Lorraine Remer, Joint Center for Earth Systems Technology.
Travis Twigg, Department of Physics.
Aerosols, such as particulate pollution, dust, smoke and sea salt, play important roles in Earth’s radiative balance, cloud and precipitation formation, fertilization of ecosystems, air quality and human health. Retrieval of aerosol parameters from space-based spectral radiometric measurements has been based on physical understanding of radiative transfer, particles and the environment that these particles inhabit. For decades physically-based methods have produced a suite of highly useful products. However, physically-based methods work best when their assumptions are accurate, and tend to fail in situations where the real world and retrieval assumptions do not match well. Especially difficult for aerosol retrieval algorithms is the separation of aerosol from clouds, identifying the type of aerosol from space and parameterizing the surface reflectance beneath the aerosol layers. In this project, a group with experience in physically-based aerosol remote sensing want to explore Machine Learning techniques with a student experienced in Data Science and Machine Learning. The project will be to take one component of the retrieval, such as separating aerosols from clouds, and apply Machine Learning to that problem. Then compare results with current state-of-the-art physically-based techniques that are now being used operationally. For this project, the student must be fluent in computer processing and demonstrate previous projects that made use of Machine Learning techniques.