A Study on Enhancing the Audit Efficiency of Natural Resource Asset Management Using Artificial Intelligence

This study introduces a dual-cycle artificial intelligence-audit framework that integrates multiple data sources and policy updates to enhance natural resource asset auditing. It addresses high anomaly miss rates and delayed feedback by embedding a Light Gradient Boosting Machine (LightGBM)-convolutional neural network fusion model within data-driven and decision-feedback loops. The framework processes remote sensing imagery, Internet of Things streams, and contract texts through adaptive attention-based fusion and synchronizes model retraining with audit cycles. Experiments comparing random sampling, robotic process automation-based rules, standalone LightGBM, and the proposed fusion model using coastal province data (2019–2022) show a 69% reduction in processing time (78 to 24 min), a detection rate increase from 45% to 74%, and an interpretability score of 8.7/10. Field pilots in two counties confirm a 29% reduction in audit cycles and a doubling of parcel coverage per auditor. Results demonstrate that coupling artificial intelligence with governance feedback transforms audits into dynamic, efficient, and transparent processes—guiding policymakers and practitioners toward scalable, machine-assisted environmental governance.