Uncertainty Quantification Working Group
February 6, 11:30 AM, CNLS Conf. Room, TA-3, Bldg. 1690

Uncertain reconstruction in remote-sensing due to 3D variability and the associated risk of false model validation: Illustration with clouds or plumes

Anthony Davis (NIS-2)


We survey the state-of-the-art in cloud/plume structure determination from remotely-sensed radiance fields in the visible and near-IR. These are popular wavelengths because of the availability of optical elements and of efficient and reliable, yet cheap, focal plane detectors. However, there is a cost to pay in the radiative transfer modeling: scattering makes the problem inherently non-local. Clouds or plumes and bright surfaces are therefore notorious complications whether they are the focus of the data collect or a contamination. So what you see in any direction is generally not what is there. It is important to understand at what scales and illumination conditions simple 1D radiative transfer can be used. It is presently abused quite systematically in operational remote-sensing.

In view of this situation, we can try to quantify and mitigate the damage caused by the 3D variability, or we can take evasive action:
* I will report on my quantification/mitigation work over the last few years which prompted a systematic investigation of 3D radiative transfer phenomenology across many scales. We now understand the problem much better and have elements of answers. We have even at times turned the multiple scattering/reflection into a valuable and yet under-exploited resource, in particular, by using a pulsed laser source and a new lidar equation.
* `Evasive action` is to move to a completely different spectral region where opacity is minimized and scattering is under control. Emissivity can be preferred as a source of signal but, not infrequently, the observer will now provide the source. This can be costly but advantageous, i.e., radar techniques yield information in the 3rd dimension. There is often still a price to pay in modeling since the quantities of interest are native to the original scattering wavelengths, for instance, clouds in climate studies. Sampling and post-processing strategies are also critical; I will demonstrate how `natural` assumptions can lead to false validation in dynamical cloud modeling.