Characteristics of Landsat 8 OLI-derived NDVI by comparison with multiple satellite sensors and in-situ observations

Abstract

Vegetation indices are important remotely sensed metrics for ecosystem monitoring and land surface process assessment, among which Normalized Difference Vegetation Index (NDVI) has been most widely used. The newly launched Landsat 8 Operational Land Imager (OLI) sensor, together with its predecessor Landsat 7 Enhanced Thematic Mapper Plus (ETM +), provides continuous earth observations with an 8-day interval. The design improvements of the new sensor, including narrower near-infrared waveband, higher signal-to-noise ratio (SNR), and greater radiometric sensitivity highlight the need for investigating the land surface observation properties, especially its consistency with data from its predecessors and other satellite sensors. This study aims to evaluate the characteristics of Landsat 8 OLI-derived NDVI against Landsat 7 ETM + by cross-comparison and by comparing with Moderate Resolution Imaging Spectroradiometer (MODIS) and Geostationary Ocean Color Imager (GOCI)-derived NDVIs as well as in-situ NDVI measurements. Simulations of Top of Atmosphere (TOA) reflectance and surface reflectance of broadleaf trees and water were conducted for Landsat 8 OLI, Landsat 7 ETM +, and MODIS in order to evaluate the impact of band pass difference on NDVI calculation. Four consecutive pairs of Landsat 8 OLI and Landsat 7 ETM + data over China and Korea were examined, and NDVIs derived from TOA reflectance and surface reflectance by three atmospheric correction methods were evaluated. Both simulations and comparisons showed that NDVIs derived from atmospherically-corrected surface reflectance had good consistency, while the simulation showed that the agreement varied with atmospheric characteristics. The four pairs of Landsat 8 OLI and Landsat 7 ETM + NDVI had a mean bias error within +/- 0.05, and R-2 from 0.84 to 0.98. Vegetated land cover types were found to have better NDVI agreement than non-vegetated land cover types. Especially, Landsat 8 OLI consistently generated lower NDVI values in water area than Landsat 7 ETM +, which resulted from higher aerosol optical thickness in atmosphere. Landsat 8 OLI-derived NDVI showed better agreement with MODIS and GOCI NDVI than Landsat 7 ETM +, mainly on vegetated surfaces. Both Landsat 8 OLI and Landsat 7 ETM + surface reflectance-derived NDVI agreed well with in-situ light emitting diode (LED) NDVI measurements at a homogeneous deciduous forest site. Landsat 8 OLI was also found to produce higher spatial variability of NDVIs than Landsat 7 ETM + at vegetated and urban areas, but lower variability on water area. The overall good agreement between Landsat 8 OLI NDVI and Landsat 7 ETM +, MODIS and GOCI NDVIs as well as in-situ measurements ensures that it is reliable to integrate the new sensor observations with those from the multiple satellite sensors, given that the same atmospheric correction methods are applied. Furthermore, the greater NDVI contrast between vegetated areas and water areas, and the higher spatial variability of Landsat 8 OLI NDVI indicated that the new sensor has better capability in land surface process monitoring, such as land cover mapping, spatiotemporal dynamics of vegetation growth, and drought assessment.