Assessment of cross-sensor NDVI-variations caused by spectral band characteristics

Abstract

Remote sensing-based vegetation indices are widely used for vegetation monitoring applications. The NDVI is the most commonly used indicator for spatial and temporal vegetation dynamics. For long term or multitemporal observations, the combined use of multisensoral NDVI data is necessary. However, due to different sensor characteristics NDVIvariations occur. The sensor geometry, like viewing- and solar angle, atmospherical conditions, topography and spatial or radiometric resolution influence the data. This study contributes to another important factor, the spectral characteristics of different sensors, in particular the relative spectral response (RSR) functions. In order to analyze the NDVI variations caused by different RSR functions, the multispectral bands of Landsat 5 TM, QuickBird, Aster and SPOT 5 were simulated by the use of hyperspectral data of the airborne HyMap sensor. The observed NDVI differences showed a non-linear but systematic NDVI offset between the sensors. Results indicate that the NDVI differences decrease significantly after cross-calibration. A gradual cross-sensor calibration of NDVI taking first spectral characteristics into account is essential. Residual factors could be calibrated in a second step. Such an inter-calibration is desirable for multisensoral NDVI- analyses to ensure the comparability of achieved results.