Date of Award
Master of Science
Agricultural and Biosystems Engineering
Agricultural and Biosystems Engineering; Environmental Science
Amy L Kaleita
Historically, the geologically-young Des Moines Lobe of Iowa was a complex wetland system. These geographically-isolated upland wetlands, frequently referred to as prairie potholes, have been methodically drained since the late 19th century to increase arable land for row crop agriculture. The complex and varying hydrology of prairie potholes, largely classified as semi-permanent wetlands, including their connection to downstream waters, has been the subject of many legal debates without strong scientific consensus regarding their impacts. The investigations of prairie potholes in this thesis attempt to build a cohesive narrative and develop tools to help clarify the impacts of agricultural management actions on pothole hydrology and subsequently, the general hydrologic behavior of farmed prairie potholes. Specifically, our investigations focus on, 1) Improving on the methodology of the AnnAGNPS modeling framework, used to model the hydrology of individual prairie potholes; 2) Generating a broader characterization of hydrologic responses to land management and climate variables within potholes using AnnAGNPS simulations; 3) Creating a simplified machine learning model to assess the relative flood risk of individual prairie potholes; and 4) Developing an accessible tool to educate and inform agricultural decision makers on the impacts of their management decisions on prairie pothole hydrology. First, we successfully calibrated 6 prairie pothole watershed models in the DML. Observations from these efforts include the following: 1) Updates to AnnAGNPS source code allow for wetland volumetric calibration as opposed to depth calibration. 2) As watersheds grow, AnnAGNPS becomes limiting in modeling a connected, spill-and-fill network of potholes. 3) Equifinality is a major issue but can be addressed by multiple statistical parameters and reproducible methods. Second, we modeled each prairie pothole using 28 different scenarios and 25 years of climate data and assessed changes in flooding. Results suggest that the opposing actions of drainage investment and significant land retirement both provide substantial reduction in pothole flooding, while tillage practices and pothole retirement provide moderate to minimal reduction. However, the frequency, extents and duration of flooding displayed in simulations suggest that many farmed potholes would continue to experience significant crop loss despite drainage investment. We also note that this flooding is highly exacerbated in high-precipitation years and that changing climate could continue to reduce the viability of farming these marginal areas. Third, we calibrated and validated a random forest machine learning model to predict pothole flood risk using a unique flood risk metric. Our risk prediction model outputs a numerical rank to characterize specific land management scenarios which were previously modeled. This strongly performing regression is constrained to potholes which are semi-permanent and may be actively farmed under natural or altered conditions. The model tree structures and predicted risk values are able to be utilized for assessing relative impacts of management actions on characteristic, general flood risk for a farmable prairie pothole. Finally, we incorporated the random forest model into an interactive web-based R Shiny Application available for public use. Dynamic interaction with inputs and outputs via this app, named the Prairie Pothole Management Support Tool (PPMST), enable users to assess prairie potholes individually and under multiple alternative scenarios. The PPMST is a capable tool for educational and information purposes with the audience being primary agricultural decision makers in the Des Moines Lobe of Iowa, whether landowners, renters, conservationists, or other stakeholders in row crop agricultural production. Cumulatively, this work improves our understanding of field-scale prairie pothole inundation dynamics across a range of management scenarios while improving and developing tools for quantitative and qualitative flood risk assessments.
Nahkala, Brady, "Watershed modeling and random forest flood risk classification of farmed prairie potholes" (2020). Graduate Theses and Dissertations. 18367.