Meter-Scale UAV Hyperspectral Remote Sensing for Pollution Source Attribution: Dynamic Reference Spectrum Optimization and Surface Albedo Correction

We overcome retrieval limitations by introducing a systematic framework that combines UAV-based hyperspectral sensing with dynamic reference spectrum optimization. This study addresses the retrieval accuracy limitations in differential optical absorption spectroscopy (DOAS) for low-altitude unmanned aerial vehicle (UAV) remote sensing under dynamic observation conditions by proposing a systematic innovation framework. We demonstrate the first implementation of quadrotor-based hyperspectral remote sensing for quantitative monitoring of NO2 and HONO with meter-scale resolution, achieved through a dynamic reference spectrum optimization scheme incorporating a weighted multicriteria scoring model (spectral similarity: 0.35, atmospheric condition matching: 0.25, temporal correlation: 0.20, geometric consistency: 0.15, and noise level: 0.05). This methodology resolves aerosol–surface coupling-induced spectral deviations, reducing NO2 retrieval errors by 18–24% compared with conventional approaches. Integration of DeepLabV3+ semantic segmentation with hyperspectral imaging enabled dynamic meter-scale surface albedo correction, reducing biases induced by surface heterogeneity in NO2 vertical column density (VCD) retrievals from 28% to 12%. Field validations at a Hefei coal-fired power plant (detected NO2 plumes: 2.98 × 1016 molecules/cm2) and Changfeng agricultural fields (captured HONO emissions during fertilization: 9.02 × 1015 molecules/cm2) confirm its ability to dynamically detect pollution events in space and time. The research establishes theoretical and technological foundations for transitioning atmospheric monitoring from "end-pipe treatment" to "process control", while providing novel tools for microemission source identification and high-resolution pollution inventory development.