Forecasting of Tropospheric Delay Using AI Foundation Models in Support of Microwave Remote Sensing
Abstract
Accurate tropospheric delay forecasts are imperative for microwave-based remote sensing techniques, playing a pivotal role in early warning and forecasting of natural disasters such as tsunamis, heavy rains, and hurricanes. Conventional methods for forecasting tropospheric delays entail substantial computational resources and high network transmission speeds, restricting their real-time applicability. This study introduces a novel approach to derive forecasted tropospheric delays using artificial intelligence weather forecast foundation models, exemplified by Huawei Cloud Pangu-Weather, Google DeepMind GraphCast, and Shanghai AI Lab FengWu. Forecast accuracy is assessed globally using ERA5, ground-based Global Navigation Satellite System, and in situ radiosonde measurements as reference data. Results show that the foundation-model-based scheme outperforms traditional methods in both forecast accuracy and length, can provide high-accuracy tropospheric delay parameters locally for 15-day forecasts at any location within minutes, and remains more accurate than empirical models when forecasting up to ten days in advance.
