The performance and response of the MODIS Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI) were evaluated over the Arctic coastal plain of Alaska. At the 250–1000-m resolution of moderate resolution sensors, a substantial portion of this landscape is a mixture of vegetated and non-vegetated cover types. Single-date MODIS swath scenes were used because of the higher geolocation accuracy, lack of radiometric artifacts, and temporal specificity. A higher resolution earth cover classification was used to sample pixels with a mixture of vegetation, water, and barren ground. The MODIS NDVI and EVI were compared to aggregated Landsat ETM+ NDVI. The subpixel ETM+ NDVI was a good predictor of the MODIS EVI in all mixed pixels, and of the MODIS NDVI in mixed vegetation and barren ground pixels. In these cases a simple linear relationship between subpixel ETM+ NDVI and the MODIS vegetation indices was observed. In the mixed pixels with vegetation and water, the MODIS NDVI had a curvilinear response to the ETM+ NDVI and the performance decreased as the subpixel water fraction increased.
Spectral mixture modeling was then applied to synthesize mixed pixel spectral values plot the response of the MODIS vegetation indices to subpixel non-vegetated fractions. The MODIS NDVI had a highly variable response to subpixel fractions of different non-vegetated backgrounds, while the MODIS EVI was fairly insensitive to background type. The models also suggest that large changes in observed NDVI values could occur due to changes in the spectral characteristics of the non-vegetated portion of a pixel—in particular, the conversion of ice to water in subpixel water fractions.