Investigation of the Utility of Spectral Vegetation Indices for Determining Information on Coniferous Forests

This work presents an investigation of the forest information content of spectral vegetation indices including the SR, NDVI, PVI, SAVI, TSAVI, and GEMI. The results were derived from two separate reflectance models including a geometric–optical and a Monte Carlo ray-tracing model, assuming nadir viewing. The effects of the atmosphere on the indices were also examined, using the 6S atmospheric code, and a short section is devoted to an analysis of Landsat Thematic Mapper data. Overall, the indices were not linear with respect to forest cover and, in many situations, an index did not have a unique value for a given cover. The general form of the indices’ responses to changes in cover could be explained by shadowing effects, linked to the heterogeneous nature of forest. The indices were also significantly affected by perturbations including solar zenith angle, background reflectance, stand structure, and leaf-area index. The relative importance of these perturbations depended on both the cover and on the particular spectral index tested. Existing indices should be selected according to the characteristics of the site under investigation, since no single index performed well over all covers. At low covers, GEMI performed best, where a decrease in the value of GEMI corresponded to an increase in crown cover. There was some evidence that the soil-adjusted indices (PVI, SAVI, TSAVI, and GEMI) partially reduced background reflectance effects in the data, although the reductions were not always associated with increased sensitivity to crown cover. At high covers, SAVI and TSAVI were found to perform best, since these indices had large dynamic ranges and small susceptibility to atmospheric perturbations. Further work is required to develop spectral indices specifically for conifer forests.