Аңдатпа
In arid and semiarid regions, flooding and land use change pose challenges to ecological risk (ER) assessment and ecosystem security. This study developed an integrated Bayesian network (BN)–Gaussian mixture model (GMM) framework for Kazakhstan (2000–2022) to characterise time-varying probabilistic dependencies of ER and transitions among ecological risk states. Results revealed spatial decoupling between disturbance intensity and ecosystem services (ESs). High-disturbance zones were fragmented across central and southern agricultural oasis systems, whereas high ES provision was concentrated in northern and eastern regions. Nationally, ER was dominated by a low-risk background with localised high-risk clusters. BN analysis revealed a reorganisation of the probabilistic dependency structure around 2010, after which flood- and runoff-related variables became increasingly prominent, indicating a stronger association between ER dynamics and hydrological variability. GMM analysis showed substantial risk-state reorganisation, with 54.99% of the area transitioning among ER states, while flood-risk areas exhibited the highest persistence (64.10%). Transitions between stable ecological and anthropogenic activity-enhanced states further highlighted human–environment interactions. Overall, Kazakhstan exhibited a relatively stable broad ER structure alongside substantial local state transitions and stage-dependent dependency reorganisation. These findings highlight the need for hydrologically informed and spatially differentiated ER management in arid regions.