NEW PARADIGMS FOR APPLICATION OF INTELLIGENT TECHNIQUES IN HYDROLOGY: BRIDGING MONITORING, MODELING AND UNDERSTANDING

Abstract

This thesis explores the application of two novel intelligent techniques (namely computer vision and explainable AI) in hydrology, as a complementary research avenue to traditional hypothesis-driven research. The new avenue starts from data acquisition, uses intelligent techniques to assist hydrological modeling, and applies interpretation techniques to extract physical insights. Specifically, the thesis is comprised of four topics: (1) using computer vision techniques to supplement current hydrological monitoring networks, (2) using computer vision to exploit information involved in data fields to empower hydrological modeling capacities, (3) incorporating physical knowledge into deep learning approaches to fill the gap between process-based and data-driven models, and (4) extracting physical knowledge from deep learning by using interpretation techniques to advance hydrological understanding. The results demonstrate the great potential and substantial promise of intelligent techniques for advancing hydrology. We argue for open minds and coordinated efforts from the community toward AI-powered scientific advances in hydrology.