Abstract
In drylands, abrupt vegetation changes may signal declining resilience and regime shifts, yet global assessments have largely focused on greenness trends while overlooking the role of spatial organization. Here we analyzed 15,898 1-km² landscape units across global drylands (aridity index < 0.65) using 30-m land cover and NDVI time series. We quantified multidimensional spatial structure via landscape metrics, patch co-occurrence networks, semivariogram-based indicators, and texture features, then examined associations with abrupt NDVI step shifts using PCA, XGBoost, and correlation networks. Spatial structure consistently resolved into two orthogonal axes: fragmentation–connectivity and configurational complexity/interspersion. Associations were strongly state-dependent: increasing fragmentation and complexity were more consistently linked to negative steps (abrupt browning) than to positive steps (abrupt greening). Positive steps were explained by fewer predictors, dominated by stable contiguity, whereas negative steps involved broader predictors including area proportion and fragmentation trajectories. Regime-dependent rewiring of correlation networks further indicated coordinated spatial breakdown under browning versus constrained reorganization under greening. We conclude that dynamic, land-cover-specific spatial structure provides mechanistic and predictive information on abrupt vegetation change beyond static fragmentation or mean greenness. Browning and greening represent distinct modes of spatial reorganization, supporting a shift from static assessments toward trajectory- and network-informed monitoring of dryland vulnerability and resilience.