Interspecific acoustic diversity in soniferous sand gobies
Eight soniferous sand gobies, namely Ninnigobius canestrinii ,Pomatoschistus marmoratus (two populations; five individuals originating from Venice lagoon, Italy, and three from Portugal),P. minutus , P. microps , P. pictus ,Orsinigobius punctatissiumus , O. croaticus andKnipowitschia panizzae , produce pulsatile sounds, thus enabling acoustic interspecific comparisons. Interspecific pairwise comparisons revealed interspecific differences in the acoustic features DUR, NP, PRR, and FM (Kruskal-Wallis H test, χ2 = 15.97 ̵̶ 30.19; d.f. = 8; N = 36; P < 0.05), while they did not differ in PF (Kruskal-Wallis H test, χ2 = 11.54; d.f. = 8; N = 36; P > 0.05 (Figure 8a-f ). On average, P. marmoratus (Portuguese), O. punctatissimus , P. microps and K. panizzae were the smallest in size (34 ̵̶ 43 mm LT), while P. minutus , N. canestrinii and P. marmoratus (Italian) were the largest species (50 ̵̶ 59 mm LT). In most cases,K. panizzae differed significantly from other species, especially in DUR and NP (Dunn’s multiple comparison test, P < 0.05). Regarding PF, P. microps had the highest mean values, alongside with K. panizzae (Dunn’s multiple comparison test,P < 0.05). Finally, P. marmoratus (Italian population) and P. pictus differed significantly from the rest of the species having lower values of FM, while other species presented upward or downward modulated sounds (Dunn’s multiple comparison test,P < 0.05) (Figure 8a-f ).
Results of multivariate statistics were summarised by a PCA and LDA, based on the five transformed (log10 or square root) and standardised acoustic features (DUR, NP, PRR, PF and FM), highlighting the acoustic differences (heterogeneity) between species according to sound feature. Due to the FM’s negative raw values, we introduced a positive factor so that we could use this feature in the analyses. In PCA, the first two principal components of the PCA explained cumulatively 69.79% of the variation, with PC1 and PC2 explaining 39.28% and 30.51% of the variation, respectively. On the PC1 and PC2 scatterplot, although several species are clearly separated based on the acoustic features of their sounds, most of the plots overlap (Figure A2 ). PC1 was strongly associated with DUR (-0.69) and NP (-0.67), while PRR (-0.65) and PF (0.58) mostly contributed to PC2. We performed two LDA analysis, first with the complete dataset (five acoustic variables DUR, NP, PRR, PF and FM) and the second excluding the temperature-dependent features (DUR and PRR), to test for sound classification into correct groups (i.e. , species). In the first LDA, the first two axes accounted for a discrimination of 83.36%, with LD1 accounting for 61.09% and LD2 for 22.27%. LDA successfully attributed the most sounds of a sand goby to the correct species according to five acoustic parameters, with a correct interspecific classification rate of 86.11%. For some goby species, a contingency table supports the 100% level of correct classification of sounds (N. canestrinii , O. croaticus , O. punctatissimus, P. pictus and P. microps ), while for the remaining species lower levels were achieved (67% for K. panizzae , 80% for Italian and 67% for Portuguese P. marmoratus , 60% for P. minutus ). In the LDA bi-plot, species clusters overlap, but not significantly, with some taxa occupying relatively isolated positions along the LD axes (Figure 9 ). LD1 supports differentiation of N. canestrinii and P. marmoratus (Italian) from the rest of the species within this acoustic space, while LD2 separated K. panizzae and P. pictus from the other taxa. Plots of other species means show overlapping patterns (Figure 9 ). Interestingly, two populations of P. marmoratus (Italian and Portuguese) overlap substantially on the scatterplot, even though the LDA successfully separated the species. LD1 was significantly loaded with FM (0.32), while LD2 with PRR (-0.47) and PF (-0.28). To explore the effect of water temperature on the interspecific acoustic classification success by LDA, we carried out a second LDA, including only the three acoustic features that are known to be unaffected by water temperature, namely NP, PF and FM. In this second LDA, axis 1 and 2 accounted for 95.49% of discrimination, with LD1 axis accounting for 66.88% and LD2 for 28.61%. However, the second LDA was less successful than the first LDA in accurately classifying the sounds of sand gobies, with a 69.44% rate of correct interspecific classification. Again, some species (K. panizzae , P. microps and N. canestrinii ) achieved 100% classification, while the remaining species were misidentified in different percentages in comparison to first LDA (75% for O. croaticus and 80% O. punctatissimus , 66% for K. panizzae , 40% for Italian and 33% for PortugueseP. marmoratus , 40% P. minutus and 80% for P. pictus ).