A Monte-Carlo Based Approach for Estimating Remote Sensing Reflectance Uncertainty
A Monte-Carlo Based Approach for Estimating Remote Sensing Reflectance Uncertainty
Abstract:
An important drawback of the otherwise highly successful ocean color missions is the difficulty in obtaining uncertainties in satellite measurements. Compounding the problem is the fact that these data are variable in time and space. Therefore to be useful, any estimation of uncertainty must be performed repeatedly, at the pixel level.
Uncertainties in ocean color remote sensing arise at three levels of the processing chain; uncertainty in top-of-the-atmosphere radiance ( LTOA ); propagation of LTOA radiance uncertainty to remote sensing reflectance ( Rrs ), chiefly through atmospheric correction; propagation of Rrs uncertainty to derived products, including band ratio-based chlorophyll, particulate organic carbon, and model inversion-based inherent optical properties.
We present a novel approach based on Monte-Carlo simulations to estimate Rrs uncertainty. We begin with the development of a band-dependent LTOA noise model. This noise model is then used to perturb LTOA relative to a baseline value. This perturbation is propagated through atmospheric correction to obtain a corresponding change in Rrs.
Repeated a sufficient number of times, this procedure yields a pixel-by-pixel spectral distribution of Rrs uncertainties. We showcase results of this approach, discuss their validity and offer some perspective on future developments.
Uncertainties in ocean color remote sensing arise at three levels of the processing chain; uncertainty in top-of-the-atmosphere radiance ( LTOA ); propagation of LTOA radiance uncertainty to remote sensing reflectance ( Rrs ), chiefly through atmospheric correction; propagation of Rrs uncertainty to derived products, including band ratio-based chlorophyll, particulate organic carbon, and model inversion-based inherent optical properties.
We present a novel approach based on Monte-Carlo simulations to estimate Rrs uncertainty. We begin with the development of a band-dependent LTOA noise model. This noise model is then used to perturb LTOA relative to a baseline value. This perturbation is propagated through atmospheric correction to obtain a corresponding change in Rrs.
Repeated a sufficient number of times, this procedure yields a pixel-by-pixel spectral distribution of Rrs uncertainties. We showcase results of this approach, discuss their validity and offer some perspective on future developments.