Talk by Aditya Nagarajan

The talk will take place at 1634, LGRT, 19 Nov. 2015, 17:30. All are welcome to join.

Our speaker this week is Aditya Nagarajan, a MSc student in Industrial Engineering at UMass Amherst.

Title: A Machine Learning approach to Precipitation Nowcasting using GPS derived Integrated Precipitable Water Vapor


Weather surveillance radars are an excellent source of information for making real-time assessments and short-term predictions of rainfall - where it is occurring, its intensity, how much has accumulated so far, where it is moving, and so on. However they are limited in that, under normal operation, they can only sense active rainfall; they cannot sense the buildup and advection of the atmospheric water vapor commonly termed as precipitable water which are the driving force behind severe storms. We thus explore this complimentary nature in the atmosphere that the build up of precipitable water which can potentially lead to precipitation. We look at a novel way to predict short term (0-2 hours) precipitation commonly termed as nowcasting, by supplementing weather radar based reflectivity products with GPS derived Integrated Precipitable Water (IPW) vapor products from a network of GPS stations. We propose to develop a software system that computes point measurements of IPW from raw GPS signals and meteorological data from publicly available databases, to build spatial IPW fields and study its spatial and temporal properties with regard to evolving weather radar reflectivity fields. We explore the spatiotemporal relationship between IPW and reflectivity fields by analyzing severe storm events in the Dallas Fort-Worth region during the year of 2014. We then propose an approach involving state of the art machine learning and image processing techniques to develop a tool that takes sequences of gridded atmospheric IPW and radar reflectivity fields to predict the future gridded rainfall field. We develop a Machine learning algorithm which takes spatiotemporal evolutions of IPW and reflectivity to predict rainfall fields one hour into the future.

comments powered by Disqus