Deep neural network modeling of river discharge in a tropical humid watershed

Abstract

Precise forecast of river discharge is crucial for a variety of sectors, from human activities to the control of environmental hazards, considering growing need for water resources and the effects of climate change. Despite the development of various discharge forecasting models, real-time projections are still difficult. This has necessitated the application of Artificial Intelligence to predict river discharge using satellite data since there is paucity of gauged records in most developing countries. In this research, a 38-year data, obtained from the National Aeronautics and Space Administration (NASA)/Goddard Space Flight Center using the Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2), was used to model the discharge of five selected rivers from South Eastern Nigeria watershed. Deep Neural Networks (DNN) modeling technique was engaged. Back propagation learning algorithms of various network topologies were developed for predicting the river’s discharge with respect to other hydrological properties. The developed model was trained and validated with the raw dataset. Results indicated that relative humidity, atmospheric pressure, wind speed, rainfall intensity, radiation, air temperature, and soil temperature govern the discharge of river. The DNN model accurately predicted the river discharge with the 7–25-25–25-1 network structure, as evidenced by 99.91, 99.62, and 99.01% R for the training, validation, and test. The results of this analysis showed that DNN approach is effective at forecasting river discharge with respect to the hydrological characteristics. Decision-makers in the water and environmental sectors can utilize this knowledge in making an informed sustainable development plan.