Developing an Original Nowcasting System for Severe Convective Weather Over Nigeria Using Deep Learning and Dual-Polarization Radar Data

Nowcasting of severe convective weather events including hail, tornadoes, and damaging wind gusts at lead times of 10 to 60 minutes represents the most challenging operational meteorological forecasting problem and a critical unmet need for aviation, agriculture, and disaster management in Nigeria. This study develops an original nowcasting system for severe convective weather over Nigeria using deep learning architectures applied to dual-polarization radar data, representing a novel contribution at the frontier of operational meteorology and artificial intelligence. The nowcasting system is developed using NiMet's C-band dual-polarization radar data from the Abuja, Kano, and Makurdi radar sites, covering the 2020 to 2023 period. A hybrid deep learning architecture combining 3D convolutional neural networks for radar storm cell feature extraction and LSTM networks for temporal storm evolution prediction was designed and trained on 8,400 labeled severe convective events. The system outputs 10-minute resolution probabilistic nowcast fields of severe convective cell location, intensity, and motion at lead times of 10, 20, 30, 45, and 60 minutes. Radar reflectivity, differential reflectivity, and correlation coefficient polarimetric variables are all incorporated as model inputs, providing microphysical context unavailable to reflectivity-only nowcasting approaches. The deep learning nowcast system achieves a critical success index of 0.58 for severe cell detection at 30-minute lead time, outperforming a traditional cross-correlation advection benchmark by 24 percent. An original polarimetric signature library for Nigerian severe convective weather is developed as a companion scientific contribution. Keywords: nowcasting, deep learning, dual-polarization radar, severe weather, Nigeria.