ANALYZING HYDROMORPHODYNAMICS AND SEDIMENTATION VARIATIONS IN THE LOWER APALACHICOLA RIVER SYSTEM

Abstract

Rivers in the environment have been vulnerable to human impact since the first settlement, as humans relied on them for their services. Industrialization, urbanization, and agriculture, including irrigation, dredging, and damming, have been practiced for centuries. Those practices induced alterations in water flows and led to continuous impacts on ecosystem services. This research focuses on one of the larger rivers in the southeastern United States, the Apalachicola River, which has historically been a site of problematic sedimentation but needs more scientific investigation. Consequently, this dissertation examines the hydromorphodynamic and sedimentation variations of the Lower Apalachicola River. We used bathymetry survey data from 1960 and 2010 and conducted Digital Elevation Model (DEM) of Differences (DoD) analysis from the 50-year period when the Navigation Project was conducted. It included dredging and disposal, artificial cutoffs, snag removal by the United States Army Corps of Engineers. The river's net sediment gain and loss patterns were assessed using high-resolution DEMs and geostatistical approaches. It was noted that the DoD map showed an unusual elevation change, suggesting a pool on the riverbed between RM 29 and 27 when the Chipola River joins with the Apalachicola River. The cumulative sediment volume change and gross change (cumulative absolute change) per river mile of the lower Apalachicola River between 1960 and 2010 were quantified. The entire reach (RM ~45-0) has a loss of -8.36 million m3, a gain of 6.99 million m3, a net change of -1.37 million m3, and a gross change of 15.35 million m3. The upstream Chipola juncture has a loss of -6.41 million m3, a gain of 1.89 million m3, a net change of -4.52 million m3, and a gross change of 8.30. On the other hand, the downstream Chipola juncture has a loss of -1.95 million m3, a gain of 5.1 million m3, a net change of 3.14 million m3, and a gross change of 7.04 million m3. Comprehending the floodplain inundation of the Lower Apalachicola River can provide crucial knowledge for river ecosystem management. The second research topic of this dissertation used hydrodynamic modeling and remote sensing to improve the accuracy of inundation mapping. The LiDAR and multibeam sonar were fused to generate the DEM of the riverbed that was used to create inundation depth maps for low-flow (175 m3/s) and high-flow (4361 m3/s) conditions during the super-flood period of 2015 and 2016, through a comprehensive analysis of 2D hydraulic model HEC-RAS and Relative Elevation Model (REM). Satellite images were used for geospatial analysis to estimate the inundated areas and results were compared with the outcomes from the HEC-RAS and REM models. This research also identified and analyzed the natural and human factors responsible for the riverbed deformation using the DoD DEM and machine learning algorithms, including the random forest model and the XGBoost model. The research examined the factors that influenced riverbed chnages in the study region between 1960 and 2010, when the USACE conducted the Navigational Project. Using the comparative analysis of two machine learning regression models to determine the long-term riverbed change, we employed the random forest regression model and the Extreme Gradient Boosting regression model (XGBoost). The models were conducted with 10 factors for the given period, including neutral factors such as floodplain width, bank vegetation density, river curvature, Stream Power Index, Junctures, and Tidal and human factors such as Cutoffs, Dikes, Dredging and disposal, from 1960 to 2010. The study identified potential drivers of riverbed changes using machine learning algorithms. The random forest model outperformed the XGBoost model with the former having R-square values of 0.95 and 0.93 for the validation and testing sets, respectively, indicating high predictive accuracy. The XGBoost model was slightly less accurate, with R-square values of 0.75 and 0.74 for the validation and testing sets, respectively. In the random forest model, variables Floodplain Width, Dikes, and Junctures were the most influential factors on the riverbed, whereas Dredging was the most influential factor in the XGBoost model. The research provides decision-makers and local communities with the knowledge to prepare for the future of the river in the face of both natural and anthropogenic changes, mitigate potential dangers, and effectively manage land recovery.