WRS Doctoral Defense

Data Driven Discoveries in Streamflow, Vadose Zone, and Baseflow

With burgeoning “big data” in recent decades, a few advanced data driven techniques including machine learning and deep learning became popular and impactful tools to advance scientific discovery. One of these disciplines is hydrology, where a vast amount of data from various sources provides abundant digitized information. This information characterizes and models complex terrestrial hydrological processes, thus allowing opportunities for hydrologic discoveries. In this dissertation, I’ll show how big data drive scientific discoveries in streamflow generation, vadose zone processes, and baseflow recession analysis. In particular, random vector strategy is a viable approach that yields reasonably well hydrology regionalization performance in gauged scenarios in contrast to catchment physical descriptors. Besides, the recent machine learning trend in vadose zone processes shows a strong trend in classic machine learning models with a lack in deep learning. Last, one of the reasons that contribute to the recognized inconsistency in baseflow recession characteristics is that the selected recessions do not exclude quick flow.