The fairy pitta ( Pitta nympha ), listed as Vulnerable by the IUCN, breeds in forested valleys across East Asia, yet its nesting habitat in South Korea remains poorly characterized. This study aimed to identify key environmental drivers of nest-site selection and predict suitable breeding habitats in Geoje-si, a core breeding region in southern Korea. Field surveys conducted from 2019 to 2023 confirmed 47 nests, primarily located on rocky substrates or in the forks of large-diameter trees (≥60 cm DBH). Using presence-only data, we developed a MaxEnt species distribution model incorporating ten ecologically relevant variables. Model optimization identified the LQH feature class with a regularization multiplier of 4 as the best-performing configuration, yielding strong predictive accuracy (AUC = 0.881 ± 0.026). The most influential predictors were the Topographic Wetness Index (TWI), dominant tree species, and slope aspect, suggesting a preference for moist, broadleaf-dominated forests on southwest-facing slopes. Habitat suitability was concentrated in low-disturbance, structurally complex valleys, yet only 2.25 km² (23.8%) of the predicted high-suitability area overlapped with legally protected zones. These findings underscore a spatial mismatch between ecological value and conservation coverage, highlighting the need to expand protection to climatically buffered breeding habitats outside current protected areas.

Amphibians are highly sensitive to habitat disturbance because of their permeable skin and reliance on both aquatic and terrestrial environments. To evaluate how human-driven land-use change affects a forest-stream amphibian, we analysed long-term population dynamics of Rana uenoi in the Gayasan region of South Korea, with a focus on breeding habitats. Between 2007 and 2023, annual egg-mass surveys were conducted at 14 breeding sites, and land-use composition was extracted each year within a 500 m buffer from national geospatial databases. Land-use categories were reclassified into ecologically relevant groups (breeding, high-quality, low-quality and non-living/high-mortality habitats) and their areas were quantified for use in regression models. Breeding populations persisted at four focal sites, where annual egg-mass counts declined by up to 98.5% over 17 years and breeding ceased entirely at the remaining ten sites. At the four persistent sites, a multiple regression model relating log₁₀-transformed egg-mass counts to habitat composition and site identity explained a large proportion of the variation in reproductive output. Breeding habitats (wetlands and rice paddies) and high-quality habitats (forests) were positively associated with egg-mass production, whereas non-living/high-mortality habitats (roads, urban and barren land) tended to have negative or non-significant effects once other covariates were included. Low-quality habitats (cultivated fields and artificial grasslands) showed weaker and context-dependent associations. Model comparisons indicated that land-use variables alone provided stronger explanatory power than models including breeding-season climate, and time-series analyses showed no strong directional trend in temperature or precipitation over the study period. Although other unmeasured factors may also contribute to declines, our results underscore the predictive value of habitat-specific land-use metrics and highlight the importance of maintaining wetland-forest mosaics and limiting new road and urban development within a few hundred metres of breeding sites to support R. uenoi populations.