Weather4cast at NeurIPS 2022: Super-Resolution Rain Movie Prediction under Spatio-temporal Shifts

Aleksandra Gruca, Federico Serva, Llorenç Lliso, Pilar Rípodas, Xavier Calbet, Pedro Herruzo, Jiřı́ Pihrt, Rudolf Raevskyi, Petr Šimánek, Matej Choma, Yang Li, Haiyu Dong, Yury Belousov, Sergey Polezhaev, Brian Pulfer, Minseok Seo, Doyi Kim, Seungheon Shin, Eunbin Kim, Sewoong Ahn, Yeji Choi, Jinyoung Park, Minseok Son, Seungju Cho, Inyoung Lee, Changick Kim, Taehyeon Kim, Shinhwan Kang, Hyeonjeong Shin, Deukryeol Yoon, Seongha Eom, Kijung Shin, Se-Young Yun, Bertrand Le Saux, Michael K Kopp, Sepp Hochreiter, David P Kreil
Proceedings of the NeurIPS 2022 Competitions Track, PMLR 220:292-313, 2022.

Abstract

Weather4cast again advanced modern algorithms in AI and machine learning through a highly topical interdisciplinary competition challenge: The prediction of hi-res rain radar movies from multi-band satellite sensors, requiring data fusion, multi-channel video frame prediction, and super-resolution. Accurate predictions of rain events are becoming ever more critical, with climate change increasing the frequency of unexpected rainfall. The resulting models will have a particular impact where costly weather radar is not available. We here present highlights and insights emerging from the thirty teams participating from over a dozen countries. To extract relevant patterns, models were challenged by spatio-temporal shifts. Geometric data augmentation and test-time ensemble models with a suitable smoother loss helped this transfer learning. Even though, in ablation, static information like geographical location and elevation was not linked to performance, the general success of models incorporating physics in this competition suggests that approaches combining machine learning with application domain knowledge seem a promising avenue for future research. Weather4cast will continue to explore the powerful benchmark reference data set introduced here, advancing competition tasks to quantitative predictions, and exploring the effects of metric choice on model performance and qualitative prediction properties.

Cite this Paper

BibTeX
@InProceedings{pmlr-v220-gruca23a, title = {Weather4cast at NeurIPS 2022: Super-Resolution Rain Movie Prediction under Spatio-temporal Shifts}, author = {Gruca, Aleksandra and Serva, Federico and Lliso, Lloren\c{c} and R\'ipodas, Pilar and Calbet, Xavier and Herruzo, Pedro and Pihrt, Ji\v{r}\'{\i} and Raevskyi, Rudolf and \v{S}im\'{a}nek, Petr and Choma, Matej and Li, Yang and Dong, Haiyu and Belousov, Yury and Polezhaev, Sergey and Pulfer, Brian and Seo, Minseok and Kim, Doyi and Shin, Seungheon and Kim, Eunbin and Ahn, Sewoong and Choi, Yeji and Park, Jinyoung and Son, Minseok and Cho, Seungju and Lee, Inyoung and Kim, Changick and Kim, Taehyeon and Kang, Shinhwan and Shin, Hyeonjeong and Yoon, Deukryeol and Eom, Seongha and Shin, Kijung and Yun, Se-Young and {Le Saux}, Bertrand and Kopp, Michael K and Hochreiter, Sepp and Kreil, David P}, booktitle = {Proceedings of the NeurIPS 2022 Competitions Track}, pages = {292--313}, year = {2022}, editor = {Ciccone, Marco and Stolovitzky, Gustavo and Albrecht, Jacob}, volume = {220}, series = {Proceedings of Machine Learning Research}, month = {28 Nov--09 Dec}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v220/gruca23a/gruca23a.pdf}, url = {https://proceedings.mlr.press/v220/gruca23a.html}, abstract = {Weather4cast again advanced modern algorithms in AI and machine learning through a highly topical interdisciplinary competition challenge: The prediction of hi-res rain radar movies from multi-band satellite sensors, requiring data fusion, multi-channel video frame prediction, and super-resolution. Accurate predictions of rain events are becoming ever more critical, with climate change increasing the frequency of unexpected rainfall. The resulting models will have a particular impact where costly weather radar is not available. We here present highlights and insights emerging from the thirty teams participating from over a dozen countries. To extract relevant patterns, models were challenged by spatio-temporal shifts. Geometric data augmentation and test-time ensemble models with a suitable smoother loss helped this transfer learning. Even though, in ablation, static information like geographical location and elevation was not linked to performance, the general success of models incorporating physics in this competition suggests that approaches combining machine learning with application domain knowledge seem a promising avenue for future research. Weather4cast will continue to explore the powerful benchmark reference data set introduced here, advancing competition tasks to quantitative predictions, and exploring the effects of metric choice on model performance and qualitative prediction properties.} }
Endnote
%0 Conference Paper %T Weather4cast at NeurIPS 2022: Super-Resolution Rain Movie Prediction under Spatio-temporal Shifts %A Aleksandra Gruca %A Federico Serva %A Llorenç Lliso %A Pilar Rípodas %A Xavier Calbet %A Pedro Herruzo %A Jiřı́ Pihrt %A Rudolf Raevskyi %A Petr Šimánek %A Matej Choma %A Yang Li %A Haiyu Dong %A Yury Belousov %A Sergey Polezhaev %A Brian Pulfer %A Minseok Seo %A Doyi Kim %A Seungheon Shin %A Eunbin Kim %A Sewoong Ahn %A Yeji Choi %A Jinyoung Park %A Minseok Son %A Seungju Cho %A Inyoung Lee %A Changick Kim %A Taehyeon Kim %A Shinhwan Kang %A Hyeonjeong Shin %A Deukryeol Yoon %A Seongha Eom %A Kijung Shin %A Se-Young Yun %A Bertrand Le Saux %A Michael K Kopp %A Sepp Hochreiter %A David P Kreil %B Proceedings of the NeurIPS 2022 Competitions Track %C Proceedings of Machine Learning Research %D 2022 %E Marco Ciccone %E Gustavo Stolovitzky %E Jacob Albrecht %F pmlr-v220-gruca23a %I PMLR %P 292--313 %U https://proceedings.mlr.press/v220/gruca23a.html %V 220 %X Weather4cast again advanced modern algorithms in AI and machine learning through a highly topical interdisciplinary competition challenge: The prediction of hi-res rain radar movies from multi-band satellite sensors, requiring data fusion, multi-channel video frame prediction, and super-resolution. Accurate predictions of rain events are becoming ever more critical, with climate change increasing the frequency of unexpected rainfall. The resulting models will have a particular impact where costly weather radar is not available. We here present highlights and insights emerging from the thirty teams participating from over a dozen countries. To extract relevant patterns, models were challenged by spatio-temporal shifts. Geometric data augmentation and test-time ensemble models with a suitable smoother loss helped this transfer learning. Even though, in ablation, static information like geographical location and elevation was not linked to performance, the general success of models incorporating physics in this competition suggests that approaches combining machine learning with application domain knowledge seem a promising avenue for future research. Weather4cast will continue to explore the powerful benchmark reference data set introduced here, advancing competition tasks to quantitative predictions, and exploring the effects of metric choice on model performance and qualitative prediction properties.
APA
Gruca, A., Serva, F., Lliso, L., Rípodas, P., Calbet, X., Herruzo, P., Pihrt, J., Raevskyi, R., Šimánek, P., Choma, M., Li, Y., Dong, H., Belousov, Y., Polezhaev, S., Pulfer, B., Seo, M., Kim, D., Shin, S., Kim, E., Ahn, S., Choi, Y., Park, J., Son, M., Cho, S., Lee, I., Kim, C., Kim, T., Kang, S., Shin, H., Yoon, D., Eom, S., Shin, K., Yun, S., Le Saux, B., Kopp, M.K., Hochreiter, S. & Kreil, D.P.. (2022). Weather4cast at NeurIPS 2022: Super-Resolution Rain Movie Prediction under Spatio-temporal Shifts. Proceedings of the NeurIPS 2022 Competitions Track, in Proceedings of Machine Learning Research 220:292-313 Available from https://proceedings.mlr.press/v220/gruca23a.html.

Related Material