Author(s): Hoa Xuan Nguyen; Ana Margarida Bento; Tiago Ferradosa; Biswajit Basu

Linked Author(s):

Keywords: Data imputation; Water discharge flow; Missing data; Machine learning; Coefficient of proportionality

Abstract: In recent years, the data-driven approach has emerged as an increasingly popular paradigm for ensuring the stability of hydraulic infrastructures in the face of high uncertainty. However, data-driven problems often encounter the issue of missing data, which can be caused by a variety of factors, including a lack of available information, such as instantaneous streamflow data from a river dam. The presence of missing data can significantly affect the performance of predictive algorithms, necessitating the use of efficient data imputation methods that can be promptly applied to address this issue. In this paper, we present an analysis of several state-of-the-art methods for imputing missing values, evaluated in the context of a real case study in Portugal. Specifically, we apply four algorithms, namely Mean Imputation, k-NN Imputation, missForest, and Generative Adversarial Imputation Nets, to extrapolate missing data for the instantaneous discharge of two Portuguese dams, namely Carrapatelo and Torrao. We collected streamflow data from gauging stations near the new Hintze Ribeiro bridge over the Douro River, obtained from the “Water Resources Information System” (SNIRH, Portugal) platform. Comparing the performance of the algorithms with the coefficient of proportionality results, we analyzed the collected datasets. The results of our research indicate that the missing value imputation algorithms are effective in dealing with the same dataset and exhibit an attractive degree of computational efficiency.

DOI: https://doi.org/10.3850/978-90-833476-1-5_iahr40wc-p1746-cd

Year: 2023