Reviews: ExtremeWeather: A large-scale climate dataset for semi-supervised detection, localization, and understanding of extreme weather events

This paper presents a new dataset, a model and experimental results on this dataset to address the task of extreme weather events detection and localization. The dataset is 27 year weather simulation sampled 8 times per day for 16 channels only the surface atmospheric level. The proposed model is based on 3D convolutional layers with an autoencoder architecture. The technique is semi-supervised, thus training with a loss that combines reconstruction error of the autoencoder and detection and localization from the middle code layer. In general the paper is very well written and quite clear on most details.