This is a collaborative project with University of Iowa (Co-PI) and U.S. Army Corps of Engineers (sponsor).

Background: Reservoirs are a vital component of our nation’s water-resources infrastructure, yet many reservoirs across the nation are slowly filling with sediment, reducing their effectiveness and increasing maintenance costs. Water-resources managers must develop sustainable sediment management plans for reservoirs to ensure the continuation of reservoir functions, which require reservoir capacity surveys to assess lost storage capacity and sedimentation rates.

The U.S. Army Corps of Engineers (USACE) has developed the Enhancing Reservoir Sedimentation Information for Climate Preparedness and Resilience (RSI) system to help evaluate aggradation trends, life expectancy, and reservoir vulnerabilities to climate change. Survey data collected entail multiple methods, instruments, and measurement protocols, which can lead to considerable differences in data that result in anomalies that require detection and correction before being permanently stored for further usage. Due to the large number of reservoirs in the RSI system and the numerous parameters that influence sedimentation, manual detection of data anomalies is a challenging, tedious, and costly task.

Objective: The primary objective of the research project was to develop methods to identify anomalous data (likely erroneous) within the RSI system using machine learning algorithms. A secondary goal of the study was to use the RSI data along with supplementary data sources in conjunction with machine-learning to estimate sedimentation rates.

Approach: Data from the RSI system were analyzed to quantify capacity loss between consecutive reservoir surveys. A filtering process was developed to identify potentially erroneous survey data. Then, a survey of available supplemental data was conducted, and a composite dataset was assembled. To further identify anomalous data, two machine learning methods were used: the automated anomaly detection (AAD) method and the Kolmogorov-Smirnov and Efron (KSE) method. Finally, the composite dataset was used with regression, machine learning, and artificial intelligence methods to develop models for predicting capacity loss as a function of several reservoir and drainage basin properties.