A Large Language Model-Oriented Framework for Effectiveness Assessment of Whole-Region Comprehensive Land Consolidation: A Case Study of Guangdong Province

Whole-Region Comprehensive Land Consolidation (WRCLC) faces persistent evaluation bottlenecks in practice: traditional quantitative monitoring captures physical land-cover changes but fails to interpret cross-sectoral policy semantics, whereas manual administrative review suffers from high operational costs, subjective inconsistency, and severe information asymmetry. To address these challenges, this study develops a domain-adapted Large Language Model (LLM) framework for structured preliminary effectiveness assessment of WRCLC, using Guangdong Province as a case study. Following the standardization and integration of multi-source heterogeneous data across county-level units, a domain-specific model is established using a knowledge distillation framework based on a hard-constraint mechanism combined with LoRA fine-tuning. A six-dimensional evaluation system is established to evaluate project progress, factor endowment guarantee, structural optimization, value realization, bottom-line constraints, and regional characteristics. Qualitative and quantitative validation across typical county-level administrative divisions in Guangdong demonstrates high alignment with expert judgment and robust capabilities in contextual narrative generation. Rather than replacing administrative decision-making or expert review, this framework demonstrates the feasibility of generative artificial intelligence as an auxiliary decision-support tool. Operating within a human-in-the-loop governance paradigm, the proposed approach effectively bridges quantitative spatial metrics and qualitative governance semantics, significantly reducing administrative drafting burdens while enhancing process monitoring, compliance review, and adaptive land governance.