Landscape fragmentation in arid inland river basins, driven by intensifying human-environment conflicts, severely compromises ecosystem service (ES) functionality and stability, hindering regional sustainable development. Traditional static assessments map ES supply-demand patterns but overlook the spatio-temporal ecosystem service flows (ESFs), which determine actual ES realization in fragmented landscapes. This study introduced a spatio-temporal ESFs framework to address this gap. Supply and demand for two ESs in the Shiyang River Basin (SRB) were assessed using InVEST and Carnegie-Ames-Stanford Approach (CASA) models. Network models were then constructed to simulate the changes of ESFs under landscape fragmentation, which was calculated through landscape pattern indices and moving window analysis. Results revealed that high-supply areas for both services concentrated in vegetation-rich upstream regions, while high-demand areas predominated in desert and urban zones. The carbon sequestration ESF network remained largely intact, whereas soil conservation ES flows from upstream failed to meet midstream and downstream demand. Moderate fragmentation, comprising over 30% of the study area, represented the dominant landscape type. Critically, landscape fragmentation exerted predominantly negative effects on ESF connectivity, though impacts varied by transmission medium. Water-mediated soil conservation flows proved highly sensitive to structural disruption, while wind-mediated carbon sequestration flows showed greater resilience. These findings underscore that ESF quantification must account for service-specific transport mechanisms and that fragmentation thresholds differ between water and wind driven processes. The network-based approach provides decision-makers with spatially explicit insights for sustainable landscape management, enabling targeted interventions to restore biophysical connectivity and enhance ES provision in vulnerable arid inland river basins.