The trees selected for this study grew in the river floodplains of the Manzanares and Henares rivers, in the provinces of Guadalajara and Madrid (central Spain, Table 1). The regional climate is continental Mediterranean, with hot and dry summers and cold winters. Mean annual temperature and precipitation are 14.2 °C and 333.1 mm, respectively (data from Torrejón de Ardoz weather station, National Institute of Meteorology, 1961-2011; Figure 1). Soils are luvi and fluvisols (Monturiol & Alcalá, 1990).

The dominant native tree species in the study area are Fraxinus angustifolia and Ulmus minor, together with Populus alba L., Populus nigra L. and different species of the genera Salix and Tamarix (Blanco et al., 2005). The most abundant exotic tree species are Ailanthus altissima and Robinia pseudoacacia. Ailanthus altissima is native to southeastern Asia and was introduced in the Iberian Peninsula in the end of the 18^th century. In its native range, climate is characterized by a long and warm growing season and by annual precipitation over 500 mm (Kowarik & Säumel, 2007). In contrast, R. pseudoacacia is native to central and eastern North America, where the climate is temperate humid or even hyper-humid (Huntley, 1990). It was introduced in the Iberian Peninsula during the 19^th century (Sanz Elorza et al., 2004). Robinia pseudoacacia and A. altissima have been previously described as pioneer, shade intolerant, fast growing species (Knapp & Canham, 2000; Sanz Elorza et al., 2004; Kowarik & Säumel, 2007). In contrast, F. angustifolia and U. minor are able to establish successfully under a wide range of light and soil moisture conditions (González Muñoz et al., 2011; González Muñoz et al., 2014).

We selected 10-12 dominant adult trees per species. Trees were chosen following 4 criteria: they should not show evidences of growth suppression by competition along the tree ontogeny; should show straight and regularly shaped stems (to avoid any tension on the rings) and should have a large DBH (to obtain long chronologies). Trees following the last requirement were difficult to find. Native riparian forests are largely impacted by human activities. Moreover, the frequency of adult individuals of U. minor has been severely reduced by the Dutch elm disease in the study area (Brasier et al., 2004; Martín et al., 2006). Similarly, the invasive trees are frequently cut-off when they reach a certain size. As a consequence, we had to sample trees placed at certain distances in some cases. However, our sampling procedure aimed to ensure the maximum habitat homogeneity in any case (trees not close to river courses, roads, paths, crops or populated places).

In each selected tree, we extracted one increment core, measured the perimeter at breast height (PBH, 130 cm above the base) and estimated the tree height (TH, m, from the angle between the distance observer-tree base and the distance observer-tree top) (Table 1). Increment cores were glued on wooden holders and dried and surfaced with a microtome to make tree rings clearly visible. All four study species are deciduous and form clear ring boundaries of ring-porous wood (http://insidewood.lib.ncsu.edu). Cores with rotten segments, knobs or unclear boundaries were rejected. In total, cores from 40 trees were considered for analysis. All samples were visually cross-dated following Yamaguchi procedures (1991). Total ring width (RW) was measured with an accuracy of 1/100 mm using a LINTAB measuring table and the software TSAP (Rinn, 1996). Dating errors of the individual series were corrected using the program COFECHA (Grissino-Mayer, 2001).

In order to remove the trend of decreasing RW with increasing tree size, we converted the data of raw RW into basal area increment (BAI, cm² year^–1), as follows:

In each wood sample, we identified the last 25-30 annual tree rings and separated them into wood segments containing 5 contiguous rings with a razor blade. Wood segments were ground to a particle size <1 mm (PM100, Retsch Haan, Germany). We did not extract cellulose as both whole wood and cellulose isotope time-series show similar long-term trends related to atmospheric CO₂ and climate (Saurer et al., 2004; Taylor et al., 2008). The isotopic ratio ¹³C/¹²C of each wood sample was determined from tin capsules containing aliquots of 0.5-0.7 mg on a mass spectrometer (Thermo finningan Delta Plus XP, UC Davis Stable Isotope Facility, Davis, California, USA).

According to Farquhar et al., (1982), the total discrimination against ¹³C during carbon fixation is expressed as:

Climate data were provided by the National Institute of Meteorology (AEMET, Government of Spain). Among the existing climate stations, we selected Torrejón de Ardoz (40°29´00´´N 3°27´12´´W), as it has the longest and most complete record available and was located close to all the sampled trees. As F. angustifolia and U. minor trees were located closer to river courses than the invaders, we also evaluated groundwater effects on their BAI, using river flow data (m³/s) as an indicator of the groundwater water availability. We used river flow data provided by the Government of Spain (www.cedex.es) from the Espinillos station for F. angustifolia (40°27´40´´N 3°25´13´´W) and the Vaciamadrid station for U. minor (40°19´15´´N 3°30´32´´W).

We used monthly and seasonal average data of the current and previous years for the following parameters: average, maximum, and minimum temperatures (T, Tmax and Tmin), precipitation, number of days with frost (DF) and river flow (RF). Seasonal data were calculated as the average (temperature and RF) or sum (precipitation and DF) of winter (December_{(t – 1)}-February_(t)), spring (March_(t)-May_(t)), summer (June_(t)-August_(t)) and autumn (September_(t)-November_(t)).

We evaluated differences between origins (native or exotic) and species (nested in origin) on the average BAI and _iWUE of the last 20 years of tree growth (1987-2007) with two-way nested ANOVAs. We only considered the last 20 years for this comparison because it was the period in which we had the highest replication. Trees were used as replicates. Post-hoc Tukey tests were applied for multiple comparisons among species. We checked for homoscedasticity and for normality of the residuals using the Bartlett and Shapiro-Wilk tests respectively.

For each origin and species, we examined the relationship between BAI and _iWUE along the last 20 years of tree growth, with simple Pearson correlations. To perform these correlations, we averaged the BAI in five-year periods, as our values of _iWUE were estimated from segments of five contiguous rings.

We used linear mixed-effect models to identify the effects of environmental variables on the species BAI. In these models, the environmental variables were fixed factors and the tree the random factor. We modelled the period comprises between 1963 to 2007 for A. altissima and R. pseudoacacia, 1967 to 2007 for U. minor, and 1981 to 2007 for F. angustifolia. For F. angustifolia and U. minor, river flow was also included in the model. The selection of climate and river flow variables to include in the model and the random factor structure were performed manually using backward stepwise regression, beginning with variables and then removing progressively less significant variables to minimize the value of Akaike Information Criterion (AIC; Burnham & Anderson, 2002). The first order autocorrelation structure of BAI (the growth in the previous year or BAIp) was modelled using the non-climate/river flow residuals (i.e., difference between observed and predicted values) as a response variable. Thus, part of the unexplained variance in the climate-growth model was accounted with the individual autocorrelation structure. The same procedure was carried out for accounting the tree-age effects. As trees were not dated, we used tree age at sampling height (number of rings contained and measured in each core) as an indirect estimate of tree age. We used the Akaike information criterion corrected for small sample size (AIC_c) for multi-model selection. The model with the lowest AIC was selected (Burnham & Anderson, 2002). Models with ∆AICc between 0 and 2 were considered to have equivalent empirical support. In the case of equivalent models, we selected the model explaining the highest variance. If the explained variance was also equivalent, the simplest model (i.e., the model with less explanatory variables) was selected.

All statistical analyses were performed using R software (R Development Core Team 2013).

Significant differences were found in the 20-year mean BAI between origins (F_1,35 = 13.66, p < 0.001) and species (F_3,35 = 15.56, <0.001) (Figure 2a). However, the average highest BAI of the natives was mostly due to U. minor (approximately 29 cm²year^–1) (Figure 2a). Significant differences were also found among origins in the last 20-year mean _iWUE (F_1,35 = 41.20, p < 0.001), with invaders showing higher _iWUE on average than natives (Figure 2b).

No significant correlations were found between BAI and _iWUE in all cases (p > 0.05) (Suppl. Figure S1 [pdf on line]).

The invasive species showed a relatively steady temporal pattern in BAI. Ailanthus altissima BAI showed a slight decline since the beginning of the 1990s (Figure 3). In contrast, the native species showed higher temporal variance in BAI (Figure 3). Fraxinus angustifolia had a very slow growth rate during its juvenile phase (approximately 5 cm²year^–1), increasing after the 1990s to relatively steady values (about 15 cm²). In contrast, U. minor showed a relatively slow growth rate until the onset of the 1980s and then reached a maximum between 1978 and 1993 (Figure 3).

For the invader A. altissima, the selected model explained the 67.92% of the variance and indicated a negative effect of TmaxOct on BAI, whereas positive effects of TminFeb and PJul (Table 2). TmaxOct accounted for the highest relative weight of the variance (42.50%, table 2). For R. pseudoacacia, the selected model explained the 67.53% of the variance and indicated a negative effect of TminOct and a positive effect of PApr on BAI (Table 2). BAIp, closely followed by TminOct, explained the highest amount of the variance (32.09 and 26.70%, respectively) (Table 2). For F. angustifolia, the selected model explained the 67.69% of the variance and indicated a positive effect of TmaxSep (31.34%), TminPrevApr, RFAug, RFMar and TminJan (Table 2). For U. minor, the selected model explained the 57.89% of the variance. This model indicated positive effects of TSep, TmaxDec, TmaxNov, RFPrevMay and PApr and negative effects of DFMay and DFOct (Table 2). BAIp explained the highest proportion of the variance of BAI of U. minor (38.73%), whereas environmental variables explained very little in this species (19.17% in total) (Table 2). The tree age at coring height was not significant in any case.

(See Suppl. Table S2 [pdf on line] for the model selection procedures)

We hypothesized a higher BAI in the invaders than in the natives, as high growth rates have been previously related to invasive species success. Instead, the native U. minor showed the highest average BAI, mostly due to an increase in growth between 1978 to 1993 (Figure 3). This could be attributed to: 1) long-term warming trends (discussed below); 2) a reduction in stand density occurred due to the Dutch elm disease (Martín et al., 2006). The stand where we collected U. minor samples is one of the best preserved elm forests of the Iberian Peninsula. However, since 1956, some events of stand reduction have been detected from aerial photographs and related to peaks of this disease (Martín et al., 2006). Also in contrast to our hypothesis, the invasive R. pseudoacacia showed a very low BAI (Figures 2a and 3), perhaps as a result of the dry conditions imposed by the Mediterranean climate (also discussed below).

Climate significantly and differentially influenced the BAI of all study species. However, despite the differences among species, it is possible to point out some general trends in climate-growth patterns for natives and invaders (see Table 2). For instance, we did not find any negative effects of warm temperatures on the BAI of the natives. Moreover, warm autumns yielded opposite effects on native and invasive trees, enhancing BAI in natives but hampering it in invaders (Table 2). High temperatures in autumn can extend the length of the growing season, by extending the canopy lifespan and favoring plant growth. In this sense, Peñuelas & Filella (2001) suggested that the accelerated tree growth observed throughout Europe is caused by the extended growing season consequent to global warming, together with atmospheric N deposition and increased CO₂. Therefore, the negative effect of warm autumn temperatures on both invaders was surprising, especially in the case of A. altissima. The native distribution of this invader presents warm conditions and, in the coldest areas of its invaded range, it is confined to the mild conditions of urban areas (Kowarik & Böcker, 1984; Gutte et al., 1987). The negative effect of warm autumns on the BAI of A. altissima may be related to an extension of summer drought stress or to a decrease in the amount of stored reserves for the next year (Atkin & Tjoelker, 2003).

The effect of precipitation on BAI was smaller than that of temperature. July precipitation had a positive effect on A. altissima BAI (Table 2). In our study site, the maximum leaf area of A. altissima coincides with the driest period (July), as this species has a delayed leaf budburst (González-Muñoz et al., 2013). Therefore, summer precipitation may compensate for evapotranspiration losses associated with high summer temperatures. However, in contrast to our initial hypothesis, spring (April) and not summer precipitation boosted R. pseudoacacia radial growth (Table 2). The positive effect of spring precipitation, together with the high effect of BAIp on the secondary growth of this species, suggest a mobilization of stored carbohydrates and an enhanced sap flux at the beginning of the growing season, which may contribute to the fast and early leaf bud-burst of R. pseudoacacia in our study area (González-Muñoz et al., 2013). The low BAI achieved by this species suggests a short growing season, which is confirmed by the short leaf lifespan of R. pseudoacacia (González-Muñoz et al., 2013). Koretsune et al., (2009), in a study conducted with adult trees of R. pseudoacacia in China, found that high precipitation during the early growing season increased R. pseudoacacia earlywood formation. However, they also found a higher average ring width than us. This suggests that R. pseudoacacia has a higher growth potential in a more humid climate, which cannot be expressed in our study area.

Regarding the natives, spring river flow and spring precipitations increased F. angustifolia and U. minor BAI, respectively. High water availability in early spring may favor a high sap flux at the beginning of the growing season. Besides, summer river flow also favored F. angustifolia radial growth, likely due to an effect of water stress alleviation through a high water table during the dry summer. Nevertheless, it was BAIp and not climate which explained the most of the variance of U. minor BAI, suggesting that its secondary growth may be highly dependent on the carbohydrates stored during the previous growing season, a low climate sensitivity, or both (Fritts, 1976).

Our models explained from the 57.89 to the 67.92% of the BAI variance of studied species. Other factors not evaluated here can be responsible of the rest of the variance, as micro-environmental conditions, groundwater availability, management activities or stand structure (competition or stand reductions by pathogens) (Fritts, 1976). In addition, we indirectly estimated the tree age as the number of rings measured in each core. Despite it was not significant, still some signal due to ontogenetic differences between species can remain in our growth models. All these are common limitations of dendrochronology studies. However, our sampling procedures and modelling approach still provide robust conditions about the climatic factors affecting the BAI of studied species and allow us to evaluate differences in their growth strategies.

We expected to observe a fertilization effect of atmospheric CO₂ on the secondary growth of invasive species, as invaders are generally considered opportunistic, and then able to quickly take advantage of available resources. However, our results do not support this hypothesis, as we did not find a significant correlation between the BAI and the _iWUE. Then, although the invaders had higher _iWUE than natives, this higher _iWUE did not translate into increased radial growth. Higher _iWUE can be achieved by increasing carbon uptake or by decreasing water loss due to lower stomatal conductance (Francey & Farquhar, 1982). For the invaders, drought stress may have reduced stomatal conductance, and thus prevent them of taking advantage of increased atmospheric CO₂ for a faster growth.

Our results suggest that neither climate change nor rising CO₂ will enhance the success of study invasive species over the natives in riparian forests of central Spain. Firstly, the contrasting effect of autumn temperatures on native and invasive species BAI suggests that study invaders will be more hampered by increasing temperatures. Secondly, we did not find a fertilization effect of CO₂ on the BAI of the invasive species. Moreover, these findings, together with the low BAI of R. pseudoacacia, suggest that higher temperatures may especially hamper this invader, limiting its presence to areas where water stress is less severe.