3.1 Niche differentiation among the two populations
To minimize the dimensionality of the niche space, we initially
conducted a principal component analysis (PCA) on environmental
variables. The top four principal components were retained as they
collectively explained 81.7% of the total variance (Fig. 3A). The EIOS
(1992.44) had a larger four-dimensional hypervolume than the WPI
(1588.38). There was a modest level of niche differentiation between the
two populations (βTotal = 0.465), primarily driven by niche shift
(0.180), which accounted for 49.32% of the observed difference, and
niche contraction/expansion (0.184), which accounted for 50.41% of the
difference. Analyzing the niche centroids for the two population
comparisons revealed that PCA1 played a crucial role in explaining the
distinct niches of the EIOS and WPI
populations (Fig. 3B). This difference was mainly attributed to
variations in mean temperature, mean dissolved oxygen, phytoplankton
abundance, and mean salinity (Fig. 3C).
3.2 Model training fitting
and environmental variable assessment
To improve model performance, a
10-replicate five-fold cross-validation procedure was employed. During
simulation, 20% of the known species distribution data was used for
model validation, while the remaining 80% was used for training. The
predictive performance of the 10 modeling algorithms varied, as
indicated by different TSS and AUC values. The modeling technique with
the highest predictive performance, determined by TSS and AUC values,
was selected. Eight out of the ten single models (excluding MAXENT and
SRE) were chosen to create weighted ensemble models at both the species
and population levels (Fig. 4A, 3B). Overall, the ensemble models
demonstrated good accuracy for most species, with AUC values above 0.80
and TSS values above 0.70. The high AUC and TSS values across all three
ensemble models indicated strong predictive performance (Table 2).
The habitat suitability models developed in this study incorporated nine
different predictors, each contributing differently to the modeled
species (Fig. 4C). In general, land distance and light at bottom emerged
as the most important predictors for T. maxima , contrasting with
ocean depth, current velocity, dissolved oxygen, salinity, mean
temperature, and phytoplankton (Fig. 4C). The species-level model
revealed that the distribution of T. maxima was primarily
influenced by land distance and light at bottom. The model predicted
that T. maxima tends to prefer environments with land distances
between 0 and 30 km (Fig. 5A1) and light at bottom between 5 and 45
(Fig. 5A2). At the population level, land distance and mean temperature
were the most significant predictors for the distribution of EIOS, while
light at bottom and land distance played key roles in predicting the
distribution of WPI. For the EIOS population, occurrence probability
decreased with increasing distance from land and was highest when the
mean temperature ranged from 10 to 30 °C (Fig. 5B1, 4B2). As for the WPI
population, preferred habitats exhibited light at bottom above 20 and
were located near the coast within approximately 100 m (Fig. 5C1, 4C2).
3.3
Habitat suitability in scenarios of the current and future climatic
conditions
The modeling analyses were conducted at both the species level (species
model) and population level (EIOS model and WPI model). Under present
conditions, potential habitat for T. maxima exhibited higher
suitability indices in the Indo-Pacific core area. In terms of
distribution ranges, the predictions from the species-level and
population-level models frequently demonstrated good agreement (Fig.
6A). T. maxima ’s preferred habitats occurred primarily in shallow
coastal waters, with little of it in deep ocean regions. Notably, the
species-level prediction showed the largest suitable habitat area
(1,519,764.73 km2), while the two population-level
predictions indicated the areas of 1,326,478.08 km2and 1,204,511.84 km2, respectively. Despite limited
ecological niche differentiation, certain variations in the projected
results were identified between two populations that were predicted by
EIOS model and WPI model (Fig. 6B and C). Moreover, distinct differences
are observed in certain regions. For instance, in the South China Sea,
the species-level model predicts the smallest suitable habitat area forT. maxima , whereas the WPI population-level model predicts the
largest area. In the Strait of Malacca, the EIOS model predicts a larger
suitable zone compared to the species-level and WPI models.
The extent of habitat change depends on the scenario of climate change.
Under the pessimistic scenario of uncontrolled greenhouse gas emissions
(RCP 8.5), significant changes in the suitable range are projected
(Table 3, Fig. 7). The species-level and population-level models predict
different impacts of climate change on potential suitable habitats. The
species-level model presents a more pessimistic outcome with greater
loss of potential suitable habitats for T. maxima in the Nansha
Islands, Strait of Malacca, and Java Sea regions under RCP 8.5
scenarios. The EIOS model predicted moderate loss of suitable areas forT. maxima among the Indo-Pacific core area, while the WPI model
forecasted substantial loss of suitable areas in shallow waters
surrounding the Philippine Islands, Sumatra Island, and Java Island.
Unlike the species-level and WPI models, the EIOS model predicted a gain
(approximately 2.43%) of spotted suitable habitats in coastal waters of
the Nansha Islands, Mindoro Island, Eastern and Southern Indonesia under
RCP 8.5 scenarios.