Large-scale onshore wind potential assessment based on a physics-guided framework coupling aerodynamic mapping and graph learning

High-resolution onshore wind potential assessment is essential for energy transition and macro-siting of large wind power bases. Existing integrated systems often rely on low-resolution land-cover products, empirical weights, and static constraints, which may bias wind-speed correction and limit cross-regional applicability. This study proposes PRM-PIGAT, an end-to-end framework for technical-spatial suitability assessment, coupling frequency-domain phase-rectified landform mapping, aerodynamic-parameterized wind-speed correction, and physics-guided graph learning. PRMNet suppresses texture interference through a semantic-detail dual-stream structure to generate 1 m cross-domain landform maps, which are converted into roughness length and zero-plane displacement for sub-grid wind-speed correction. A semi-supervised graph attention network integrates self-organizing maps, turbine-distribution topological priors, and explicit physical constraints to estimate suitability probabilities, followed by LISA-DBSCAN clustering for high-potential-area extraction and theoretical capacity estimation. Validation across six Chinese cities demonstrates robust transferability. PRMNet achieves a source-domain mean Intersection over Union (mIoU) of 83.39% and improves cross-domain overall accuracy (OA) by 9.77%, while providing clearer boundaries than open-source products. The 1 m landform-driven correction reduces the station-scale mean absolute relative error of GWA wind speed from 32.39% to 6.30%. In leave-one-city-out validation, SOM-PIGAT achieves source- and target-domain Turbine Capture Rate (TCR) of 94.59% and 92.39%, exceeding the second-best model by 5.30 and 10.03 percentage points; meanwhile, full-load-hour (FLH) errors decrease by 56.8% and 32.4%. Ablation, sensitivity, and uncertainty analyses confirm the contribution of the multi-stage modules and the stability and traceability of PRM-PIGAT under data, constraint, and wind-speed perturbations. The framework provides a technical-spatial suitability basis for onshore wind macro-siting and planning-priority identification.