Complex networks to identify the merging of patches of coherent vorticity dynamics during tropical cyclones in the Bay of Bengal

Predicting the intensity of cyclones a few days in advance during the formation as well as intensification of the cyclone is an open challenge. Heating from moist convection within the cyclone is considered the primary driver for the large-scale cyclonic vortex. However, the effect of interactions between small-scale vortices within the cyclone environment on the intensification of the cyclone vortex and its event-to-event variability are poorly considered. To enable skilful cyclone predictions, it is essential to first understand the local interactions in the atmosphere that facilitate self-sustained rotation and updraft of moist air.

We present a novel approach using complex networks to study atmospheric interactions and identify vortical perturbations that influence the formation of a depression and eventually a cyclone. We analyze the atmospheric flow over the Bay of Bengal (BoB) during different category-5 cyclones, namely, Amphan (2020), Sidr (2007) and Bangladesh (1991). Relative vorticity is obtained at hourly temporal resolution from the ERA5 reanalysis dataset (ECMWF reanalysis project). Nodes are locations between the equator to 30°N and 75°E to 105°E with a spatial resolution of 0.5°. We construct time-varying networks where each network corresponds to a short time period of 29 hours. Consecutive networks are separated by a difference of three hours. In each network, links are established between two nodes if (i) the time series of relative vorticity at both locations are correlated in a 24-hour window with a maximum of five-hour lag, and (ii) the two nodes are in spatial proximity of 2° latitude-longitude width centred at any one of the nodes. Note that, the spatial proximity is approximately 200 km that is comparable to the gale force wind radius of category-5 cyclones in BoB.

Through this approach, we decipher the relation between the local flow interactions and the global emergence of order in the form of a cyclone in the atmosphere. Regions of high connectivity in the network represent patches of locally coherent vorticity dynamics. Multiple such patches emerge throughout the life of a cyclone. Initially, these patches revolve around a developing low-pressure system, merging and intensifying the low-pressure system into a tropical depression and eventually into a tropical cyclone. Our approach helps identify prominent mesoscale convective systems that can form away from the low-pressure system but are entrained towards the depression and help intensify the storm at different stages. 

Next, we use Broadcast Mode Analysis (BMA), an advanced tool to identify the critical nodes that influence information propagation in time-varying networks. The analysis reveals nodes (locations) from where the most influential patch of coherent vorticity dynamics emerges that will eventually propagate, merge with and intensify the storm. We find the most influential region (the broadcast mode) and the most influenced region (receiving mode) in every 56-hour period corresponding to 10 networks. The receiving mode of one 56-hour period is approximately similar to the broadcast mode of the next 56-hour period. Broadcast mode analysis highlights the potential of tracking local interactions and mesoscale patches of coherent vorticity dynamics to improve the prediction of cyclone intensity.