Spectral Unmixing Model for Precision Monitoring of Vegetation Using Multi/hyperspectral Imaging

A pixel over a vegetated surface is divided into five fractions, including photosynthetic vegetation (consisting of sunlit and shaded canopy components), non-photosynthetic vegetation, bright bare soil, and shadowed soil. The proposed model will be developed based on the linear spectral mixture theory to utilize field/unmanned aerial vehicle hyperspectral remote sensing data. Using the model, spectral signatures in mixed pixels caused by the mixtures of the five fractions mentioned above can be unmixed at high spatial resolutions (10- 30 m). Meanwhile, from spectral features of the five components and their spatial distribution patterns, the sunlit and shaded canopy components can be separated.

Red edge parameters and NDVI of two species of coniferous trees were studied using an unmanned aerial vehicle's hyperspectral visible-to-near infrared images. Positive correlations between vegetation red edge slope and reflectance with different illuminated/shaded canopy proportions were obtained. we devised a new vegetation index named normalized difference canopy shadow index (NDCSI) using red edge's reflectance and the NDVI. Combined with the bare soil index (BSI), NDCSI was applied for linear spectral mixture analysis (LSMA) using Sentinel-2 and Landsat-8 multispectral imaging. We can calculate and extract the illuminated vegetation canopy from satellite images.

Because the variation in soil endmember spectrum signatures significantly changes with soil moisture (SM), spectral unmixing with fixed endmember spectra can lead to poor accuracy of the abundance of the spectral constituents of pure crop residue. Herein, this paper presents a dynamic soil endmember spectrum selection approach for improving the performance of soil and rice residue spectral unmixing analysis in rice residue cover (RRC) estimation. This new approach uses SM and soil spectral reflectance model to modify soil endmember spectra in each spectral unmixing analysis. Results indicated that the distribution of SM in farmland was crucial for RRC estimation. Our proposed approach can be used to improve RRC estimation accuracy in harvest field.