Can machine learning help us to discover more about clouds?

The difference between the amount of energy entering and leaving the Earth's atmosphere in the form of radiation, (the Earth radiation budget) can be altered by changes in atmospheric composition or land use. This is called radiative forcing. A positive difference translates into an increase in Earth's temperature. Among the human-generated influences on radiative forcing, aerosol-cloud interactions are the least understood. Satellite observations are perhaps the most promising way to learn about such effects at global scale.  Satellite data is often collected in several channels, each of them focussing on the different physical properties of the environment targeted. 

The study is based on two bulk qualities describing the properties of a liquid cloud: the liquid water path (LWP), and the droplet concentration (Nd). To analyze these properties, we use data from the MODIS satellite instrument; however, the dependence on different physical properties of droplets and optical factors of clouds generate errors in the interpretation of qualities of interest.

We hypothesize that machine learning techniques can help us find patterns in the data gathered from several channels to understand more about cloud properties and their co-variation in different weather conditions.

We also plan to integrate statistical models generated in an emulator.  These should be able to approximate the Radiative Transfer Model, and will then be used to analyse the statistical relationships between aerosol and clouds directly, eliminating the need to perform the retrievals first.