A Three-dimensional Convolutional-Recurrent Network for Convective Storm Nowcasting
In one sentence. A hybrid 3D CNN and ConvLSTM encoder-decoder fuses radar and 3D re-analysis fields to nowcast convective storms, outperforming extrapolation baselines.
Abstract
Very short-term convective storm forecasting, termed nowcasting, has long been an important issue and has attracted substantial interest. Existing nowcasting methods rely principally on radar images and are limited in terms of nowcasting storm initiation and growth. Real-time re-analysis of meteorological data supplied by numerical models provides valuable information about three-dimensional (3D), atmospheric, boundary layer thermal dynamics, such as temperature and wind. To mine such data, we here develop a convolution-recurrent, hybrid deep-learning method with the following characteristics: (1) the use of cell-based oversampling to increase the number of training samples; this mitigates the class imbalance issue; (2) the use of both raw 3D radar data and 3D meteorological data re-analyzed via multi-source 3D convolution without any need for handcraft feature engineering; and (3) the stacking of convolutional neural networks on a long short-term memory encoder/decoder that learns the spatiotemporal patterns of convective processes. Experimental results demonstrated that our method performs better than other extrapolation methods. Qualitative analysis yielded encouraging nowcasting results.
Labels
- Topic
- Nowcasting, Extremes & hazards
- Type
- New model or method
- Method
- CNN / U-Net, RNN / LSTM / ConvLSTM
- Time scale
- 0–6 h
- Domain
- Regional
- Data
- Radar, NWP analysis or forecasts (IFS, GFS, ...)
- Status
- Research
- Identifiers
- arXiv:1910.00527 DOI:10.1109/icbk.2019.00052
- Added
- 2026-10-22
Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.