End-to-End Prediction of Lightning Events from Geostationary Satellite Images
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 exclus...
Main Authors: | , , , , |
---|---|
Format: | Article |
Language: | English |
Published: |
MDPI AG
2022-08-01
|
Series: | Remote Sensing |
Subjects: | |
Online Access: | https://www.mdpi.com/2072-4292/14/15/3760 |
_version_ | 1797440749309001728 |
---|---|
author | Sebastian Brodehl Richard Müller Elmar Schömer Peter Spichtinger Michael Wand |
author_facet | Sebastian Brodehl Richard Müller Elmar Schömer Peter Spichtinger Michael Wand |
author_sort | Sebastian Brodehl |
collection | DOAJ |
description | 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. |
first_indexed | 2024-03-09T12:13:50Z |
format | Article |
id | doaj.art-680ee98cfecd4663b4647be7ce811f89 |
institution | Directory Open Access Journal |
issn | 2072-4292 |
language | English |
last_indexed | 2024-03-09T12:13:50Z |
publishDate | 2022-08-01 |
publisher | MDPI AG |
record_format | Article |
series | Remote Sensing |
spelling | doaj.art-680ee98cfecd4663b4647be7ce811f892023-11-30T22:49:36ZengMDPI AGRemote Sensing2072-42922022-08-011415376010.3390/rs14153760End-to-End Prediction of Lightning Events from Geostationary Satellite ImagesSebastian Brodehl0Richard Müller1Elmar Schömer2Peter Spichtinger3Michael Wand4Institute of Computer Science, Johannes Gutenberg University Mainz, 55128 Mainz, GermanyGerman Weather Service, 63067 Offenbach, GermanyInstitute of Computer Science, Johannes Gutenberg University Mainz, 55128 Mainz, GermanyInstitute for Atmospheric Physics, Johannes Gutenberg University Mainz, 55128 Mainz, GermanyInstitute of Computer Science, Johannes Gutenberg University Mainz, 55128 Mainz, GermanyWhile 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.https://www.mdpi.com/2072-4292/14/15/3760neural networkssatellite imagesclass imbalancefeature attributionlightning predictionnowcasting |
spellingShingle | Sebastian Brodehl Richard Müller Elmar Schömer Peter Spichtinger Michael Wand End-to-End Prediction of Lightning Events from Geostationary Satellite Images Remote Sensing neural networks satellite images class imbalance feature attribution lightning prediction nowcasting |
title | End-to-End Prediction of Lightning Events from Geostationary Satellite Images |
title_full | End-to-End Prediction of Lightning Events from Geostationary Satellite Images |
title_fullStr | End-to-End Prediction of Lightning Events from Geostationary Satellite Images |
title_full_unstemmed | End-to-End Prediction of Lightning Events from Geostationary Satellite Images |
title_short | End-to-End Prediction of Lightning Events from Geostationary Satellite Images |
title_sort | end to end prediction of lightning events from geostationary satellite images |
topic | neural networks satellite images class imbalance feature attribution lightning prediction nowcasting |
url | https://www.mdpi.com/2072-4292/14/15/3760 |
work_keys_str_mv | AT sebastianbrodehl endtoendpredictionoflightningeventsfromgeostationarysatelliteimages AT richardmuller endtoendpredictionoflightningeventsfromgeostationarysatelliteimages AT elmarschomer endtoendpredictionoflightningeventsfromgeostationarysatelliteimages AT peterspichtinger endtoendpredictionoflightningeventsfromgeostationarysatelliteimages AT michaelwand endtoendpredictionoflightningeventsfromgeostationarysatelliteimages |