Optimization of multiple ecosystem service flows for ecological restoration using Bayesian networks: A case study of the Wujiang River Basin in Southwest China

Facing ecosystem service imbalances, rocky desertification, and soil erosion under growing human pressure, China's Karst region requires spatial planning to restore supply-demand balance, optimize ecosystem service flow (ESF), and ensure ecological security. Using breakpoint and field strength models, we delineated the spatial flows of water retention (WRF), soil retention (SRF), and carbon sequestration (CSF) in the Wujiang River Basin in Southwest China. Although the supply-demand ratio of ecosystem services (ESDR) indicated that supply exceeded demand basin-wide, deficits emerged in some county-level divisions within the cities of Guiyang, Zunyi, and Liupanshui. WRF showed medium-to-high levels (ESDR >0.06), SRF remained high (ESDR >0.08) along the main river channel and its surrounding counties, and the CSF linked the nine southwestern demand counties to the adjacent supply counties. Redundancy analysis identified precipitation (51.5 %), NDVI (34.1 %), and population density (8.5 %) as key drivers of ESF variability. The integration of Bayesian network with ESF analysis enabled the precise delineation of optimization areas (WRF: 10712 km 2 ; SRF: 12163 km 2 ; CSF: 14233 km 2 ). Scenario analysis delineated distinct restoration priorities: Scenario I (single-service decrease) targeted 9811 km 2 (CSF), 9728 km 2 (WRF), and 8545 km 2 (SRF), whereas Scenario III (simultaneous decrease) highlighted a critical 1296 km 2 zone for integrated intervention. The novel integration of ESF mapping with Bayesian network yielded a powerful analytical framework, informing targeted policy recommendations and providing a scientifically-grounded strategy for the transformative management of basin-scale ecological security and ecosystem services. • Diagnosed a basin-scale ES supply-demand imbalance, with core deficit areas in Guiyang, Zunyi, and Liupanshui. • Discovered that precipitation, NDVI, and population density are pivotal drivers of ESF spatial patterns. • Constructed a novel optimization framework integrating Bayesian networks with ESF modeling. • Quantified key optimization zones (10712–14,233 km 2 ) for enhancing water, soil, and carbon service flows. • Simulated restoration priorities (1296–28,084 km 2 ) under three ESF-decline scenarios.