Results
All variables together explained 50.37% of total inertia of the floristic table on the first six axes (P < 0.01 for each axis). Landscape variables together, i.e. the three scales together, contributed to 15.46% of total explained variance after local factors effects were removed (P < 0.001), and local factors together contributed to 78.71% of total explained variance after landscape effects were removed (P < 0.001), which was expected. Explained variance increased with scale, ranging from 5.36% for the 25-ha scale to 7.72% for the 400-ha scale (P < 0.001, see Table 1). Passing from 25-ha scale to 100-ha scale added 4.72% of explained variance, i.e. about 30% of the landscape part (15.46%), and passing from 100-ha scale to 400-ha scale added 5.57% of explained variance, i.e. about 36% of the landscape part. The 25-ha scale shared only 0.68% of explained variance with the 400-ha scale, while the 100-ha scale shared 1.27% of explained variance with the 25-scale and 2.15% of explained variance with the 400-ha scale (results not shown). This result underlined that it was important to analyze one scale effect independently of the other two scale effects. The pure contribution of 25-ha, 100-ha and 400-ha scales in determining plant assemblages accounted for 3.9%, 3.06% and 5.38% of explained variance, respectively. Moreover, the overdominance of the 400-ha scale in predicting plant assemblages was revealed by a highly significant pure part (P < 0.001); the pure part of the 400-ha scale was 38% higher than the pure part of the 25-ha scale (5.38% versus 3.9%). The pure part of the 100-ha scale was not significant, and came to half of the global part ( Table 1). The pure part of the 25-ha scale was slightly significant (P < 0.01) which suggested that the 25-scale might best predict plant response for some species or certain habitat types.