Page 6 - FATTORINI2011
P. 6

Fig. 2. The results of the CCA of the effect of island charac-  tionships were negative and positive. No variable was
teristics in determining the species composition of tenebrionid   distinctly associated with axis 4. Of these variables, dis-
beetles on circum-Sicilian islands. The relative importance of    tance to Sicily and distance to Africa have significant
individual geographical variables is expressed by the length of   conditional effects, whilst water depth to Africa was mar-
the respective vectors. A – Area; Ds – Distance to Sicily; Da –   ginally non-significant. Area and distance to Africa
Distance to North Africa; Di – Distance to the nearest island;    showed similar increases in eigenvalue, whereas distance
SDs – Sea depth to Sicily; SDa – Sea depth to Africa; SDi –       to Sicily had a substantially higher value of additional fit.
Sea depth to the nearest island.                                  Island distance to Sicily was particularly important in
                                                                  determining species composition on Pantelleria and the
Variation in species composition                                  Pelagie Islands. By contrast, area was particularly impor-
                                                                  tant for the Maltese Islands.
  The constrained ordination (CCA) biplot for geo-
graphical variables (Fig. 2) resulted in relatively high            Using raw values of land cover categories, the con-
eigenvalues and cumulative percentage variances, indica-          strained ordination biplot resulted in low eigenvalues and
tive of a well structured data set (Table 4). Moreover,           cumulative percentage variances, indicative of a “noisy”
there were strong species-geography correlations with all         data set (Table 5). However, there were strong species-
four axes, which together accounted for about 90% of the          environment correlations with all three axes, which
variance explained by the geographical data. Table 4 also         together accounted for 84% of the variance explained by
shows the correlation coefficients of the geographical            the environmental data. Use of percentage values of land
variables with four axes of the ordination, the results of        cover produced even worse results (Table 5). No variable
the automatic forward selection of the geographical vari-         had a significant effect when raw values were used,
ables, additional fit given by each step and their statistical    whilst the percentage of cultivated and grassland areas
significance. Distance to Sicily was very strongly related        was the only variable that had a (marginally) significant
to axis 1, whereas island area was associated with axis 2.        value (F-ratio = 1.55, P = 0.05) when percentage data
Axis 1 was also negatively related to maximum water               were used.
depth between the island in question and Africa (referred
to as “depth to Africa”). Distance to Africa and the              Inter-island biogeographical similarity
nearest island were related to axis 3; the respective rela-
                                                                    Cluster analysis based on Kulczynski 2 inter-island
                                                                  faunal similarity and UPGMA clustering method pro-
                                                                  duced a dendrogram that reflects the geographical group-
                                                                  ings of the islands (Fig. 3). The first basic split separates
                                                                  the Maltese Islands from all other islands. The latter are
                                                                  subdivided into two main clusters: in one are the islands
                                                                  of the Sicilian Channel (Pantelleria and Pelagie) and in
                                                                  the other larger one are the islands closer to Sicily. In this
                                                                  large cluster, two smaller clusters can be identified: one
                                                                  grouping the Egadi Islands with Ustica and another
                                                                  including the Aeolian Islands and islets. Although three
                                                                  dimensions were retained from NMDS and introduced in
                                                                  CANCORs, a biplot of the first two dimensions indicates
                                                                  that these are sufficient to reflect inter-island relationships
                                                                  (Fig. 4). The third dimension was also not significant in
                                                                  CANCORs (see below).

  TABLE 4. Results (F-statistics) of CCA for the geographical variables. O a indicates the increase in eigenvalue (additional fit). P
indicates the significance level of the conditional effects based on Monte Carlo tests (999 random permutations). Variables are the
same as in Table 1.

     Variable  Oa  P                         F                            Weighted correlations

                                           2.682                  Axis 1  Axis 2  Axis 3         Axis 4
                                           1.900                  0.901                          –0.177
        Ds 0.51 0.004                      1.930                  0.211   –0.051  –0.142         –0.336
         A 0.35 0.083                      1.720                  –0.594                         –0.355
        Da 0.34 0.016                      1.516                  –0.744  0.687   0.221          –0.420
       SDa 0.29 0.060                      0.987                  0.139                          –0.327
        Di 0.25 0.134                      0.845                  –0.255  0.208   –0.614         –0.231
       SDi 0.16 0.480                                             –0.414                         –0.447
       SDs 0.14 0.576                                             0.555   –0.224  –0.297         0.207
Eigenvalues                                                        11.3                           32.6
Cumulative % variance                                              27.1   –0.224  0.682           77.8
Cumulative species-geography relationship                         0.964                          0.899
Species-geography relationship                                            –0.281  0.000

                                                                          –0.415  –0.228

                                                                          0.456   0.376

                                                                          20.6 28.3

                                                                          49.3 67.7

                                                                          0.961   0.963

664
   1   2   3   4   5   6   7   8   9   10   11