The focus of this research is on the modeling of hydrologic and transport processes in watersheds at various spatial scales. The hydrologic processes include infiltration, surface runoff, evapo-transpiration, groundwater recharge, and groundwater discharge into streams. The scale of analysis spans from the soil pedon to the scale of a river basin.
Various approaches are used for models. One approach includes models based on the numerical solution of partial differential equations. Examples of this type of model are the Gridded Surface Subsurface Hydrology Analysis (GSSHA) tool, MODFLOW/MODPATH, COMSOL-MP, and GSFLOW. The GSSHA model has been implemented on the supercomputer and applied to Dobbins Creek near Austin, Minnesota where the researchers are modeling the transport of sediment and nutrients out of the watershed. MODFLOW/MODPATH and COMSOL-MP are currently being applied to selected watersheds in southeastern and north-central Minnesota. An alternative watershed modeling approach to these is the semi-distributed hydrologic model represented by the HSPF model, or the SWAT model. For these two models the watershed is subdivided into "homogeneous" land units and mass balances are conducted on each of the land units. One application of all these models will be to provide a conservation of mass constraint in the analysis of combined ground-based hydrologic measurements and satellite-based water volume measurements for a current research project. A second application will be to the simulation of transport of nitrate in groundwater over large areas in the southeast region of Minnesota, and in the north-central sandplains regions of Minnesota.
Parallel to the effort to application of physically based models, the researchers are currently conducting research on application of machine learning algorithms to emulate hydrological processes. Several manuscripts coauthored by hydrology researchers working with computer scientists have been published in the last two years. In addition to the development and testing of models for emulating hydrologic processes, the researchers are also investigating the use of ML-based models for causal analysis. At times the ML models will be run without direct coupling with physically based models, but the researchers are also conducting KGML where the physically based models are coupled to the machine learning models. The training effort for the ML models or the KGML models requires significant computational resources, and also high capacity storage. MSI resources meet the demands of these models. The supercomputer resources are needed to allow the simulation of hydrologic processes over large land surfaces while also allowing high enough spatial resolution to be able to model the processes realistically.