We used data of 7655 day-traps, finding 378 species of 159 genera (Online resource 2, 3). The most abundant species were Nessaea obrinus with 1277 individuals and Harjesia obscura with 1273 individuals (Online resource 3). We had 43 species with only unique specimens. The adjusted linear model was built using a λ of -0.343 to the richness and -0.303 to the abundance (Fig. 3). In the linear model, we obtained that at least one slope is statistically significant in the richness analysis (F = 27.48, Degrees of freedom = 26 and 7628, P < 0.05), and in the abundance analysis (F = 29.74, Degrees of freedom = 26 and 7628, P < 0.05). The residuals show a median near zero, and symmetric minimum and maximum unlike the unadjusted linear model (Table 1, Online information 1). In the same way, the dispersion and QQ plots show the data is more adjusted to normality and is homoscedastic than the unadjusted model, but they are not normal nor homoscedastic (Fig. 3, Table 1).
Table 1. Summary values of the linear model associated with the richness and the abundance, 1°Q. first quartile; 3°Q. third quartile; Med. Median; K. KSL value; K_P. KSL significance; B. Bartlett test value; B_P. Bartlett test significance value
Formula
|
Min
|
1°Q
|
Med.
|
3°Q
|
Max
|
K
|
K_P
|
B
|
B_P
|
Richness = f (Season x Stratif x Dist)
|
-1.11
|
-0.5
|
-0.008
|
0.5
|
1.4
|
0.1
|
<0.05
|
101.7
|
<0.05
|
Abundance = f (Season x Stratif x Dist)
|
-1.2
|
0.6
|
0.005
|
0.5
|
1.7
|
0.1
|
<0.05
|
119.4
|
<0.05
|
According to the type III ANOVA (Table 2), all the interactions with the vertical stratification were statistically significant (P < 0.05). The interaction of the stratification with the disturbance gradient (Fig. 2a, 2c) produces changes in the diversity difference between disturbance gradients at different strata, but the order is the same, being first SLR, followed by PCR, and finally CCR at all strata. The diversity difference between the disturbance of the highest stratum is wider than the difference in the other strata. Similarly, the interaction between the stratification and the climatic seasonality (Fig. 2b, 2d) changes the diversity difference. The difference between the intermediate season and the other is wider in the low stratum and narrows when we pass to the higher stratum. The richness and abundance difference between the dry and rainy seasons is not significant at the low stratum and increases when going up to the higher stratum, but this is not wider than the differences with the intermediate season. The order is the same in each stratum, with greater diversity in the intermediate season, followed by the dry and rainy seasons respectively.
Table 2. Type III ANOVA summary of the linear model associated with the butterflies richness and abundance, showing the significance of each variable and their interactions. Df. Degrees of freedom, F. Fisher statistic, P. significance value
|
Richness
|
Abundance
|
Variables
|
SSq
|
Df
|
F
|
P
|
SSq
|
Df
|
F
|
Pr..F.
|
(Intercept)
|
3123.9
|
1
|
9192.29
|
<0.05*
|
4104.3
|
1
|
9900.89
|
<0.05*
|
Seasonality (S)
|
17.8
|
2
|
26.16
|
<0.05*
|
24.4
|
2
|
29.43
|
<0.05*
|
Disturbance (D)
|
68.9
|
2
|
101.33
|
<0.05*
|
81.2
|
2
|
97.98
|
<0.05*
|
Stratification (St)
|
151.9
|
2
|
223.50
|
<0.05*
|
214.6
|
2
|
258.80
|
<0.05*
|
S:D
|
2.9
|
4
|
2.19
|
0.067
|
2.8
|
4
|
1.70
|
0.148
|
S:St
|
9.9
|
4
|
7.26
|
<0.05*
|
9.5
|
4
|
5.75
|
<0.05*
|
D:St
|
9.6
|
4
|
7.03
|
<0.05*
|
12.2
|
4
|
7.33
|
<0.05*
|
S:D:St
|
6.9
|
8
|
2.57
|
0.008*
|
7.8
|
8
|
2.36
|
0.015*
|
Residuals
|
2592.4
|
7628
|
|
|
3162.1
|
7628
|
|
|
* Significant
Analyzing only the vertical stratification independently, we found statistically significant differences in the richness (K = 332.36, P < 0.05) and abundance (K = 389.69, P < 0.05), being higher at low stratum (mean richness = 5.01, median richness = 3, mean abundance = 6.75, median abundance = 4 ), followed by the medium stratum (mean richness = 3.46; median richness = 2, mean abundance = 4.24, median abundance = 2), and finally the highest (mean richness = 2.76; median richness = 2, mean abundance = 3.23, median abundance = 2). This decay in the richness and abundance when the stratum increased was also verified in the violin plots (Fig 4). The pairwise analysis revealed a statistical difference (P<0.05) at Dunn post hoc analysis in the richness and abundance (Fig 4). There is a slightly negative correlation between the vertical stratification with the richness (S = -0.22, P < 0.05), and with the abundance (S = -0.12, P < 0.05).
In the same way, we observed statistical differences between the levels of human disturbances in the richness analysis ( K = 121.774, P < 0.05) and abundance ( K = 114.189 , P < 0.05), being higher at SLR (mean richness = 4.52, median richness = 3, mean abundance = 5.87, median abundance = 3), followed by PCR (mean richness = 3.88, median richness = 2, mean abundance = 4.87, median abundance = 3), and finally CCR (mean richness = 3.39, median richness = 2, mean abundance = 4.45, median abundance = 2). A decay in the richness and abundance is observed when moving from the least disturbed to the most disturbed at violin plots (Fig. 4). The pairwise analysis with the Dunn post hoc test revealed statistically significant differences between all the levels (P < 0.05). There is a slightly negative correlation between the human disturbances with the richness (S = -0.125, P < 0.05) and with the abundance (S = -0.121, P < 0.05).
The dry, intermediate, and rainy seasons also showed statistical differences in the richness (K = 66.237, P < 0.05) and the abundance (K = 67.5105, P < 0.05), being higher at intermediate season (mean richness = 4.807, median richness = 3, mean abundance = 6.42, median abundance = 3), followed by the dry season (mean richness = 3.68, median richness = 2, mean abundance = 4.769, median abundance = 3), and finally the rainy season (mean richness = 3.645, median richness = 2, mean abundance = 4.464, median abundance = 2). A clear higher richness and abundance in the intermediate season was shown in the violin plot (Fig. 4). The pairwise analysis with the Dunn post hoc test revealed statistically significant differences between all the levels (P < 0.05).