Cascading failure mechanisms in the fresh agricultural product supply chain of China’s Yangtze River Delta

Owing to product perishability and tight inter-firm coupling, fresh agricultural product supply chains are exceptionally vulnerable to cascading failures. Drawing on complex adaptive systems and supply chain resilience theory, this study constructs a real supply chain network of China’s Yangtze River Delta, comprising 1,669 firm nodes and 6,139 supply–demand edges, from enterprise registration records and listed-company disclosures. Perishability is embedded directly in the cascading failure model: a Sigmoid function characterizes overload and underload failure probabilities continuously, load redistribution weights decay with physical distance, and a cumulative load threshold captures progressive deterioration under sustained mild shocks. The DebtRank algorithm, combined with load weights, re-identifies core risk nodes; a dynamic recovery indicator characterizes post-cascade self-repair; and optimal values of the inventory elasticity coefficient ( ε ) and distance decay coefficient ( λ ) are solved through a cost–benefit trade-off. Simulations under random and deliberate attacks show that the network is highly heterogeneous ( γ = 1.214), robust to random attacks yet fragile to deliberate ones: removing merely 10% of the highest-degree nodes destroys roughly 80% of connectivity efficiency, a nonlinear critical phase transition. The cumulative threshold mechanism contributes a real but small additional loss under sustained mild shocks, essentially symmetric between attack types rather than a dominant driver of resilience loss; DebtRank uncovers hidden risk nodes underestimated by degree centrality; post-cascade recovery restores failed nodes at a similar rate regardless of attack type, but restores connectivity efficiency mainly after random attacks, since an importance-blind recovery process is unlikely to reinstate the specific hubs that connectivity depends on; ε exhibits a plateau-type optimum at 0.35–0.45, whereas the resilience benefit of geographic concentration ( λ ) saturates rapidly. Four quantitative strategies (tiered hub protection, moderate inventory buffering, diversified emergency collaboration, and pre-positioned recovery) directly support SDG 12 and SDG 13.