Estimation and mapping of NDVI uncertainty from Landsat 8 OLI datasets: An operational approach

Within remote sensing applications aimed at monitoring vegetation, spectral indices represent an effective and widely used tool. Unfortunately, in the most of cases users do not take into account any estimation of index uncertainty. This information can be useful and desirable especially in multi-temporal analysis to define index sensitivity with the aim of identifying significant differences between pixels of the same scene or of the same pixel in time. The goal of this work is to investigate potential uncertainty affecting spectral indices, with particular focus on NDVI (Normalized Vegetation Index). An 'open' (entirely controllable) self-developed radiative transfer model is considered for this study. Uncertainty concerning factors involved in the model was considered to estimate its effects on NDVI final accuracy. For this task the statistical model of the variance propagation law was adopted. Two Landsat 8 OLI images acquired over a sample study area sited in Piemonte (NW Italy) were used to compute NDVI images at two different dates, estimate its uncertainty and investigate the way this information can be exploited during a change detection analysis.