MS-nowcasting: Operational Precipitation Nowcasting with Convolutional LSTMs at Microsoft Weather
This work improves operational weather forecasting for meteorologists and the public, though it is incremental as it builds on existing deep-learning and physics-based methods.
The authors tackled precipitation nowcasting by developing an encoder-forecaster convolutional LSTM model that predicts radar reflectivity up to 6 hours ahead, outperforming optical flow and HRRR baselines by 20-25% on multiple metrics.
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.