EO-ALERT: Very-Low Latencies for Convective Storm Nowcasting Based on a Next Generation Satellite Processing Chain

I. Bravo(1), A. Fiengo(1), R. Hinz(1), M. Kerr(1), C. Marcos(2)

(1)DEIMOS Space S.L.U., Tres Cantos – Madrid, Spain, Email: {juan-ignacio.bravo, aniello.fiengo, robert.hinz, murray.kerr}@deimos-space.com
(2)Agencia Estatal de Meteorología, Spain, Email: cmarcosm@aemet.es


Nowcasting and very-short range forecasting products play an important role for emergency management and civilian security, and the climate change related increase of extreme weather events will further intensify the need for the rapid detection and alert of severe weather conditions. While EO-satellite products are becoming ubiquitously available, their potential for short-latency alert systems is limited by the bottleneck created by the classical EO-data chain, which involves the acquisition, compression and storage on-board and the transmission to ground station of the raw EO data, followed by ground processing, thus introducing a long delay between observation and alert delivery.

The EO-ALERT project (http://eo-alert-h2020.eu/), an H2020 European Union research activity led by Deimos Space, provides a solution not only for meteorological nowcasting applications but for generic, satellite-data based applications demanding low latencies, such as extreme event and other enhanced-NRT scenarios. By transferring optimised key EO data processing elements, like image generation and processing, from the ground segment to on-board the satellite, and thus reducing both the amount of data sent to ground and the processing time necessary at ground station, the EO-ALERT architecture is able to deliver products to ground with a global very low latency (<5 minutes).

Following last year’s presentation of the overall concept and algorithmic approaches at the European Nowcasting Conference, this communication gives an overview of the implementation and first results of EO-ALERT’s system for nowcasting and early warnings of convective storms. Inspired by NWCSAF’s Rapid Developing Thunderstorms (RDT-CW) product, and optimized for on-board satellite processing, its 3-step-approach consists of: Candidate cell extraction from SEVIRI images of the MSG satellites, tracking of their position and characteristics over time, and the discrimination of convective cells in their different stages of evolution.

The results are validated against a specifically created dataset, corresponding to a number of severe weather events that occurred over 2016-2018 convective season, composed of MSG images and OPERA weather-radar network composites. The performance is further compared to the RDT product. Despite the current restriction to on-board raw data for candidate convective discrimination, the solution based on machine learning classifiers shows promising results in terms of the common classification scores. This is further improved and surpassed using deep learning models (CCN, LSTM) applied to the same data, which preliminarily show the benefit of this AI-based approach. The results confirm EO-ALERT’s ability to complement ground-based nowcasting systems, like the RDT product, via very short latencies and reliable very-short forecasts.

© 2020 EO-ALERT All Rights Reserved