A translation algorithm of NDVI to minimize biases caused by differences in spectral band-pass filters

This paper addresses the issue of NDVI continuity across sensors by focusing on minimization of NDVI differences caused by sensor spectral differences. Two sets of hyperspectral data were used to simulate a sensor´s NDVI value by band convolution with appropriate spectral filters, thus eliminating the spatial issue related to sensor spatial response curves in this study. One is a set of airborne radiometric data obtained by a spectroradiometer from 330 m above ground level over grass dominated Cerrado region In Brasilia, Brasil. The other data set is an AVIRIS Image over a semi-arid grass dominated region in Arizona, USA. The translation algorithm is briefly explained and its performance was demonstrated with the airborne hyperspectral data sets. The values of translated MODIS-equivalent NDVI from simulated ETM+ NDVI were compared to the value of simulated-MODIS NDVI by direct convolution from hyperspectral data with the MODIS band-pass filters. The differences of NDVI values between the translated and simulated MODIS NDVI (~3.0e-4) were reduced by two orders smaller than the original differences of 0.02 represented by (ETM+-MODIS). The proposed technique showed good performance in minimizing the NDVI differences between the two sensors over the entire NDVI range