Data-driven framework for forecasting sedimentation at culverts

<p>The increasing intensity and frequency of precipitation in recent decades, combined with the human interventions in watersheds, has drastically altered the natural regimes of water and sediment transport in watersheds over the whole contiguous United States. Sediment-transport related concerns include the sustainability of aquatic biology, the stability of the river morphology, and the security and vulnerability of various riverine structures. For the present context, the concerns are related to the acceleration of upland erosion (sediment production) and in-stream sediment-transport processes that eventually lead to sediment accumulation at culverts (structures that pass streams under roadways). This nuisance has become widespread in many transportation agencies in the United States, as it has a direct bearing on maintaining normal culvert operations during extreme flows when these waterway crossings are essential for the communities they serve. Despite the prevalence of culvert sedimentation, current specifications for culvert design do not typically consider aspects of sediment transport and deposition.</p> <p>The overall study objective is to systematically identify the likelihood of culvert sedimentation as a function of stream and culvert geometry, along with landscape characteristics (process drivers of culvert sedimentation) in the culvert drainage area. The ideal approach for predicting sedimentation is to track sediment sources dislocated from the watershed, their overland movement, and their delivery into the streams using physical-based modeling. However, there are considerable knowledge gaps in addressing the sedimentation at culverts as an end-to-end process, especially in connecting the upland with in-stream processes and simulating the sediment deposition at culverts in non-uniform, unsteady flows, while also taking into account the vegetation growth in culverts’ vicinity. It is, therefore, no surprise that existing research, textbooks, and guidelines do not typically provide adequate information on sediment control at culverts.</p> <p>This dissertation presents a generalizable data-driven framework that integrates various machine-learning and visual analytics techniques with GIS in a web-based geospatial platform to explore the complex environmental processes of culvert sedimentation. The framework offers systematic procedures for (1) classifying the culvert sedimentation degree using a time-series of aerial images; (2) identifying key process-drivers from a variety of environmental and culvert structural characteristics through feature selections and interactive visual interfaces; (3) supporting human interactions to perceive empirical relationships between drivers and the culvert sedimentation degree through multivariate Geovisualization and Self-Organizing Map (SOM); and (4) forecasting culvert sedimentation potential across Iowa using machine learning algorithms. Developed using modular design and atop national datasets, the framework is generalizable and extendable, and therefore can be applied to address similar river management issues, such as habitat deterioration and water pollution, at the Contiguous US scale.</p> <p>The platform developed through this Ph.D. study offers a web-based problem-solving environment for a) managing inventory and retrieving culvert structural information; b) integrating diverse culvert-related datasets (e.g., culvert inventory, hydrological and land use data, and observations on the degree of sedimentation in the vicinity of culverts) in a digital repository; c) supporting culvert field inspections and real-time data collection through mobile devices; and d) hosting the data-driven framework for exploring culvert sedimentation drivers and forecasting culvert sedimentation potential across Iowa. Insights provided through the data-driven framework can be applied to support decisions for culvert management and sedimentation mitigation, as well as to provide suggestions on parameter selections for the design of these structures.</p>