularly on their responses to the changing hydrogeomorphic regimes, has been a serious drawback for the success of conservation and restoration projects (Adams and Perrow, 1999; Buijse et al., 2002; Tockner and Stanford, 2002; Lake et al., 2007; White and Stromberg, 2010). The insufficient knowledge of riparian vegetation has been historically more important in the Mediterranean region, where few studies have been devoted to floodplain forest ecology (e.g., Salinas et al., 2000; Ferreira and Aguiar, 2006; Rodríguez-González et al., 2008) compared to other arid and semi-arid world regions such as the Southwestern US (see reviews by Webb and Leake, 2006; Stromberg et al., 2007a; 2007b).

3I present here a summary of a four-year work aimed at getting insight into the ecology of the floodplain forest of one of the largest Mediterranean rivers: the Ebro River (NE Spain). This study was focused on the woody compartment but approached from a broad and integrative perspective. The aim was to include different aspects of the tree structure, functioning (nutrient use) and functions (litter production) and the control exerted by their primary driver, the hydrogeomorphic regime, at different spatial and temporal scales. The ultimate purpose of this work was to give guidelines for management, conservation and restoration practices in the Ebro that might be applied to other Mediterranean watersheds with similar hydrogeomorphic constraints.

1 - Materials and methods

1.1 - Study area

4The study area was the 10 y floodplain of an 8-km river segment, located 12 km downstream of the city of Zaragoza (NE Spain, 41º36’ N, 0º46’ W, elevation 175-185 a.s.l.) including the Galachos Alfranca Natural Reserve (Fig. 1).

5The study segment belongs to the middle section of the river (Middle Ebro River), which is a 346-km reach between the towns of Logroño and La Zaida. The Ebro River is one on the largest Mediterranean rivers in terms of length (~930 km), annual average discharge (12 000 hm³ y^-1) and drainage area (85 534 km²). The climate of the Ebro River Basin is semi-arid, characterized by precipitation scarcity (400 mm y^-1) and temperatures often exceeding 35 ºC in summer. At the gauging station of Zaragoza, the mean monthly discharge of the Ebro is 230 m³ s^-1 (period 1981–2003). Low water levels (30-50 m³ s^-1) occur in summer, frequently as long (4–5 months) and severe drought episodes, but the surface flow remains permanent. Both the selected river segment and the middle reach of the Ebro have a very limited channel migration since the 1970’s and 80’s, as a result of two decades of intense lateral dike construction along the river course. However, the main channel still maintains a meandering shape (sinuosity = 1.39, longitudinal slope = 0.50 ‰, Ollero, 1995) as a relict of a former more dynamic geomorphic regime. The hydrologic regime is pluvionival but also controlled by several dams that were built during the 1950’s and 60’s for irrigation and flood protection purposes. Although highly regulated, the river maintains a certain hydrogeomorphic dynamism. In fact, the chosen river segment in particular is today one of the most dynamics parts of the Ebro (Ollero, 1995). The river still floods the zones adjacent to the main channel, covered by natural vegetation in a 500 – 1000 m wide belt, with a relatively high frequency (1-3 y) (Ollero, 2007; 2010; Cabezas et al., 2009; Magdaleno and Fernández, 2010). The creation of new barren sites, however, is limited to narrow areas inside levees (between the constructed levee and the main channel). Phreatophytic species, such as white poplar (Populus alba), European black poplar (P. nigra), white willow (Salix alba) and saltcedar (Tamarix gallica, T. africana and T. canariensis) dominate the overstory of the Ebro floodplain forests. The presence of hardwood species such as narrow leaf ash (Fraxinus angustifolia) and field elm (Ulmus minor) is also significant. All these species are autochthonous, whereas the occurrence of invasive tree species (Acer negundo and Robinia pseudoacacia) is very small.

1.2 - Research approach

6The riparian vegetation in the study area was digitized in a series of aero-photographs (1927, 1957, 1981 and 2003) to analyze its recent historical evolution (González et al., 2010b) and to select 43 forest patches of different age covering the geomorphic gradient existing in the floodplain (gravel bars, levee of permanent and intermittent channels, active floodplain terrace). The patches were therefore subjected to different hydroperiods and groundwater depths, which were monitored in a piezometer network covering the 10 y recurrence interval floodplain (Losada, 2004), with a higher concentration within the surface occupied by natural vegetation (Fig. 1). Piezometers were 6 m depth and installed using a rotary drill-rig. Water levels were registered manually at weekly intervals from their installation in 2006 to the beginning of the 2008 hydrologic year in October 2007 when an integrated absolute pressure sensor and data logger (van Essen DI502 TD Diver®) were submerged to the bottom of each piezometer to record pressure values daily, every 30 min from October 2007 to September 2009 (i.e. 2008 and 2009 hydrologic years). For each piezometer, the measurements were converted to water table level above sea level (a.s.l.) (±1 cm) after subtracting the atmospheric pressure recorded with an on-site barometer (van Essen DI250 BaroDiver®). The water table measurements were used to calculate different flooding regime surrogates at the patch level, namely the flood duration (in weeks y^-1 and % of the time), the depth to groundwater (in cm) and the flood frequency (number of events y^-1).

7Some of the physical and chemical soil properties were surveyed as complementary descriptors of the flood inundation regime. Thus, in each forest patch, three topsoil (0–10 cm) samples were collected in late‑summer 2006. The silt + clay fraction (%) was calculated gravimetrically. Total organic carbon (%) was measured with a LECO SC 144 DR elemental analyzer. Total N (%) was measured by combustion using a varioMAX N/CN elemental analyzer. Total P (%) was analyzed using an inductively coupled plasma emission spectrometer (ICP) after digesting soil subsamples in a double-acid solution in a microwave digester MWS-3Berghof®.

8One study plot was randomly placed within each forest patch to perform a tree survey in 2006 and 2007, including the diameter and species record of every woody stem ≥ 0.3 m height (total = 6891 stems) (González et al., 2010b). A health status was qualitatively assigned to every stem recorded according to the following scale: dead, dying (evidence of mortality affecting the main stem), die-backing (affected by branch dieback sensu Rood et al., 2000) and healthy. The recorded data served to calculate the stem density (stems ha^-1) and basal area (m² ha^-1) of each dominant tree species and plot. Seedlings were excluded from basal area calculations since their diameters were not quantified (the consequent sub-estimation was assumed not to be significant due to their small size). The relative abundance of each tree species in each plot was summarized by their importance values (IV), which were calculated using a modification of the S.A. Gergel et al., (2002) formula, ranging from 0 (absence) to 1 (complete dominance): IV = [(number of stems of one species on a plot/total number of stems on a plot) + (basal area of a species on a plot/total basal area on a plot)]/2. A mortality rate (MR) for each species at each plot was also computed as the frequency of dead plus dying stems, including seedlings (%).

9All this information was used to assess the effects of regulation on the abundance of the different tree species and to discuss their future perspective in the light of their regeneration capacity and mortality patterns under the dry hydrological conditions typical of the study area (González et al., 2010b; 2012). A greater level of detail was attained for the regeneration strategies of Populus alba, an autochthonous species in the Mediterranean region whose reproduction ecology had remained largely unstudied. In particular, we used a greenhouse experimental facility that simulated alluvial water table declines (based on the “Rhizopod” design developed by Mahoney and Rood, 1991) to evaluate the survival and growth of 9-day-old P. alba seedlings subjected to five hydrological treatments (permanent saturation, drawdown rates of 1, 2.5, 5 cm day^-1 and immediate drainage) in two different substrates (coarse and sandy) (González et al., 2010a). Also, P. alba seed dispersal was monitored in the field for the period 2006-2008, and seed germination and longevity tests were performed in the laboratory after collecting seeds in the field (González et al., 2010a).

10In 2007, a sub-selection of 12 plots was made to monitor litterfall patterns in a series of 120 litterfall traps (10 traps / plot) (González et al., 2010d; González, 2012) and nutrient use (N and P resorption proficiency sensu Killingbeck, 1996) by the dominant tree genera Populus and Tamarix (González et al., 2010c). Linear mixed effect models (LME, Pinheiro and Bates, 2000) were used to explain to what extent those processes were controlled by the hydrogeomorphic regime. LME models were selected because they allow introducing fixed and random effects, thus accounting for pseudoreplication. For litter production, a full model was run using 17 fixed factors classified in the following four groups: (1) forest structure (basal area, stem density and tree species richness), (2) flooding regime (flood duration, depth to groundwater and river distance), (3) soil conditions (silt+clay fraction, total organic carbon, total nitrogen and total phosphorus) and (4) litterfall chemistry variables (phosphorus resorption proficiency, nitrogen resorption proficiency, phosphorus use efficiency, nitrogen use efficiency, litter N:P, C:P and C:N; see González et al., 2010d for definitions and calculations). For nutrient use, six models were run for the corresponding six dependent variables: leaf litter %N, %P (i.e., N and P resorption proficiency) and N:P in Populus and Tamarix genera. The three abovementioned variables characterizing the flooding regime (flood duration, depth to groundwater and river distance) and the four variables describing the soil conditions (silt+clay fraction, total organic carbon, total nitrogen and total phosphorus) were used as fixed effects. To correct the regression models for spatial pseudoreplication, the variable ‘‘study plot’’ was introduced as a random effect. As a high level of multi-collinearity was expected among the explanatory terms in all models, Spearman’s rank correlation coefficients were used as preselection of predictors, selecting those having r_s < 0.7. In this sense, when a pair of predictors had r_s ≥ 0.7, the one registering a lower correlation with the response variable was removed from the analysis. After that, a backward selection of predictors was performed to retain only the most significant ones (p < 0.05) in explaining the response of the dependent variable, testing each step with a likelihood ratio-test (Crawley, 2002). Pearson’s product–moment correlation between the observed and predicted values was used to assess the goodness of fit of the final models. The LME models were carried out using the functions available in the package ‘‘nlme’’ (Pinheiro et al., 2007), R 2.8.1 software (www.R-project.org). Once the main predictors were identified, the sum of independent and shared variance in litterfall explained by each significant predictor (i.e., the relative contribution) in each model was assessed through hierarchical partitioning (Chevan and Sutherland, 1991). To that end, we used the functions available in the package ‘‘hier.part’’ (Walsh and McNally, 2007), R 2.8.1 software.

2 - Results

11The digital analyses of the aerial photographs showed that potential recruitment sites (barren sites) decreased progressively during the 20^th century (Table 1).

12Young forests (i.e., < 25 years old) were more frequent than old forests (i.e., > 25 years old) until 1981, year considered as the beginning of a new geomorphic period in the recent history of the river characterized by a restricted meandering channel and stable riverbanks (Ollero, 2007). Then, a shift towards dominance of old forests occurred during the period 1981‑2003 (Table 1). According to the tree survey carried out in 2006 and 2007, young forests were dominated by the pioneer trees Populus nigra, Salix alba and Tamarix spp. They were in a very active recruitment phase and exhibited good health (Fig. 2a).

13In old forests, larger individuals of those species also dominated the overstory layer of the forest but usually with a declining canopy, while Ulmus minor and Fraxinus angustifolia were frequent but only as small-size stems (Fig. 2b, 2c). Although usually dominating and no exhibiting die-backing, P. alba was only present in the oldest forests (> 50 years old) (Fig. 2c). This fact was in clear contradiction with a sexual regeneration of P. alba that had traits typical of other pioneers of the Salicaceae family (Karrenberg et al., 2002), as shown by our field and greenhouse observations (González et al., 2010a). That is, the species exhibited a massive annual production of seeds that peaked in mid-April, a period of frequent floods in the Ebro. Seeds had high initial seed germination (~90%) but lost viability in less than one month. No seedlings survived the water table declines in the coarse substrate although survival was high (85%) under saturated conditions. In the sandy soil, survival was significantly greater in the permanent saturation (87%) and 1 cm day^-1 (88%) treatments than in the 2.5 cm day^-1 (58%), 5 cm day^-1 (25%) and immediate drainage (22%) treatments. The lowest root and shoot growth rates occurred under the saturated and immediate drainage conditions.

14All the phreatophytic species present in the Ebro exhibited a negative response to dry hydric conditions in terms of survival of their mature populations in the 43 surveyed plots. In particular, S. alba was found to be the species with lowest tolerance to drought, followed by P. nigra, Tamarix spp. and P. alba (Fig. 3a). However, only S. alba exhibited a decrease in relative abundance across the hydric gradient, as shown by the significantly increasing importance values with the wetter sites (Fig. 3b).

15The mean litter production recorded in 2007 over the sub-selection of 12 study plots was 563 ± 55 g of dry matter m^–2 year^–1 (mean ± 1 SE). The range of variation was rather wide, with a factor of 4.6 between the minimum (199) and maximum (916 g of dry matter m^–2 year^–1). The linear mixed effect model that was adjusted to the data using the full set of independent predictors explained 51% of the variance in litter production. In particular, it predicted higher litter production with decreasing river distance, which accounted for the highest contribution to the model, and with increasing total P in soil and stem density (Table 2). Alternate models that were run with a fewer number of variables also explained a high percentage of variability in litter production. However, the relative contribution of explanatory terms that were directly or indirectly related to the hydrogeomorphic regime was more than >70% in all cases (Table 2).

16The terminal N and P concentration in Populus and Tamarix leaves, and leaf litter N:P ratios also varied according to the hydrogeomorphic regime, exhibiting lower concentrations (i.e., greater nutrient resorption proficiency) and higher N:P ratios in the drier sites along the surveyed hydrogeomorphic gradient (Table 3).

3 - Discussion

3.1 - Floodplain forest structure,composition and functioning driven by a changing hydrogeomorphic regime

17The impacts of flow regulation on riparian vegetation observed in this study (detailed in González et al., 2010b) corresponded to a common alteration within transition from braided-meandering channel morphologies to stabilization caused by decades of dam operations and river embankment (Scott et al., 1996; Johnson, 1998). That is, in a first step (1957‑1981 in the Ebro) the surface occupied by pioneer forests increased because trees may colonize barren zones that were no longer subject to a high degree of flow disturbance. Then, at the first stages of regulation, the floodplain forest regenerated even more effectively than in the absence of regulation (see ratios young/old forests in Table 1). In a second step (1981‑2003 in the Ebro), however, new colonization declined sharply because there were almost no sites uncovered by vegetation and the new hydrogeomorphic regime only created recruitment sites marginally (in narrow bands running parallel to the main channel and in-channel areas). In parallel, forest patches that had occupied the stabilized zones during the first step got older and eventually declined. Therefore, old forests that historically only occupied the outer floodplain have today become dominant all over the active floodplain. I suggest that, if the hydrogeomorphic dynamics are not reactivated, a future third step in the alteration process of forest structure could be characterized by a reduction in the functional width of the river, with an eventual colonization of the outer floodplain by steppic vegetation. This was the case of the Salt River near Phoenix (Arizona, USA) where dewatered areas once covered by Salix-Populus forests were replaced first by Tamarix communities and later by sparse patches of short xerophytic shrubs such as Bebbia spp. (Stromberg et al., 2007a). In the Gila River (New Mexico, USA), as another example, Prosopis velutina forests died as water tables dropped and today these former riparian sites are sparsely vegetated by small desert shrubs (Stromberg et al., 2007a).

18Under such a dramatic scenario, P. alba might act as an alternate successional pathway to the declining P. nigra, S. alba and, eventually, Tamarix spp. because its sensitivity to drought is less than that of the other phreatophytes (Fig. 3, González et al., 2012), and its vegetative regeneration by root suckering (not dependent on hydrological disturbance) could be more effective (González, personal observation). According to this hypothesis, with the decline of P. nigra, S. alba and, to a lesser extent, Tamarix spp., the composition of the floodplain forest might change in the following decades towards a dominance by P. alba and late-seral species, such as U. minor and F. angustifolia. Other space-for-time substitutions have been used to suggest future changes in the composition of riparian forests with similar riparian tree genera (e.g., in the Southwestern US, Stromberg et al., 1996; Lite and Stromberg, 2005; Stromberg et al., 2007b). However, contrary to our hypothesis, those studies essentially predicted a shift from Salicaceae- to Tamaricaceae-dominated communities with the intensification of the dry hydrological conditions, probably because there is not any facultative phreatophyte among the US Salicaceae species that could cope with drought conditions better than their coexisting Tamarix spp., as might be the case for P. alba in the Mediterranean systems (González et al., 2012). It may be also that the naturalized Tamarix species in America (especially T. ramosissima and T. chinensis) could tolerate more extreme drought conditions than their Mediterranean counterparts (T. gallica, T. africana and T. canariensis). Unfortunately, any comparison among these systems must be taken with caution as there is still a lack of intercontinental stress tolerance scales in literature (but see Niinemets and Valladares, 2006). On the other hand, more studies on the phenotypic plasticity of the different riparian tree species to drought stress should be done to predict their adaptability to environmental changes, which could in turn compensate for the tendencies observed in the Ebro presented here and detailed by E. González et al. (2010b; 2012). This is especially relevant for Mediterranean communities that, according to climate models, will be greatly affected by the increased potential evapotranspiration linked to global warming and by decreased precipitation (Prieto et al., 2009), and by a subsequent increasing water demand by society (Tockner and Stanford, 2002). A priori, according to the observations made by S.J. Lite and J.C Stromberg (2005) and by J.C. Stromberg et al. (2007b), Tamarix could be a more plastic genus than Salicaceae for water supply, as it seems to be the case for nutrient use as well (González et al., 2010c). Another question that remains unsolved and deserves further research is the abovementioned contradiction of P. alba being a species with a high sexual regeneration potential but not colonizing new forests since river stabilization began. I suggest that this paradox may be related to the different resistance of species to hydraulic forces, a topic that has not received enough attention yet (Puijalon et al., 2011).

19As occurred for floodplain forests structure and composition, the aspects concerning their functioning and functions studied here appeared to be highly controlled by the hydrogeomorphic regime. In a river system such as the Ebro, where N availability is high (medium‑high NO₃^- levels in superficial- and ground-waters, Torrecilla et al., 2005; González et al., 2010c), the input of P by sedimentation seemed to play an essential role in litter production. Unlike N, the majority of the P flowing in river waters is transported bound to particles (Vought et al., 1994). In the Ebro, sites with a lower concentration of P in the topsoil (or with P sequestered by a high total soil organic carbon), located at longer distances from the main channel and subject to shorter hydroperiods (where sediment deposition could be lower) had a lower litter production (Table 2, González et al., 2010d). Similarly, a higher P-limitation in the less flooded sites, mediated by a lower sedimentation, could also underlie the higher nutrient resorption proficiency (i.e., lower terminal foliar concentrations) in Populus and Tamarix and the higher leaf litter N:P ratios observed as the floodplain got drier (Table 3, González et al., 2010c). These results suggest that litter production, nutrient use by plants (uptake, resorption, storage and loss) and nutrient cycling in the riparian ecosystem are intimately connected by positive feedbacks within a biogeochemical template established and controlled by the hydrogeomorphic regime. In fact, 28% of the variability in litter production could be explained only by P use efficiency (Table 2), that is the total litterfall divided by total P content in litterfall (sensu Vitousek, 1982) and was significantly correlated to other nutrient use descriptors such as P resorption proficiency and N:P (González et al., 2010d). As a result, the synergic effect of a low litter production and a high efficiency and proficiency in the use of nutrients limited their return to the soil at the more disconnected sites (Fig. 4), and evidenced the importance of flood pulses in maintaining the ecological functions of litter production and nutrient cycling in the Ebro forests.

3.2 - Implications for management and restoration

20Fortunately, the resilience of riparian systems to disturbance and, consequently, the potential for successful ecological restoration is very high. Even in unregulated rivers, the hydrologic conditions for effective recruitment of new individuals on a large scale are only occasionally met and thus, it is not uncommon for regeneration of pioneers to fail over several successive years or even decades (Stromberg, 1993; Cordes et al, 1997; Mahoney and Rood, 1998). Therefore, riparian species have developed the faculty of keeping an enormous reproductive potential (Gom and Rood, 1999; Karrenberg et al., 2002) and, when appropriate restoration measures are taken, successful regeneration occurs rapidly and effectively (Rood and Mahoney, 2000; Stevens et al., 2001; Stromberg, 2001; Rood et al., 2003; 2005). Restoration in the Ebro may be implemented by different actions: removal and set back of flood defense structures; reclamation of agricultural lands which are no longer profitable; reconversion of poplar crops to natural floodplain after harvest; lowering floodplain heights to improve the erosive capacity of floods and in-channel reallocation of extracted sediments to facilitate the creation of nursery sites and the provision of P and other nutrients to vegetation; local controlled flooding to promote regeneration of desirable tree species, especially P. nigra, S. alba, Tamarix spp. and P. alba, preceded by mechanical disturbance of the substrate and removal of ruderal vegetation (or even of old forests) in stabilized sites; and prohibition of grazing in potential nursery sites (González et al., 2010a; 2010b; González 2010). Although the regeneration of new cohorts should be prioritized in case of economic shortage in restoration and management plans of the Ebro, our results (Fig. 3) also showed that, in the meantime, river flows should be managed to avoid drought stress in the declining old forests as much as possible (especially in summer), thus protecting the existing phreatophytic populations from further degradation. A proposal of hydrologic thresholds for riparian forest conservation in the Middle Ebro was given by E. González et al. (2012) (see example in Table 4).

21Previous research also identified a higher mortality and canopy dieback in floodplain forests subject to extreme, subtle and chronic human-induced drought stress and designed some flow management prescriptions to minimize impacts on the adult riparian trees (e.g., Stromberg and Patten, 1991; 1992; Scott et al., 1999; 2000; Amlin and Rood, 2003; Williams and Cooper, 2005). All these experiences share the principles of restoration as a process of assisting the self-recovery of a degraded ecosystem (SER, 2004). Their differences and particularities must be seen as an evidence of the need to define the restoration and conservation strategies locally, depending on the kind and degree of human impacts, the available data and scientific knowledge, and the desired restoration outcomes agreed by society (Dufour and Piégay, 2009).

Conclusion

22The results of this study showed the close relationship between the floodplain forest ecology and the hydrogeomorphic regime of the river. It follows that the restoration of certain hydrogeomorphic dynamism, including a more re-naturalized flooding regime, an enhancement of lateral channel migration, and a partial restitution of the sediment flow; as well as the control of low water levels in summer, should be included in the Ebro management plans to guarantee the ecological integrity of their floodplain forests. Integrative studies (i.e., including different aspects of the structure, functioning and function of a biotic compartment) such as the presented here support the idea of raising the hydrogeomorphic drivers as the central element of management in river restoration and conservation projects (Poff et al., 1997; Rood et al., 2005; and many others). I suggest that similar adjustments of the floodplain forest ecology to the new hydrogeomorphic regimes may be occurring in other large Mediterranean regulated rivers subject to similar regulation constraints. Therefore, coordinated management strategies and studies across Mediterranean rivers (including reference unregulated or minimally perturbed systems, if available) would be very beneficial to be carried out in the near future.