Analyzing the effect of shadow on the relationship between ground cover and vegetation indices by using spectral mixture and radiative transfer models

Abstract

We present and evaluate an experimental relationship between the fraction of ground cover (FV) and multispectral vegetation indices (VI) derived from medium resolution images (Landsat 5-TM) in vertical shoot trellised vineyards. The results indicate a strong linear relationship between FV and the Normalized Difference Vegetation Index (NDVI) and Soil Adjusted Vegetation Index (SAVI), resulting in correlation coefficients greater than 0.90. These relationships were evaluated for the effect of variations in illumination angles and shadow enlargement using two analytical approaches: the Linear Spectral Mixture Analysis techniques and a radiative transfer approach with the Markov-chain canopy reflectance model, with additions to simulate the row structure. Previous to this analysis, both models were evaluated by comparing the model results with VIs in row vineyards obtained from satellite images, performing fairly well. The exploratory analysis demonstrated that the use of a single relationship based on the NDVI index could result in significant inaccuracies for larger zenith angles and row directions perpendicular to the sun azimuth at the satellite acquisition time. In contrast, the SAVI improved the linearity of the relationships and resulted in less sensitivity to changes in the sun angles and row directions.