Vegetation index suites as indicators of vegetation state in grassland and savanna: An analysis with simulated SENTINEL 2 data for a North American transect

Grasslands and savannas form a heterogeneous and patchy mosaic of spectral properties that are challenging characterization of vegetation states. This study examines the potential for use of suites of vegetation indices (VIs) from the proposed Sentinel 2 sensor to describe vegetation states in grasslands and savannas for a North American transect. Hyperion hyperspectral data from the EO-1 satellite were used to simulate Sentinel 2, MODIS (Moderate Resolution Imaging Spectroradiometer) and VIIRS (Visible Infrared Imaging Radiometer Suite) images for field sites in Alberta, North Dakota and Texas that represent the continuum from short grass prairie to oak savanna and are intermingled with agriculture. Indices representing photosynthetic pigments (Normalized Difference Vegetation Index, Carotenoid Reflectance Index, Anthocyanin Reflectance Index and Red-Green Ratio), vegetation and landscape water content (Normalized Difference Infrared Index), senescent vegetation and soil (Short Wave Infrared Ratio and Plant Senescence Reflectance Index) and herbaceous biomass (Soil Adjusted Total Vegetation Index) were used. There were distinct differences among sites in the relative sensitivity of different VIs depending upon moisture status, tree cover and type of grassland. Simple multi-variate models based on mean values or VIs showed limited ability to predict land cover classes and nominal vegetation states. However, analysis of sample areas using pixels as individual observations within a statistical distribution indicated that subtle variation and gradients within management or land units could be used to characterize fine differences in selected nominal states at each site. Despite some differences in band locations, all VIs except the anthocyanin reflectance index were scalable between Sentinel 2 and MODIS and VIIRS data. A framework for using suites of VIs as indicators of vegetation states that could be applied to the state and transition model approach-applied by the US Natural Resource Conservation Service is described. Land types can be effectively characterized by pixel value distribution histograms, and statistical metrics may be used as indicators of status and change. However, time series are needed to fully capture states and state changes, since grasslands and savannas have such high levels of spectral and phenological variation. (c) 2013 Elsevier Inc. All rights reserved.