Joint Seminar: Estimating Balloon-observed Gravity Wave Momentum Fluxes using Machine Learning with Inputs from Reanalyses
The superpressure balloons of the Stratéole 2 project have provided precise estimates of the gravity wave momentum fluxes (GWMF) in the tropical lower stratosphere (M. Corcos, 2021). In this talk, several ensemble machine learning methods such as random forests, extra trees and adaptive boosting are explored to estimate the GWMF from inputs describing the underlying or surrounding flow, obtained from the ERA5 reanalysis. The aim is to probe the relationship between the ‘large-scale’ flow, as described by the reanalyses, and the GWMF, as sampled locally by the balloons. At best, our approach could constrain the upper bound for the stochastic component of GWMF.
Our analysis shows encouraging results (correlations larger than 0.7) in some cases. We observe that the most informative variables generally include those characterizing precipitation in the environment of the balloon, and those characterizing winds at the balloon level or slightly below. However, their order and relative importance vary significantly depending on the test dataset and on the approach used.
- Bundesstr. 53, room 022/023
- Seminar Room 022/023, Ground Floor, Bundesstrasse 53, 20146 Hamburg, Hamburg
- Claudia Stephan