Leveraging Machine Learning to Exploit SmallSats

Machining learning (ML) has become one of the most active areas of research in Environmental Science owing to its powerful applications to aid many aspects of data processing, numerical weather prediction (NWP), and post processing. This study will focus on the emerging SmallSats in the global observing system. SmallSats provide high-temporal resolution data, fill data gaps, mitigate risk via data redundancy, and can be readily launched alone or in a constellation. SmallSats typically have a highly modular design and use available commercial off the shelf (COTS) components, but they often lack high instrument precision or sufficient information on the sensor components and their noise performance. Full utilization of these data in the NWP systems requires carefully estimating their uncertainties and selecting or thinning the data to maximize the information content from them. This study will present preliminary exploration of data selection and uncertainty characterization using informed statistics and machine learning for NWP applications.