The area of land covered with shrubs and taller trees within the urban park (= area of vegetation cover) was calculated using aerial photographs taken in 1998. Surrounding land uses were computed with GIS (geographic information systems) using topographic maps (1:2,500) and the aerial photographs. All land uses were classified into nine land-use types: rice field, dry field, orchard, woodland, bamboo forest, residential, road, wasteland and water. The proportion of each land-use type within 500 m of the edge of park was calculated for each urban park.
First the correlation of the size of the urban park and the area of vegetation cover to species richness was confirmed by regression analysis. In order to find patterns of bird species compositional change among the urban parks, TWINSPAN (two-way indicator species analysis; Hill 1979) was carried out. Bird species that were recorded in only one park were omitted from the analysis. As the result, 16 species (out of 21 species observed) were included in the analysis. The total number of recorded individuals by species was used in the calculation. The pseudospecies cut levels were defined as follows: 0, 1, 2-9, 10+ individuals. The groups that represented less than five parks were not split further.
Classification-tree analysis was performed to determine factors relating to compositional change. Classification-tree analysis has recently been used to determine major environmental factors of variation in plant/animal communities (e. g., De’ath and Fabricius 2000). Park grouping obtained by TWINSPAN was used as the dependent variable for classification-tree analysis. Explanatory variables of the analysis were the size of the park, the area of vegetation cover, PBSUM, SPBSUM, TPBSUM, PBHD and the proportion of different land-use types within 500 m.