The Loire valley, in France, is enclosed by the old Hercynian folds of the Armorican peneplain in the northwest, the Massif Central in the south and the Morvan in the east. From the secondary era, successive sea deposits of sand and clay interspersed with limestone and loess, leading to the formation of a vast alluvial plain. The Loire carries sand and gravel that form changing sandbanks and islands. Water level varies greatly; receiving little rainwater supply in summer, the river reaches its lowest level in September (160–$180{\mathrm{m}}^3{\mathrm{s}}^{-1}$), while shoreline woods are flooded in the winter (1320–$1500{\mathrm{m}}^3{\mathrm{s}}^{-1}$ February in Saumur-Montjean; [19]).

Riparian woods are characterised by three main plant communities. The small Willow Grove (less that 5 to 6 m high) includes pioneer species (Osier Salix purpurea L., Salix triandra L. and Salix viminalis L.). The tall Willow Grove (15–20 m high) is dominated by white Willow S. alba L., crack Willow Salix fragilis L. and black Poplar P. nigra L. often combined with the american box Elder Acer negundo L. [19,20]. On the top of the riverbank, narrow-leaved Ash (Fraxinus angustifolia Vahl) and smooth-leaved Elm (Ulmus minor Mill.) dominate accompanied by common Oak (Quercus robur L.). Banks and islands are sometimes protected from erosion by ripraps.

The studied part of the Loire River began 127 km downstream from the city of Blois. It was divided in two areas, one located east of Angers city (U-area) and the other located west of Angers (D-area). The U-area began at Montsoreau and ended at Bouchemaine, a 80-km long strech, where the first lodge was found in 1985. In 1996, five sites with several active lodges were recorded on this area [16]. D-area, located between Chalonnes-sur-Loire and Ancenis (36 km), was a part of the river where beavers have not established yet.

The field study was conducted by canoe and by foot from July to September 2000 and 2001. Low water level allowed an easy access to riparian woods, and a better detection of beaver signs and of sandbanks. Both sides of the Loire were surveyed, as were the numerous islands. In U-area, 90 kilometers of contiguous shoreline were surveyed in 2000, with approximately 20 km of diked bank. In D-area, 88 km of shoreline were surveyed in 2001. In streches, the whole bankline was divided into 10-m wide contiguous plots (or sites), parallel to the stream. Plots varied from 0.6 to 1.5 km in length and corresponded to zones where vegetation (structure and composition) appeared similar. On each plot, plants were identified and categorized based on their morphology: (TT) 10–15-m tall trees with large trunks and no branch touching the water at flood time, (ST) 5–10-m small trees, (B) 5–12 m tall bushes with basal branches accessible to beavers even at low water level, (H) herbs and forbs. Within each class of woody plants, Salicaceae were respectively called TTS, STS and BS. At each plot, canopy cover by each plant category was estimated using an Abundance-Dominance scale (AD), ranging from 1 to 5: (1)<5%, (2) 5–25%, (3) 25–50%, (4) 50–75%, and (5) 75–100% [21]. The number of lodges was recorded, as the signs of beaver activity (cut trees, refectories, bark stripping, trunk gnawing, tracks...). The bank slope was classified as: (1) <20%, (2) 20–50% or (3) >50%. The percentage of bank length protected from erosion by ripraps, the characteristics of these ripraps and the presence of a contiguous sandbank in front of the site were also recorded. Human frequentation (mainly fishers and strollers) was assessed using a qualitative scale ranging from 1 to 5.

Data collected in U-area were analysed to compare plots with lodges, plots with other beaver signs and plots without beavers. The 20-km diked bank was excluded. As the number of plots within each group was small, we performed a Kruskall–Wallis nonparametric ANOVA. Multiple comparison Dunn's post test was used for means separation. We then conducted multivariate analysis (Automatic Interaction Detection, AID) with lodge occurrence as dependant variable, in order to identify factors influencing lodge location (minimum split value 0.05, Phi-square fitting method, Instat software). A first AID was conducted with all U-plots. A second was conducted after removal of the plots without beavers. Student's t-tests were performed to compare habitat variables between up- and downstream sites. Welch t-test was used when variances were unequal. U- and D-plots were examined regarding their suitability for lodge construction according to the AID results. Plots that fitted the model within both datasets were compared using a Mann–Whitney nonparametric test.

The surveyed U-area shoreline was divided into 86 plots. One to four lodges were recorded at 19 plots and other beaver signs were observed at 36 plots. All these signs of activity were mainly recorded on island edges: 70% of the lodges, 65% of cut trees, refectories, bark stripping or tracks.

Analysis of vegetation structure (Table 1) indicated that the canopy cover by tall trees was higher in the plots with lodges (35.1%, SD 19.3%, KW 9.62, P=0.008, df=2). Plots with lodges have significantly higher tall Salicaceae canopy cover than other plots (11.8%, SD 10.3%, KW 9.5, P=0.009, df=2). Beavers settled down in rather quiet places (mean human frequentation 1.2 in sites with lodges, SD 0.4, KW 12.5, P=0.002, df=2). Even though ripraps were most found in sites without beavers, the animals did not always avoid bank protection to build lodges. Most of ripraps were made of non-cemented stones, which length ranged from 20 to 40 cm and width from 15 to 30 cm (n=23). Stones never covered more than 40% of the plot bank, and lodges were observed in the stoneless areas. At two sites, beavers managed to dig between adjacent stones: in both cases, 100% of the bank was covered with bigger stones (50–100 cm long, 30–35 cm wide, n=23) that were not cemented to each other and interstices were large enough (mean 27.3 cm, SD 9.7, n=13) to allow an access to the bank substrate. No beaver signs were observed in two other sites, continuously covered with a flagstone pavement.

Among the 34 lodges, 12 were burrows, 20 hut-burrows, and two were temporary huts built on the river edge. Beavers always dug their lodges into abrupt banks (slope>50%), with the absence of sandbank (Table 1). Rooms and galleries were always burrowed between roots of a 10–15-m tall tree. A lot of these trees were poplars (about 44%, n=32); only one willow was used (S. alba, 3%). However, most trees were non-Salicaceae (53%): F. angustifolia (31%), Q. robur (13%), U. minor (6%), and A. negundo (3%).

The AID carried out with all U-plots (R²=0.468; Fig. 1a) clearly related plots with lodges with a canopy cover by tall trees higher than 37.5% (AD⩾3), sandbank absence, and canopy cover by tall Salicaceae higher than 12.5% (AD⩾2). The second AID with only sites with beaver signs (R²=0.484) added as factor bushy Salicaceae higher than 2.5% (AD⩾1; Fig. 1b). Among the 19 plots with lodges, only six of them did not fit this model. Regarding the plots without beavers, only one fitted the model, but at that site, the bank was protected by a continuous small-stoned riprap.

In D-area, the surveyed part of shoreline was divided into 89 plots. The canopy cover by tall trees and bushy plants did not differ between up- and downstream riparian woods (Table 2). Regarding tall Salicaceae, values were higher in D-area compared to U-area (mean 13.5%, SD 16.2%, P=0.0009, t=3.42, df=128). With a mean cover percentage of 35.8% (SD 25.1%), D-plots were also significantly richer in small trees than U-plots (mean 9.7%, SD 13.7%, P<0.0001, t=8.54, df=132). This difference was also verified for small Salicaceae trees, whose mean cover reached 9.0% in D-area (SD 12.3%), and 2.1% in U-area (SD 4.9%, P<0.0001, t=4.86, df=112). Both areas did not differ in the abundance of bushy plants. Moreover, from U-plots to D-plots, comparison of canopy cover by woody species, showed a significant decrease of S. fragilis (P=0.0031, t=3.03, df=105) and an increase of S. alba (P<0.0001, t=4.028, df=136). P. nigra cover markedly rose up, reaching 9.42% in U-plots (SD 13.2%), against 23.3% in D-plots (SD 18.7%, P<0.0001, t=6.71, df=158). Regarding non-Salicaceae species, F. angustifolia was more dominant in D-area (P<0.0001, t=6.44, df=139), likewise U. minor (P=0.0027, t=3.045, df=168). No significant differences in bank slope, the presence of sandbank and human frequentation was found between both areas. Ripraps were however more abundant in D-plots (mean 44.8%, SD 32.7%, P<0.0001, t=5.9, df=173).

Twenty-eight D-plots fitted the model. Sixty four percent of them were located on islands. Compared with U-sites with lodges that fitted the model, these plots were markedly richer in tall Salicaceae (mean 31.2%, SD 15.2%, P=0.004, U=222; Table 3), in small trees (mean=25.4%, SD 21.9%, P<0.0001, U=258.5), small Salicaceae (mean=6.1%, SD 10.4%, P=0.001, U=232.5), and in herbs (mean=27.6%, SD 20.8%, P<0.0001, U=252). With a mean of 1.92 (SD 0.76), human disturbance at these plots was higher than at U-plots with lodges (P=0.016, U=210).