Lake Level Estimation using Machine Learning and Physically-based Approaches in Lake Tana, Ethiopia

Abstract

This study aims to estimate the monthly water levels of Lake Tana, Ethiopia, which is the source of the Blue Nile and as such is of great importance not only for the local communities but for all the countries depending on its waters. Two different approaches are used for this purpose: a physically-based model and a data-driven algorithm which uses support vector regression. A comparative analysis of their errors, applicability, ease of use and computational speed is performed. Although the physically-based model outperformed the data-driven model in all but the bias metric, the latter has multiple competitive advantages such as reduced computational effort, shorter training/calibration time, and the fact that it requires the selection of fewer model parameters.