Designing Ecological Networks to Foster Regional Economic Sustainability: source identification in the Longdong Loess Plateau using self-organizing map and complex network theory

Developing robust ecological networks (ENs) is critical for sustaining ecosystem function and biodiversity in the ecologically vulnerable Loess Hills of the Central Yellow River Basin—a region increasingly fragmented by intensive agriculture and infrastructure expansion. Conventional methods for identifying ecological sources often depend on weighted overlays of ecosystem services (ESs), introducing subjectivity and limiting replicability. To address this, the present study combines a Self-Organizing Map (SOM) neural clustering model with complex network analysis to identify ecological sources and enhance overall network structure. Using a 50 km² threshold to define ecologically functional patches, the analysis identified 42 ecological sources—comprising 23 climate regulation-type sources and 19 agricultural provisioning-type sources—accounting for 26.8 % of the total landscape. These nodes were connected through 91 ecological corridors, which were classified into three types: 26 climate corridors, 29 provisioning corridors, and 36 integrated multifunctional corridors. Following optimization, the network exhibited a 6.5-fold increase in total source area and a 2.7-fold rise in corridor density. Quantitative improvements in structural indices were observed, including increased connectivity (α rising from 0.51 to 0.68), greater complexity (β from 1.81 to 2.34), and higher efficiency (γ from 0.69 to 0.82). Robustness simulations under both random and targeted disturbances demonstrated significant gains in network resilience after the addition of eight strategic corridors guided by node betweenness centrality. This research introduces a transferable, data-driven framework that merges machine learning and systems theory for ecological network construction, with implications for spatial planning and environmental resilience in erosion-prone, agriculturally dominated landscapes.