ADVANCING AIRBORNE HYPERSPECTRAL DATA PROCESSING AND APPLICATIONS FOR SUSTAINABLE AGRICULTURE USING RTM-BASED MACHINE LEARNING

Airborne hyperspectral remote sensing data can provide rich spatial and spectral information at very high resolutions (spatial: <1 m; spectral: <5 nm) to accurately monitor diverse agroecosystem variables such as crop traits, soil properties, and agricultural practices. This high-resolution information is beneficial to agricultural stakeholders for decision-making, timely management, and precision evaluations, which can further advance sustainable agriculture. However, it is challenging to process large-volume airborne hyperspectral data and fulfill the potential of airborne hyperspectral information for sustainable agriculture. To overcome these challenges, we developed an operational framework to process and utilize airborne hyperspectral images using Radiative Transfer Modeling (RTM)-based Machine Learning (RTM-ML) approaches. The framework can accurately and efficiently conduct atmospheric correction of and retrieve agroecosystem variables such as crop nitrogen, cover crop aboveground biomass, and crop residue from airborne hyperspectral data. Extensive validation against ground truth data indicated that the framework can efficiently derive surface reflectance from airborne raw data with mean absolute errors<0.03 and cosine similarities>0.99; and reliably quantify crop nitrogen, cover crop aboveground biomass, and crop residue with an R-2 of 0.85, 0.72, and 0.82, respectively. Our framework demonstrated strong capabilities in processing and utilizing airborne hyperspectral data for sustainable agriculture.