A Generalized Model for Intersensor NDVI Calibration and Its Comparison With Regression Approaches

Satellite remote sensing has accumulated decades of normalized difference vegetation index (NDVI) data. For long-term environmental studies, multisensor NDVI discrepancies should be initially corrected via intersensor NDVI calibration. This paper proposes a generalized model and its simplified form for the calibration by incorporating sensor-, atmosphere-, and observational geometry-related parameters into an analytical function. The models are first validated using the tandem Landsat-5 Thematic Mapper and Earth Observing One (EO-1) Advanced Land Imager (ALI) NDVIs and then used to evaluate current intercalibration methods. The results show that surface reflectance is the most critical factor for linear intercalibrations followed by water vapor. While other atmospheric parameters as well as solar zenith angle have minor effects on linear intercalibrations. The strong nonlinear relationships between the spectral band adjustment factors and NDVI may explain why quadratic functions are used for NDVI intercalibration. Although quadratic intercalibrations are generally more accurate than linear intercalibrations, they may still perform poorly when water vapor exhibits significant spatial variations. In addition, the uncertainties associated with current NDVI intercalibrations are discussed based on the proposed models. This paper furthers the understanding of multisensor NDVI intercalibration and provides insight into intercalibration of other remote sensing indices.