Master of Science
Water quality impairment in small tributaries due to soil erosion and stream degradation of West Tennessee is an ongoing problem. A method to model streamflow permanence can assist stream restoration work by supplementing ground monitoring and providing better targeting of conservation attempts in the most vulnerable areas. This project applied a random forest model by incorporating climatic and landcover data as predictors to create streamflow permanence data for the West Tennessee tributaries (Lower Mississippi-Hatchie Hydrologic Unit, HUC 4-801). Specifically, the applicability of the Flow Conditioned Parameter Grids (FCPG) process is tested to study if the process improves prediction results compared to raw predictor results. In addition, the model’s ability to capture the effect of headwater lakes in increasing the probability of streamflow permanence in downstream reaches is investigated using two pairs of streams from the Tennessee Department of Environmental Conservation (TDEC) database of watershed water quality assessments. With the various predictor variable configurations tested in the model, an average of 25 percent Mean Squared Error (MSE) accuracy is acquired in the prediction of the streamflow permanence status of west Tennessee streams. The results showed processing FCPG layers did not provide an increase in prediction accuracy for this study. Validation of the model results using the test stream pairs was inconclusive. While the model did predict streamflow permanence downstream of one headwater lake and intermittent streamflow in the other stream of the pair. it predicted perennial flow for both streams in the second pair, regardless of the presence of a headwater lake. This project provides scalable and replicable methods using machine learning and remote sensing data to predict streamflow permanence at a 25% MSE rate.
Dissertation or thesis originally submitted to ProQuest
Tok, Berkay, "Random Forest Modeling Approach to Predict Streamflow Intermittence of Tennessee Headwaters using Flow Conditioned Parameter Grids" (2022). Electronic Theses and Dissertations. 3167.