Warming of waters and intensification of coastal rains: The case of cyclone Shaheen in Oman and their implications for weather forecasting

In October 2021, cyclone Shaheen traversed northern Oman, bringing heavy rains along the coast and highlighting the vulnerability of arid regions to tropical storms. A team of researchers from Sultan Qaboos University used this event as a test case to explore how warming waters and local topography influence cyclone simulations, particularly in terms of intensity, trajectory, and rainfall distribution.

What the study reveals: Precise results and key figures

Simulations show that as sea surface temperature (SST) increases, the model predicts intensified convection at the cyclone's core, resulting in heavier rainfall over coastal areas. Furthermore, Oman's mountainous topography, characterized by abrupt ranges rising rapidly from the coastline, alters air flow and leads to notable differences in the simulated cyclone trajectory. Thus, variations of a few degrees in SST and adjustments in topography can shift the cyclone's path by tens of kilometers, radically altering the areas most affected by downpours.

These differences are particularly pronounced when comparing standard ECMWF simulations with those enriched by Copernicus satellite data and advanced neural network models like GraphCast and Pangu-Weather. Machine learning models better capture fine-grained interactions between atmospheric humidity and oceanic heat, reducing prediction uncertainty related to cloud formation processes and surface dynamics.

How: Research method and data used

The researchers conducted a series of numerical experiments by varying two key parameters: sea temperature and topography. SST data was obtained from the Copernicus Marine Service, providing a spatial resolution of 0.083°, while the terrain model was constructed using topographic data provided by the Omani geographical service. For atmospheric dynamics, they used the ECMWF predictive model as a reference framework, to which they added machine learning modules based on GraphCast and Pangu-Weather architectures, trained on decades of satellite data and in-situ measurements.

Each simulation was evaluated using prediction uncertainty indicators, comparing model outputs to real rainfall observations recorded by local weather stations and radars. The hybrid approach, combining classical physical models and neural networks, allowed for quantifying the marginal impact of each factor (SST and topography) on the cyclone's trajectory and intensity, while also highlighting the current limitations of purely deterministic models.