Relationships Between Landbird Densities and Forest Structure in the National Park of American Samoa, 2011–2023

Hunt, Noah; Judge, Seth; Camp, Richard

Due to differences in the resolution of habitat data collected in 2023 from 2011 and 2018, a qualitative analysis comparing vegetation composition and structure across survey years was provided in Appendix D of the report “Population Densities and Trends of Landbirds in the National Park of American Samoa” (Hunt et al. 2026). Here, we compare bird densities per station within each unit across the three surveys relative to habitat structure (canopy cover and tree height) using principal component analyses (PCA). Principal component analysis is used to reduce the number of variables to a few indices (i.e., combinations of variables) that are linear combinations of the original variables (Sharma 1996). Principal component analysis provides an objective way of combining indicators so that variation in the data is minimized. While typically applied to datasets with many variables, we use it here to aggregate two habitat indicators of bird densities for the three surveys and two units, which we treat as independent. That is, we use PCA as a descriptive, exploratory data analysis tool, rather than an inferential analysis (Jolliffe and Cadima 2016). This allows us to assess changes among years and units using the same component structure. The 2011 survey included sampling points along legacy and random transects (Judge et al. 2013), which were resampled during the 2023 survey, except for the random transect on Mount Lata in the Taʻū Unit. However, sampling in 2018 was conducted only on the legacy transects and excluded sampling above 850 m in the Mount Lata region in the Taʻū Unit (Judge et al. 2021). Therefore, we conducted the PCA with data from (1) the transects sampled in all three surveys, and (2) data collected from all transects surveyed, regardless of being repeatedly sampled. This preliminary analysis allows us to determine if changes in the spatial coverage yields non-representative sampling of canopy cover and canopy height and produces spurious results. Moreover, this analysis allows for assessing the relative magnitudes and sign patterns of the principal components (PCs) loadings and PC scores (Jolliffe and Cadima 2016). We fitted the PCAs using the stats package (version 3.6.2) in base R, which standardizes the data to zero mean and unit variance. We converted the categorical canopy cover and canopy height variables to ordinal numerical values, and defined survey year and island as factor variables. We fitted a singular value decomposition of the centered and scaled data matrix using the prcomp function. Only two PCs were retained, and eigenvalues and eigenvectors of the correlation matrix and percent variance explained were output. Representations of the fitted PCA to the datasets in two-dimensional space accounting for variance with variable vectors were produced with the ggbiplot package (version 0.6.2; Vu et al. 2024). We assessed the strength of correlation between measurements of canopy cover, canopy height, and overall bird density based on their coefficient values with the following criteria: 0.0 = none, >0 to 0.3 = weak, >0.3 to 0.5 = moderate, >0.5 to 0.7 = strong, >0.7 to 1 = very strong.

Relationships Between Landbird Densities and Forest Structure in the National Park of American Samoa, 2011–2023

See also