EGU25-10561, updated on 15 Mar 2025
https://doi.org/10.5194/egusphere-egu25-10561
EGU General Assembly 2025
© Author(s) 2025. This work is distributed under
the Creative Commons Attribution 4.0 License.
Exploring the Dynamics of Climatological Mean Monsoon Using a Machine Learning Based Empirical Leading Order Analysis
Arijeet Dutta1, Ruth Geen1, and Maike Sonnewald2
Arijeet Dutta et al.
  • 1University of Birmingham, School of Geography, Earth and Environmental Sciences, United Kingdom of Great Britain – England, Scotland, Wales (a.dutta.2@bham.ac.uk)
  • 2University of California Davis, USA

The global monsoon circulation, which governs the subtropical rainband, can be interpreted as a manifestation of the seasonal migration of tropical overturning circulation (Hadley cell). However, the dynamics of regional monsoons is additionally controlled by zonal asymmetries occurring from land sea distribution, zonal gradients in sea surface temperature, and other stationary wave forcings. Despite its importance, the dynamics of regional monsoons remain poorly understood. Here, we demonstrate, using a machine learning guided empirical leading order analysis, emergence of distinct dynamical regimes that describe the complex evolution of regional monsoons. Conservation of angular momentum plays an important role in our understanding of the climatological and zonal mean picture of monsoon. It suggests, during the solstitial seasons the dominant balance in the momentum budget comes from the mean meridional circulation and the advection of mean zonal wind by the divergent wind. However, for regional monsoons the resulting angular momentum budget now includes many terms arising from the drivers mentioned above. We deploy an unsupervised machine learning algorithm to find the dominant balances in the momentum budget. This enables us to find spatio-temporal clusters characterized by distinct balances in the momentum budget and to study how they evolve throughout the seasonal cycle. The inherent stochastic nature of the algorithm is leveraged to find the robustness of the identified clusters. Entropy is used to measure uncertainty for the clusters recognized by the algorithm. The algorithm is successfully applied to idealized simulations with varying complexities ranging from aquaplanet to different distributions of land-sea-topography. Consistent with zonal mean theory, resulting clusters capture the dominant tropical overturning. However, zonal asymmetries result in additional clusters with distinct dynamical regimes

How to cite: Dutta, A., Geen, R., and Sonnewald, M.: Exploring the Dynamics of Climatological Mean Monsoon Using a Machine Learning Based Empirical Leading Order Analysis, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10561, https://doi.org/10.5194/egusphere-egu25-10561, 2025.