Combining process understanding and machine learning: Two hydrologic case studies

Department of Hydrology and Atmospheric Sciences

4 pm on Thursday, November 19, 2020
Contact the department for zoom details or to subscribe to the department email list

Tianfang Xu
Assistant Professor, School of Sustainable Engineering and the Built Environment, Arizona State University

Abstract

Knowledge-driven reasoning and process-based models have been the primary tools to analyze and quantify hydrologic processes as well as inform decisions with significant social and economical implications. Recent years have seen dramatically increased data availability, and existing knowledge about the hydrological processes are no longer adequate to represent the full range of variability observed in data. Machine learning and data-driven reasoning it enables provide exciting opportunities to get the most out of data and improve prediction capability. We discuss two hydrologic applications in which machine learning are used in combination with process understanding. In the first application, we use machine learning to integrate satellite imagery and various hydrometeorological data to create high-resolution maps of irrigated row crops in southwestern Michigan. Guided by process understanding, a set of novel features based on weather-sensitive scene selection are developed to enhance the contrast between neighboring rainfed and irrigated areas, thus overcoming a key challenge in remote sensing-based irrigation mapping in subhumid to humid areas. In the second application, we present a hybrid modeling approach to simulating streamflow in snow-dominated mountainous karst watersheds. A high-resolution snow model captures spatiotemporally varying snowmelt, which is then digested by a deep learning karst model based on CNN and LSTM architectures. The hybrid models are applied to a watershed in northern Utah with seasonal snow cover and variably karstified carbonate bedrock.

Bio

Dr. Tianfang Xu is an assistant professor in School of Sustainable Engineering and the Built Environment at Arizona State University. She holds a bachelor degree in Geotechnical Engineering from Nanjing University, China, and master’s and doctoral degrees in Civil Engineering from University of Illinois at Urbana-Champaign. Before joining ASU, she was a postdoctoral researcher at Michigan State University and a research assistant professor in Department of Civil and Environmental Engineering and Utah Water Research Laboratory, Utah State University. Her research interests include numerical simulation of groundwater flow and solute transport, uncertainty quantification, and machine learning.