Assessment of spectral vegetation indices for riparian vegetation in the Colorado River delta, Mexico

This study tested the relationship between three, commonly-used vegetation indices (VIs), percent vegetation cover (% cover) and leaf area index (LAI) over a complex riparian landscape in the Colorado River delta, Mexico. Our objective was to correlate vegetation and soil features with VIs using low-level aerial photography, in preparation for scaling up to analysis of vegetation features using satellite imagery. We used a three-band digital imaging camera (Dycam) to collect data from an aircraft flying at 150 m. A series of 84 images (67×100 m) were analysed. Nine of these sites were ground-truthed; the species, % cover, and LAI were determined. Measured LAI (nine sites) from tree, shrub, and groundcover categories were used to determine a global (GLAI) value for 63 images. We conducted both VIs:% cover and VIs:GLAI regression analyses. The normalized difference vegetation index (NDVI) was the VI that best predicted % cover (r2=0•837), but the soil adjusted vegetation index (SAVI) and enhanced vegetation index (EVI) gave nearly equal results (r2=0•807 and 0•796, respectively). Normalized difference vegetation index, SAVI and EVI were less useful in predicting GLAI (r2=0•73, 0•65, 0•64, respectively). Variability in GLAI was due mainly to differences in % cover among images rather than differences in LAI among vegetation types. We also measured reflectance values of the major plant types between 450 and 900 nm, and found small but significant (p<0•05) differences among some of the species. The results support the conclusion that vegetation indices are most simply related to % vegetation cover, rather than species differences in LAI or VIs, even in this mixed riparian biome. There was also a near 1:1 correspondence between the Dycam and Thematic Mapper (TM) NDVI values over a wide range of landcover types (water, bare soil, partial and complete vegetation cover), which indicate that reflectance-based NDVI values can be scaled from low-level aerial Dycam images to satellite images for this ecosystem.