Effect of canal blocking on biodiversity of degraded peatlands: Insight from West Kalimantan

PLOS One | https://doi.org/10.1371/journal.pone.0334014 October 8, 2025 6 / 23 minutes continuously throughout 24 hours. Effective total duration in data logging was also set to coordinate with PAM as well as human survey. Environmental data collection was done in 6–25 May 2023 and 4 to 24 August 2023 using the same tools setting and placement. Data analysis Wildlife attributes. We gathered data on both individual counts and species richness to assess biodiversity within different habitats. Various indices were computed based on this data to evaluate biodiversity levels. Menhinick’s and Margalef’s indices were used to determine species richness, while Pielou’s Evenness Index measured evenness across habitats and the Berger Parker Index assessed dominance. Additionally, the Shannon and Simpson Indices provided insights into diversity levels within specific areas. After obtaining index values, we compared them using one-way ANOVA for normally distributed data and Kruskal-Wallis tests for non-normally distributed data. Prior to comparisons, we tested for normality using the Shapiro-Wilk test. Subsequently, we conducted post-hoc tests using Tukey’s honestly significant difference (HSD) to identify specific groups with significant differences following the detection of overall mean variations. We utilized Generalized Linear Model (GLM) to identify relationships between each index and habitat parameters. Here, the indices served as dependent variables, while habitat parameters (including temperature, humidity, canopy cover,

ground cover, tree density, pole density, sapling density, seedling density, understory density, vegetation richness, month, and habitat type) acted as independent variables. Model selection was guided by comparing the Akaike Information Cri- terion (AIC) values, with lower AIC values indicating better model fit. We systematically eliminated independent variables that did not significantly affect the dependent variables, as indicated by high p-values, to produce the best regression model with the lowest AIC. To identify wildlife from acoustic data, we calculated acoustic indices, i.e., biodiversity calculation based on acoustic cues [52], using Kaleidoscope Pro 5.6.2 software (Wildlife Acoustics, USA). All audio recordings retrieved from the first and second field surveys were used without any filtering or special selection. We selected four acoustic indices as prox- ies to quantify biodiversity: Acoustic Complexity Index (ACI), Acoustic Diversity Index (ADI), Acoustic Evenness Index (AEI), and Biodiversity Index (BI). An increased value in ACI, ADI, and BI can be understood as a rise in biological activity, which may further suggest greater biological diversity. The AEI value needs to be carefully interpreted, as a lower number suggests a more even distribution of sound energy across all frequency bands, indicating reduced dominance by a single species or a few species. The numbering of acoustic parameters followed the default settings from Kaleidoscope Pro.

Statistical analyses were done in Rstudio 2023.06.2 + 561 (Posit Software, PBC). Bio-acoustic data were fitted into a Linear Mixed Model (LMM) using the ‘lmer’ function from the ‘lme4’ package (Bates, 2015). The Y variables were acoustic indices and macro taxa detections. Firstly, we included month, transect code, and transect position as a nested random factor, with habitat as a predictor. We ran an ANOVA (type II analysis of variance) from the ‘car’ package [53] to observe the differences in Y variables between habitat types. Pairwise comparisons of two habitats were conducted using the emmeans test from the ‘emmeans’ package [54]. We ran a second linear mixed model, which included environmental variables (temperature) and vegetation features (canopy cover, ground cover, number of plant individuals, and Shannon diversity of plants) as fixed factors. Acoustic indi- ces were the Y factors, while the nested random factors comprised month, habitat, transect position, and transect code. Predictors were selected based on multicollinearity tests and VIF values below 3. We set all p-values significant at 0.05 or a 95% confidence interval. Habitat features. To visualize plant community compositions across different habitats, we employed a Non-Metric Multidimensional Scaling (NMDS) using the vegan package in R [55]. We applied metaMDS function utilizing Bray-Curtis dissimilarity. The NMDS analysis yields a stress value of 0.112, indicating an adequate fit to the original dissimilarity data.

In addition, we presented temperatures and humidity data for all habitats using boxplots to provide an overview about the data distribution.