• Pallavi kumari
• Dr.Anvesha Katti
• Dr. Priyanka Vashisht

Abstract: Effective predictions of the weather over the next 12-48 hours are important for protecting people and assets in sudden changes. An original model is introduced in this research that uses the combination of CNN and LSTM for local Doppler radar nowcasting. At first, convolutional layers examine radar data to find features such as reflectivity cores and the edges of storms. After that, the features are sent to LSTM units that learn about the way convective systems increase, transition and end. Using data from 5-minute S-band Doppler observations over one year, the hybrid setup achieves forecasts accurate up to three hours ahead, with a mean accuracy of 93.2% and an average RMSE of 0.45 mm. It can detect rain events greater than 40 dBZ with a 89% hit rate, making few false warnings. The application achieves the final stage within just 30 rounds of data on an ordinary GPU, showing prompt training and fast forecasting (below 200 milliseconds). Using effective spatial coding and secure memory for timing events means this approach works efficiently and exactly as needed in delivering urban and regional weather services.

Keyword: NN–LSTM hybrid, Doppler radar nowcasting, lightweight deep learning, real-time weather prediction, short-term precipitation forecasting, spatiotemporal modelling

"CNN–LSTM Hybrid for Doppler Radar Nowcasting A Lightweight Framework for Local Weather Forecasting", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.12, Issue 5, page no.i689-i696, May-2025, Available :http://www.jetir.org/papers/JETIR2505976.pdf

"CNN–LSTM Hybrid for Doppler Radar Nowcasting A Lightweight Framework for Local Weather Forecasting", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.12, Issue 5, page no. ppi689-i696, May-2025, Available at : http://www.jetir.org/papers/JETIR2505976.pdf