The general process of thermophilization and the hypothesized framework of thermophilization in fragmented habitats.

Two general processes of thermophilization are shown on the top: increasing colonization rate of warm-adapted species, and increasing extinction rate of cold-adapted species over time (a). Compared to continuous habitats, a hotter microclimate caused by lower buffering ability on fragmented patches may attract more warm-adapted species to colonize while causing cold-adapted species to extirpate or emigrate faster (b). Loss of cold-adapted species can also be exacerbated when highly fragmented patches harbor lower habitat heterogeneity (e.g. resource, microrefugia), which will also reduce colonization of warm-adapted species (c). Isolated patches due to habitat fragmentation will block warm-adapted species’ colonization and cold-adapted species’ emigration under warming. Relative species richness is shown by the number of bird silhouettes in the community (d).

The location of Thousand Island Lake and the experimental islands.

The map was created using ESRI (Environmental Systems Resource Institute) ArcMap software (version 10.3). The base map sources include Esri, Maxar, Earthstar Geographics, and the GIS User Community.

Temporal trend in CTIoccur and CTIabun in the TIL.

Vertical lines in (a) are posterior estimated CTI (mean ± SD); solid black line and shaded area are predictions and 95% credible intervals extracted from JAGS modeling posterior mean CTIoccur as a function of year, and using island identity as a random effect while accounting for the variation in posterior CTIoccur. Points and error bars in (b) are observed CTIabun (mean ± 2SD); solid black line and shaded area are predicted values and 95% confidence intervals estimated from LMM modeling observed CTIabun as a function of year, and using island identity as a random effect.

Relationship between STI and occupancy trend and the relationship between species thermal preference and trend in occupancy, colonization, and extinction rates.

Each point and error bar in (a) represents the temporal trend of occupancy rate (posterior mean and 95% credible interval of year effect on occupancy rate) for each species, with the filled dots indicating a significant year effect while the hollow dots indicating nonsignificant year effect. The number of species with significant occupancy trends in each quadrant was added to the plot. The black dashed line and shaded area are the predicted values and 95% confidence intervals of the weighted linear regression model. Each point in (b) represents the posterior mean estimate of the colonization trend and extinction trend for each species. The color of the point indicates the temporal trend in occupancy.

Relationship between STI and abundance trend.

Each point and error bar represents the temporal trend of abundance (year effect and 95% confidence interval of year effect on occupancy rate) for each species, with the filled dots indicating significant year effect while the hollow dot indicating nonsignificant year effect. Cold-adapted species are plotted in blue and warm-adapted species are plotted in orange. The number in each quadrant indicates the number of species with significant occupancy trends. The black solid line and shaded area are the predicted values and 95% confidence intervals of the weighted linear regression model.

Posterior estimates related to cold-adapted species’ extinction rate and warm- adapted species’ colonization rate.

Points are species-specific posterior mean. Only cold- adapted species with positive trends in extinction (a) and warm-adapted species with positive trends in colonization (b) are shown, which are two main processes contributing positively to thermophilization. Points in blue indicate significant effects (95% or 80% credible intervals) and points in white indicate non-significant effects. All points were jittered slightly by 0.22 in width.

Breeding season temperature from the year 2012 to 2021 in the Thousand Island Lake, China. Breeding season temperature was calculated as the average of the mean monthly temperature for breeding seasons (April – June). Black lines and grey shade indicate the regression line and confidence interval of the linear regression model (slope = 0.078, r2 = 34.53%, p = 0.04).

Robustness of STI to changes in data resources.

The units on the horizontal and vertical axes are degrees Celsius. Pearson correlation coefficients between STI used in our study and STI calculated with different time-windows: March to August (a) and January to December (b), were significantly high. Pearson correlation coefficients between STI used in our study and STI calculated with different distributional ranges: the distributional range in China (c) and distributional range across the world (d), were significantly high.

Species Temperature Index of 60 species in the Thousand Islands Lake, China. The dotted vertical line indicates the median of STI values; the dashed vertical line indicates the mean of STI values. Grey points are Oriental species, red points are Palearctic species and purple points are Eurytopic species. The bottom right presents the frequency of STI values (right) and the overview relationship between total abundance (natural logarithm transversion of total abundance over 10 years and on all islands) and STI.

Relationship between STI and mean estimates and 95%, and 80% credible intervals of posterior distribution for temporal trends of species extinction rate

(a) and species colonization rate (b). Points and error bars in blue or orange indicate significant posterior estimates and error bars in grey indicate non-significant effects. Results of linear regression models modeling posterior mean of temporal trends in colonization or extinction weighted by standard deviations were added to plots. Partial regression effects of year effect on total species richness (a) and total abundance (b) for warm versus cold-adapted species groups. The solid lines indicate significant effects. The shaded bands are 95% confidence intervals of fitted values with other variables at their mean values. Total richness was calculated as the mean of the posterior estimate of species absence/presence, while total abundance was calculated from raw maximum abundance data.

Relationship between island area and understory breeding season air temperature of 20 islands monitored in the year 2022.

The understory air temperature was monitored in 20 selected study islands differing in island area. Each plot represents the average air temperature (April – June) of a site within the islands. Lines and shaded areas are predicted values and 95% confidence intervals from LM models.