Group seminar on 26. January, 16:00 CET
Developments on Atmospheric Gravity Waves parameterization
One of the most interesting parts of atmospheric dynamics is the parameterization of subgrid-scale phenomena, such as Atmospheric Gravity Waves (hereafter GWs) in General Circulation Models. This aim has been surveyed in two ways using a traditional parameterization scheme and investigation of modern methods like Machine Learning (ML). In the first approach, the relations previously developed on the f plane for tropospheric sources of GWs including jets, fronts, and convection in terms of associated secondary circulations strength are generalized for application over the globe. Then, the connection between parameterized energy and diagnosed GW energies explored using regression analysis. The discovery of the best performance in regression analysis and obtaining the proper coefficients in terms of time and location is the next purpose of the current study. The best performance in regression analysis is obtained by using a combination of power and exponential functions, which suggests evidence of exponential weakness of radiation.
The second approach looks into new methods like ML to provide novel techniques to increase the performance of parameterization schemes by learning from observations and targeted high-resolution simulations. For this purpose, the advantages of Random Forest Regressor have been exploited to reconstruct GW signals at the lowermost stratosphere by feeding coarse-grained description of the flow in the troposphere. In this point of view, there is no predefined relation for parameterization then the model tries to learn from the data only. Therefore, two ideas have been explored for the development of GW parameterization. First, to indicate how much of the GW signal can be reconstructed from the low-resolution resolvable flow. Second, to rank the variables used to train the model by order of relevance when the model presumes no relation between variables in advance. The results of this preliminary experience showed the positive and relative success of this approach to predict the variation of three targets including high-resolution vertical velocity, horizontal divergence, and absolute momentum flux.