Coupling ecosystem services with machine learning to construct ecological security patterns: a case study of Jiangxi Province, China

Rapid urbanisation threatens regional sustainability in Jiangxi Province, a critical ecological barrier in southeastern China. This study establishes a high-precision, long-term (2005–2024) Ecological Security Pattern (ESP) framework, addressing conventional limitations in subjective weighting and spatial heterogeneity. Coupling the InVEST model with Morphological Spatial Pattern Analysis (MSPA), we found that ecological sources encompass ~ 40% of the provincial territory. However, driven by urban expansion, core source areas contracted by 29,844.69 km 2 (from 67,567.91 km 2 in 2005 to 37,723.22 km 2 in 2024), with the severest reduction (16,903.20 km 2 ) occurring during 2010–2015. For resistance surface construction, a Random Forest (RF) machine learning approach replaced expert weighting. The RF model successfully captured an ‘urban–rural dichotomy’ and outperformed alternative classifiers (Accuracy: 0.8921, F1-score: 0.8929). Circuit theory was then employed for corridor extraction. Results show that core low-resistance corridor density expanded 3.5-fold (from 0.0068 to 0.0238 km/km 2 ). Notably, the Poyang Lake urban agglomeration emerged as the most critically obstructed zone, where high-current-density pinch points peaked at 8,163.25 km 2 in 2020. This data-driven framework provides scientific guidance for enforcing ecological redlines and advancing high-quality, green development.