
The vast majority of physics-based retrieval algorithms used in remote sensing of atmosphere and/or surface properties are multi- or hyper-spectral in nature, some use multi-angle information as well; recently, polarization diversity has been added to the available input from sensors and accordingly modeled with vector radiative transfer codes.
At any rate, a single pixel is processed at a time using a forward radiative transfer model predicated on 1D transport theory. Neighboring pixels are sometimes considered but, in general, only to formulate a statistical constraint in the inversion based on spatial context. We demonstrate here the potential power that could be harnessed by adding bona fide multi-pixel techniques to the mix.
To this effect, we use a forward radiative transfer model in 2D—sufficient for this demo, and easily extended to 3D—for the response of a single wavelength imaging sensor.
The data, an image, is used to infer the position, size and opacity of an absorbing atmospheric plume somewhere in a deep valley in the presence of a partially-known/partially-unknown aerosol assumed to have an exponential profile with altitude. We first describe the necessary innovation to speed-up forward multi-dimensional radiative transfer. In spite of its notorious reputation for inefficiency, we use a Monte Carlo technique.
However, the adopted scheme is highly accelerated without loss of accuracy by using efficiently “recycled” Monte Carlo paths, a methodology adapted from ongoing research in biomedical imaging. This forward model is then put to work in a novel Bayesian inversion adapted to this kind of forward model where it is easy to trade precision and efficiency.
The retrievals target the key plume properties and the specific amount of background aerosol. In spite of the limited number of pixels and low signal-to noise ratio, we show that there is added value for certain kinds of nuclear treaty verification applications.