End-to-End Prediction of Lightning Events from Geostationary Satellite Images

Abstract

While thunderstorms can pose severe risks to property and life, forecasting remains challenging, even at short lead times, as these often arise in meta-stable atmospheric conditions. In this paper, we examine the question of how well we could perform short-term (up to 180 min) forecasts using exclusively multi-spectral satellite images and past lighting events as data. We employ representation learning based on deep convolutional neural networks in an “end-to-end” fashion. Here, a crucial problem is handling the imbalance of the positive and negative classes appropriately in order to be able to obtain predictive results (which is not addressed by many previous machine-learning-based approaches). The resulting network outperforms previous methods based on physically based features and optical flow methods (similar to operational prediction models) and generalizes across different years. A closer examination of the classifier performance over time and under masking of input data indicates that the learned model actually draws most information from structures in the visible spectrum, with infrared imaging sustaining some classification performance during the night.

A Survey of Deep Learning-Based Lightning Prediction

“…In modern meteorological systems, geostationary satellites hold a pivotal role, with satellite images often serving as a common medium for lightning prediction. A noteworthy example of this can be seen in the work of Sebastian Brodehl et al, who proposed an end-to-end CNN-based lightning prediction method that employs geostationary satellite images [25]. This approach allows for direct prediction of lightning events from satellite images.…”

A Survey of Deep Learning-Based Lightning Prediction

“…In modern meteorological systems, geostationary satellites hold a pivotal role, with satellite images often serving as a common medium for lightning prediction. A noteworthy example of this can be seen in the work of Sebastian Brodehl et al, who proposed an end-to-end CNN-based lightning prediction method that employs geostationary satellite images [26]. This approach allows for the direct prediction of lightning events from satellite images.…”

“…U-Net + ResNet-v2 [26] Geostationary satellite images KL + CNN + MLP [29] Time-series observation meteorological parameters Spectrograms + CNN [31] Random noise, lightning sounds, and background noise Sliding window + CNN [32] 3D weather radar data The most commonly used deep learning method for lightning prediction primarily processes sequential data, often combining with existing methodologies in a variant form to achieve more accurate predictions.…”

Current and Near-Term Earth-Observing Environmental Satellites, Their Missions, Characteristics, Instruments, and Applications

“…This new instrument is likely to be of interest to environmental scientists: the Lightning Imager (LI) on the MTG satellites. The lightning imagers (Geostationary Lightning Mission, GLM) on the two GOES satellites were the first with this technology ( Table 16 ), starting in 2016 [ 290 , 291 , 292 ]. The Lightning Imager (LI) on the MTG-I will, along with the GOES GLM, monitor lightning intensity above a background radiance per cell (8km x 8km) over a specified time interval (e.g., 5 min).…”