Understanding how functional trait dynamics respond to long-term agricultural disturbance is central to understanding long-term agroecosystem trajectories. Weed seed banks, often dismissed as static repositories of future weeds, can retain disturbance history and act as reservoirs of taxonomic and trait diversity. Here, we analyzed a 12-year dataset from a long-term field experiment comparing continuous cropping, continuous grassland, and crop-grassland rotation systems. We quantified temporal shifts in trait composition using community-weighted means and functional diversity indices, and examined ecological memory, here interpreted as species-level temporal dependence and temporal reconfiguration of community composition, through species-level autoregressive models and temporal beta-diversity. Continuous cropping was associated with increasing seed density, expansion of occupied functional space, and increasing representation of disturbance-associated traits (e.g., annual life cycles, transient seed banks, and ruderal strategies), consistent with directional filtering and comparatively limited temporal legacy signals. Continuous grasslands combined mainly non-directional and oscillatory functional responses with taxon-specific temporal dependence, consistent with persistent filtering under comparatively low disturbance. In contrast, crop-grassland rotations combined directional and cyclical functional reconfiguration, moderate-to-high turnover, and the highest number of taxa with significant autoregressive effects, consistent with a distinctive combination of temporal reorganization and comparatively frequent legacy signals across alternating phases. These findings support viewing the soil seed bank as a dynamic reservoir that retains the outcomes of past ecological filtering and supplies propagules for subsequent community assembly. By integrating perennial phases, crop-grassland systems may help regulate weed communities while maintaining functional variation, representing a potential pathway to resilience that requires testing across additional contexts.
No takes yet. Share an insight, caveat, or question.
Schuster et al. (2026) studied this question.
Synapse has enriched 5 related papers. Consider them for comparative context: