Deep learning for weather forecast




  • Description of the data and preprocessing pipeline.
  • Description of the used methodology and frameworks.
  • Introduction to the challenge deep learning for weather forecasting

What: The weather is a chaotic system. Small errors in the initial conditions of a forecast grow rapidly, and affect predictability. Furthermore, predictability is limited by model errors due to the approximate simulation of atmospheric processes of the stateof- the-art numerical models. These days, on the other hand, we gather a lot of data thanks to modern IoT technologies richly occupied in the fields. Incorporation of global weather data among data collected from sensors in the same field can contribute to accurate weather forecast in the local environment. Preparation of training data i.e. creation of preprocessing pipeline where the global weather data would be correctly enhanced by in local data from sensors. Using RNNs (LSTM) with supporting Vowpal Wabbit or Prophet.

Why: Adaptation of deep learning algorithms specialized for time-series prediction can be beneficial or more accurate for weather forecasting in the local environment for farmers than the publicly available global forecast model.

Who (is the webinar for): Researchers who are interested in time-series and high-dimensional function modelling/prediction via deep learning.

About Your Presenters

Ondrej Kaas was born in 1991, currently researching doctorate in the area of clustering algorithms at Computer Science at the University of West Bohemia, Faculty of Applied Science in Pilsen, Czech Republic. Employed history contains more than six years as a machine learning researcher of big data in-car telemetry, internet advertisement, and remote sensing of Earth. His professionalism of interest is time series prediction, object detection, and semantic segmentation in satellite data based on deep neural networks and related distributed computing. His technical skills cover the development of real-time cloud-based applications, IoT, and sensors.

Ondrej Kaas

Ondrej Kaas

Data Science Enthusiast, Plan4all

Amit Kirschenbaum is a research associate at the Institute for Applied Informatics (InfAI), Leipzig. He currently works on the H2020 STARGATE project, where he focuses on climatic data analysis and data integration for smart agriculture.

Amit Kirschenbaum 150

Amit Kirschenbaum

Research Associate, Leipzig University

Bente Lilja Bye has been a member of the GEO community since 2004, engaged both as representative in the GEO plenary, in committees and contributing to the GEO Work Programme, and currently represents Norway on the GEO Programme Board. Bente runs a small research and consultancy company, BLB, focusing on transforming Earth observation data to information and knowledge for societal benefit. She is responsible for Communication, Dissemination and Assessment as partner in NextGEOSS.

Bente Lilja Bye 150

Bente Lilja Bye CEO, BLB & webinar host on behalf of Plan4all