Seminar on Computational Learning and Adaptation


  Automating the estimation of meteorological parameters from multiple data sources using machine learning techniques

Mike Hadjimichael
Naval Research Laboratory, Monterey
hadjimic@nrlmry.navy.mil

U.S. Navy weather observing and forecasting operations would be greatly assisted with the immediate assessment of meteorological parameters when ground observations are not available. We attempt to solve this problem by combining four types of coincident data: numerical weather model, satellite, climatological, and observed, into a single database to which machine learning methods are applied. Supervised machine learning techniques, including decision tree, neural network, K-nearest neighbor, etc., are used to discover patterns in the data and develop associated classification and parameter estimation algorithms. The parameters of initial interest are cloud ceiling height and rain accumulation. There are 45 locations studied, over three geographical regions: the U.S. west coast, the Adriatic Sea, and the Korean peninsula. Initial results show that analysis using data from a combination of model and satellite parameters yields a better estimation than either individual source, and better than current methods. This presentation will discuss the problem to be solved, the technical obstacles involved in working with such data, solution methods applied, and initial results.



Date: Thursday, February 7

Time: 4:15-5:30PM

Place: Cordura 100


Return to the seminar schedule