Results

Lineage genetic diversity

We found significant genetic divergence between the populations ofT. hemprichii in the Western Tropical Indo-Pacific and Central Tropical Indo-Pacific regions. Although we found some minor discrepancies (see Data availability) between the two datasets after carefully inspecting the calibrated fragment lengths of the microsatellites (Hernawan et al. 2017; Jahnke et al.2019a), even after deleting a few microsatellites (e.g., Thh41, TH07 and TH37), two genetic lineages in T. hemprichii remained significantly diverged (i.e., CTIP and WTIP) across the Tropical Indo-Pacific (Fig. 1b, 1c). Genetic variation among lineages accounted for 43.42% of the total genetic variation (Ф CT = 0.43, P < 0.0001; Supporting Information Table S1). Very limited genetic admixture was observed between the CTIP and WTIP lineages. The CTIP lineage harbored strikingly rich genetic diversity, with three times more alleles and allelic richness, and eight times fewer private alleles than the WTIP lineage (Supporting Information Table S2).

Niche differentiation between hypervolumes

The size of the realized niche of the CTIP lineage was one order of magnitude greater than that of the WTIP lineage (CTIP lineage: 17295.6; WTIP lineage: 2273.2) (Fig. 2). Niche differentiation between the two hypervolumes (0.97) was mainly due to variation in niche size (0.79), whereas niche shift contributed only marginally (0.18). Difference in realized niches was easily distinguished via water depth and distance to land, with the WTIP lineage selecting a narrow range of water depth and distance to land (Fig. 2). The two lineages also exhibited niche differentiation with respect to annual mean sea surface salinity. In addition, the CTIP lineage niche was broader with respect to annual mean SST and annual range SST, whereas that of the WTIP lineage was broader for annual mean current velocity, minimum current velocity, and annual range of sea surface salinity (Fig. 2). Niche differentiation between the two hypervolumes was also high (0.86) when we considered only marine environmental predictors (i.e., excluding water depth and distance to land) (Supporting Information Fig. S2).

Model performance

The tuning parameter settings with optimal complexity for the species-level and lineage-level models ranged from relatively simple to complex. The optimal species-level model was the most complex (hinge features and 0.5 RM), while those for the lineage-level models were simpler (CTIP: linear/quadratic/hinge features and 2.5 RM; WTIP: linear/quadratic features and 0.5 RM) (Table 1). The average 10% omission rate was considerably lower for the WTIP lineage-level model (3.57%) than for the other models (CTIP: 26.69%; species: 17.93%; Table 1) — as this was lower than the expection of average 10% omission for the metric, it indicates that the optimal settings results in models that may over-predict to some extent for WTIP. Although omission rate was used primarily for model selection, the average validation AUC scores used to break ties were very high for all optimal models (Table 1); we think this is due to the fact that a majority of presence data are in near-shore waters (Fig. 1a), which likely inflated the model’s ability to discriminate between these presences and background records in deeper water. In addition, all three optimal models had relatively high continuous Boyce index scores (over 0.90; Table 1), indicating that final model predictions matched the presence data well. The eight predictors had different levels of importance in the three models, but water depth and distance to land consistently played important roles (Table 2). In particular, these two predictors accounted for more than 95% of permutation importance in the WTIP model (Table 2). For the CTIP and species models, annual mean SST also had a high permutation importance (~29% and ~24%, respectively) (Table 2). Response curves for water depth and distance to land suggest that shallow coastal waters are more suitable for T. hemprichii (Supporting Information Fig. S3).

Present-day habitat suitability projections

Under present-day conditions, species and lineage models projected similar but not identical habitat suitability patterns, with a large part of the East African coast and the Pacific region as suitable habitat for this species (Fig. 3). Compared with the species model, the CTIP model predicted more southern distribution in Australia (Fig. 3c, 3d). In particular, the CTIP model predicted suitable conditions in the Spencer Gulf, Southern Australia, where the species does not naturally occur (Fig. 3a, 3c). The species model did not capture this pattern (Fig. 3b, 3d). Moreover, the WTIP model identified more suitable habitat in the Red Sea than the species-level model (Fig. 3c, 3d). Overall, species- and lineage-level models predicted comparable suitable areas for T. hemprichii in the WTIP region (species model: 302,800 square km; WTIP model: 315,000 square km), while the species model predicted broader suitable area for the CTIP region (species model: 1,873,800 square km; CTIP model: 1,757,900 square km).

Climate change impacts on habitat suitability

Species- and lineage-level models resulted in different future habitat suitability projections in the CTIP region, with the lineage-level model resulting in predictions of more loss of suitable areas (Table 3; Fig. 4). Both species- and lineage-level models predict considerable future loss of suitable area in the CTIP region, especially on the Sunda Shelf (i.e., Indonesia and Malaysia) (Table 3; Fig. 4). Compared with the species model, the CTIP model projected more extensive range loss under all climatic scenarios (Table 3). Interestingly, both models predicted that the species will shift slightly southwards in Australia.
Species-level and lineage-level models predicted different impacts of climate change on habitat suitability for T. hemprichii in the WTIP region (Table 3). The WTIP model predicted range expansion (except under the RCP 2.6 scenario for the 2050s), whereas the species model consistently indicated range contraction (Table 3). Overall, both species- and lineage-level models predicted that future climate change marginally affects habitat suitability in the WTIP region and that changes in range size were mostly < 15%, with the exception of a higher value (~24%) for the species model in the 2100s for the RCP 8.5 scenario (Table 3). The WTIP model predicted that habitat suitability of T. hemprichii in the WTIP region will remain stable in the future, while the species model predicted range contraction in the Red Sea and expansion in southern Madagascar and South Africa (Fig. 4).
Both species and CTIP models consistently showed that MESS values in the Sunda Shelf were slightly negative, which demonstrates small differences in climatic conditions between the present-day and future scenarios for this region (Supporting Information Fig. S4). For the WTIP region, the lineage and species model showed high environmental similarity except slight environmental dissimilarity in the Red Sea between present-day and future scenarios (Supporting Information Fig. S4). These results indicate a low degree of extrapolation in our model predictions.