Watershed Alchemy: Transforming Data into Hydrologic Understanding

Data driven models are emergent tools in the field of hydrology that can enable researchers to leverage the vast quantities of available data to skillfully predict hydrologic responses without a priori knowledge of hydrologic processes. Large, in-situ hydrologic datasets as well as novel datasets such as the Surface Water and Topography Mission (SWOT) present an opportunity to turn the problem of hydrologic modeling on its head by extracting information embedded in the data in an effort to improve our understanding of hydrologic processes. A key unanswered question in hydrologic science is how simultaneous, complex climate changes will collectively affect hydroclimatic extremes in different regions and across varying spatial and temporal scales. For example, a regional trend toward a wetter climate could result in higher average soil moisture, and if this increased soil moisture is combined with a trend toward more intense storms, the combined effects on runoff rates could be significantly greater than the impact of either change alone. In this research, we are using data driven models such as recurrent neural networks (e.g. Long Short Term Memory Networks) to investigate the watershed-scale streamflow response to changes in underlying hydrologic variables. By examining a large range of watersheds, we aim to quantify how this response differs across different spatial and temporal scales. A better understanding of these and other factors influencing extreme runoff generation will enable us to better understand the implications of climate change and improve our physics-based hydrologic models.