Earth climate research crucially depends on measurements of the atmospheric distribution of CO2, which is largely obtained using satellites. But satellites cannot directly measure CO2—they capture photographs at different wavelengths that must be mathematically processed to obtain this information. Current methods for solving this inverse problem are unaware of many aspects of the images including topography, cloud shape, shadowing, etc. The UN Intergovernmental Panel on Climate Change (IPCC) has identified the resulting errors as the main cause of discrepancies between different climate sensitivity models.
This proposal in the area of computer graphics introduces methods for inverting the physics of light at unprecedented scales that will address these inaccuracies. However, the scope of our contribution extends far beyond climate modeling: it will have a revolutionary impact on all scientific disciplines that involve the analysis of images, including biology, computer vision, architecture, and many others.
In this project,
- we will establish the first framework for inverting light simulations with billions of parameters.
To demonstrate its generality, and to realize the impact of this framework, we will specialize it to three areas:
- we will develop the first invertible atmospheric optics simulator for earth climate monitoring that accounts for 3D structures, addressing severe inaccuracies of current methods.
- we will create an invertible virtual microscope that will open the door to fundamentally new reconstruction techniques in the area of biology.
- we will design architectural light simulations that are able to adapt buildings so that they make ideal use of naturally available daylight.
To achieve these goals, we must unravel the messy physics of light to either reveal or control the properties of visible and invisible objects. Our approach will solve this impossible-seeming problem at large scales with substantial impact across disciplines.
Field of science
- /natural sciences/computer and information sciences/artificial intelligence/computer vision
Call for proposal
See other projects for this call