On the relevance of radiometric normalization of dense Landsat time series for forest monitoring

Abstract

Several radiometric preprocessing strategies to adjust multiple images are reported in the literature. These include absolute and relative correction methods. Dense time series comprising data from different seasons, have rarely been assessed so far for their sensitivity to the radiometric preprocessing. Time series are required to fully understand forest development. In this paper, we explore the effects of relative radiometric normalization on dense Landsat time series of a forested study site on southern Vancouver Island, British Columbia, Canada. A comparison of using absolute radiometric correction alone and the additional step of relative atmospheric correction was performed. We can show that relative atmospheric correction blurs seasonal signals and obscures long-term trends. Relative radiometric normalization is strong in eliminating non-surface related image noise, which is important for bitemporal studies. The exploration of dense time series with images from different seasons, however, should not include relative atmospheric correction in order to preserve process-related dynamics at different temporal scales.