A deployment-oriented quality-control framework for integrating UAV LiDAR and field inventory in precision forestry

Reliable integration of UAV-LiDAR and field inventory is essential for operational forest monitoring, but heterogeneous data quality can destabilize plot-level predictions. Conventional quality-control procedures generally focus on filtering individual records or applying fixed pass-fail rules, providing limited support for differentiated decisions at the plot level. We developed a plot-level quality governance framework centred on QC_score. Unlike record-level screening or binary exclusion alone, QC_score combines spatial matching quality, field-LiDAR structural consistency, and data completeness into a continuous measure of plot-level risk, which is then used to prioritize review, determine plot admission, and support quality-adaptive uncertainty reporting. The framework also integrates staged quality control and low-cost repair assessment and was evaluated using 26,767 tree records from 133 plots in Guangdong Province, China. Prediction instability was driven mainly by plot-level structural inconsistency rather than isolated tree-level errors. Excluding structurally inconsistent plots outperformed both full retention and soft quality weighting. For dominant plot height, hard exclusion after QC.3 reduced RMSE from 1.927 to 1.421 m. In the separate cost-based admission analysis, a QC_score threshold of 0.579 retained 95 plots and yielded an RMSE of 1.146 m. In a separate quick-repair experiment, QC123-hard reduced RMSE from 2.597 to 1.459 m relative to the QC12 baseline. Mondrian conformal calibration achieved more balanced coverage across quality grades while assigning narrower intervals to higher-quality plots. The framework therefore provides auditable rules for plot admission, anomaly handling, and uncertainty-aware deployment of UAV-LiDAR forest inventory products.