We present the encoder-forecaster convolutional long short-term memory (LSTM) deep-learning model that powers Microsoft Weather's operational precipitation nowcasting product. This model takes as input a sequence of weather radar mosaics and deterministically predicts future radar reflectivity at lead times up to 6 hours. By stacking a large input receptive field along the feature dimension and conditioning the model's forecaster with predictions from the physics-based High Resolution Rapid Refresh (HRRR) model, we are able to outperform optical flow and HRRR baselines by 20-25% on multiple metrics averaged over all lead times.
@article{arxiv.2111.09954,
title = {MS-nowcasting: Operational Precipitation Nowcasting with Convolutional LSTMs at Microsoft Weather},
author = {Sylwester Klocek and Haiyu Dong and Matthew Dixon and Panashe Kanengoni and Najeeb Kazmi and Pete Luferenko and Zhongjian Lv and Shikhar Sharma and Jonathan Weyn and Siqi Xiang},
journal= {arXiv preprint arXiv:2111.09954},
year = {2022}
}
Comments
Minor updates to reflect final submission to NeurIPS workshop