Use of vegetation indices in forested regions: Issues of linearity and saturation

Numerous problems and difficulties have been reported with the use of vegetation indices in high biomass, forested regions. In this study we analyzed Landsat-5 Thematic Mapper (TM) scenes from various temperate and tropical forested biomes, representing needleleaf and broadleaf canopy structures in the Pacific Northwest (Oregon), Eastern U.S. (Harvard Forest), southern Chile, the Amazon, and Central America. The TM scenes were atmospherically corrected and reduced to MODIS surface reflectance data at 250 m pixel sizes. Various vegetation indices (VIs) were then computed including the normalized difference vegetation index (NDVI), simple ratio, soil-adjusted vegetation index (SAVI), enhanced vegetation index (EVI), and green vegetation index (GVI). The NDVI was also tested utilizing the green and middle-infrared (MIR) bands. All of the NDVIs were non-linear and were fairly saturated across the forested biomes. In contrast, the remaining indices remained sensitive to canopy structure variations over all of the forested biomes with minimal saturation problems. The high `penetrating' capability of the near:infrared band through forested canopies was the dominant factor in vegetation index sensitivity and performance. We found that indices with higher weighing coefficients in the `near-infrared' to be the best approach in extending vegetation index performance over forested and dense vegetated canopies.