Unveiling the restoration effectiveness of China's Shan-Shui Initiatives: Insights from dynamic reference ecosystems and geographically explainable machine learning

Ecological degradation threatens global sustainability and has prompted major investment in landscape restoration. China's Shan-Shui Initiatives (SSIs), integrating Mountain-River-Forest-Farmland-Lake-Grassland (MRFFLG) management, have rehabilitated millions of hectares of degraded land. However, systematic evaluation remains constrained by static benchmarks, limited treatment of spatial heterogeneity, and poor understanding of nonlinear ecological responses. To address these gaps, we developed an integrated framework combining dynamic reference ecosystems (DREs), spatiotemporal counterfactual analysis (SCA), and geographically explainable machine learning (GeoXAI) to evaluate 25 SSI pilot projects implemented during 2016–2020 and elucidate the underlying mechanisms. Results showed that the identified DREs exhibited high ecological integrity (90th–95th percentile of relative ecosystem quality, REQ) with remarkable temporal stability (SD = 0.001–0.002), providing reliable dynamic baselines. From 2015 to 2020, restoration reversed persistent degradation: ecological restoration project areas (ERPA) gained 1449.41 km 2 of ecological space, 85.6% more than the 781.17 km 2 gained in adjacent non-restoration areas (NERPA). Excellent-grade ecosystems increased by 3.14%, while REQ rose slightly more in ERPA (+0.020) than in NERPA (+0.018). The GeoRF model performed best (R 2 = 0.6984, RMSE = 0.0116). Geographic location was the leading predictor (24.40%), and hydrological and climatic factors jointly contributed over 35%. The largest predicted gains were associated with moderate runoff reductions (∼10 mm), slight precipitation increases (∼10 mm), minor cooling (−0.3 °C), baseline REQ of 0.18–0.65, and soil thickness > 100 cm. This framework offers a transferable approach for restoration assessment and supports climate-adaptive, region-specific management.