Key points are not available for this paper at this time.
Small urban wetlands face intensifying pressures from rapid urbanization, yet their biodiversity patterns remain underexplored due to data scarcity and spatial complexity. We established a Monte Carlo-augmented geographical random forest (GRF) framework to model avian diversity across 45 small wetlands in Hefei, China, documenting 102 bird species (31 waterbirds, 71 terrestrial) along urban-suburban gradients. This methodology addresses concurrent challenges of limited sample sizes and spatial heterogeneity by integrating Monte Carlo simulation with spatially explicit machine learning. The GRF model demonstrated superior performance over conventional approaches, reducing RMSE from 15.234 (ordinary least squares, OLS) to 2.863 while achieving R2 = 0.854 for waterbird species richness compared to R2 = 0.765 for the standard random forest. Monte Carlo simulations stabilized after 100 iterations for waterbird richness and 150 for density metrics. Critical ecological thresholds were identified: Waterbird richness peaked at wetland sizes of 2 ha then declined, while construction intensity exceeding 40% caused a 75% reduction in waterbird diversity (from ∼80 to ∼20 species). Terrestrial birds exhibited consistent positive responses to wetland area, with this factor dominating in 86.67% of sites for both richness and density. Spatial analysis revealed distinct urban gradient patterns—The urbanization index influenced waterbird richness in 35.56% of wetlands primarily located in urban cores, whereas vegetation cover affected terrestrial birds in 28.89% of sites mainly situated in peripheral areas. Factor synergies were strongest in low-quality waterbird habitats (+80.2% above expected richness) and high-quality terrestrial sites (+30.4%), indicating disproportionate conservation benefits from coordinated management strategies. These findings provide quantitative guidelines for urban wetland design: Networks of 1.5–3-ha wetlands optimized for waterbirds, larger refugia (>5 ha) for terrestrial species, and construction intensity limits below 30% within 500 m buffers. This methodological framework offers a transferable approach for biodiversity modeling in data-limited urban systems while establishing actionable targets for wetland conservation planning.
Xiao et al. (Wed,) studied this question.