Multi-Horizon Typhoon Wind Field Prediction via a Lightweight CNN–LSTM Network: Error Growth and Cross-Year Robustness at 6–24 h Lead Times
- 1School of Computer, Electronics and Information, Guangxi University, Nanning 530004, China
- 2China Meteorology Administration Earth System Modeling and Prediction Centre, Beijing 100081, China
Abstract
Short-term typhoon wind-field prediction across multiple lead times is challenging because forecast error grows with lead time and inappropriate sample construction may leak future storm-position information. We adopt STL-Net, a lightweight spatiotemporal framework integrating convolutional encoding, long short-term memory (LSTM) temporal modeling, squeeze-and-excitation recalibration, and multi-branch fusion for 10 m wind-field prediction at 6, 12, and 24 h lead times. A fixed-t₀ protocol anchors the target patch at the initialization-time storm position, avoiding future best-track leakage. Trained in 2020–2021, validated in 2022, and tested in 2023, STL-Net achieves a root mean square error (RMSE) of 3.11, 4.06, and 5.31 m·s⁻¹ at 6, 12, and 24 h. Among deep-learning baselines, U-Net achieves the lowest errors at 6 and 12 h and convolutional neural network (CNN) the lowest 24 h RMSE and mean absolute error (MAE), while STL-Net remains competitive with 1.673 million parameters and low inference latency. Ten-seed ablation identifies multi-resolution fusion as the most consistently beneficial component. An out-of-year evaluation with 2020 as the test year reproduces the error-growth pattern, supporting cross-year robustness. Models are deterministic without uncertainty quantification; error growth is deterministic, not calibrated uncertainty. Diagnostic interpretation links the error structure to the multi-scale organization of typhoon winds, storm translation, and wind-field asymmetry. Results constitute an offline proof-of-concept, not an operational system.
Related
- Ensemble fire probability prediction by integrating meteorological AI models with lightweight prediction networks
- Should All Noises Be Treated Equally: Impact of Input Noise Variability on Neural Network Robustness
- Spatial association network and structural robustness of human-water system resilience in the Yangtze River Economic Belt, China
- Adaptive Horizon-Aware Model Selection and Routing for Multi-Step Water Quality Prediction: An Integrative Review
- A transparent constrained stochastic framework for urban expansion forecasting and future service-network screening: Cape Town demonstration using annual building evidence, empirical land constraints, spatial hindcasting, and multi-horizon ensembles
BibTeX
@article{liu2026multihorizon,
title = {{Multi-Horizon Typhoon Wind Field Prediction via a Lightweight CNN–LSTM Network: Error Growth and Cross-Year Robustness at 6–24 h Lead Times}},
author = {Jun Liu and Jie Cui and Yan Liu},
journal = {Atmosphere},
year = {2026},
month = sep,
doi = {10.3390/atmos17100950},
url = {https://doi.org/10.3390/atmos17100950},
note = {CC BY},
}
RIS
TY - JOUR
TI - Multi-Horizon Typhoon Wind Field Prediction via a Lightweight CNN–LSTM Network: Error Growth and Cross-Year Robustness at 6–24 h Lead Times
AU - Jun Liu
AU - Jie Cui
AU - Yan Liu
PY - 2026
DA - 2026/09/01
JO - Atmosphere
DO - 10.3390/atmos17100950
UR - https://doi.org/10.3390/atmos17100950
AB - Short-term typhoon wind-field prediction across multiple lead times is challenging because forecast error grows with lead time and inappropriate sample construction may leak future storm-position information. We adopt STL-Net, a lightweight spatiotemporal framework integrating convolutional encoding, long short-term memory (LSTM) temporal modeling, squeeze-and-excitation recalibration, and multi-branch fusion for 10 m wind-field prediction at 6, 12, and 24 h lead times. A fixed-t₀ protocol anchors the target patch at the initialization-time storm position, avoiding future best-track leakage. Trained in 2020–2021, validated in 2022, and tested in 2023, STL-Net achieves a root mean square error (RMSE) of 3.11, 4.06, and 5.31 m·s⁻¹ at 6, 12, and 24 h. Among deep-learning baselines, U-Net achieves the lowest errors at 6 and 12 h and convolutional neural network (CNN) the lowest 24 h RMSE and mean absolute error (MAE), while STL-Net remains competitive with 1.673 million parameters and low inference latency. Ten-seed ablation identifies multi-resolution fusion as the most consistently beneficial component. An out-of-year evaluation with 2020 as the test year reproduces the error-growth pattern, supporting cross-year robustness. Models are deterministic without uncertainty quantification; error growth is deterministic, not calibrated uncertainty. Diagnostic interpretation links the error structure to the multi-scale organization of typhoon winds, storm translation, and wind-field asymmetry. Results constitute an offline proof-of-concept, not an operational system.
SN - 2073-4433
ER -