Skip Navigation


Inversion of Computer Models of Physical Systems

Remote sensing scientists have long estimated parameters associated with our biosphere using multispectral visible-near-infra-red remote sensed data. Often these estimates have been based on a radiative transfer model of the process being modeled, however the inversion has often been based either on a look-up-table or on a fallback to a simpler, heuristic algorithm. Uncertainty quantification in the estimates has been poor, often only a qualitative quality label.

In support of NASA, scientists at USRA's Research Institute for Advanced Computer Science and university partners have developed a methodology for the analysis of computer model experiments with full uncertainty quantification. The methodology takes the radiative transfer model of sunlight reflecting off a forest canopy as a computational black-box, and uses a Gaussian Process model to approximate the radiative transfer model in a computationally efficient manner to produce estimates of the inversion of the radiative transfer model with full uncertainty quantification (in this case for leaf area index (LAI)).


Inversion of Computer Models of Physical Systems

Traditional satellite-based mapping of vegetation vigor and amount is based on the way vegetation interacts with red and infrared light.