Proceedings of the 2023 Mississippi Water Resources Conference

Year: 2023 Authors: Roland V., Crowley-Ornelas E.

In recent history, interest in the conservation of riverine ecosystems has grown as people have become more knowledgeable about the functions of these habitats. Anthropogenic hydrologic alteration is a direct threat to the health of these ecosystems because it often triggers a range of negative effects on the biological, physical, chemical, and hydrologic characteristics of impacted waters. Understanding important factors and drivers of hydrologic alteration is essential to the planning effective conservation action plans. This study explores the application of machine learning to predicting hydrologic alteration and identifying important predictors of hydrologic alteration in the Pearl and Pascagoula River Basins in Mississippi. Modeled daily streamflow for 12-digits hydrologic unit code (HUC12) watershed pour points was used to compute the net change in streamflow volume and to conduct a confidence interval hypothesis test across pre- and post-alteration periods between 1950 and 2009. Cubist models were developed for each basin to predict the p value of the confidence interval test as a function of the net change and a range of other physical and meteorological watershed parameters. Analysis of the net change and confidence interval test results indicated the basins had similar amount of altered HUC12 watersheds. Moreover, patterns of altered watersheds tended to coincide with the locations of densely populated areas, dams, and in areas with substantial land cover change in both basins. The cubist models developed for the basins produced accurate predictions of the confidence interval test results in most HUC12 watersheds. The importance of model predictors demonstrated differences in the relationships between basin geomorphology, land cover, and hydrologic alteration in the basins. The results of this study are evidence of the potential of the cubist algorithm in hydrologic alteration assessments. More broadly, machine learning and other data driven approaches can be applied to a variety of complex water resources issues to inform local, state, and federal resource managers.