Precipitation Nowcasting: Leveraging bidirectional LSTM and 1D CNN
In one sentence. Bidirectional LSTM is compared with a 1D CNN for multivariate precipitation nowcasting, with sequence models favoured over feed-forward networks; no quantitative results are given.
Abstract
Short-term rainfall forecasting, also known as precipitation nowcasting has become a potentially fundamental technology impacting significant real-world applications ranging from flight safety, rainstorm alerts to farm irrigation timings. Since weather forecasting involves identifying the underlying structure in a huge amount of data, deep-learning based precipitation nowcasting has intuitively outperformed the traditional linear extrapolation methods. Our research work intends to utilize the recent advances in deep learning to nowcasting, a multi-variable time series forecasting problem. Specifically, we leverage a bidirectional LSTM (Long Short-Term Memory) neural network architecture which remarkably captures the temporal features and long-term dependencies from historical data. To further our studies, we compare the bidirectional LSTM network with 1D CNN model to prove the capabilities of sequence models over feed-forward neural architectures in forecasting related problems.
Labels
- Topic
- Nowcasting
- Type
- New model or method
- Method
- RNN / LSTM / ConvLSTM, CNN / U-Net
- Time scale
- 0–6 h
- Data
- Other
- Status
- Research
- Identifiers
- arXiv:1810.10485 OpenAlex:W2896555060
- Added
- 2026-10-22
Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.