Authors: Reyhaneh Rahimi, Praveen Ravirathinam, Ardeshir Ebtehaj, Ali Behrangi, Jackson Tan, Vipin Kumar

Published on: July 20, 2023

Impact Score: 8.35

Arxiv code: Arxiv:2307.10843

Summary

• What is new: The novel architecture combines U-Net and LSTM networks for precipitation nowcasting with a 4-hour lead time on a near-global scale, examining the impact of training loss functions.
• Why this is important: Precipitation nowcasting on a global scale with accurate prediction of both light and extreme precipitation rates.
• What the research proposes: A deep learning architecture that fuses U-Net and LSTM networks, utilizing IMERG data and key GFS precipitation drivers, tailored with specific loss functions for different precipitation intensities.
• Results: The regression network excels in predicting light precipitation, whereas the classification network is superior for extreme precipitation events. The inclusion of physical variables enhances nowcasting accuracy.

Technical Details

Technological frameworks used: U-Net, LSTM

Models used: Regression network, Classification network

Data used: IMERG, Global Forecast System (GFS)

Potential Impact

Weather prediction services, outdoor event planning businesses, emergency management agencies, and agricultural sector.

Want to implement this idea in a business?

We have generated a startup concept here: PreciseWeatherNet.