Search papers, labs, and topics across Lattice.
This study enhances physics-constrained neural networks (PCNNs) for short-term weather forecasting by introducing three key innovations to the WeatherGFT architecture. An upgraded numerical solver allows for a fourfold increase in integration time steps while significantly reducing daily mean squared error, and a unified autoregressive hybrid block mitigates overfitting across lead times. The integration of advanced neural backbones with the physical core results in models that outperform purely neural approaches, achieving 8-22% reductions in root mean squared error at 1-12 hour lead times while maintaining physical consistency.
Hybrid models that blend physics and advanced neural architectures can achieve up to 22% better accuracy in short-term weather forecasting without sacrificing physical realism.
This study introduces enhancements to physics-constrained neural networks (PCNNs) that improve the accuracy and stability of hybrid short-term weather forecasting models. Building on the WeatherGFT architecture, three innovations are proposed. First, an upgraded numerical solver, combining a fifth-order weighted essentially non-oscillatory scheme (WENO-5), a beta-plane approximation, and subgrid-scale viscosity, permits a fourfold increase in the integration time step to 1200 s while reducing the daily mean squared error by up to 26%. Second, a unified autoregressive hybrid block replaces the original chain of 24 specialised modules, eliminating overfitting to specific lead times. Third, the physical core is integrated with two state-of-the-art neural backbones, resulting in PI-PredFormer and PI-IAM4VP. Evaluation on the WeatherBench South Pacific subset from 2000 to 2004 shows that these hybrids reduce root mean squared error at 1-12 h lead times by 8-22% compared to purely neural counterparts, while better preserving physical consistency. These results demonstrate that incremental refinement of hybrid components offers a practical route toward more accurate and efficient short-range weather forecasting.
