SII-NowNet

PhD student, Joey Smith, has developed a machine-learning-based weather prediction tool for the Tropics. Simple Initiation and Intensification Nowcasting neural Network (SII-NowNet), uses freely-available satellite data to generate short-term predictions (0 – 6 hours ahead) of intensifying and initiating convective storms. The tool is ‘simple’ in the sense that it only requires satellite data as input, meaning it can be easily applied to different regions of the Tropics.
Since the development of SII-NowNet, Joey has focussed on translating the tool from research into operationalisation at national weather services.

In November 2023, Joey visited Jakarta, Indonesia to present the capabilities of SII-NowNet to Badan Meteorologi, Klimatologi, dan Geofisika (BMKG), the national Indonesian weather service. Following this, a collaboration has been setup to implement, evaluate and improve SII-NowNet for BMKG’s use as a nowcasting (short-term forecasting) product within their operational bench. BMKG have now completed implementation of SII-NowNet, with planning for the evaluation stage underway. Gathering valuable user-feedback, as well as quantitative analyses, is crucial for improving the practicality of the tool and generating greater impact.

Through involvement in the WISER EWSA testbed, the University of Leeds has developed a network of meteorological partners in Southern Africa. In particular, strong links with the South African Weather Service (SAWS) has led to the creation of a new partnership project with Joey, which will take place in the summer 2025. It will focus on implementing the SII-NowNet tool into the SAWS operational nowcasting bench and involves a period of co-development to improve their machine learning capability. Finally, a SII-NowNet training period will also be provided, enabling SAWS full control and flexibility of the tool themselves.

Left: Observed locations of intensifying convection (green boxes) over Sumatra, Indonesia. Right: SII-NowNet’s corresponding prediction of intensifying convection, with a +1 hour lead time.