Convective storm prediction
The thesis project was performed by Feraz Younis
These are Feraz’s comments about his thesis work:
“My master’s thesis focused on developing a dual deep learning framework for the early identification of convective storms and the short-term prediction of their evolution. By fusing high-resolution radar reflectivity data with ground-based meteorological sensor networks, this work tackled three major challenges: estimating storm severity, predicting storm trajectories, and forecasting meteorological changes such as pressure, temperature, humidity, and rainfall.
Read more
The core of this thesis was implemented through a hybrid architecture combining Feedforward Neural Networks (FFNNs) for storm classification and a fusion of FFNN and LSTM (Long Short-Term Memory) networks for storm trajectory and weather forecasting. This dual-model approach outperformed conventional methods, offering improved predictive accuracy and lead time, especially for flood-prone areas like the Seveso River Basin in Northern Italy.
From my perspective, this thesis was made possible thanks to the excellent supervision and constant support of Prof. Giovanna Venuti and Dr. Xiangyang Song, as well as the training provided by several key courses in the Geoinformatics program at Politecnico di Milano”