Data-driven Seasonal Hydrologic Prediction Using Earth Observing Satellites

点击翻译按钮获取中文摘要

Timely prediction of flood and drought greatly minimize the related losses. It requires precise precipitation estimates as a predictor of terrestrial processes across spatiotemporal scales. In this fiscal year, we proposed a novel precipitation retrieval framework in which regression and classification tasks are simultaneously trained using multi-task learning approach. Satellite-based precipitation estimations provide frequent, large-scale measurements. Recently, deep learning has shown significant potential for improving estimation accuracy. In this project, we designed a novel network architecture and loss function to maximize the benefits of multi-task learning. The proposed multi-task (i.e., two-task) model successfully achieved a better performance than the conventional single-task model possibly due to efficient knowledge transfer between tasks. The product intercomparison showed that our product outperformed existing products in rain rate retrieval and also yielded better skills in the rain/no-rain retrieval task.