Addressing the risks of functional land degradation caused by farmland fragmentation and misaligned agricultural inputs, this study investigates the spatial coordination between land consolidation and agricultural service systems in the Beijing–Tianjin–Hebei (BTH) region—a critical area facing both urban expansion and rural transition. Historically, asynchronous development between these systems has led to structural mismatches in Chinese agriculture, increasing the risk of land underutilization, ecological inefficiency, and productivity decline. This study evaluates the degree of coordination between land scale and service scale, and explores their nonlinear relationship with grain yield to inform region-specific optimization strategies for sustainable land use. Farmland Landscape Pattern Index (FLP) and a TOPSIS-based service input index were used to quantify land and service scaling. The Coupling Coordination Degree model assessed land–service synergy, while a generalized additive model (GAM) captured the nonlinear effects of coordination on yield. K-means clustering identified three dominant agricultural systems: (1) low-coordination, low-yield zones in ecologically sensitive areas; (2) moderate-coordination, high-yield peri-urban zones; and (3) high-coordination, moderate-yield traditional farming regions. Within each cluster, XGBoost and SHAP analyses revealed spatial heterogeneity in the yield impacts of key inputs, including irrigation, labor, fertilizer, protected agriculture, and FLP. A second level of clustering was conducted within each major system to identify internally differentiated patterns of input configuration and yield performance, enabling the formulation of fine-grained, county-level land management strategies. Results highlight that excessive input intensification under poor structural coordination may lead to declining efficiency and latent land degradation risks. Enhancing productivity thus depends more on optimizing land structure and aligning service provision than on simply increasing input volume. By integrating coordination analysis with interpretable machine learning, this study provides a hierarchical classification framework—first distinguishing macro-regional systems and then refining intra-cluster strategies—that supports spatially targeted, sustainability-oriented land governance in the BTH region.