A stochastic technique for remote sensing of ocean color

摘要

Remote sensing of ocean color from space aims at retrieving from a noisy top-of-atmosphere radiance the values taken by some relevant quantities like the chlorophyll-a concentration or the marine reflectance. From a mathematical perspective, it is an ill-posed inverse problem with a highly nonlinear operator. Few techniques are available in the case of a nonlinear inverse problem; even its theoretical study is far from easy, yet some techniques may be used in a practical setting when the noise distribution is known. However in the case of ocean color remote sensing, the noise encompasses several types of error owing to the forward operator approximation (radiative transfer model) as well as to calibration and pure measurement noise. Hence the noise distribution is unknown. In this work, a stochastic technique is proposed to first infer a noise distribution, which is next used to retrieve the marine reflectance in a least-square prediction setting by a regression model. The methodology is illustrated on actual data originating from the SeaWiFS sensor.