


This session features an in-depth discussion of the Weather4cast 2021 competition results with the CORE and TRANSFER challenge winners as well as selected participants from the leaderboard.
The goal of the competition was a short-term prediction of selected weather products based on meteorological satellites data obtained in collaboration with AEMET/ NWC SAF. The competition presented weather forecast as a video frame prediction task. The challenge was to predict the encoded weather products (cloud properties, temperature, turbulence, and rainfall) in various regions including mountains, deserts, islands and seas, and others.
The competition dataset offers real-world benchmark for few shot and transfer learning and allows testing multi-sensor data fusion.
Program
1st session:
Wednesday, 2021-12-15 11:00-12:50 EST (17:00 – 18:40 CET)
chair: Michael Kopp
Time (EST) | Title | Presenter | Affiliation |
---|---|---|---|
11:00 | Keynote talk: Insights on end-to-end weather forecasting | Nal Kalchbrenner | Google Brain, Netherlands |
11:40 | First insights on transfer learning from the Weather4cast Competitions 2021 | Pedro Herruzo | IARAI, Austria |
12:00 | 1st prize: Improvements to short-term weather prediction with recurrent-convolutional networks | Jussi Leinonen | MeteoSwiss, Switzerland |
12:20 | 2nd prize: UNet based future weather prediction on Weather4cast 2021 stage 2 | Sungbin Choi | South Korea |
12:40 | Coffee Break |
2nd session:
Wednesday, 2021-12-15 13:00-15:00 EST (19:00 – 21:00 CET)
chair: David Kreil
Time (EST) | Title | Presenter | Affiliation |
---|---|---|---|
13:00 | Invited talk: The NWC SAF satellite derived weather nowcasting & data source for ML | Llorenç Lliso | AEMET |
13:20 | 3rd prize: Enhanced variational U-Net for weather forecasting | Pak Hay Kwok and Qi Qi | UK |
13:40 | Submission: Spatiotemporal Vision Transformer for short time weather forecasting | Alabi Bojesomo | Khalifa University, UAE |
14:00 | Closing session keynote talk: Skilful precipitation nowcasting using deep generative models of radar | Piotr Mirowski | DeepMind, UK |
14:40 | Discussion and closing remarks | David Kreil | IARAI |
Keynote Speakers

Keynote talk:
Google Brain, Netherlands
Insights on end-to-end weather forecasting
The problem of forecasting weather has been scientifically studied for centuries due to its high impact on human lives, transportation, food production and energy management, among others. Current operational forecasting models are based on physics and use supercomputers to simulate the atmosphere to make forecasts hours and days in advance. Better physics-based forecasts require improvements in the models themselves, which can be a substantial scientific challenge, as well as improvements in the underlying resolution, which can be computationally prohibitive. An emerging class of weather models based on neural networks represents a paradigm shift in weather forecasting: the models learn the required transformations from data instead of relying on hand-coded physics and are computationally efficient. For neural models, however, each additional hour of lead time poses a substantial challenge as it requires capturing ever larger spatial contexts and increases the uncertainty of the prediction. We present a neural network that is capable of large-scale precipitation forecasting up to twelve hours ahead and, starting from the same atmospheric state, the model achieves greater skill than the state-of-the-art physics-based models HRRR and HREF that currently operate in the Continental United States. Interpretability analyses reinforce the observation that the model learns to emulate advanced physics principles. These results represent a substantial step towards establishing a new paradigm of efficient forecasting with neural networks. The talk will cover these results as well as additional insights stemming from the research.
Nal Kalchbrenner is a research scientist and co-organizer of the Google Research Brain team in the Netherlands. Nal has contributed deep learning algorithms for various domains, from machine translation and NLP, to speech generation, image generation, deep RL for games and, most recently, also weather modeling. Nal was previously a researcher at DeepMind and before that he obtained his PhD in Computer Science from the University of Oxford.

Keynote talk:
DeepMind, UK
Skilful precipitation nowcasting using deep generative models of radar
Precipitation nowcasting, the high-resolution forecasting of precipitation up to two hours ahead, supports the real-world socioeconomic needs of many sectors reliant on weather-dependent decision-making. State-of-the-art operational nowcasting methods typically advect precipitation fields with radar-based wind estimates, and struggle to capture important nonlinear events such as convective initiations. Recently introduced deep learning methods use radar to directly predict future rain rates, free of physical constraints. While they accurately predict low-intensity rainfall, their operational utility is limited because their lack of constraints produces blurry nowcasts at longer lead times, yielding poor performance on rarer medium-to-heavy rain events. Here we present a deep generative model for the probabilistic nowcasting of precipitation from radar that addresses these challenges. Using statistical, economic and cognitive measures, we show that our method provides improved forecast quality, forecast consistency and forecast value. Our model produces realistic and spatiotemporally consistent predictions over regions up to 1,536 km × 1,280 km and with lead times from 5–90 min ahead. Using a systematic evaluation by more than 50 expert meteorologists, we show that our generative model ranked first for its accuracy and usefulness in 89% of cases against two competitive methods. When verified quantitatively, these nowcasts are skillful without resorting to blurring. We show that generative nowcasting can provide probabilistic predictions that improve forecast value and support operational utility, and at resolutions and lead times where alternative methods struggle.
Piotr Mirowski is a staff research scientist at DeepMind, where he is a member of Deep Learning department and Dr. R. Hadsell’s team. His work focuses on navigation-related research and scaling up autonomous agents to real world environments. He obtained his Ph.D. in Computer Science at New York University in 2011. His previous work experience encompasses epileptic seizure prediction from EEG, the inference of gene regulation networks, WiFi-based geo-localization, simultaneous localization and mapping on a smartphone, robotics, natural language processing, and search query auto-completion.